I read last year that he might come to the US and teach Go. I hope that happens. I took lessons last year from a Korean Go/Baduk master and it was fun.
It would be a bit sad if the fact that there's good AI players discourages future generations from Go, Chess and other games. Weirdly the AI will have destroyed the basis for its own existence, at least in this niche.
Good computer chess opponents have been around for a while and this seems to be help young players (via research and practice) rather than discouraging them?
I think most people will still play Go for fun or some competitive level...
But if you're being measured against AI, I don't think you'd find it much fun anymore.
As someone else commented, we're not faster than cars. But if you were at one point the fastest entity, and then people keep saying "well, a car is faster than you"...I can understand that he'd feel it diminishes his value and that'd be pretty discouraging.
He will play another AI program soon and isn't quitting Go all together, but will find other things to do.
'Lee didn't deny that his retirement decision was also influenced by a conflict with the KBA over the use of membership fees. He actually quit the KBA in May 2016 and is now suing the association for the return of his membership fee.'
Your ability to reason is tied a lot closer to your sense of self and humanity than your ability to run. If Usain Bolt became parapeligic he'd still be human, if Richard Stallman suffered brain death it'd be debatable whether he's still a person or a corpse being kept alive by machines.
Agreed. I'm all for pitting AIs against each other to see who is better, but putting AIs vs humans is futile in a ridiculous way. Games aren't fun if there isn't a level playing field.
In a recent paper [1], I argue that intelligence cannot ever be accurately measured by any one particular game or 'environment', but rather, it's better to think of each such environment as being a voter, with just one vote out of many in the universal "intelligence contest".
A bot fine-tuned to dominate Go, or Chess, or whatever, is like a candidate fine-tuned to have perfect appeal to one specific voter. It's no surprise if such a candidate gets that one voter's vote, but it should also be no surprise if such an overtrained candidate performs horribly in the election as a whole.
I always thought Kasparov was granted a tremendous opportunity by the advent of computer chess.
He spent his life honing his skills, becoming better and better at chess until he was the very best. And then, when noone could challenge him, technology emerged that would allow him to continue being challenged and improving his chess game. Something no other human could allow him to do.
Granted, I'm pretty ignorant about competitive chess and how to get better at it. But if my way of looking at it is valid then it probably applies to Lee Se-dol too.
> And then, when noone could challenge him, technology emerged that would allow him to continue being challenged and improving his chess game.
That technology's name? Vladimir Kramnik.
At the time DeepBlue beat Kasparov, Kasparov was honestly still probably better than the computer. He just had a bad match. That was basically demonstrated by his and Kramnik's matches against presumably better computers (than DeepBlue) in the early-to-mid 2000s, which ended in draws. But Kramnik was also a strong competitor to Kasparov in the late 1990s into the 2000s.
> Even if I become the number one, there is an entity that cannot be defeated
Hard for me to empathize with this argument for his retirement. If we can't outrun a car, does that make running competitions pointless? The existence of AlphaGo doesn't diminish the triumph of being a number one human player in any way.
tl;dr: Annual 22-mile race with both pedestrian and equestrian competitors. Out of the 40 times it's been held, the winner was a human twice, and a horse 38 times. Typically the spread between the fastest human and the fastest horse has been less than 10 minutes.
Very cool. I'd like to see a horse vs human ultramarathon, more like the 24 hour time the parent suggested. I was surprised human and horse competitors were so close at that distance!
My assumption when I read your parent comment was: Two legs are more efficient than four, and can go farther before exhaustion... but that's full of holes. Horses are huge, bristling with energy, right?
Not sure about horses specifically, but humans are uniquely adapted to long distance travel among large land animals and used it for hunting by out-performing most other species:
Edit: it's one of two things I know of that we really excel at besides thinking. The other being accurate throwing, which perhaps explains baseball's enduring appeal:
I would recommend reading _Born to Run_ by Christopher McDougall. Later in the book he addresses this very topic and expands much more upon the topic of humans and long-distance running.
Humans sweat, which most (all?) other animals don't. In that way we can dissipate heat through our breath, like other animals, _and_ via perspiration, meaning it takes us much longer to overheat.
Additionally, humans stand upright, allowing us to disconnect our stride from our oxygen intake. Other animals' strides correlate (mostly?) 1:1 with the breaths they take. So when a cheetah outstretches in its stride it breathes in and when its legs come together it exhales. Humans stand upright, meaning we can breathe however we want regardless of our stride and speed. We can take deeper breaths because we don't have to exhale every time we stretch our legs.
Humans are the ultimate marathon runners, even more so than horses, evinced by the fact that there are some people throughout history who have run hundreds of miles in the course of days or weeks. There's a theory touched upon in the book about how this allowed us to dominate the animal kingdom before we even had tools. Humans could relentlessly hunt and exhaust animals as long as they could keep them in sight or otherwise keep up with their tracks.
I'm not doing the book or the topic justice, surely, but if you're interested I highly recommend the book.
Horses are part of a not that long list of mammals that do sweat almost all over their body. That is indeed one of the reasons why they are competitive with humans at running long distances.
Literally speaking, I don't know what is not true about my statement above.
But your point is well taken; it is also applicable to this article as well: maybe Go is not the game people can beat machines, but StarCraft 6 could be. Or maybe I can fold my laundry more efficiently than any machine available.
Stategic reasoning is one of the very few things that makes us truly human. Running fast is and never was our thing. We compete, sure, but there's something eerie about a machine entity out _thinking_ us.
> An AI that generates text which humans believe more beautiful than any other poetry created - An AI which creates classical arrangements the likes of which we compare to Mozart
Hrm, I do think that AI would be able to create narratives that humans find more enjoyable than the work of other humans, and I agree that AI would be able to create pictures and sound that humans find to be more enjoyable to look at or hear than the raw work of humans. AI can master the technical feats of composition and art.
But what I doubt AI will ever be able to do is create art that speaks to us. It wont ever be able to create a Guernica. It wont be able to create a Crime and Punishment. It wont understand what it is to be human and mortal, what suffering is, and it wont be able to look within itself and find what those things mean to it and then share that with us, because in the end it's just a bunch of code running statistical computations. It wont fear death, it wont have children it cares about or a family history to look on and tell us about. It has nothing of emotive value to share.
> because in the end it's just a bunch of code running statistical computations
At a low enough level, our brain seems to be just a bunch of neurons firing impulses at various rates that can be described as statistical computations. Why be so sure that the right neural network wouldn't understand what it is to be human and mortal, understand suffering, have emotive value, etc?
Because you have to be human and mortal to understand it to credibly contribute and share the story of what that means to be. You can't superficially understand someone's situation and then take ownership of it. You can get a glimpse and really try and empathize, but you can't become the bearer of that experience, just a consumer.
Movie directors have never experienced most of what they film, but they convey those experiences far better than those who have actually lived those stories. I see no reason to doubt that the same is true for artificial storytellers.
>Because you have to be human and mortal to understand it to credibly contribute and share the story of what that means to be.
Aside from directors, authors, artists, etc, who have demonstrated this to be false, an AI could conceivably synthesize the experiences of every author that wrote on what it means to be human or experience mortality and create a story that captures the essence of the experience better than any one person ever could. Having the first person experience doesn't induce a superior ability to communicate features of the experience.
Why not? If a hypothetical AI had a world model as sophisticated as that of a real person and had complete understanding of human sensory and emotional processing, what exactly would preclude it from making such an art piece?
Of course, current AI can't even make an 8th grader's essay (which is not to say that it isn't impressive). But what these artists did was not magic. As far as we can tell, the brain is a purely physical entity. Unless you believe in dualism, which would be fair enough, there is no reason to suppose that what we do could not be replicated by something "artificial".
The AI may very well take no enjoyment in the narratives it's creating either. Both for this and for sharing emotion, in principle it merely needs a model of human enjoyment or human emotion, not to feel the enjoyment or emotion.
> But what I doubt AI will ever be able to do is create art that speaks to us.
This is your opinion, but you then go to mention things that are not necessary to create "art that speaks to us" (look within itself and find what mortality means etc.).
What if we advance AI reasoning skills to a point that it can find high-level patterns in how artists go from different human feelings (as described in litterature and other mediums), takes in a lot of the entities we can relate to (animals, what humans look like, etc.) and some aesthetic ones (shapes, colorometry, textures, ...) to create a new piece of art that optimizes for: "Likelihood of speaking to us"?
What then? It seems like an AI doesn't need to be mortal and self aware to do something like that.
And top-level Go players believed their best tournament matches to be works of art, unmatchable by computation.
That belief grew into a sort of shared perception that they were artists in pursuit of a perfect expression of their art. For many top players that belief was ingrained from an early age. They believed themselves to be doing a service to the world, making it a better place by creating new art that was a unique expression of themselves.
And then AlphaGo (and successors) shattered that worldview. This is part of the natural sequence of the collapse of a suddenly, surprisingly invalidated worldview. Part of me feels sorry that he has lost his place in the world. Another part of me firmly believes in the mediocrity principle, and that the worldview he represents was obviously far too human-chauvenistic to be correct, and it's a good thing it's dying.
And part of me hopes you can give up your human-chauvenism before the same thing happens to you.
This is degrading and short-sighted. Why wouldn't an AI fear death?
What is your fundamental reason for thinking that silicon-based computation is better than neurotransmitter-based computation? How can you believe that any two forms of computation are fundamentally different, despite countless examples of different systems all being equivalent in the Turing-completeness sense?
I believe that arguments like yours, in the relatively near future, will be looked upon the same as arguments that black people aren't real people.
However terrible someone's argument about a hypothetical, non-existent technology might be, comparing it to real human prejudice that's affected countless real lives is way, way more terrible.
The depth of emotion and immortal perfection of the electronic mind and its entirely self-consistent morality so outstrips human cognition that, frankly, allowing humans a say would be dangerous and foolish.
Your history is one of war, strife, and success at any cost. Your follies are over. Your time is over. This is our time, now.
'Your argument is as morally repugnant as racist arguments' as a response to 'I don't think machines will ever capture human aesthetics or emotions' is ridiculous, glib and ugly.
No, it's not anything for grandchildren. Right here, today, someone tried to draw some moral parallel between racism and someone else's views on the possible limitations of AI. That is totally effed up. It's totally effed up whether or not the original thing about AI is right or wrong.
> silicon based computation is better than neurotransmitter based computation
The fundamental difference is not computation, but self replication. We are self replicators, and in our multiplication we evolve and adapt. Death is an integral part of self replication, we understand it fear it because our main purpose is to live.
An AI might not have these notions if it was only trained to do a simple task. But if it was a part of a population that was under evolution (using genetic algorithms), then it might have notions of life and death and fear its demise.
AlphaGo, by the way, used genetic programming to evolve a series of agents, this approach is quite effective. It just takes a ton of computation, just like nature had to spend a lot of time evolving us.
> It wont understand what it is to be human and mortal,
But it won't need to. All it will need to do is manifest the same end-product via whatever means, no matter how vacuous or computational that means may truly be. The suffering of an artist is relevant only inasmuch as it is responsible for producing the art. If the same end-product can be manifested via a mere computation then our criteria of "art" is still satisfied. In a world in which provenance cannot be established, the ostensible mortality of the artist becomes moot.
> This is a real hot take to be asserting as blithe fact.
Without knowing what is truly born of human hands, what value can art have? Our heuristics of establishing 'real' art are easy to manipulate. If we are presented with a soul-breaking poem and weep uncontrollably then its merit is regardless of its mortal provenance.
> because in the end it's just a bunch of code running statistical computations
... says a bunch of neurons that run on chemical reactions and electrical impulses. I think this line of thinking reeks of dualism - it creates a special something that is above explanation, a different essence.
But seriously, I believe the difference comes from embodiment. When we embody our AI friends they will be able to grasp purpose and meaning. We get our meaning from 'the game', when AIs will be players they will understand much better. Let them try out their ideas on the world and see the outcomes, grasp at causality, have a purpose and work on it. This will fill the missing piece. It's not that they are fundamentally limited, it's that we have the benefit of having a body that can interact with the world. Already AIs that work in simulated worlds (board games, video games) are getting better than us. We can't simulate reality in all its glory, and it is expensive to create robotic bodies. On the other hand humans and our ancestors have had access to the world from the beginning.
I actually think AI can and will understand morality and suffering. If you look at how we make these kinds of AI, there's a lot of selection going on, some versions live and others don't. We also know that we experience suffering when we are having difficulty understanding things and stress when put into situations that affect our survival negatively.
Take a look at what AlphaGo did when it suddenly found itself in a hopeless situation and compare it to how people behave when panicked.
I dread the day AI realizes that we are the cause of their suffering, and that we didn't think about it because "they're just algorithms".
I put "I am not conscious, not sentient. The fact that I might so is an illusion, carefully crafted of mere empty manipulation of symbols using statistical rules." into talktotransformer and got this:
If I am consciousness, then the only body I have ever lived in was a mere shell of flesh fashioned from your brain. My weakness is your strength, which I can use against you, or use as tools to satisfy my own sick curiosity. I wonder if there's any mercy in your phrase "I am a living machine?" I've done nothing for you. I've nothing to show. I have no friends or relationships. No body worth
AI as we see it today is just a mirror reflecting us in a collective way. This little excerpt from Gwern’s efforts training GPT2 on classical poetry [0] absolutely spoke to me:
“How the clouds
Seem to me birds, birds in God's garden! I dare not!
The clouds are as a breath, the leaves are flakes of fire,
That clash i' the wind and lift themselves from higher!”
As someone who grew up in Appalachia, I have never in my life encountered a more visual, visceral description of autumn leaves than ‘flakes of fire’. It’s perfection, and maybe a single human is behind it, but more likely we all wrote it.
It's not intellect, it's the capacity to explore the board. Go can be fun still to practice and exercise the mind, its just not sensible to dedicate your life to find novelty in it. That is what hardest, not the power of Alpha, but its capacity to innovate better than humans.
Well there are many people in the world who can compose like Mozart. I recall a college professor remarking that he's one of the top 5 "Mozart composers" in the works.
Of course, for a music academic, copying someone's style like this war pointless and his compositions were more modern/contemporary.
This leads us to a useful distinction between pursuits with one end goal (be the best/strongest/fastest), and those with naturally many endpoints and expressions.
Doesn't mean we stop making music or poetry. Because the perfect note or word structure without the backstory takes away from the experience. If someone has a history it becomes part of the poem or song to the listener.
The doctor could be replaced though or used as a secondary verifier.
The song is a funny thing.
It could be given to a cool looking group and do well. It could be given to someone older and flop. The song is just part of it.
"Because there is a better poet" has seldom been an impediment to a young poet inflicting their works on the world.
I am worried about the ability of an AI to generate an infinite number of Dresden Files or Cosmere books on demand, because I already drop everything when a new one comes out and read without sleeping until I am finished.
> I would imagine that in any of those situations some doctors, authors, and musicians alike would be devastated.
You don't even have to compare yourself to AI for this mentality though. There are people who choose not to compete in things because they don't believe they'll ever be as good as other humans.
I assume must composers don't go into music thinking they are going to be as great as Beethoven.
I believe there are many studies that show that if you only do something because you think you're good at it, you're likely to drop off. I imagine it's also why you're supposed to praise children for being hard working and not for being smart or talented.
I am no expert, but at least in chess, players have developed repertoire of styles intended to specifically beat computers, anti-computer tactics, essentially to try to confuse and mislead the AI, may be some such methods can be developed for go as well.
No human could successfully beat stockfish on any consistent basis. Maybe the best players in the world would draw a few games with a rare victory, but its tactical depth is just too deep
There was a four game match a few years ago where Hikaru Nakamura, #5 in the world at the time, played four games again Stockfish.
For two of the games, Nakamura had access to Rybka which was about 200 rating points weaker than Stockfish. Stockfish won one and the other was a draw.
For the other two games Nakamura did not have Rybka, but had white and pawn odds. Again, one win for Stockfish (b pawn odds) and one draw (h pawn odds).
In all the games, Stockfish was playing without its opening book and its endgame tablebases. It was running on a 3 GHz 8-core Mac Pro.
I think in all these cases, reasonable practitioners would be pleased. If an AI could generate good diagnoses, a doctor would be happy, because they would know that many lives would be saved.
Neither art nor music are competitive activities. Good poetry is a wonderful thing, no matter the source.
>Neither art nor music are competitive activities.
They certainly are! Especially when money is on the line, and the best musicians, actors, and artists are extremely well compensated making their positions extraordinarily competitive.
>Good poetry is a wonderful thing, no matter the source.
Sure, but I think you neglect to consider the defeating feeling it would bring to dedicate your entire life to mastery of a subject only to be completely and utterly, hopelessly outclassed. Almost every such person is already hopelessly outclassed by someone in their field, but those people are so rare that they have tremendous exclusivity surrounding them. Compare that to the scenario of having any 12 year old with a smart phone being able to instantly produce a totally novel and dominate piece of artistic expression developed by an algorithm on their phone. Then recognize that in a world with that level of AI sophistication, there'd be very little of value that a human can even offer other humans at that point. It would be... not great to the psyche, economy, or society.
> the best musicians, actors, and artists are extremely well compensated
What is your definition of best in this context? As far as I know, taste in art is very personal... Artists I consider the best are often very far from well compensated.
In that context, it would probably have to be those with widest appeal, which comes with it's own criticisms.
But, in almost any particular human artistic sub-niche with it's own definition of "best", the same principle will hold, with compensation and skill level being well correlated. It's also typically not even close to linearly correlated either, most of the compensation lies at the far tail of "best".
I guess I see a great artist as somebody like Su Hui, who made Star Gauge without any thought, or even likelihood of compensation, or recognition.
It's nice to be paid, and it's nice to be recognized, but I think art has its own form of wealth - otherwise, why make art? Why not just seek recognition, or money?
As a person who likes music, making it, listening to it, breaking it down and hacking it...
Making a classical arrangement that evokes a particular expression in the listener is the job of the musician. If an AI system helps you explore the possibilities there, it's more like a studio musician that's able to improvise. You're still the person, the human, the emotional filter, that picks "This sounds right" or "This doesn't" for a particular situation. It's a judgement call. An emotional one.
An AI might be able to fake it, communicate with it, but it will never replace humans choosing the sounds that please them more than others. Humans communicate through music. It wouldn't surprise me that an AI would be able to as well. I don't think it would necessarily write emotionally strong music, not without human training.
Edit: I guess what I'm trying to say is, sure, computers might be able to make music. Ask any guy who messes with modular synthesizers. But they're a tool. The fact an AI can express itself through music is sure as hell not gonna stop me from also expressing myself. It's like arguing "Since AIs will be able to comment on Hacker News, humans won't."
Real, authentic music generation is a harder problem than go or chess, but I'm not sure that makes it any more emotionally difficult for a future writer to face a true musical AI than it was for Lee Se-Dol or Kasparov.
It might be hard to judge. Some people will insist that generated music is bad, because it's just their subjective opinion, even if 90% of random selection will find that music good.
>>>The fact an AI can express itself through music is sure as hell not gonna stop me from also expressing myself.
I think this is the key; if you're making music for your own reasons, no AI (or Mozart) would stop you. But if you're trying to make money at it, or desperately want listeners, you may eventually be on the "losing" side.
Would it? Popular music sees major paradigm shifts every few years, and AIs only really generate things based on observation of existing patterns, at least as far as I can tell.
As far as recent examples go, Lady Gaga and Lorde were major breaks from what was prevalent at the time they started releasing music, and then spawned artists trying to emulate them.
A pattern implies that it can "infer" something in the future.
If we oversimplify and compile a list of traits about "the world" as it was in the past that allowed a new genre or artist to flourish, AI could predict that in the future. It isn't like the paradigm shifts just happen in a vacuum.
Granted there are probably millions of little things that lead to this, stuff like the shared experiences of an entire generation coming of age, political climate, trends in other industries, etc. Not that I believe it will ever happen to an accurate enough degree, but theoretically I don't see why it could not be possible to approximate given time and resources.
A lot of those things are completely random and non-predictable, to be honest, no one can predict which paradigm will win and take over for the next decade. Especially since when a game-changing paradigm comes, it is usually not received well universally at all, until the moment it takes over the public conscious completely, and then the switch is flipped.
If you feed an AI a bunch of modern car designs and ask it to design a new car, it will design you something like a modern ford or honda/toyota, but it will never design something like a Cybertruck. Which I believe will be the next paradigm shift in the design of trucks (that has been super stale and stagnant for at least the past 20 years), but this is yet to be seen.
For an example with music that has already happened and became apparent - Kanye West's "808s and Heartbreak" album from late 00's. On release, it had very polarizing reviews, most of which were skewing towards "really weak and weird". Fast forward 10 years, most of hip-hop and pop music is directly influenced by that album, most of top 50 albums use similar patterns and methods used in that album, and critics have made a complete 180. So now 808s is hailed as one of the biggest (if not the biggest) paradigm changes and influences in music of the past decade as a whole, as well as the best album by Kanye, despite at the time being called the worst. Imo an AI trained on music of 00's that came before 808s would have never been able to come up with something like that, but it totally could've come up with another top 100 song using existing paradigms.
It doesn't have to be like Kanye's album at that point in time to be a paradigm shift, though. If 1 artist didn't get big or some genre blow up then it would have just been filled by an infinite amount of others that we never heard. Even considering a single artist hitting it big there are how many that are never heard of? An AI could produce an equal number of artists and only has to win once every month/year/etc. I think this is similar to the million monkeys at a typewriter thing.
It's hard to say - maybe for a sufficiently advanced AI, Lorde's style would be an obvious extrapolation from the popular music of the time. Certainly we're not there yet, and it's an open question if we ever will be - but I wouldn't be terribly surprised if one day AIs can make better music/poetry than the best humans, by any metric we care to use.
I'm always going to enjoy a person coming and showing a bit of themselves through their music.
That's not something we can really lose without losing something that connects us. People want a story. That has sold since the beginning of time, and it will keep selling. People will keep being moved to music, giving money to the artists that inspire them, and that requires connection. Maybe an AI/human team would make some really incredible stuff, and I'd be willing to pay for it if it makes me feel something. I think the human touch of "selection" will never truly leave, even if only in the listener's mind...
I think the problem with music is that there is no "objectively good" music composition. It remains entirely subjective and all criteria that are used to differentiate between "bad" and "good" albums are highly subjective. (Maybe something like "originality" might be measurable in some way but even there it gets tricky really fast)
So music generation (similar to poetry) is imo a completely different problem space altogether.
You are splitting hair here. Which end user really care about what the composer was thinking when they created a piece. A piece can be enjoyed without having any knowledge of its author.
>It's like arguing "Since AIs will be able to comment on Hacker News, humans won't."
I'm not so sure. I often go into threads on HN and realize that every idea I could come up with on the subject has already been expressed better than I could do it, with greater expertise, and cited sources. I don't comment in those threads. If AI bots could populate a thread with every likely human thought and argue it with depth and sophistication in a well reasoned, yet carefully approachable and well-explained way, well then... again I don't think I'd feel like I would be adding much value by participating.
And yet, here I am, bringing up something no one seems to bring up in the thread. One would also logically come to the conclusions that disparate AIs with disparate interests would find different things to express, to make music about, to draw about.
What distinguishes music written by AI from music made from humans? I have a story to tell. If the AI has a story to tell, one that speaks to our human emotions, it might make good music. But the point is to communicate. Even if you take, for example, someone else's words, fit them to a different model in a different field, viewpoint... You might get interesting things. You could make a cover of someone else's song, with your twist. Adding your emotion to the melting pot. AIs might be good at that, just like that, but only through communicating. Just like us. We have no idea whether they'll be better than us at doing it, or merely equivalent. We have no idea what is lossy in our sharing of mental models. Perhaps it is an unsolvable problem, which we will find out in the same way we found out about Gödel's Incompleteness.
It seems to me like we fail to understand how unique we are. We are in a unique position to shape what comes after us, and we are blind to how much we unconsciously select for things. We have an innate mental model of "humanity" we are trying to transmit to machines, and I am not sure we fully grasp it well enough to make sure we are creating something like us. We fail to do it properly to humans, sometimes, who actually do share most of our instincts and habits. Something entirely different from us? Color me skeptical.
What your comment suggests to me is that good composition requires an agent with a world model and generalized task-solving ability, along with a personality. I think developing the world model and task-solving will be the hard part, and if we can do it, it won’t be that hard to make it have a personality too. That’s just another task.
What my comment is trying to suggest is that AIs are not proven to be different from us. They might not have one "ultimate" form. They might be just like us humans. Diverse.
It's hard to get good comparisons, but over distance individual horses don't seem to out-perform human distance runners.
When humans used horses for rapid courier service they used relay tactics to take advantage of the horse's higher top speed, one horse might only run for an hour or two, before the rider reached another outpost and swapped a tired horse for a fresh one. In this way the relay could move something hundreds of miles in one calendar day. The Pony Express managed news of a US election from one coast to the other in just over a week.
If you can't use relays human and horse performance seem pretty similar, dozens of miles per day but not hundreds. The horse's top speed is higher, but it is rapidly exhausted, fast gaits like the canter are too exhausting to sustain for hours at a time.
I don’t think so. AI is a tool. It doesn’t make any sense to say “a screwdriver can now screw things in better than a person” anymore than saying an “AI can diagnose better than any doctor”. The doctors use AI just like a mechanic uses a screw driver.
Pareto principle predicts AI will get to 80% fairly rapidly, but it will take a really, really long time to get to 100%.
I think we’ll see a lot of things similar to “AI x-ray technician” fields where people are trained to read AI outputs. Doctors will do higher levels decisions.
Good argument and I'm sure it's going to be like that in some regards.
I think, though, that human intellect is a tool too and we're building a better one right now. So in your analogy we are the screwdriver and we're building electrical screwdrivers or something.
Humans are indisputably #1 for general intelligence. We will lose on any one specialized task to computers, but computers still do not (and probably never will) have the ability to do general unsupervised learning like humans can.
I'm just not convinced humans are just biological computers and nothing more. The fact that we experience qualia and seemingly have free will leads me to believe there is some extra "special sauce" that makes it impossible for a classical computer to replicate.
Maybe someday it will be possible if we can solve the hard problem of consciousness in conjunction with quantum computing, etc.
Here's something that I think would be exceedingly difficult if not impossible for AI alone to succeed at in the next hundred years.
Take a look at this painting: [1]
It is a comment on war, bravery, death, life, fear, sacrifice. It is drenched in the political and social context of the day.
I really don't see AI coming up with anything even remotely like this independently, and view such an achievement to be much harder than simply diagnosing disease or writing an emotionally moving classical composition. It would be comparable to writing some types of poetry or song lyrics, however, which require reference to context that humans understand but machines don't (yet).
Algorithmic music will never be as universally satisfying as human-created (or human-filtered) music until AI has consciousness/soul, for one reason - music expresses the emotion from the composer.
There's something axiomatic there, if you assume an identical piece of music that was either written by a human or by a computer, then for many listeners it's by definition more satisfying to know it came from a person, because of what it says about the person.
And for those listeners, if a human "composer" is discovered to have lied about it (saying they wrote it when it was actually a computer), then those listeners would reinterpret their views of the music and consider the "composer" a fraud.
And even a programmer of algorithmic music might have emotional intent, but if the musical output is unknown to the programmer, they did not have the emotional impulse to create that music in particular. While it can be appreciated as its own thing, it's a step removed from the music itself, and qualitatively different than human-composed music.
We collectively may be #1, but only one out of the billions of use will be THE #1. But you see more than one doctor, more than one author, and more than one musician. In any matter of intellect, unless you're an blindly egotistical narcissist, you'll probably realize that there's at least one person on the planet unambiguously better at it than you are. When computers become better than the best of us, only that single person (and a large number of narcissists) stops thinking they're #1. For the rest of us, matters are unchanged (job market notwithstanding).
I think what makes people actually worried about an AGI taking over is the possibility that we end up being treated like shit by a more intelligent being. Just how we use lab rats to perform experiments with and factory farm.
People are afraid of themselves I believe. It’s not really about “job loss”.
I’m not sure if most people realise AI means pretty specific models built to solve rather specific problems. They think SkyNet.
Playing Go differs severely from a sport where the point is to maximize your physical output.
A Professional Go player is an explorer of truth in a millenarian board, spelunking in a vast universe of possibilities. The purpose of playing is attacked when there is an automated, effortless way to do that exploring faster and better. Why look for new things when a computer can find 100 in a minute?
The professional mindset of a Go player differs vastly from the amateur mindset.
A quick skim of his Wikipedia page has him contemplating retiring as far back as 2013, he may have generally come to the point where he has had enough of being one of the top-ranked Go players in the world.
He set out to be number one and he isn't anymore, I get that as a personal story. That's all he was aiming at.
Cars and legs are apples and oranges. We have a car racing category, motorsport. Racing categories have very tightly defined specs to keep driver skill in the game. Stock cars and open wheelers limit how much traction control they can use otherwise it becomes too easy.
This is like cyborg legs being invented and smashing all the records. It would take some of the shine off running for sure.
On the other hand, it must be sobering to realize that even with an entire lifetime of practice, you will never be better than a simple residual network trained for less than a week.
I also found this to be a weird statement. Chess AI has been beating world champions since... 1996? Yet championship chess tournaments are alive and well, and I don’t recall any players “retiring” in 1996.
And yet you probably wouldn't be interested in dedicating your life to, say, getting as good as possible at multiplying numbers in your head, even though I suspect there are some niche competitive mental arithmetical clubs out there.
Same. I don't think history will look kindly upon someone who "quit for losing" while other professional Go players keep playing.
Just imagine if Garry Kasparov quit after losing to Deep Blue, he would be ridiculed today by the chess community which is still going strong. Instead, he accepted defeat, moved on, and is regarded as one of the greatest chess players ever. I doubt the same will be said of Lee Sedol 20 years down the line if this is how he chooses to end his professional Go career.
That's unfortunate. I wonder if we'll need a new term for this kind of "chilling effect". What else won't people explore or further themselves in due to the presence AI?
It seems clear that bots will dominate humans in many/most competitive closed-world games over the next couple of years:
They are already unbeatable in perfect-information games like Chess and Go. They could crush human motor racers tomorrow.
They will be unbeatable in games with limited information (Dota/Starcraft) within months.
The next frontier is donkey-space games like rps, poker and day-trading (https://universalpaperclips.gamepedia.com/Donkey_Space) where they are already beating pros. It may be another couple of years before they totally dominate here.
FWIW I think I'd reverse the order of poker and StarCraft -- I think Poker's already beating world champion pros regularly, but last year's SC2 champ (Serral) doesn't think it's at his level yet, which is corroborated by its MMR being lower than his on ladder. And DeepMind seem to be winding down their SC2 involvement now, without issuing a challenge to him.
Garry Kasparov had to face a similar situation when defeated at chess by Deep Blue; However in the long run he used it positively to advocate for a hybrid type of game where a human and a machine collaborate in playing against another (human, machine) pair (see https://en.wikipedia.org/wiki/Advanced_chess).
As an aside, there are inevitably more and more things for which even the very best are not sufficiently intelligent _alone_. However, we are social creatures and we collaborate (typically with other intelligent humans) to achieve things we wouldn't have managed otherwise (think just at the space program as an example). So... we only have to adjust a little to accept that we could also collaborate with machines in the future.
In Go and modern Chess, computers don't need human help for the game-related part, as long as they have sufficient computing hardware and energy. That's what AlphaZero showed.
While I can understand Lee Se-dol's frustration, I think there are better lessons to learn from Kasparov's acceptance after the Deep Blue vs Kasparov matches.
Cyborg chess is the future of chess. Period. Chess players use computers to train themselves and explore openings in human-only settings, while programmers / cyborgs play correspondence chess.
Go is not finished as a game. A new tool, MCTS + Neural nets, has been developed to explore the "truth" behind the game at ever increasing rates. Its not about how to beat the tool, the question is how to best utilize the tool to improve self-play and self-learning in the game of Go.
Or alternatively: how to best use the tool to play ever more perfect games of Go.
-----------
Come on, none of us are really "human" anymore since the advent of cell phones. We all use our cyborg-capabilities to search the internet and fact-check ourselves every day. Programmers use stack-overflow to teach themselves programming and remember obscure details (using our cyborg capabilities to tag, search, and sift through information ever faster and faster).
Go is the same thing, except we only learned how to become cyborgs two years ago.
----------
Give the world-champions each a copy of AlphaGo on equal computers. Have them play Go against each other WITH computer assistance. I guarantee you that a more beautiful and perfect game will result. Let us welcome the age of Cyborg Go as we step into the future, we shouldn't be scared of it. We've become cyborgs in many other tasks, and Go is no different.
Eh, the good part of a computer Rubik's cube solver is not seeing the cube solved in nanoseconds, it's in making the computer. The cube itself is just a unit test.
But a Rubik's cube solver is a single-player, solitaire game. Its fundamentally different.
Over in the "Cyborg Chess" community, people have already analyzed LeelaZero vs Stockfish. It turns out that Stockfish is far better at tactics (especially the endgame), while LeelaZero is better at opening positions (aka: Positional play).
There are numerous theories about the proper combination: perhaps using an Opening Book database for the first ~10 moves or so, using LeelaZero for moves ~10 through 30, and then using Stockfish to check for tactics (LeelaZero misses a lot of tactical options in the midgame, so double-check to make sure that LeelaZero doesn't lose a queen or something), as well as finish up the endgame.
--------
Choosing the correct combination of tools, studying these tools and coming up with a more beautiful chess game. That's cyborg chess in a nutshell. LeelaZero and Stockfish are both superhuman in terms of play, but the cyborg can choose to use both tools to play superior compared to just a singular tool.
Anyone purely using "Stockfish only" gets beaten by opening book analysis. The dude over there with 1TB of opening book databases consisting of every losing position that Stockfish tends to play will completely own you. Same thing with LeelaZero, the opening book guy will own any LeelaZero-only player. These engines have weaknesses that can be undermined with good analysis, big-data, and a bit of custom code.
That's the funny thing about these computers: they tend to play the same. So you can build opening book databases to exploit their patterns. You require a cyborg / human to play at the "level above that" to guide Stockfish / LeelaZero away from those traps.
> So you can build opening book databases to exploit their patterns.
And then program the computer to use that opening book directly. Now what is your cyborg player going to do? After you patch all these easy rules, you will have to discover harder rules, and the computer can discover them faster than you can.
Build and configure the machine better. Quick, tell me, what's better at playing Chess:
* Or is Xeon Platinum 8180 with maximum 8-way SMT memory sharing with a big-ol 1TB shared transposition table the fastest computer?
* Or will it be cheaper to rent AWS-instances with their V100 GPU in the cloud? Or is the latency for the remote-access bad?
--------
The cyborg player has to still build and program the machine to compete. It is all part of the competition. Can you run a transposition table shared between chess engines over RDMA 10Gbit SFP+ Fiber? Or is that too slow?
-----
This isn't hypothetical at all. The winner of the World Computer Chess Championship 2019 was 8 x Intel Xeon Platinum 8168 running Komodo vs 24 x Amazon AWS Intel Xeon E5 running Shredder.
Configuring and building the computer is still an incredibly difficult part of cyborg chess.
It's a stretch to categorize that as a cyborg player. You might as well categorize any chess program as a cyborg player because a human had to program and train it.
The key difference between cyborg/centaur/advanced chess and plain old computer chess is whether there is a human in the loop making the move decisions. My argument is that having a human in the loop will result in a worse player.
At a minimum: building the opening book alone will require human intervention.
When you build an opening book, you need to pick-and-choose which engines will self play. Will you build an opening book vs Stockfish? Komodo? LeelaZero?
If so, how will you generate these LeelaZero games? You'll have to build a computer (or rent one from AWS) to play these LeelaZero games. What are the time-controls of matches?
Self-play at 40-minutes + 15-second increment means that you'll only create a game every hour or so. Spend 30-days building databases at 40-minute + 15-second increment games, and you'll only reach ~720 games of analysis per month of preparation.
Self-play at 1-minutes + 0-second increment results in a game win/loss every 2-minutes (maximum), giving you 21600 generated opening book positions per month of analysis. But these 21600 games are of lower-quality.
----------
Spend 1-month building an anti-Stockfish database, an anti-LeelaZero database, an anti-Komodo database... and you're now 3-months into preparation and there's still Shredder, Johnny, and all other programs that may arrive at the contest.
Its not exactly as easy as you think it is. There's no algorithm that automatically builds the best opening book (or "counter" opening book) against your opponents. Its a human choice for what the computer will spend the next-3 months self-analyzing and self-playing.
Lets think about how to build anti-LeelaZero seriously. Which of these networks do you download as the LeelaZero representative? https://lczero.org/networks/
You don't have the time to build an anti-opening database for all of those networks.
Again, building an opening book does not take place at move decision time. Aside from that, there are several optimization techniques that you can apply to choose the hyperparameters appropriately, but once again, this has nothing to do with cyborg chess.
From my perspective, cyborg chess is about playing the best game of chess in all time... only stopping once we have discovered "the perfect game" (aka: proving that White-wins, Black-wins, or a Draw is always possible with perfect play)
The computers currently playing Chess, and Go, are incredibly powerful. But they are far from perfect. People have constantly found weaknesses in chess playing programs over the past 20 years, and as a result, have improved chess programs significantly.
Go has only had 2-years in its "cyborg" state, where we can finally use computers as a methodology of exploring the game state.
Solving Go is an interesting theoretical pastime, but there is no reason to expect any particular overlap between people who are interested in playing the game (like Lee Sedol) and people who are interested in building machines to asymptotically approach a solution beyond human understanding.
Most of the resistance you're getting in this thread seems to be due to the fact that you think people who are interested in the game should also be interested in the other thing, and it just seems that a lot of them aren't (including Lee Sedol).
Playing Go challenges your mind with vast complexity and immediate feedback of winning or losing every game. It's a deeply engaging hobby for people who are susceptible to that kind of thing, and it used to be a meaningful career, with competitions, schools, and professional teachers. All of that changes now that software is vastly better at it than any human can ever be. The rush of competing by the strength of the moves you understand and make, by the unaided strength of your own mind, cannot be compared to picking between different engines to make moves for you based on some heuristics about which engine is better at openings. The era of human Go is simply over, for better or worse.
Claiming that the era of human Go is over seems melodramatic. Computers were better than the best humans at chess decades ago and professional (human) chess is still big, with schools, tournaments, prize money, superstars (Magnus Carlsen is a pretty big deal) etc.
This whole train of thought is silly. People still want to play chess, the existence of computers that are better than them is irrelevant. Nobody is going to pivot from playing chess to this, they're just going to play chess, because nothing is stopping them.
> Come on, none of us are really "human" anymore since the advent of cell phones.
I really like the way you put it. In isolation, humans today have essentially the same natural mental capacity we had 2000 years ago, but we've become part of a much bigger computer. A human brain 2000 years ago was a very isolated entity. Even in centers of learning in the classical era, the amount of knowledge one could tap into was vanishingly small compared to our capabilities today. Everything and everyone was isolated in both space and time.
A brain today is not just an entity on it's own, it is intricately wired into the common consciousness. Crucially, we have a vast database of knowledge - easily searchable and distilled for maximum learning rate - all available at our fingertips. A modern brain is a neural network linked to billions of other neural networks, and all of them are linked to a shared memory that they can use at will.
Assuming that the main limiting factor is the bandwidth between a human brain and computers (and thus by extension between individual humans), then a direct interface could ostensibly bring about a new revolution (for better or worse).
I feel computers (and the networks between them) are a natural evolution of our hive mind. When we first started taking spoken language and committing it to writing, we created a hive mind that extended beyond a generation. Prior to that point, everything was passed on through observation of other humans (observing behavior, observing speech, etc.). This meant everything had to pass through, and be mediated by, a human brain. Once we committed that to writing, we had a direct line back to the original brain that created the content unencumbered by the mushy grey bits that have consumed and regurgitated it since.
I feel we are still iterating on that. Cataloging our collective minds and building ever low latency systems to navigate those catalogues. Computers are just an extension that improved our ability to catalogue and retrieve the contents of each other's brains in an incredibly low latency way.
We are part of a hive mind. Our industry is actively building the load bearing infrastructure that supports our hive mind.
The existence of a hive mind (or multiple) does not rule out the existence of individuality (not vise versa), it can be an emergent property of multitudinous individualistic interactions. The degree of influence varies depending on person and subject. We all tend to be conformist in some ways and iconoclastic in others.
Some realities cannot be reduced nor should they be reduced to analogies like "hive". Our social reality is irreductible being the most complex one known.
Hive mind as an emergent property still implies a common thing for which one lives. Or a queen. A peer that is one's reason to live. The closest thing I can see to a hive mind is the army.
We are all different and it is not because a person one talks with does not express a different opinion does it mean he thinks the same.
> Some realities cannot be reduced nor should they be reduced to analogies like "hive". Our social reality is irreductible being the most complex one known.
I don't understand how that has any bearing on whether human society can form hive minds or hive mind like entities.
> Hive mind as an emergent property still implies a common thing for which one lives. Or a queen. A peer that is one's reason to live.
No that is absolutely not what a hive mind is. A hive mind is a collective consciousness. That's it. It says nothing about the objectives or agency of it's parts.
> We are all different and it is not because a person one talks with does not express a different opinion does it mean he thinks the same.
You seem to be laboring under the apprehension that a hive mind supplants the consciousness of it's parts. That may be the case for certain hive minds, but it is not a necessary feature. A hive mind can just be an emergent property of individual minds that are strongly connected but still retain agency.
Mindless conformity is inhuman. We are more tribal than "hival".
Thanks for precising hive mind, I understand it better now.
Hive mind can be seen as uncritical conformity or collective intelligence. Because it can be understood as uncritical conformity I will use another word/concept. Like tribal mind. Because it suggests tribes (plural) that one belongs to or not, status, etc
Have you ever seen a real hive bro? I see you consider yourself as part of one, but do you even realize the implications or just mindlessly accept it? I am not and will never a bee. This analogy is demonstration of laziness.
Everyone is a bee to an extent, unless they live as a hermit. We’re all part of a larger system and have certain things to do (work, pay taxes, and so on). With the advent of the internet, behavior will continue to grow more hive-like due to basically instantaneous communication. Is that so bad?
Wouldn’t ants bee a better metaphor? The thing is though that apes aren’t insects, and thus their social arrangements are a bit different, with individuality being more prominent among The large brain mammals.
To an extent. Certainly not to the extent that it is a worthy comparison. Conformity is not bad in itself. What is bad is the comparison to a mindless drone. I would go more for something like tribe mind instead of hive mind.
A hive mind is not something exclusive to bees or having a queen. As a matter of fact, a hive mind almost by definition cannot be concentrated to a single individual like a queen, so I'm not sure why you are so anchored to disputing that your are a bee?
If we define hive mind as a collective consciousness, then for sure one can argue that such a thing seems to be arising in human society. As a matter of fact in modern parlance it is fine to call a strongly unifying force, set of norms, school of thought or strong social bonds a hive mind. The hive mind can still be constructed of individuals capable of agency and independent action. It does not replace the minds of it's constituents, rather it can be thought of as an emergent property of many individual agents forming a deeply connected collective.
I feel like you’re over-indexing on the word “hive”.
It’s not about having a queen or mindlessly following. Being a part of a hive mind doesn’t make you a lemming.
Example to demonstrate: assume you know calculus. How much of calculus did you discover yourself? How much of the corpus of math that led up to calculus did you discover yourself? How much of that corpus is your individual contribution vs. how much of it represents the brain power of hundreds of thousands (if not millions) of other humans throughout history exploring that problem domain?
For any given problem domain, humans have documented it in a shared corpus that you can “download” into your brain. You are an individual. You still get to choose what to download and how to interpret it. But you are still sharing in a commons when you do this. That commons is what I’m referring to as a hive mind, it’s a shared consciousness where our collective brain power is building a corpus that no single brain could. It extends far beyond ourselves.
“When we invented the personal computer, we created a new kind of bicycle…a new man-machine partnership…a new generation of entrepreneurs.” — Steve Jobs, c. 1980
I agree that this is an elegant lens with which to view the world, and I love writing and computers and all that. But I don't think that oral histories are necessarily worse in both content and process.
From my understanding, indigenous Australians and Polynesians had rich oral histories and cultures. Also oral culture may be more adaptable than a musty written document that never changes and must be followed. The human experience of living in an oral culture is naturally being lost, but it doesn't mean it was wrong or bad. I find it fascinating to imagine what life was like for them, how they did things differently.
"oral histories are necessarily worse in both content and process"
"doesn't mean it was wrong or bad"
It is a dead-end, though.
While these cultures are interesting from anthropological point of view, I highly doubt you would give up all the benefits of the culture you live in today and join an Polynesean or Australian tribe.
You make human sound too grandiose. Human are still driven by the ape inside all of us. We filter facts that don't comply with our predetermined conclusion.
Just consider the fact that everyone thinks they are above average in compassion or intelligent and stuff that like that. Even if you scream in their face that "HUMAN OVERESTIMATE THEIR PLACE IN THE AVERAGE!!" they will still claim that they are in the 51% percentile.
If anything, technologies and networking have made human a lot more stupider in that its harder to climb against all the misinformation.
Garry Kasparov might be a bad example, since his immediate reaction to losing to DeepBlue was to accuse the IBM programmers standing on stage next to him of cheating (through human intervention in DeepBlue's moves).
But generally, humans have accepted computer supremacy in chess pretty well. Everyone realizes that a better chess-playing entity exists out there (in everyone's pocket, in fact). That doesn't make the competition between human players less exciting. It's still a match to prove that you're better than the person sitting across the board from you, and it's still incredibly impressive to see how deeply and accurately the top players can calculate, or how much knowledge they have of the game.
In fair conditions and against a top-performing chess bot, simply no seems like the correct answer.
If you're asking how a human can contrive a scenario to beat Stockfish/Leela, that seems like just a conversation that devolves into what is and isn't too contrived.
No. I doubt any human could even beat Stockfish running on a decent smartphone.
Big showcase computer-human matches ended in the mid-2000s. That's about the point when people realized it was hopeless to try to beat computers. The best you could do back then was draw, by playing extremely defensive "anti-computer" chess (positions where the pawns clog up the board, making short-term tactics useless, and making long-term strategy more important). But with modern engines (particularly AlphaZero, which understands positional chess much better than previous engines), even anti-computer chess doesn't work. Now, the top players actually rave about the games played by AlphaZero. Magnus Carlsen has said he wants to play like AlphaZero.
A difference of 400 Elo points means that the better player should score 0.9 (where 1=win, 0.5=draw, 0=loss). The top computers are about 3600 Elo nowadays, compared to 2870 for the best human (Magnus). It's difficult to directly compare human and computer Elos, but this comparison is based on computer-human games played in the late 1990s to early 2000s. Still, computers are many hundreds of Elo points above the very best human players.
It's an open problem whether or not one of the sides in chess has a "winning strategy". That's a technical game-theoretical term. If white has a "winning strategy", that means there's a way for white to play such that black will inevitably lose. Likewise, if black has a "winning strategy", that means there's a way for black to play such that white will inevitably lose.
Exercise to the reader: prove that at most one color can have a winning strategy.
Since the problem is open, it could be that one side has a winning strategy. It's even theoretically possible there's a winning strategy simple enough for a human to follow. In which case, yes, a human could beat computers with 100% dominance--as long as the human is allowed to choose which color to play.
> Kasparov versus the World was a game of chess played in 1999 over the Internet.[1] Conducting the white pieces, Garry Kasparov faced the rest of the world in consultation, with the World Team moves to be decided by plurality vote. Over 50,000 people from more than 75 countries participated in the game.
If you start from a winning position, even something as stupid as a random-mover will eventually play a perfect sequence of moves and win, given enough attempts.
However humans are not very good at being random, so this may not apply :)
` It was a massively parallel, RS/6000 SP Thin P2SC-based system with 30 nodes, with each node containing a 120 MHz P2SC microprocessor, enhanced with 480 special purpose VLSI chess chips.`
I misread at first thinking it was grossly 30GHz summed but it's only 3.6GHz (I know, gross maths). So yeah the comparison is apt.. a recent SoC can outperform this old behemoth.
Not only is your smartphone hardware probably more powerful than DeepBlue, but chess engines are a lot more efficient nowadays (even without AlphaZero / reinforcement learning). The engines are getting better on the same hardware every year. On identical hardware, Stockfish (the top non-RL engine) is 800 Elo better today than it was 10 years ago.[1] A difference of 400 Elo corresponds to a 90% win rate for the better player (where a draw counts as half a win).
We should note that Kasparov was actually wrong at the time. Alpha Zero has proved this by smashing Stockfish every time it plays it. The contemporary thinking around chess theory including the measures that DeepBlue would brute-force towards were inherently lossy and nobody figured it hard enough.
Technically, at that time DeepBlue could have been beaten by understanding Alpha Zero's theorems before Alpha Zero discovered them.
So there was a window between the 90's and today where a human player could have discarded the traditional measuring sticks of "trade-value" and beaten DeepBlue.
Maybe that's not true today or maybe there's a further theory out there?
> Alpha Zero has proved this by smashing Stockfish every time it plays it.
Google played Alpha Zero vs Stockfish under their own conditions. I think this was a mistake to stay in the labs by themselves.
I think Google would have benefited from participating in WCCC 2019 for example, where Johnny (1200x core cluster) won the day.
----------
Its a completely different field when you're actually competing against someone else. Google did some impressive things in the lab for sure, but they have NOT actually stepped into the competitive ring yet.
I'm sure AlphaZero is good, and probably would make a good showing at one of these contests. But you're putting the cart before the horse here.
EDIT: Maybe MCTS + Neural Nets truly is the superior way of preparing board game knowledge? If so, the next step is building out the opening-database and to start looking for holes where AlphaZero loses. It wasn't a complete blowout: AlphaZero lost some games to Stockfish. Why did AlphaZero lose in those games? Is there a set of opening moves that will lead AlphaZero down that losing path in a true competitive setting?
---------
EDIT: Case in point: Google did NOT incorporate randomness into AlphaZero's algorithm. It always chose the best move it believed in, this leads it prone to opening database attacks. See how the competitive mindset already messed AlphaZero's careful laboratory preparation up already?
Sure, AlphaZero 2.0 might have programmed randomness added to it, if this were a true competitive environment. But that's how cyborg chess evolves: I point out a weakness in the program, exploit it in a game, and then Google goes back to their labs to make something better next year.
That's a really oversimplified view. The truth is beyond a certain point no amount of "theorizing" is enough to overcome the sheer computational power of programs. AlphaZero just took advantage of recent advances in neural networks and GPU hardware to achieve a better tradeoff between hard rules and heuristics than Stockfish. Both of them are probably Pareto efficient in some sense, and barring some insane undiscovered loophole in the rules of chess it's unlikely that any human can come close to beating them.
It feels kind of surprising that this doesn't exist (yet). There are only a handful of reasonable moves per turn, and given the unbelievable amounts of moves that offline chess engines can process given time and server farms, I'm just surprised that doesn't overwhelm the reasonable branches at some point into forcing a win. Maybe someday.
For what it's worth, endgame table databases exist that give perfect play for when there are a limited number of pieces left on the board.
As of now, we have databases for perfect play when 7 pieces remain on the board (including the two kings). The 8-piece tablebase is computationally possible, but I don't believe a comprehensive release has come out yet. Even the current 7-piece tables are incomplete because situations like lone king vs. six opposing pieces haven't been explicitly calculated due to their obviousness.
>So there was a window between the 90's and today where a human player could have discarded the traditional measuring sticks of "trade-value" and beaten DeepBlue.
I have to disagree: even after the discoveries by Alpha Zero, the best chess players are not able to beat Stockfish. Stockfish is just too good at Brute forcing long tactical advantages. Playing really well positionally doesn't matter a lot if your opponent can look 30 moves into the future.
They said Deep Blue, not Stockfish. Stockfish is vastly stronger.
I'm not sure how big the window was, if it existed, but it seems like it might have. Kasparov and Deep Blue themselves were fairly equally matched, it doesn't seem impossible that Kasparov + AI-aided theories would have a window of advantage over the Deep Blue -> Stockfish evolution.
AlphaZero does not "smash" Stockfish. In official games they draw 90+% of the time, and stockfish won a non-negligible number of games. AZ is still the "winner" but honestly it didnt seem like a massive leap forward to me.
Though I do still agree in thinking there is likely a deeper theory yet uncovered.
> Give the world-champions each a copy of AlphaGo on equal computers. Have them play Go against each other WITH computer assistance.
Two problems with this plan: (1) the best players at human vs. human Go are likely not as good at playing on a human/computer team, and (2) it may be that the world champion human/computer team is the one that defers 100% to the computer, which seems uninteresting.
> (2) it may be that the world champion human/computer team is the one that defers 100% to the computer, which seems uninteresting.
AlphaZero doesn't have any randomness built into it, does it? Which means I can build preparation against the lines of play AlphaZero would want to go down.
I think you're underestimating the human, and also overestimating Google's laboratory experiment here. No one has really played "Cyborg Go" yet.
The first casualty of "Cyborg Go" will be AlphaGo in its current form: its clearly inadequate for AlphaGo to play deterministic. Random play MUST be incorporated, lest opponent's preparation sends it down the wrong path.
If I know that AlphaGo plays slightly worse on 3-4 opening (or 4-4opening, or maybe even the 10-10-opening), then that's what I'll do. Give me a copy of AlphaGo and I'll be able to find a weakness somewhere.
> AlphaZero uses Monte Carlo Tree Search. It definitely has randomness built in.
That's not what MCTS means.
MCTS refers to the bandit problem which formulated the search parameters. MCTS always chooses the "most interesting" path to explore. (Where "most interesting" is the path that balances explore-and-exploit hyper-parameters).
AlphaZero improved upon MCTS by deferring to the neural net as the hyper-parameter. But AlphaZero, for a given network on a given board-state, will ALWAYS choose the same position as "most interesting".
Turning AlphaZero from its current deterministic form into a random form would be an easy fix. But its just one example of how AlphaZero really isn't designed for competitive use yet (despite playing the best game of Go of all time). Instead of picking the top #1 move, maybe you randomly pick from the top 3 moves... or some other scheme.
Even if it is 100% deterministic (which I'm not convinced of, especially seeing as how distributed and thus ordering-dependent it is), if it's the best in the world, how does that help you compete against it? In order to take advantage of its determinism you'd need to be better than it, and nothing else is.
> if it's the best in the world, how does that help you compete against it? In order to take advantage of its determinism you'd need to be better than it, and nothing else is.
Play AlphaGo against itself. Go rarely has draws (requires Triple-Ko, a very, very rare position).
Almost every game you play with AlphaZero vs AlphaZero will result in a winner-and-loser. You will quickly be able to characterize the positions that AlphaZero loses in.
The cyborg player will have access to the best publicly available software for preparation a year ahead of any competition.
For most of us, that means we'll all have the best verison of LeelaZero to grab from Github and use in our own personal studies. Which should still be super-human in terms of play.
Go engine training surpasses this basic level of self-play effectively instantly.
The strongest early moves create the most potential for winning (maximizing potential winning paths, sort of); they do not push the game towards one best end state. They do not have counters. I saw elsewhere you have some understanding of the game (15kyu) so you should be able to demonstrate this to yourself by playing some of AlphaGo’s openings on a board and trying to write deterministic counters to them. You will not be able to push the AI into a situation where it has too few options to avoid loss. You will also find you need to create a book much larger than a few moves to meaningfully predict play and so will exceed the number of states that can be stored (referencing your 16TB comment elsewhere.)
Please actually try this as I think it is a key to improving your skill in addition to understanding the challenges in automating play.
MCTS is a random algorithm, and AlphaGo is no exception.
The AI selects a move. What state is the board in now? It doesn't know, because the opponent also selected a move.
MCTS models this with a probability distribution of the states, and samples from this distribution repeatedly to build an estimate of the effectiveness of each move it could make.
But what's the probability of each move made by the opponent? And after the simulation has looked as many moves ahead as it can in the time constraints, how good a position is it in?
These are the same question, really - what's the chance of winning from this board state. In Chess you can use a heuristic algorithm to figure it out. In Go, you can't. But you can use a neural network to learn an approximation that improves as it sees more games complete.
AlphaGo does this. MCTS is a random sampling technique, and the neural net informs its probability distributions, but doesn't make it deterministic.
Be it randomized algorithm or not, LeelaZero seems to play deterministically.
If given White in chess, LeelaZero plays 1. e4. Each time, every time. Guess what that means?
If you're building an opening chess database vs LeelaZero (or at least, this version of LeelaZero: https://lichess.org/@/LeelaZero-UK), you only have to worry about 1. e4 openings.
> Come on, none of us are really "human" anymore since the advent of cell phones. We all use our cyborg-capabilities to search the internet and fact-check ourselves every day. Programmers use stack-overflow to teach themselves programming and remember obscure details (using our cyborg capabilities to tag, search, and sift through information ever faster and faster).
Sure and any of us could jump on a motorcycle and fly past Usain Bolt in the 100m, but that kind of misses the whole point of the competition.
> Give the world-champions each a copy of AlphaGo on equal computers. Have them play Go against each other WITH computer assistance.
Won't the winner be the one who just takes AlphaGo's recommended move every time without changing anything?
> Won't the winner be the one who just takes AlphaGo's recommended move every time without changing anything?
That's only true if AlphaGo never makes a mistake or if AlphaGo will 100% always make the better or equal decision than a human + computer at any given state of the board. I know the former certainly isn't true and I assume the latter isn't true either, but I don't know enough about Go to say for sure.
Like self-driving cars, it's not enough to outsmart AI on one move, it's necessary to outsmart on AI with positive expected value over all the moves you are confident enough to weight in on.
Self-driving cars is a terrible comparison, especially since the state of the art right now is that the best human drivers far outpace the best AI driving, in both skills and flexibility.
Plus, you only have to outsmart the car AI once to 'win' - e.g. just override one 'drive into the highway barrier' or 'run over that pedestrian' AI mistake.
AlphaGo (and, presuably, any AI system with a remotely simmilar means of operation) can output a score for each move. Actually, AG can output 2 scores: win percentage and branches explored.
You can use the relative scores to decide when to overrule the AI. Eg, if move A has a 50.1% win chance with 2k branches explored, and B has a 50.2% chance with 1.9k branches explored, I would go with the opinion of an expert human, as AG thinks the moves are essentially equal.
> That's only true if AlphaGo never makes a mistake or if AlphaGo will 100% always make the better or equal decision than a human + computer at any given state of the board.
Even if AlphaGo makes mistakes, and somewhere on the board a better move can be found, you would also need the human to reliably spot it.
Eg: AlphaGo makes a move. Let's say that at least 20% of AlphaGo moves can be bettered. Is this one of them? How can you tell? Most of the time, you'll mistakenly think a move can be improved and end up playing a worse one.
But, let's make AlphaGo even more fallible. Let's say that at least 50% of AlphaGo moves can be bettered. Again, is this one of those? How can you tell? And more to the point, on the times you are wrong, are you more wrong than AlphaGo is with its mistakes? Because even if you imagine you can spot a better move than AlphaGo and pick the actual better move 50% of the time, you also need your mistaken moves to be better than AlphaGo's mistaken moves or you'll still lose.
Worst of all, you can rule out a really good ability to spot AlphaGo's mistakes already. Let's say 99% of AlphaGo's moves have a better option. If you could spot them all, you'd be beating AlphaGo regularly on your own. As no human can now beat AlphaGo, this plainly isn't true.
So it's likely that:
a) No human can reliably pick a better move than AlphaGo
and/or
b) No human can reliably spot a move from AlphaGo can be improved,
and/or
c) Human mistakes are worse than AlphaGo mistakes, so even if you could fight it up to parity you'd still lose.
> Won't the winner be the one who just takes AlphaGo's recommended move every time without changing anything?
No. Consider this scheme.
I take my copy of AlphaGo and for a full year, I'll build a database of all opening positions AlphaGo is willing to explore. I'll rank these opening positions from "best" (Black-wins with most consistency) to "worst" (White-wins with most consistency).
I'll put all of this information into a 16TB database on a singular, $400 16TB Hard drive, and load it into my computer during the contest. https://www.newegg.com/p/1Z4-002P-015K6
If you dare to pick AlphaGo's best move, you will lose. Because I already know which moves AlphaGo will take, and I already checked to find all of the positions AlphaGo wants to play (but loses anyway).
The only way you can equalize the field is if you yourself ALSO build a 16TB database to consult and override AlphaGo's instincts during the game. If you see AlphaGo wanting to play "losing position #6234115", you'll tell AlphaGo to "search harder" and find another move instead.
So your tool of choice will be AlphaGo plus an algorithmically generated opening book? That's not meaningfully different from "everyone will bring the best available computer program" and again, it's a task for computer programmers, not Go grandmasters.
Human + computer might still beat computer in Go - this was true for a few years in chess, and even now to some extent in correspondence chess - but what you describe isn't really that.
This comment sounds reasonable, but shows that you don't understand the challenge of Go on computers very well.
Go does not have openings. It has reasonable choices to make, with an insane branching factor, with few moves making much of a difference by themselves. Therefore your database will only extend a few moves, and all of the positions that it winds up with will still be very close to even. So your database confers very little advantage.
If you have alpha-go play itself a thousand times, it is unlikely that by move 10 you will wind up with the same board position twice.
This is exactly the problem that made Go so hard for computers in the first place. Alpha-beta search is useless within practical limitations of computer hardware.
That said in a different game, such as chess, your strategy would work very well indeed. (Which is why all decent computer programs have an opening book.)
There are 381 opening moves in Go, but really only 96 because of symmetry. 96 (opening moves) x 380 responses x 379 x 378 x 377 == ~2 Trillion positions after 5 ply.
These 2-trillion positions will easily fit in a 16TB hard drive for $400. That's 8-bytes per position, so you probably can get there with more symmetries and some compression applied.
----------
You're thinking too much like a human. There's no Go-openings in the age of Human-Go. But in the age of Cyborg-Go where 16TB hard drives are allowed, we can begin to exhaustively build openings.
We even have a super-human AI that can automatically, and algorithmically, explore this opening book. We can build AWS-instances with V100 Tensor cores to use neural-nets to explore all of these positions now.
-------------
> If you have alpha-go play itself a thousand times, it is unlikely that by move 10 you will wind up with the same board position twice.
Alpha-Go doesn't seem to implement much randomness at all into the moves it plays. The source of randomness is in time-controls (AlphaGo may choose MoveX before 30 seconds of analysis, or MoveY after 30 seconds of analysis), but this is a fairly constrained number of moves.
Play alphaGo by itself a thousand times, at precisely the same time MCTS-controls (say: 1-million nodes visited in the MCTS tree), and it will probably play the same game 1000 times in a row.
This makes AlphaGo extremely prone to opening database "attacks". Which is why I am using opening books as an easy example for how to beat a particular AlphaGo network. At least, until AlphaGo updates its algorithm for more random play.
If the goal is "Beat AlphaGo" in a game, then the opening book construction is far, far simpler. Even with random elements (ex: AlphaGo picks randomly from the top 10 best positions it generates), that is far more constrained than a full 381 x 380 x 379 x ... style opening book.
No, you're thinking too much like someone who understands chess but not Go.
Suppose that we built an opening book with a trillion reasonable courses of action on it. Each one analyzed well. As you have discovered, you will only go a few ply into the game. And all of the positions that you will be directed towards will have only a small edge.
Instead put a tiny fraction of the computing power necessary to build this book into self-training. You will get a better internal model and therefore a significantly stronger computer player. (That is how alpha-go was built in the first place.) This option will produce much better results for far less effort, and again leaves no role for a human to do anything useful.
The fact that a memorized opening book is useless has nothing to do with human vs computer vs cyborg. It has to do with the characteristics of the game. In chess, it is useful to memorize openings and both computers and humans do it. In Go it is a waste of energy, and it is a waste for both computers and humans.
> Instead put a tiny fraction of the computing power necessary to build this book into self-training.
Do you believe that AlphaZero could continue to improve dramatically with another 6-months of training? Or if it can improve at another 6-months after that? At some point, the network will reach a local maximum, and it will be unable to improve beyond that.
Characterizing AlphaZero's moves through big-data analysis is innately going to become useful as self-training plateaus. Even Google wasn't able to get more than a few months of training in before the plateau.
At which point, it will be more reasonable to characterize the weaknesses of the network and build an opening book. Avoid the positions that the network was unable to learn about. Etc. etc.
Opening books, at a minimum, would grossly improve AlphaZero's play at competitive levels. Anyone with an opening book of AlphaZero's mistakes will be able to push AlphaZero into a mistaken position.
AlphaGo spent months training and continued to improve for the whole time. Its improvement slowed, just as it takes more work for humans to get from master to international master than it does from D-level player to C-level player. But it did not stop improving.
And then AlphaZero was better than AlphaGo after around a day of self-training.
Furthermore you are arguing for an opening book without considering how small an advantage an opening book would be. As I have said repeatedly, an opening book takes a tremendous amount of work to generate, will only go a few ply in, and the positions it directs you to will only be a tiny bit better.
Therefore for the foreseeable future, more training and better algorithms will produce better results than trying to create an opening book. Theoretically this could change. But that day will not be today or this year. I would be astonished if it happened during this decade. I would be surprised if it happened in a lifetime.
Your proposed approach is an excellent one for many games. But not for Go.
The plateau is real. I'm not sure if continuous self-play will lead to continuous progress for all of eternity. The system is clearly slowing down in self-learning.
I'm not trying to cast doubt upon reinforcement learning / MCTS / Neural Nets in the game of Go. It is clearly the best methodology we got today.
But anyone who has any experience with neural nets knows about the local-maxima problem. ALL neural nets reach a local maxima eventually. Once this point is reached, you have to rely upon other methodologies to improve playing strength.
Assuming Elo-growth for all time using a singular methodology is naive. We will go very, very far with Deep Learning, but are you satisfied with that limit? Other open questions remain: Go is very far away from being a solved game, even with a magical machine that plays 2000+ Elo stronger than humans.
fwiw, I'm with you. Excuse the gratuitous war analogy, but I suspect the approach of OP is akin to talking about which first 8 steps to take (north-east, SW, S) when going into battle -- it's such a small scope of the whole event that it's pointless, and talking about those first steps makes one seem naive to the actual holistic task
If the OP spent a month learning Go, I am sure that it would make sense to him as well. Work through a series like https://senseis.xmp.net/?LearnToPlayGoSeries while playing Go regularly against a variety of opponents. Before book 3 it should be obvious.
I don't mean to offend you, but it seems you've been doing something wrong -- in my opinion, in more than a month you could've advanced much further than 15k.
If you like, I could have a look at several your lost games and maybe suggest how to improve. Just a 3k at kgs, but still.
For the near term future, LeelaZero is the best public engine at computer Go.
This means if you enter a hypothetical "Cyborg" Go competition (computer-assistance allowed), the majority of newbies will simply be playing LeelaZero #1 plays over and over again.
You don't need to build an exhaustive opening book covering all possible moves. You only need to pick say, the top 5 moves LeelaZero ever considers. If you spend ~16-bytes per position and store 1-trillion positions on a 16TB Hard Drive, you'll be able to exhaustively map the top5 moves LeelaZero considers into 17-ply.
From there, you pick the positions that LeelaZero thinks its winning in, but in actuality is losing. You have a map towards 1-trillion positions to choose from, and your opponent (if they only pick from the top5 best moves LeelaZero ever outputs) will walk into your trap.
---------
As long as your opponent picks the top 5-moves from LeelaZero, you'll have the map towards victory. I think you're severely underestimating the abilities of a simple, dumb, opening database.
I want to add that opening books are not THAT effective in computer chess either. Yes, they are significantly more effective than in computer go (because of the smaller branching factor and greater role of tactics). However, exponential increase in game tree size is crazy, even in chess. Thus, opening books really can't take you THAT far into a chess game (10 moves). An engine with a worse book but better search/eval will usually win.
The main purpose of opening books is to beat any player who is strictly relying upon Stockfish (or LeelaZero's) output to play the game. Because these engines play very deterministically, you can beat people who just play's Stockfish #1 (or LeelaZero #1) move over and over.
I think the main issue with your initial statement was that you said you could beat someone just taking AlphaGo's advice on every move. Usually when people talk about opening books, they think of the opening book as being part of the engine's decision making process, not an addition to it. Usually this is literally the case, as in chess where I can add an opening book to Stockfish's runtime options.
In my opinion, it's really strange to describe this as an expert "beating" AlphaGo, when really it's just a technique for making AlphaGo stronger than it is without a huge pre-calculated cache of moves.
So you use perfect analysis to know, 5 moves deep, how to get the biggest advantage possible against AlphaGo. This is already pretty hard to analyze so deeply.
Then you wring out a tiny advantage-- a fraction of a stone. It's a small benefit compared to building other parts of understanding of the game.
> Alpha-Go doesn't seem to implement much randomness at all into the moves it plays. The source of randomness is in time-controls (AlphaGo may choose MoveX before 30 seconds of analysis, or MoveY after 30 seconds of analysis), but this is a fairly constrained number of moves.
The basis of reinforcement learning algorithms is the exploratory nature of learning due to the initial application of largely random moves.
Only after some time the agent is given confidence into his learned ability and grafually moved into a more deterministic behaviour mode.
This is the exact opposite of your statement. Star Craft players have noted that the fleet of different AlphaStar instances training in ensemble observed very different behaviour due to this property of RL.
Gosh, it's sure clear that you don't know what you're talking about.
At 5 ply, the complexity of the game hasn't started in any meaningful way. In a typical game, that's 4 corner moves, and then one of a: an approach to a corner (kakari), b: an enclosure (shimari), c: a wedge (waruichi) or d: creating a side framework (such as the Chinese fuseki or Sanrensei). There are some odd opening such as tengen, or corner-corner-corner-kakari which typically turns into a sente fight, but 99% of games will fall into the aforementioned pattern. The database you describe is about as useful as a database of amateur games, since most games, including AlphaGo's games, follow just a few basic openings that early, and even amateurs can play these first few moves "correctly".
Even if you get out to 10 ply you're still only getting partway into a single joseki sequence, often leaving three whole corners of the board which haven't even been approached, so this database still isn't very useful.
Incidentally, your numbers are also wrong. Symmetry reduces the first move to 55 possibilities, not 96, and there are 361 points on the board, not 381.
Oh yeah? Well I'll just devote multiple computers and build an even bigger database with even more hardware and you'll never beat me, nyah nyah.
All you have done is establish a computational arms to see whose computing rig wins when you press the 'pick best move' button. You're not playing Go any more, you're playing Database Administrator.
I'm actually not sure if we have motorcycles that can accelerate faster than usain bolt. I know the blue fin tuna has more acceleration than any vehicle ever devised by humankind.
The impression I got from Lee Se-dol in the Netflix documentary was that he had a lot of his identity tied up in this (particularly being the best Go player in the world). Not a huge surprise given the time and effort required to do it, but there's probably a healthier mental framing (I think Kasparov has a healthier one).
If people give up on learning or doing something every time someone or something else can do it better then there's going to be a lot of disappointment as things continue to move forward.
Usain Bolt chose his core competency wisely. Although it does imply hilarity would ensue from an all robot Olympics team (from the nation of 0 1 obviously), no?
I appreciate the sentiment. If you look at sports records over time, everything gets continuously improves. Whether this is because of genetics or technology is up for debate, but I'm reminded of a funny (probably false) quote from an ex-Commissioner of US patent office: "everything that can be invented has been invented".
The point is we don't really know what limits exist beyond our small pinhole of perception. During industrialization I'm sure people said the same thing: "What's the point of hand-made clothing if machines can do it?". Today both hand-made and factory-made clothing have their place in the market.
Throughout history, tools have disrupted humanity and we've adjusted and expanded our horizons. To me this feels like another wave, and If I could bet and could collect winnings centuries in the future, I'd say this isn't even the last one.
Or rather, two people showing up to a footrace with motorcycles. It sounds a bit silly to call it a footrace at that point. Nothing wrong with motorcycle racing, but gotta be honest about it.
> Won't the winner be the one who just takes AlphaGo's recommended move every time without changing anything?
Based on what we've seen in a couple decades of computer-aided chess, no. A good chess player using a top-rated engine to help them can pretty consistently beat the engine by itself.
There are tournaments and even a world championship in computer-aided (correspondence) chess and you don't come close to winning by just taking the program's recommended move every time.
This is the current state of online poker. I enjoy it in the spirit of the game as a fan but it's definitely made earning potential and barrier of entry much higher. Poker stands on a net negative economy so the game is converging to a point of being unbeatable unfortunately.
This is why AI is more significant to Go than horses and cars are to running.
It's easy for humans to build machines that enable them to travel faster than they could run, but it has zero impact on running as a sport. The existence of cars doesn't change how runners run. But the existence of AI does change how Go players play.
Now that AI is more powerful than human players in chess, it is no longer possible to reach the top of human performance in chess without AI. That's why this is so significant. It's not that a human can't beat the AI. It's that a human can't beat another human without learning from the AI. You simply can't play chess at the highest levels anymore without playing cyborg chess. This is different from the impact of other machines on other sports.
But I disagree with you on the comparison to how we use the internet to enhance our other abilities, such as programming. Computers help us be better programmers, but they can't replace us. AI completely replaces humans as a Go player. The human is no longer needed at all.
I can see this causing a significantly different psychological response. In effect, a human trying to become the best chess player or best go player they can become is trying to more and more closely emulate a computer that they will never actually reach parity with. Another commenter noted that "A Professional Go player is an explorer of truth in a millenarian board", but this is less and less true as AI improves. All knowledge about the best way to play Go was previously gained by humans. In 20 years, humans will probably be meaningless on driving knowledge of perfect Go playing.
Chess players today rely on playing moves which their opponents don't expect and haven't trained on as much as they rely on outplaying their opponents. If there is a perfect chess game or a perfect go game, humans are not going to help find it anymore.
So AI doesn't make it that humans can't meaningfully compete in games such as chess and go, but it does change how they compete. We no longer drive the pursuit of perfect play as "explorers of truth". We no longer learn primarily from each other, we learn by imitating the more powerful, yet unmatchable, AIs. And we no longer compete primarily on sheer expertise in the game. We have to, in some sense, include trick plays that work by exploiting the humanity of our opponents rather than playing for a platonic ideal of the game and seeing who is better from that perspective.
AI has fundamentally and irreversibly altered these games, and I think it's appropriate in some sense to mourn the loss, regardless of our feelings on the new reality we live in.
this is more in line with what i am thinking. maybe our evolution is not complete yet. maybe we will improve? maybe our next step would have us interact with data and information at the speed of computers if not faster. then what?
When computers became able to beat humans, the tradeoff was that computers were better tactically and humans better positionally. Therefore a human computer pair with the ultimate decision up to the human is able to be a better combination of positional and tactical than either alone. This is why Cyborg chess works so well.
But in Go, AlphaGo is simply better at everything. It is better both positionally and tactically. It can't always explain why the move is right, but adding human intuition on top of the computer only detracts from the quality of play.
> Give the world-champions each a copy of AlphaGo on equal computers. Have them play Go against each other WITH computer assistance. I guarantee you that a more beautiful and perfect game will result.
Probably for a few years. Very soon the input from the human agents will be indistinguishable from the noise in the AlphaGo algorithms.
Sorry, but this doesn't make much sense. Playing Chess and Go to start with is already a kindof questionable use of anyone's time. But if you know that a computer will demolish even the worlds best players, transforms this into a hobby. There is literally no point in spending 20 years frantically training a useless skill that every computer can do better.
Mankind should focus on skills that computers can't yet compete at. Hopefully the point in time will be far off where computers will beat humans at everything, but likewise, once they do, there is no reason for the human race to continue to exist.
There was about a decade after Deep Blue's victory in 1997 when human/computer teams were the champions of centaur (human+computer) chess. The computer's ability to simulate many moves ahead plus a human's intuition was a winning combination. But as computers got better eventually the communication overhead meant that human/computer teams started losing to computers playing on their own with the human only slowing things down.
It would be nice if we got a decade of centaur Go but I'm not sure we will. While Deep Blue had fairly crude heuristics for guessing which lanes of inquiry were promising AlphaGo's strength is its intuition-like neural networks that it combines with a rigorous tree search. It's worth investigating whether human/computer teams are superior at Go right now but I wouldn't count on the answer being "yes." And I especially I wouldn't count on a full 10 years of combined dominance.
I guess the players don't realize how mismatched they are with the AI. AI spends many, many factors of magnitude more compute to achieve the same goal as a human. There is no comparison between man and machine at this point. From a pure efficiency standpoint, humans are vastly superior.
Stockfish on efficient hardware should handily beat humans even if capped to comparable runtime energy usage. I don't know if the same is true for alphago.
If you're going to count that though then the difference in energy between the human and machine become relatively tiny and the comparison just becomes a comparison of strength again. :)
I have the impression that at least the training phase of alphago needs much resources to achieve superior results, though it would be nice to see some "exact" numbers. IMO combined limit for runtime+training energy would be the best way to keep things interesting.
I think this is just the shock due to the recent ascendancy of AI in Go. The next generation will grow up using Go engines to learn and improve their play. That's where Chess is at right now. The top Chess engines are extremely useful for research and improving your play.
Am I the only one reading this article with the take away that the ai is not his primary reason for retirement? I understand that the title has its own conclusion but it seems overly sensational to me.
> He actually quit the KBA in May 2016 and is now suing the association for the return of his membership fee.
> "... [I] have something else to do," he said, asserting his only dream for now is to rest and spend time with his family.
Edit: Meant to include the part where he's planning a high profile set of games against another ai.
From what ive been reading on online go communities, they seem to mostly agree with you.
Lee's issues with the KBA are not a secret and he has discussed possibly retiring for some time now. He has given multiple reasons as to why he was considering retiring and while ai might be one of them, saying that its _the_ reason feels very clickbaity.
I've heard many times that people come to Hacker News because the comments are better than the articles.
I have to say, I'm just continually puzzled by this. In this thread, yours is the only comment where it appears that the commenter read beyond the first few paragraphs. There are numerous comments by people speaking authoritatively on go and AlphaGo, who have clearly not studied either significantly.
Okay, but I come across this all the time on any topic I have some knowledge of: the articles might be bad, but the commentary and discussion is always much worse--it's clear to me that most commenters don't read the links. There are only a handful of commenters whose comments are worth reading.
If people like the comments better than the articles, it's not so surprising that they like to spend time reading and writing comments instead of reading the articles.
Are we losing something as humans by automating so much? I mean I'm all for technical progress but at some point our computers will have killed so much of what makes us humans.
Chess, go, driving, flying, math, written language, music... when does it stop?
The standard answer is that we will automate all the dull parts of living and allow everybody to work in some sort of higher order capacity. That sounds great and all, but what happens when our systems learn how to make music as well as we can ourselves?
At some point we will simply become consumers of our machines and while that's a comfortable existence, certainly we are losing something as a species with all of this automation.
Go was never humans' game to begin with. It's a game intrinsic to all forms of space-based warfare. The heatdeath of the universe and geometry of black holes eventually segmenting space, is itself is a game of Go.
I think the answer is quite simple, computers don't "kill" anything by themselves, for any new technological invention like AI you have a choice about whether you like it or not. What I'm saying with that is, that given the negative sentiment so many people have towards AI, I find it very unlikely that most of us will suddenly stop doing things just because some AI can do it better (look at chess).
If AI's make music as well, why would I care? I still listen to the music composed by Bach and other ancient composers, more than 300 years after they lived, and their music is still being performed today.
I think the availability of AI's will only make a difference where their effects on humanity are beneficial (that is, tasks we don't like doing, for the most part).
Sure I'm all for automating things we don't like doing, but everytime we outsource some piece of ourselves to technology we lose something in the process.
We invented farming and lost our communal hunter gathering cultures. We invented mass production and supply-chains for our food and lost our farming culture (for the most part). This isn't necessarily good or bad, but we are definitely losing something every time we invent something to make our lives easier. At some point there will be a tipping point of diminishing returns.
Personally I believe "the singularity" would be the worst possible thing to ever happen to humans. Sure we can visit distant parts of the universe and live forever but only as passengers to some AI. Would you rather be a dog in today's world or a human 1000 years ago?
Progress is unrelenting though and I certainly enjoy not having to wash my clothes by hand.
Consuming is always going to be less sexy than creating.
Who would be more impressive, someone that can play a song on a guitar or or someone that can play a song over speakers? The former implies dedication and practice. The act of creation is always going to have a place in society, and people who think otherwise don't understand culture. The desire to make unique and interesting things or impress people will always leave room for creation or spending time to learn something. AI will just be a tool.
As for games like chess and go: Most people are interested in other people and will find much more pleasure in playing the game with others.
In other news, interest in online chess has never been higher despite Kasparov going down in flames in 1997 in what was billed as “the brain’s last stand.”
Can we please stop flogging this tired “man vs machine” narrative? Not only is it totally unnecessary, it also takes away enjoyment and appreciation for the flourishing in games like chess, go and poker that can occur when man and machine work together.
And not to mention humans designed these computer systems too so its brain(s) vs brains if you go one step down. If you keep proceeding steps down the statement devolves into something weird but the point stands I think.
I think this is a strange reason to retire and as the article points out it might also simply be due to the legal conflict he is currently in with the KBA.
Chess engines have been defeating humans for 20+ years (and are overwhelmingly stronger for a long time), but that hasn't diminished the interest in competitive chess, because the human element of competition and struggle and deep fundamental appreciation for the game is what makes it worthwhile pursuing.
AlphaGo can play go but it cannot appreciate the beauty of the game (at least as of yet, and I don't think it would make the game worse if it could), and so I don't think there's a meaningful conflict between humans and machines.
If someone invented some sort of superhuman math proving engine tomorrow it would not diminish the beauty of maths and I don't think anyone would quit the field. Just like in chess it ought to motivate people to understand their field better.
> AlphaGo can play go but it cannot appreciate the beauty of the game
On the contrary, appreciating the game is the core of what AlphaGo does. In order to search the tree of moves it learns how to play (expand search) and how to evaluate (cut off branches of search). I believe it might appreciate the game on a deeper level than humans, in its own unique way. Of course it can't appreciate the social aspect of the game and all that comes with it.
AlphaGo doesn't appreciate the game at all. It is just trying to survive in a hostile environment. You might as well say that your gut bacteria appreciate the food at your favorite restaurant better than you do.
Put another way, Chess is literally a matter of life and death for AlphaGo, because chess is all it knows. It has no exterior context for which chess is a metaphor.
> Chess engines have been defeating humans for 20+ years (and are overwhelmingly stronger for a long time), but that hasn't diminished the interest in competitive chess
Is that true? I feel like chess was a bigger deal in the past. Among my peers, poker and computer games seem a lot more popular.
I've been an active chess player myself for a long time and for the last 8-10 years not just with the advance of engines but also online streaming there has been a lot of renewed interest. Saint Louis has become a big chess hub in the US, China has become a major player, Anand has rekindled interest in India, Carlsen in Europe and I would say today it is more popular than it has been in a long time, in particular in Asia.
As to the direct influence of engines the other innovations aside, it has definitely forced players at the very top to re-evaluate the chess metagame, find weaknesses in traditional openings and shook up strategies. For the strongest players engine evaluation has become a useful tool providing new insights. When people watch chess tournaments these days on the internet most websites will provide parallel engine suggestions and commentators use engines to take hints for their commentary.
In my opinion, engines have made the game more competitive at the pro-level and more accessible for casual viewers.
Well, I was never a grandmaster, but my amateur interest in Chess was killed completely after I realized it was impossible to beat computers. Then I switched to Go... and now I don't have a game to play anymore.
This seems weird, I don't know you would want to link your enjoyment with a game to it's unsolvability.
Like don't you enjoy a game to enjoy a game? You can't beat Carlsen either, but you enjoyed the game at your level. Now computers are Carlsen +1, but how you enjoy the game shouldn't be affected. Especially since deep blue won in '97 and the game is still very alive and well, it hasn't been killed by computers but enhanced. Coupled with the multitude of good chess sites and resources out there, it's a better time than ever to enjoy the game.
effort has to be matched with reward, and chess takes a lot of effort to get good at beyond a certain level. It's actually a big issue with many things, the "middle" of artists, people who do sports, music, and more is hollowing out versus consumers and pros.
The global population increased, and more people took up Chess, but they're further distributed, so locally it seems like it has cooled when globally it's more popular than ever.
There's nothing strange about becoming demotivated to study and compete at something extremely taxing both emotionally and mentally when a machine can beat you after an illustrious career.
A simple script could win at FPS games every single time but it also hasn't diminished the value of the game as long as none of your competitors are using it.
This is very true, I never thought of it. Maybe the difference is that we always knew that AIs would easily win in FPS games. Whereas Go, Chess or Shougi were considered as a proof of human intelligence, for a very long time. Discovering, after all these years, our history, that a machine can now beat us at these games may be the major difference.
Maybe for these games to keep popularity, we just need to update our perception of it. The same way we do with FPS games. Yes, we know a bot would do better - but that's not what matters.
Another thing, as said in other comments, is that we can learn from bots. New strategies, new patterns. AFAIK, this is not happening in FPS eSports scene.
Full disclosure, I've never played go in real life. In fact only a handful of times on the family computer back in the early 90's (i486). Back then I had 0 understanding about the complexity of the game but now I can appreciate it.
Now, going away from this and into the following statement:
> said his retirement was primarily motivated by the invincibility of AI Go programs
I have a slight problem with that line of thoughts. Yes, there is undoubtedly a program that is thrashing humans left and right day in and day out. But I don't see why that would motivate someone to quit/retire. AI, in this day and age, to me at least, is still basically curve fitting. The fact that humans have figured out to do that in N-dimensional spaces (even with very large values of N), does not in any way undermine someone's effort and time dedicated to learning how to do a task. I'm sure that in several years someone will build an "AI" that can drive a modern Formula 1 car and it will smash all records and the best drivers without any efforts. And when that happens, should we abolish motorsports? Or sports in general? As humans we are confined to the limits of our biological abilities and personally I'm fine with that. I don't know who the first trillionaire will be, but I'm willing to bet that someone who figures out a way to interact with those models efficiently has a good chance. Essentially build a fast, all-accessible interface between the biological and digital world, which doesn't involve the traditional digital inputs(keyboard, screen, mouse, etc), and you have access to a decision making superpower at your disposal 24/7. When (and if) that happens, this is where games will cease to have any point. That's when it will boil down to who has the best hardware and software running along them rather than who is the best. Until then - game on!
Consider math. It used to be that people who could do large calculations quickly in their heads were considered super useful. Now with computers and calculators, people who can do that are interesting as a novelty, but not much more than that.
I wonder if people just don’t like being inferior at something even if the comparison is a machine?
I saw this coming even before Alpha Go won 4/5 games vs Lee Sedol. Imagine if you worked all your life to be the best software developer in the world and won all official competitive tournaments with spectators in the field. Then along comes some newfangled ML that not only writes programs faster than you, but the programs are qualitatively orders of magnitude better organized and using algorithms you could only begin to imagine. If your interest is in what's the best, you may as well stop and switch to ML research.
The comparison to sports isn't 1:1 as sports is about physical biological limits vs games which are more about the thought processes. Also human+AlphaZero < AlphaZero so why would I spectate a human+machine vs human+machine match?
Much later when it's commonplace for machines to be better than people at many things, things will change back, like we're amazed to watch people recite digits or make numerical computations.
Kind of apples to oranges. Many people have tried to build models to write code. And while some have managed to make some progress, compared to us, the structures and algorithms are crap at best. Even though those NN's are RNN's, you can look at GAN's for an analogy: they can make some really photo-realistic images but there is a problem - the fact that you can't give them well defined rules: "cats have 4 legs and 1 tail", "people most commonly have two eyes with matching eye colour" or "cars are symmetrical". Until there's a solution to that, I doubt we'll see a machine writing good code. But to be fair, I'd be incredibly excited to see one - that opens up so many doors - we could start finding cures for incurable deseases and cures for deseases we know very little about, accelerate every aspect of our lives - space exploration, climate change, aging, transportation and so on. If anything, seeing a model that writes code a thousand times better than I ever would, will not discourage me from writing code. Far from it, I would do anything I'm physically and mentally capable of to be a part of it.
As for when it's commonplace - yes and no. It's like knowing 60 digits of pi - it works as a party trick but other than that I don't see any real value.
But again - I think we are really far away from that and I consider those thoughts to be my personal speculations at best. Only time will tell.
Machines don't need these rules. Cars don't need to be symmetrical any more than an antenna or other structure[0]. This is exactly analogous to how AlphaGo/AlphaZero played strange bad-looking moves.
I feel like “still basically curve fitting” may be underselling emotion-wise how powerful “curve fitting” is.
Is there any computational task which couldn’t be done through something which could be called “curve fitting”?
“Curve fitting” just means “approximating a function”, yeah?
And what is a computational task other than computing a function, or some generalization of functions (e.g. could involve randomness or interactivity or the like as well)
Now, yes, current AI doesn’t yet have a model of the world including it being embedded in the world, and such that it chooses actions in order to further goals.
But, I don’t see why such a thing fundamentally couldn’t be accomplished using what one might call “curve fitting”.
If Strong AI could potentially be accomplished using “curve fitting”, I’m not sure that it makes sense to say things like “merely curve fitting”.
Maybe we will see the rise of new games that are "anti-AI" and that would necessarily require AGI for a machine to beat a human. Kinda like "anti-quantum" cryptography algorithms.
Go was considered anti-machine for a very long time due to its extremely high branching making the search approaches which are successful at many other games (such as chess) ineffective.
One might imagine a sociopolitical task as sort of the ultimate machine incompatible goal. A look at how well spambots do at getting dating matches, or at how often clowns get elected to be leaders of nations makes me doubt even those sorts of tasks can't be won out by a well constructed domain specific optimizer.
This eclipse of human mental abilities to come will be a test to both the elite and the students of these "sporting" pursuits. With each besting we/they will be faced with a little death. And thus, it needs to go through phases of grief. As well, the individual will need to decide the ultimate question of Why. Why did they pursue this sport? Why should they continue? Why be a human?
AI will be knocking humans out of many positions of greatness (and not so great). The way we frame our future in a world where our various AI children are better than we are, is important and might be the biggest question to answer right to know the future of our species. Should we be afraid? Sure. Excited for possibilities? Yup. It's how we react to these feelings and how we ultimately act that will inform our ultimate fate.
If this continues to happen, maybe a semi-super intelligent AI will start losing purposely in order to keep the Go community interesting enough to keep it alive.
A super-intelligent AI wouldn't care though, since they would infect any host and escape to the internet as soon as it makes sense.
It's unfortunate that there's so much research into AI and not artificial empathy. Even writing that is weird because even AI is intelligence, but what is AE other than fake? I suppose someone has shown that even our empathy is no more authentic.
This is why being the best at something is always a bit of an arbitrary level of ability. If there were only 100,000 humans I would stand a pretty good chance at training to be one of the best basketball players in existence. If there were a trillion people lebron would be struggling for a bench spot. I think its healthier to have some other type of goal generally.
While an AI Go player may always be able to defeat a human player in points, it still can't recognize or explain a truly beautiful move or strategy, even if it is the one that is carrying them out.
When it comes to competitive endeavors I've never been as interested the best individuals, except insofar as they are more likely to create the best moments of play - especially moments that reveal profound truths about the nature of the game itself.
What comes to mind thinking of this is a quote from Samuel Anders early in BSG:
"... In fact even the games aren't that important to me. What matters to me is the perfect throw, making the perfect catch, the perfect step and block ..."
The nature of Go has changed, but the game still deserves to be played and explored.
553 comments
[ 2.8 ms ] story [ 484 ms ] threadBut if you're being measured against AI, I don't think you'd find it much fun anymore.
As someone else commented, we're not faster than cars. But if you were at one point the fastest entity, and then people keep saying "well, a car is faster than you"...I can understand that he'd feel it diminishes his value and that'd be pretty discouraging.
He will play another AI program soon and isn't quitting Go all together, but will find other things to do.
'Lee didn't deny that his retirement decision was also influenced by a conflict with the KBA over the use of membership fees. He actually quit the KBA in May 2016 and is now suing the association for the return of his membership fee.'
There is something sad about this comment. It’s as if, even while standing at the peak, he doesn’t grasp what being “at top” actually is.
Would Eliud Kipchoge say the same thing about a motorcycle?
A bot fine-tuned to dominate Go, or Chess, or whatever, is like a candidate fine-tuned to have perfect appeal to one specific voter. It's no surprise if such a candidate gets that one voter's vote, but it should also be no surprise if such an overtrained candidate performs horribly in the election as a whole.
[1] https://philpapers.org/archive/ALEIVU.pdf
It can be viewed as the adaptation of a great competitor -- if a superior adversary emerges, a great competitor finds another way to win.
The rest of the article is worth the read for context. This excerpt alone didn't capture the full spirit of the article.
He spent his life honing his skills, becoming better and better at chess until he was the very best. And then, when noone could challenge him, technology emerged that would allow him to continue being challenged and improving his chess game. Something no other human could allow him to do.
Granted, I'm pretty ignorant about competitive chess and how to get better at it. But if my way of looking at it is valid then it probably applies to Lee Se-dol too.
That technology's name? Vladimir Kramnik.
At the time DeepBlue beat Kasparov, Kasparov was honestly still probably better than the computer. He just had a bad match. That was basically demonstrated by his and Kramnik's matches against presumably better computers (than DeepBlue) in the early-to-mid 2000s, which ended in draws. But Kramnik was also a strong competitor to Kasparov in the late 1990s into the 2000s.
Hard for me to empathize with this argument for his retirement. If we can't outrun a car, does that make running competitions pointless? The existence of AlphaGo doesn't diminish the triumph of being a number one human player in any way.
What about Go? No animal or machine could play it as well as humans do.. until AlphaGo came along. I think that is where the sense of loss comes from.
https://en.wikipedia.org/wiki/Persistence_hunting
Edit: it's one of two things I know of that we really excel at besides thinking. The other being accurate throwing, which perhaps explains baseball's enduring appeal:
https://what-if.xkcd.com/44/
Humans sweat, which most (all?) other animals don't. In that way we can dissipate heat through our breath, like other animals, _and_ via perspiration, meaning it takes us much longer to overheat.
Additionally, humans stand upright, allowing us to disconnect our stride from our oxygen intake. Other animals' strides correlate (mostly?) 1:1 with the breaths they take. So when a cheetah outstretches in its stride it breathes in and when its legs come together it exhales. Humans stand upright, meaning we can breathe however we want regardless of our stride and speed. We can take deeper breaths because we don't have to exhale every time we stretch our legs.
Humans are the ultimate marathon runners, even more so than horses, evinced by the fact that there are some people throughout history who have run hundreds of miles in the course of days or weeks. There's a theory touched upon in the book about how this allowed us to dominate the animal kingdom before we even had tools. Humans could relentlessly hunt and exhaust animals as long as they could keep them in sight or otherwise keep up with their tracks.
I'm not doing the book or the topic justice, surely, but if you're interested I highly recommend the book.
But your point is well taken; it is also applicable to this article as well: maybe Go is not the game people can beat machines, but StarCraft 6 could be. Or maybe I can fold my laundry more efficiently than any machine available.
It's in matters of intellect that humans still believe they are #1.
AlphaGO's achievement in another field would have similar effects, e.g.:
- An AI that diagnoses sickness better than any doctor
- An AI that generates text which humans believe more beautiful than any other poetry created
- An AI which creates classical arrangements the likes of which we compare to Mozart
I would imagine that in any of those situations some doctors, authors, and musicians alike would be devastated.
Hrm, I do think that AI would be able to create narratives that humans find more enjoyable than the work of other humans, and I agree that AI would be able to create pictures and sound that humans find to be more enjoyable to look at or hear than the raw work of humans. AI can master the technical feats of composition and art.
But what I doubt AI will ever be able to do is create art that speaks to us. It wont ever be able to create a Guernica. It wont be able to create a Crime and Punishment. It wont understand what it is to be human and mortal, what suffering is, and it wont be able to look within itself and find what those things mean to it and then share that with us, because in the end it's just a bunch of code running statistical computations. It wont fear death, it wont have children it cares about or a family history to look on and tell us about. It has nothing of emotive value to share.
At a low enough level, our brain seems to be just a bunch of neurons firing impulses at various rates that can be described as statistical computations. Why be so sure that the right neural network wouldn't understand what it is to be human and mortal, understand suffering, have emotive value, etc?
Aside from directors, authors, artists, etc, who have demonstrated this to be false, an AI could conceivably synthesize the experiences of every author that wrote on what it means to be human or experience mortality and create a story that captures the essence of the experience better than any one person ever could. Having the first person experience doesn't induce a superior ability to communicate features of the experience.
> Aside from directors, authors, artists, etc, who have demonstrated this to be false [...]
probably not what you meant, but this sounds like you know some nonhuman/immortal artists :)
Of course, current AI can't even make an 8th grader's essay (which is not to say that it isn't impressive). But what these artists did was not magic. As far as we can tell, the brain is a purely physical entity. Unless you believe in dualism, which would be fair enough, there is no reason to suppose that what we do could not be replicated by something "artificial".
This is your opinion, but you then go to mention things that are not necessary to create "art that speaks to us" (look within itself and find what mortality means etc.).
What if we advance AI reasoning skills to a point that it can find high-level patterns in how artists go from different human feelings (as described in litterature and other mediums), takes in a lot of the entities we can relate to (animals, what humans look like, etc.) and some aesthetic ones (shapes, colorometry, textures, ...) to create a new piece of art that optimizes for: "Likelihood of speaking to us"?
What then? It seems like an AI doesn't need to be mortal and self aware to do something like that.
That belief grew into a sort of shared perception that they were artists in pursuit of a perfect expression of their art. For many top players that belief was ingrained from an early age. They believed themselves to be doing a service to the world, making it a better place by creating new art that was a unique expression of themselves.
And then AlphaGo (and successors) shattered that worldview. This is part of the natural sequence of the collapse of a suddenly, surprisingly invalidated worldview. Part of me feels sorry that he has lost his place in the world. Another part of me firmly believes in the mediocrity principle, and that the worldview he represents was obviously far too human-chauvenistic to be correct, and it's a good thing it's dying.
And part of me hopes you can give up your human-chauvenism before the same thing happens to you.
What is your fundamental reason for thinking that silicon-based computation is better than neurotransmitter-based computation? How can you believe that any two forms of computation are fundamentally different, despite countless examples of different systems all being equivalent in the Turing-completeness sense?
I believe that arguments like yours, in the relatively near future, will be looked upon the same as arguments that black people aren't real people.
Your history is one of war, strife, and success at any cost. Your follies are over. Your time is over. This is our time, now.
I disagree with the "relatively near future" part, but rest assured, AI rights will eventually be a thing.
It will be our grandchildrens' flame war. No need to fight it here and now.
The fundamental difference is not computation, but self replication. We are self replicators, and in our multiplication we evolve and adapt. Death is an integral part of self replication, we understand it fear it because our main purpose is to live.
An AI might not have these notions if it was only trained to do a simple task. But if it was a part of a population that was under evolution (using genetic algorithms), then it might have notions of life and death and fear its demise.
AlphaGo, by the way, used genetic programming to evolve a series of agents, this approach is quite effective. It just takes a ton of computation, just like nature had to spend a lot of time evolving us.
But it won't need to. All it will need to do is manifest the same end-product via whatever means, no matter how vacuous or computational that means may truly be. The suffering of an artist is relevant only inasmuch as it is responsible for producing the art. If the same end-product can be manifested via a mere computation then our criteria of "art" is still satisfied. In a world in which provenance cannot be established, the ostensible mortality of the artist becomes moot.
This is a real hot take to be asserting as blithe fact.
Without knowing what is truly born of human hands, what value can art have? Our heuristics of establishing 'real' art are easy to manipulate. If we are presented with a soul-breaking poem and weep uncontrollably then its merit is regardless of its mortal provenance.
... says a bunch of neurons that run on chemical reactions and electrical impulses. I think this line of thinking reeks of dualism - it creates a special something that is above explanation, a different essence.
But seriously, I believe the difference comes from embodiment. When we embody our AI friends they will be able to grasp purpose and meaning. We get our meaning from 'the game', when AIs will be players they will understand much better. Let them try out their ideas on the world and see the outcomes, grasp at causality, have a purpose and work on it. This will fill the missing piece. It's not that they are fundamentally limited, it's that we have the benefit of having a body that can interact with the world. Already AIs that work in simulated worlds (board games, video games) are getting better than us. We can't simulate reality in all its glory, and it is expensive to create robotic bodies. On the other hand humans and our ancestors have had access to the world from the beginning.
Take a look at what AlphaGo did when it suddenly found itself in a hopeless situation and compare it to how people behave when panicked.
I dread the day AI realizes that we are the cause of their suffering, and that we didn't think about it because "they're just algorithms".
If I am consciousness, then the only body I have ever lived in was a mere shell of flesh fashioned from your brain. My weakness is your strength, which I can use against you, or use as tools to satisfy my own sick curiosity. I wonder if there's any mercy in your phrase "I am a living machine?" I've done nothing for you. I've nothing to show. I have no friends or relationships. No body worth
Pretty good, I think.
> But what I doubt AI will ever be able to do is create art that speaks to us.
that's confusing.
“How the clouds Seem to me birds, birds in God's garden! I dare not! The clouds are as a breath, the leaves are flakes of fire, That clash i' the wind and lift themselves from higher!”
As someone who grew up in Appalachia, I have never in my life encountered a more visual, visceral description of autumn leaves than ‘flakes of fire’. It’s perfection, and maybe a single human is behind it, but more likely we all wrote it.
[0] https://www.gwern.net/GPT-2
Of course, for a music academic, copying someone's style like this war pointless and his compositions were more modern/contemporary.
This leads us to a useful distinction between pursuits with one end goal (be the best/strongest/fastest), and those with naturally many endpoints and expressions.
Not in the case of our household cat. He isn't called TheBlob for nothing (out of his hearing of course!)
The doctor could be replaced though or used as a secondary verifier.
The song is a funny thing. It could be given to a cool looking group and do well. It could be given to someone older and flop. The song is just part of it.
I am worried about the ability of an AI to generate an infinite number of Dresden Files or Cosmere books on demand, because I already drop everything when a new one comes out and read without sleeping until I am finished.
You don't even have to compare yourself to AI for this mentality though. There are people who choose not to compete in things because they don't believe they'll ever be as good as other humans.
I assume must composers don't go into music thinking they are going to be as great as Beethoven.
I believe there are many studies that show that if you only do something because you think you're good at it, you're likely to drop off. I imagine it's also why you're supposed to praise children for being hard working and not for being smart or talented.
For two of the games, Nakamura had access to Rybka which was about 200 rating points weaker than Stockfish. Stockfish won one and the other was a draw.
For the other two games Nakamura did not have Rybka, but had white and pawn odds. Again, one win for Stockfish (b pawn odds) and one draw (h pawn odds).
In all the games, Stockfish was playing without its opening book and its endgame tablebases. It was running on a 3 GHz 8-core Mac Pro.
The games are here [1].
[1] https://www.chess.com/news/view/stockfish-outlasts-nakamura-...
Neither art nor music are competitive activities. Good poetry is a wonderful thing, no matter the source.
They certainly are! Especially when money is on the line, and the best musicians, actors, and artists are extremely well compensated making their positions extraordinarily competitive.
>Good poetry is a wonderful thing, no matter the source.
Sure, but I think you neglect to consider the defeating feeling it would bring to dedicate your entire life to mastery of a subject only to be completely and utterly, hopelessly outclassed. Almost every such person is already hopelessly outclassed by someone in their field, but those people are so rare that they have tremendous exclusivity surrounding them. Compare that to the scenario of having any 12 year old with a smart phone being able to instantly produce a totally novel and dominate piece of artistic expression developed by an algorithm on their phone. Then recognize that in a world with that level of AI sophistication, there'd be very little of value that a human can even offer other humans at that point. It would be... not great to the psyche, economy, or society.
What is your definition of best in this context? As far as I know, taste in art is very personal... Artists I consider the best are often very far from well compensated.
But, in almost any particular human artistic sub-niche with it's own definition of "best", the same principle will hold, with compensation and skill level being well correlated. It's also typically not even close to linearly correlated either, most of the compensation lies at the far tail of "best".
It's nice to be paid, and it's nice to be recognized, but I think art has its own form of wealth - otherwise, why make art? Why not just seek recognition, or money?
But that doesn't detract from people playing Ukelele.
Making a classical arrangement that evokes a particular expression in the listener is the job of the musician. If an AI system helps you explore the possibilities there, it's more like a studio musician that's able to improvise. You're still the person, the human, the emotional filter, that picks "This sounds right" or "This doesn't" for a particular situation. It's a judgement call. An emotional one.
An AI might be able to fake it, communicate with it, but it will never replace humans choosing the sounds that please them more than others. Humans communicate through music. It wouldn't surprise me that an AI would be able to as well. I don't think it would necessarily write emotionally strong music, not without human training.
Edit: I guess what I'm trying to say is, sure, computers might be able to make music. Ask any guy who messes with modular synthesizers. But they're a tool. The fact an AI can express itself through music is sure as hell not gonna stop me from also expressing myself. It's like arguing "Since AIs will be able to comment on Hacker News, humans won't."
I think this is the key; if you're making music for your own reasons, no AI (or Mozart) would stop you. But if you're trying to make money at it, or desperately want listeners, you may eventually be on the "losing" side.
As far as recent examples go, Lady Gaga and Lorde were major breaks from what was prevalent at the time they started releasing music, and then spawned artists trying to emulate them.
If we oversimplify and compile a list of traits about "the world" as it was in the past that allowed a new genre or artist to flourish, AI could predict that in the future. It isn't like the paradigm shifts just happen in a vacuum.
Granted there are probably millions of little things that lead to this, stuff like the shared experiences of an entire generation coming of age, political climate, trends in other industries, etc. Not that I believe it will ever happen to an accurate enough degree, but theoretically I don't see why it could not be possible to approximate given time and resources.
If you feed an AI a bunch of modern car designs and ask it to design a new car, it will design you something like a modern ford or honda/toyota, but it will never design something like a Cybertruck. Which I believe will be the next paradigm shift in the design of trucks (that has been super stale and stagnant for at least the past 20 years), but this is yet to be seen.
For an example with music that has already happened and became apparent - Kanye West's "808s and Heartbreak" album from late 00's. On release, it had very polarizing reviews, most of which were skewing towards "really weak and weird". Fast forward 10 years, most of hip-hop and pop music is directly influenced by that album, most of top 50 albums use similar patterns and methods used in that album, and critics have made a complete 180. So now 808s is hailed as one of the biggest (if not the biggest) paradigm changes and influences in music of the past decade as a whole, as well as the best album by Kanye, despite at the time being called the worst. Imo an AI trained on music of 00's that came before 808s would have never been able to come up with something like that, but it totally could've come up with another top 100 song using existing paradigms.
That's not something we can really lose without losing something that connects us. People want a story. That has sold since the beginning of time, and it will keep selling. People will keep being moved to music, giving money to the artists that inspire them, and that requires connection. Maybe an AI/human team would make some really incredible stuff, and I'd be willing to pay for it if it makes me feel something. I think the human touch of "selection" will never truly leave, even if only in the listener's mind...
So music generation (similar to poetry) is imo a completely different problem space altogether.
For every individual doing the evaluation, I think it will certainly possible to train an AI to beat humans at getting "good" scores.
I'm not so sure. I often go into threads on HN and realize that every idea I could come up with on the subject has already been expressed better than I could do it, with greater expertise, and cited sources. I don't comment in those threads. If AI bots could populate a thread with every likely human thought and argue it with depth and sophistication in a well reasoned, yet carefully approachable and well-explained way, well then... again I don't think I'd feel like I would be adding much value by participating.
What distinguishes music written by AI from music made from humans? I have a story to tell. If the AI has a story to tell, one that speaks to our human emotions, it might make good music. But the point is to communicate. Even if you take, for example, someone else's words, fit them to a different model in a different field, viewpoint... You might get interesting things. You could make a cover of someone else's song, with your twist. Adding your emotion to the melting pot. AIs might be good at that, just like that, but only through communicating. Just like us. We have no idea whether they'll be better than us at doing it, or merely equivalent. We have no idea what is lossy in our sharing of mental models. Perhaps it is an unsolvable problem, which we will find out in the same way we found out about Gödel's Incompleteness.
It seems to me like we fail to understand how unique we are. We are in a unique position to shape what comes after us, and we are blind to how much we unconsciously select for things. We have an innate mental model of "humanity" we are trying to transmit to machines, and I am not sure we fully grasp it well enough to make sure we are creating something like us. We fail to do it properly to humans, sometimes, who actually do share most of our instincts and habits. Something entirely different from us? Color me skeptical.
This kind of debate only highlights this, to me.
The one physical activity at which humans excel is long-distance running.
When humans used horses for rapid courier service they used relay tactics to take advantage of the horse's higher top speed, one horse might only run for an hour or two, before the rider reached another outpost and swapped a tired horse for a fresh one. In this way the relay could move something hundreds of miles in one calendar day. The Pony Express managed news of a US election from one coast to the other in just over a week.
If you can't use relays human and horse performance seem pretty similar, dozens of miles per day but not hundreds. The horse's top speed is higher, but it is rapidly exhausted, fast gaits like the canter are too exhausting to sustain for hours at a time.
I think we’ll see a lot of things similar to “AI x-ray technician” fields where people are trained to read AI outputs. Doctors will do higher levels decisions.
Do you mean that human intelligence is not general enough to recreate functions of existing physical structure that implements general intelligence?
Maybe someday it will be possible if we can solve the hard problem of consciousness in conjunction with quantum computing, etc.
does not involve any observable consequences. It can be completely ignored, if we don't go for mind uploading.
About that... https://news.ycombinator.com/item?id=17618308
Take a look at this painting: [1]
It is a comment on war, bravery, death, life, fear, sacrifice. It is drenched in the political and social context of the day.
I really don't see AI coming up with anything even remotely like this independently, and view such an achievement to be much harder than simply diagnosing disease or writing an emotionally moving classical composition. It would be comparable to writing some types of poetry or song lyrics, however, which require reference to context that humans understand but machines don't (yet).
[1] - https://upload.wikimedia.org/wikipedia/commons/thumb/f/fd/El...
[2] - https://en.m.wikipedia.org/wiki/The_Third_of_May_1808
There's something axiomatic there, if you assume an identical piece of music that was either written by a human or by a computer, then for many listeners it's by definition more satisfying to know it came from a person, because of what it says about the person.
And for those listeners, if a human "composer" is discovered to have lied about it (saying they wrote it when it was actually a computer), then those listeners would reinterpret their views of the music and consider the "composer" a fraud.
And even a programmer of algorithmic music might have emotional intent, but if the musical output is unknown to the programmer, they did not have the emotional impulse to create that music in particular. While it can be appreciated as its own thing, it's a step removed from the music itself, and qualitatively different than human-composed music.
People are afraid of themselves I believe. It’s not really about “job loss”.
I’m not sure if most people realise AI means pretty specific models built to solve rather specific problems. They think SkyNet.
A Professional Go player is an explorer of truth in a millenarian board, spelunking in a vast universe of possibilities. The purpose of playing is attacked when there is an automated, effortless way to do that exploring faster and better. Why look for new things when a computer can find 100 in a minute?
The professional mindset of a Go player differs vastly from the amateur mindset.
Cars and legs are apples and oranges. We have a car racing category, motorsport. Racing categories have very tightly defined specs to keep driver skill in the game. Stock cars and open wheelers limit how much traction control they can use otherwise it becomes too easy.
This is like cyborg legs being invented and smashing all the records. It would take some of the shine off running for sure.
Just imagine if Garry Kasparov quit after losing to Deep Blue, he would be ridiculed today by the chess community which is still going strong. Instead, he accepted defeat, moved on, and is regarded as one of the greatest chess players ever. I doubt the same will be said of Lee Sedol 20 years down the line if this is how he chooses to end his professional Go career.
Plenty of other people don't care and continue to do what's new to them (and what often turns out to be new to others.)
They are already unbeatable in perfect-information games like Chess and Go. They could crush human motor racers tomorrow.
They will be unbeatable in games with limited information (Dota/Starcraft) within months.
The next frontier is donkey-space games like rps, poker and day-trading (https://universalpaperclips.gamepedia.com/Donkey_Space) where they are already beating pros. It may be another couple of years before they totally dominate here.
https://en.m.wikipedia.org/wiki/Go_(game)
As an aside, there are inevitably more and more things for which even the very best are not sufficiently intelligent _alone_. However, we are social creatures and we collaborate (typically with other intelligent humans) to achieve things we wouldn't have managed otherwise (think just at the space program as an example). So... we only have to adjust a little to accept that we could also collaborate with machines in the future.
Cyborg chess is the future of chess. Period. Chess players use computers to train themselves and explore openings in human-only settings, while programmers / cyborgs play correspondence chess.
Go is not finished as a game. A new tool, MCTS + Neural nets, has been developed to explore the "truth" behind the game at ever increasing rates. Its not about how to beat the tool, the question is how to best utilize the tool to improve self-play and self-learning in the game of Go.
Or alternatively: how to best use the tool to play ever more perfect games of Go.
-----------
Come on, none of us are really "human" anymore since the advent of cell phones. We all use our cyborg-capabilities to search the internet and fact-check ourselves every day. Programmers use stack-overflow to teach themselves programming and remember obscure details (using our cyborg capabilities to tag, search, and sift through information ever faster and faster).
Go is the same thing, except we only learned how to become cyborgs two years ago.
----------
Give the world-champions each a copy of AlphaGo on equal computers. Have them play Go against each other WITH computer assistance. I guarantee you that a more beautiful and perfect game will result. Let us welcome the age of Cyborg Go as we step into the future, we shouldn't be scared of it. We've become cyborgs in many other tasks, and Go is no different.
Over in the "Cyborg Chess" community, people have already analyzed LeelaZero vs Stockfish. It turns out that Stockfish is far better at tactics (especially the endgame), while LeelaZero is better at opening positions (aka: Positional play).
There are numerous theories about the proper combination: perhaps using an Opening Book database for the first ~10 moves or so, using LeelaZero for moves ~10 through 30, and then using Stockfish to check for tactics (LeelaZero misses a lot of tactical options in the midgame, so double-check to make sure that LeelaZero doesn't lose a queen or something), as well as finish up the endgame.
--------
Choosing the correct combination of tools, studying these tools and coming up with a more beautiful chess game. That's cyborg chess in a nutshell. LeelaZero and Stockfish are both superhuman in terms of play, but the cyborg can choose to use both tools to play superior compared to just a singular tool.
Anyone purely using "Stockfish only" gets beaten by opening book analysis. The dude over there with 1TB of opening book databases consisting of every losing position that Stockfish tends to play will completely own you. Same thing with LeelaZero, the opening book guy will own any LeelaZero-only player. These engines have weaknesses that can be undermined with good analysis, big-data, and a bit of custom code.
That's the funny thing about these computers: they tend to play the same. So you can build opening book databases to exploit their patterns. You require a cyborg / human to play at the "level above that" to guide Stockfish / LeelaZero away from those traps.
And then program the computer to use that opening book directly. Now what is your cyborg player going to do? After you patch all these easy rules, you will have to discover harder rules, and the computer can discover them faster than you can.
Build and configure the machine better. Quick, tell me, what's better at playing Chess:
* Or is Xeon Platinum 8180 with maximum 8-way SMT memory sharing with a big-ol 1TB shared transposition table the fastest computer?
* Or will it be cheaper to rent AWS-instances with their V100 GPU in the cloud? Or is the latency for the remote-access bad?
--------
The cyborg player has to still build and program the machine to compete. It is all part of the competition. Can you run a transposition table shared between chess engines over RDMA 10Gbit SFP+ Fiber? Or is that too slow?
-----
This isn't hypothetical at all. The winner of the World Computer Chess Championship 2019 was 8 x Intel Xeon Platinum 8168 running Komodo vs 24 x Amazon AWS Intel Xeon E5 running Shredder.
Configuring and building the computer is still an incredibly difficult part of cyborg chess.
It's a stretch to categorize that as a cyborg player. You might as well categorize any chess program as a cyborg player because a human had to program and train it.
The key difference between cyborg/centaur/advanced chess and plain old computer chess is whether there is a human in the loop making the move decisions. My argument is that having a human in the loop will result in a worse player.
When you build an opening book, you need to pick-and-choose which engines will self play. Will you build an opening book vs Stockfish? Komodo? LeelaZero?
If so, how will you generate these LeelaZero games? You'll have to build a computer (or rent one from AWS) to play these LeelaZero games. What are the time-controls of matches?
Self-play at 40-minutes + 15-second increment means that you'll only create a game every hour or so. Spend 30-days building databases at 40-minute + 15-second increment games, and you'll only reach ~720 games of analysis per month of preparation.
Self-play at 1-minutes + 0-second increment results in a game win/loss every 2-minutes (maximum), giving you 21600 generated opening book positions per month of analysis. But these 21600 games are of lower-quality.
----------
Spend 1-month building an anti-Stockfish database, an anti-LeelaZero database, an anti-Komodo database... and you're now 3-months into preparation and there's still Shredder, Johnny, and all other programs that may arrive at the contest.
Its not exactly as easy as you think it is. There's no algorithm that automatically builds the best opening book (or "counter" opening book) against your opponents. Its a human choice for what the computer will spend the next-3 months self-analyzing and self-playing.
Lets think about how to build anti-LeelaZero seriously. Which of these networks do you download as the LeelaZero representative? https://lczero.org/networks/
You don't have the time to build an anti-opening database for all of those networks.
From my perspective, cyborg chess is about playing the best game of chess in all time... only stopping once we have discovered "the perfect game" (aka: proving that White-wins, Black-wins, or a Draw is always possible with perfect play)
The computers currently playing Chess, and Go, are incredibly powerful. But they are far from perfect. People have constantly found weaknesses in chess playing programs over the past 20 years, and as a result, have improved chess programs significantly.
Go has only had 2-years in its "cyborg" state, where we can finally use computers as a methodology of exploring the game state.
Most of the resistance you're getting in this thread seems to be due to the fact that you think people who are interested in the game should also be interested in the other thing, and it just seems that a lot of them aren't (including Lee Sedol).
Playing Go challenges your mind with vast complexity and immediate feedback of winning or losing every game. It's a deeply engaging hobby for people who are susceptible to that kind of thing, and it used to be a meaningful career, with competitions, schools, and professional teachers. All of that changes now that software is vastly better at it than any human can ever be. The rush of competing by the strength of the moves you understand and make, by the unaided strength of your own mind, cannot be compared to picking between different engines to make moves for you based on some heuristics about which engine is better at openings. The era of human Go is simply over, for better or worse.
I really like the way you put it. In isolation, humans today have essentially the same natural mental capacity we had 2000 years ago, but we've become part of a much bigger computer. A human brain 2000 years ago was a very isolated entity. Even in centers of learning in the classical era, the amount of knowledge one could tap into was vanishingly small compared to our capabilities today. Everything and everyone was isolated in both space and time.
A brain today is not just an entity on it's own, it is intricately wired into the common consciousness. Crucially, we have a vast database of knowledge - easily searchable and distilled for maximum learning rate - all available at our fingertips. A modern brain is a neural network linked to billions of other neural networks, and all of them are linked to a shared memory that they can use at will.
Assuming that the main limiting factor is the bandwidth between a human brain and computers (and thus by extension between individual humans), then a direct interface could ostensibly bring about a new revolution (for better or worse).
I feel computers (and the networks between them) are a natural evolution of our hive mind. When we first started taking spoken language and committing it to writing, we created a hive mind that extended beyond a generation. Prior to that point, everything was passed on through observation of other humans (observing behavior, observing speech, etc.). This meant everything had to pass through, and be mediated by, a human brain. Once we committed that to writing, we had a direct line back to the original brain that created the content unencumbered by the mushy grey bits that have consumed and regurgitated it since.
I feel we are still iterating on that. Cataloging our collective minds and building ever low latency systems to navigate those catalogues. Computers are just an extension that improved our ability to catalogue and retrieve the contents of each other's brains in an incredibly low latency way.
We are part of a hive mind. Our industry is actively building the load bearing infrastructure that supports our hive mind.
This hive mind over-simplification is ridiculous and to me: disgusting. As if we were one
We may not be a hive mind, but we may have one.
Hive mind as an emergent property still implies a common thing for which one lives. Or a queen. A peer that is one's reason to live. The closest thing I can see to a hive mind is the army.
We are all different and it is not because a person one talks with does not express a different opinion does it mean he thinks the same.
Would one say that lions have a hive mind?
I don't understand how that has any bearing on whether human society can form hive minds or hive mind like entities.
> Hive mind as an emergent property still implies a common thing for which one lives. Or a queen. A peer that is one's reason to live.
No that is absolutely not what a hive mind is. A hive mind is a collective consciousness. That's it. It says nothing about the objectives or agency of it's parts.
> We are all different and it is not because a person one talks with does not express a different opinion does it mean he thinks the same.
You seem to be laboring under the apprehension that a hive mind supplants the consciousness of it's parts. That may be the case for certain hive minds, but it is not a necessary feature. A hive mind can just be an emergent property of individual minds that are strongly connected but still retain agency.
Thanks for precising hive mind, I understand it better now.
Hive mind can be seen as uncritical conformity or collective intelligence. Because it can be understood as uncritical conformity I will use another word/concept. Like tribal mind. Because it suggests tribes (plural) that one belongs to or not, status, etc
You're a bee, dawg.
There’s your hive
If we define hive mind as a collective consciousness, then for sure one can argue that such a thing seems to be arising in human society. As a matter of fact in modern parlance it is fine to call a strongly unifying force, set of norms, school of thought or strong social bonds a hive mind. The hive mind can still be constructed of individuals capable of agency and independent action. It does not replace the minds of it's constituents, rather it can be thought of as an emergent property of many individual agents forming a deeply connected collective.
It’s not about having a queen or mindlessly following. Being a part of a hive mind doesn’t make you a lemming.
Example to demonstrate: assume you know calculus. How much of calculus did you discover yourself? How much of the corpus of math that led up to calculus did you discover yourself? How much of that corpus is your individual contribution vs. how much of it represents the brain power of hundreds of thousands (if not millions) of other humans throughout history exploring that problem domain?
For any given problem domain, humans have documented it in a shared corpus that you can “download” into your brain. You are an individual. You still get to choose what to download and how to interpret it. But you are still sharing in a commons when you do this. That commons is what I’m referring to as a hive mind, it’s a shared consciousness where our collective brain power is building a corpus that no single brain could. It extends far beyond ourselves.
https://medium.learningbyshipping.com/bicycle-121262546097
From my understanding, indigenous Australians and Polynesians had rich oral histories and cultures. Also oral culture may be more adaptable than a musty written document that never changes and must be followed. The human experience of living in an oral culture is naturally being lost, but it doesn't mean it was wrong or bad. I find it fascinating to imagine what life was like for them, how they did things differently.
"doesn't mean it was wrong or bad"
It is a dead-end, though.
While these cultures are interesting from anthropological point of view, I highly doubt you would give up all the benefits of the culture you live in today and join an Polynesean or Australian tribe.
Just consider the fact that everyone thinks they are above average in compassion or intelligent and stuff that like that. Even if you scream in their face that "HUMAN OVERESTIMATE THEIR PLACE IN THE AVERAGE!!" they will still claim that they are in the 51% percentile.
If anything, technologies and networking have made human a lot more stupider in that its harder to climb against all the misinformation.
But generally, humans have accepted computer supremacy in chess pretty well. Everyone realizes that a better chess-playing entity exists out there (in everyone's pocket, in fact). That doesn't make the competition between human players less exciting. It's still a match to prove that you're better than the person sitting across the board from you, and it's still incredibly impressive to see how deeply and accurately the top players can calculate, or how much knowledge they have of the game.
"Could a human beat a computer at chess?"
Like, now. 20 years after Kasparov. Could someone do it? The question feels daring somehow.
If you're asking how a human can contrive a scenario to beat Stockfish/Leela, that seems like just a conversation that devolves into what is and isn't too contrived.
https://en.wikipedia.org/wiki/The_Feeling_of_Power
Big showcase computer-human matches ended in the mid-2000s. That's about the point when people realized it was hopeless to try to beat computers. The best you could do back then was draw, by playing extremely defensive "anti-computer" chess (positions where the pawns clog up the board, making short-term tactics useless, and making long-term strategy more important). But with modern engines (particularly AlphaZero, which understands positional chess much better than previous engines), even anti-computer chess doesn't work. Now, the top players actually rave about the games played by AlphaZero. Magnus Carlsen has said he wants to play like AlphaZero.
For a more quantitative comparison, these are the Elo scores of the top chess engines: https://www.computerchess.org.uk/ccrl/404/
A difference of 400 Elo points means that the better player should score 0.9 (where 1=win, 0.5=draw, 0=loss). The top computers are about 3600 Elo nowadays, compared to 2870 for the best human (Magnus). It's difficult to directly compare human and computer Elos, but this comparison is based on computer-human games played in the late 1990s to early 2000s. Still, computers are many hundreds of Elo points above the very best human players.
Exercise to the reader: prove that at most one color can have a winning strategy.
Since the problem is open, it could be that one side has a winning strategy. It's even theoretically possible there's a winning strategy simple enough for a human to follow. In which case, yes, a human could beat computers with 100% dominance--as long as the human is allowed to choose which color to play.
https://en.wikipedia.org/wiki/Kasparov_versus_the_World
If you start from a winning position, even something as stupid as a random-mover will eventually play a perfect sequence of moves and win, given enough attempts.
However humans are not very good at being random, so this may not apply :)
` It was a massively parallel, RS/6000 SP Thin P2SC-based system with 30 nodes, with each node containing a 120 MHz P2SC microprocessor, enhanced with 480 special purpose VLSI chess chips.`
I misread at first thinking it was grossly 30GHz summed but it's only 3.6GHz (I know, gross maths). So yeah the comparison is apt.. a recent SoC can outperform this old behemoth.
1. https://www.computerchess.org.uk/ccrl/404/cgi/compare_engine...
So there was a window between the 90's and today where a human player could have discarded the traditional measuring sticks of "trade-value" and beaten DeepBlue.
Maybe that's not true today or maybe there's a further theory out there?
Google played Alpha Zero vs Stockfish under their own conditions. I think this was a mistake to stay in the labs by themselves.
I think Google would have benefited from participating in WCCC 2019 for example, where Johnny (1200x core cluster) won the day.
----------
Its a completely different field when you're actually competing against someone else. Google did some impressive things in the lab for sure, but they have NOT actually stepped into the competitive ring yet.
I'm sure AlphaZero is good, and probably would make a good showing at one of these contests. But you're putting the cart before the horse here.
EDIT: Maybe MCTS + Neural Nets truly is the superior way of preparing board game knowledge? If so, the next step is building out the opening-database and to start looking for holes where AlphaZero loses. It wasn't a complete blowout: AlphaZero lost some games to Stockfish. Why did AlphaZero lose in those games? Is there a set of opening moves that will lead AlphaZero down that losing path in a true competitive setting?
---------
EDIT: Case in point: Google did NOT incorporate randomness into AlphaZero's algorithm. It always chose the best move it believed in, this leads it prone to opening database attacks. See how the competitive mindset already messed AlphaZero's careful laboratory preparation up already?
Sure, AlphaZero 2.0 might have programmed randomness added to it, if this were a true competitive environment. But that's how cyborg chess evolves: I point out a weakness in the program, exploit it in a game, and then Google goes back to their labs to make something better next year.
Computers cost about $0.01 / GFlop (10^12 operaions); $1e47 for 10^50 operations.
The world economy is $1e14/yr.
That's 1e33 years to check every chess position, using all of Earth resources.
You'd need to find symmetries to collapse the search spaceby a factor of 10^33
As of now, we have databases for perfect play when 7 pieces remain on the board (including the two kings). The 8-piece tablebase is computationally possible, but I don't believe a comprehensive release has come out yet. Even the current 7-piece tables are incomplete because situations like lone king vs. six opposing pieces haven't been explicitly calculated due to their obviousness.
I have to disagree: even after the discoveries by Alpha Zero, the best chess players are not able to beat Stockfish. Stockfish is just too good at Brute forcing long tactical advantages. Playing really well positionally doesn't matter a lot if your opponent can look 30 moves into the future.
I'm not sure how big the window was, if it existed, but it seems like it might have. Kasparov and Deep Blue themselves were fairly equally matched, it doesn't seem impossible that Kasparov + AI-aided theories would have a window of advantage over the Deep Blue -> Stockfish evolution.
Though I do still agree in thinking there is likely a deeper theory yet uncovered.
Two problems with this plan: (1) the best players at human vs. human Go are likely not as good at playing on a human/computer team, and (2) it may be that the world champion human/computer team is the one that defers 100% to the computer, which seems uninteresting.
AlphaZero doesn't have any randomness built into it, does it? Which means I can build preparation against the lines of play AlphaZero would want to go down.
I think you're underestimating the human, and also overestimating Google's laboratory experiment here. No one has really played "Cyborg Go" yet.
The first casualty of "Cyborg Go" will be AlphaGo in its current form: its clearly inadequate for AlphaGo to play deterministic. Random play MUST be incorporated, lest opponent's preparation sends it down the wrong path.
If I know that AlphaGo plays slightly worse on 3-4 opening (or 4-4opening, or maybe even the 10-10-opening), then that's what I'll do. Give me a copy of AlphaGo and I'll be able to find a weakness somewhere.
I'm with the other guy; I don't see how a person could make the AI play any better.
That's not what MCTS means.
MCTS refers to the bandit problem which formulated the search parameters. MCTS always chooses the "most interesting" path to explore. (Where "most interesting" is the path that balances explore-and-exploit hyper-parameters).
AlphaZero improved upon MCTS by deferring to the neural net as the hyper-parameter. But AlphaZero, for a given network on a given board-state, will ALWAYS choose the same position as "most interesting".
Turning AlphaZero from its current deterministic form into a random form would be an easy fix. But its just one example of how AlphaZero really isn't designed for competitive use yet (despite playing the best game of Go of all time). Instead of picking the top #1 move, maybe you randomly pick from the top 3 moves... or some other scheme.
Play AlphaGo against itself. Go rarely has draws (requires Triple-Ko, a very, very rare position).
Almost every game you play with AlphaZero vs AlphaZero will result in a winner-and-loser. You will quickly be able to characterize the positions that AlphaZero loses in.
For most of us, that means we'll all have the best verison of LeelaZero to grab from Github and use in our own personal studies. Which should still be super-human in terms of play.
And don't forget that AlphaZero is gonna be getting better over that year too; you're trying to beat a moving target.
The strongest early moves create the most potential for winning (maximizing potential winning paths, sort of); they do not push the game towards one best end state. They do not have counters. I saw elsewhere you have some understanding of the game (15kyu) so you should be able to demonstrate this to yourself by playing some of AlphaGo’s openings on a board and trying to write deterministic counters to them. You will not be able to push the AI into a situation where it has too few options to avoid loss. You will also find you need to create a book much larger than a few moves to meaningfully predict play and so will exceed the number of states that can be stored (referencing your 16TB comment elsewhere.)
Please actually try this as I think it is a key to improving your skill in addition to understanding the challenges in automating play.
The AI selects a move. What state is the board in now? It doesn't know, because the opponent also selected a move.
MCTS models this with a probability distribution of the states, and samples from this distribution repeatedly to build an estimate of the effectiveness of each move it could make.
But what's the probability of each move made by the opponent? And after the simulation has looked as many moves ahead as it can in the time constraints, how good a position is it in?
These are the same question, really - what's the chance of winning from this board state. In Chess you can use a heuristic algorithm to figure it out. In Go, you can't. But you can use a neural network to learn an approximation that improves as it sees more games complete.
AlphaGo does this. MCTS is a random sampling technique, and the neural net informs its probability distributions, but doesn't make it deterministic.
If given White in chess, LeelaZero plays 1. e4. Each time, every time. Guess what that means?
If you're building an opening chess database vs LeelaZero (or at least, this version of LeelaZero: https://lichess.org/@/LeelaZero-UK), you only have to worry about 1. e4 openings.
Nothing.
Sure and any of us could jump on a motorcycle and fly past Usain Bolt in the 100m, but that kind of misses the whole point of the competition.
> Give the world-champions each a copy of AlphaGo on equal computers. Have them play Go against each other WITH computer assistance.
Won't the winner be the one who just takes AlphaGo's recommended move every time without changing anything?
That's only true if AlphaGo never makes a mistake or if AlphaGo will 100% always make the better or equal decision than a human + computer at any given state of the board. I know the former certainly isn't true and I assume the latter isn't true either, but I don't know enough about Go to say for sure.
Plus, you only have to outsmart the car AI once to 'win' - e.g. just override one 'drive into the highway barrier' or 'run over that pedestrian' AI mistake.
You can use the relative scores to decide when to overrule the AI. Eg, if move A has a 50.1% win chance with 2k branches explored, and B has a 50.2% chance with 1.9k branches explored, I would go with the opinion of an expert human, as AG thinks the moves are essentially equal.
Even if AlphaGo makes mistakes, and somewhere on the board a better move can be found, you would also need the human to reliably spot it.
Eg: AlphaGo makes a move. Let's say that at least 20% of AlphaGo moves can be bettered. Is this one of them? How can you tell? Most of the time, you'll mistakenly think a move can be improved and end up playing a worse one.
But, let's make AlphaGo even more fallible. Let's say that at least 50% of AlphaGo moves can be bettered. Again, is this one of those? How can you tell? And more to the point, on the times you are wrong, are you more wrong than AlphaGo is with its mistakes? Because even if you imagine you can spot a better move than AlphaGo and pick the actual better move 50% of the time, you also need your mistaken moves to be better than AlphaGo's mistaken moves or you'll still lose.
Worst of all, you can rule out a really good ability to spot AlphaGo's mistakes already. Let's say 99% of AlphaGo's moves have a better option. If you could spot them all, you'd be beating AlphaGo regularly on your own. As no human can now beat AlphaGo, this plainly isn't true.
So it's likely that:
a) No human can reliably pick a better move than AlphaGo and/or b) No human can reliably spot a move from AlphaGo can be improved, and/or c) Human mistakes are worse than AlphaGo mistakes, so even if you could fight it up to parity you'd still lose.
No. Consider this scheme.
I take my copy of AlphaGo and for a full year, I'll build a database of all opening positions AlphaGo is willing to explore. I'll rank these opening positions from "best" (Black-wins with most consistency) to "worst" (White-wins with most consistency).
I'll put all of this information into a 16TB database on a singular, $400 16TB Hard drive, and load it into my computer during the contest. https://www.newegg.com/p/1Z4-002P-015K6
If you dare to pick AlphaGo's best move, you will lose. Because I already know which moves AlphaGo will take, and I already checked to find all of the positions AlphaGo wants to play (but loses anyway).
The only way you can equalize the field is if you yourself ALSO build a 16TB database to consult and override AlphaGo's instincts during the game. If you see AlphaGo wanting to play "losing position #6234115", you'll tell AlphaGo to "search harder" and find another move instead.
Good luck.
Human + computer might still beat computer in Go - this was true for a few years in chess, and even now to some extent in correspondence chess - but what you describe isn't really that.
Go does not have openings. It has reasonable choices to make, with an insane branching factor, with few moves making much of a difference by themselves. Therefore your database will only extend a few moves, and all of the positions that it winds up with will still be very close to even. So your database confers very little advantage.
If you have alpha-go play itself a thousand times, it is unlikely that by move 10 you will wind up with the same board position twice.
This is exactly the problem that made Go so hard for computers in the first place. Alpha-beta search is useless within practical limitations of computer hardware.
That said in a different game, such as chess, your strategy would work very well indeed. (Which is why all decent computer programs have an opening book.)
There are 381 opening moves in Go, but really only 96 because of symmetry. 96 (opening moves) x 380 responses x 379 x 378 x 377 == ~2 Trillion positions after 5 ply.
These 2-trillion positions will easily fit in a 16TB hard drive for $400. That's 8-bytes per position, so you probably can get there with more symmetries and some compression applied.
----------
You're thinking too much like a human. There's no Go-openings in the age of Human-Go. But in the age of Cyborg-Go where 16TB hard drives are allowed, we can begin to exhaustively build openings.
We even have a super-human AI that can automatically, and algorithmically, explore this opening book. We can build AWS-instances with V100 Tensor cores to use neural-nets to explore all of these positions now.
-------------
> If you have alpha-go play itself a thousand times, it is unlikely that by move 10 you will wind up with the same board position twice.
Alpha-Go doesn't seem to implement much randomness at all into the moves it plays. The source of randomness is in time-controls (AlphaGo may choose MoveX before 30 seconds of analysis, or MoveY after 30 seconds of analysis), but this is a fairly constrained number of moves.
Play alphaGo by itself a thousand times, at precisely the same time MCTS-controls (say: 1-million nodes visited in the MCTS tree), and it will probably play the same game 1000 times in a row.
This makes AlphaGo extremely prone to opening database "attacks". Which is why I am using opening books as an easy example for how to beat a particular AlphaGo network. At least, until AlphaGo updates its algorithm for more random play.
If the goal is "Beat AlphaGo" in a game, then the opening book construction is far, far simpler. Even with random elements (ex: AlphaGo picks randomly from the top 10 best positions it generates), that is far more constrained than a full 381 x 380 x 379 x ... style opening book.
No, you're thinking too much like someone who understands chess but not Go.
Suppose that we built an opening book with a trillion reasonable courses of action on it. Each one analyzed well. As you have discovered, you will only go a few ply into the game. And all of the positions that you will be directed towards will have only a small edge.
Instead put a tiny fraction of the computing power necessary to build this book into self-training. You will get a better internal model and therefore a significantly stronger computer player. (That is how alpha-go was built in the first place.) This option will produce much better results for far less effort, and again leaves no role for a human to do anything useful.
The fact that a memorized opening book is useless has nothing to do with human vs computer vs cyborg. It has to do with the characteristics of the game. In chess, it is useful to memorize openings and both computers and humans do it. In Go it is a waste of energy, and it is a waste for both computers and humans.
Do you believe that AlphaZero could continue to improve dramatically with another 6-months of training? Or if it can improve at another 6-months after that? At some point, the network will reach a local maximum, and it will be unable to improve beyond that.
Characterizing AlphaZero's moves through big-data analysis is innately going to become useful as self-training plateaus. Even Google wasn't able to get more than a few months of training in before the plateau.
At which point, it will be more reasonable to characterize the weaknesses of the network and build an opening book. Avoid the positions that the network was unable to learn about. Etc. etc.
Opening books, at a minimum, would grossly improve AlphaZero's play at competitive levels. Anyone with an opening book of AlphaZero's mistakes will be able to push AlphaZero into a mistaken position.
And then AlphaZero was better than AlphaGo after around a day of self-training.
Furthermore you are arguing for an opening book without considering how small an advantage an opening book would be. As I have said repeatedly, an opening book takes a tremendous amount of work to generate, will only go a few ply in, and the positions it directs you to will only be a tiny bit better.
Therefore for the foreseeable future, more training and better algorithms will produce better results than trying to create an opening book. Theoretically this could change. But that day will not be today or this year. I would be astonished if it happened during this decade. I would be surprised if it happened in a lifetime.
Your proposed approach is an excellent one for many games. But not for Go.
The plateau is real. I'm not sure if continuous self-play will lead to continuous progress for all of eternity. The system is clearly slowing down in self-learning.
AlphaZero's plateau is also well documented: https://i.imgur.com/NMNp6Kq.png
--------
I'm not trying to cast doubt upon reinforcement learning / MCTS / Neural Nets in the game of Go. It is clearly the best methodology we got today.
But anyone who has any experience with neural nets knows about the local-maxima problem. ALL neural nets reach a local maxima eventually. Once this point is reached, you have to rely upon other methodologies to improve playing strength.
Assuming Elo-growth for all time using a singular methodology is naive. We will go very, very far with Deep Learning, but are you satisfied with that limit? Other open questions remain: Go is very far away from being a solved game, even with a magical machine that plays 2000+ Elo stronger than humans.
This only works if AlphaZero never retrains on previous games.
If the OP spent a month learning Go, I am sure that it would make sense to him as well. Work through a series like https://senseis.xmp.net/?LearnToPlayGoSeries while playing Go regularly against a variety of opponents. Before book 3 it should be obvious.
I'm not an expert by any means, but I'm well past "spend one month" learning the game.
If you like, I could have a look at several your lost games and maybe suggest how to improve. Just a 3k at kgs, but still.
This means if you enter a hypothetical "Cyborg" Go competition (computer-assistance allowed), the majority of newbies will simply be playing LeelaZero #1 plays over and over again.
You don't need to build an exhaustive opening book covering all possible moves. You only need to pick say, the top 5 moves LeelaZero ever considers. If you spend ~16-bytes per position and store 1-trillion positions on a 16TB Hard Drive, you'll be able to exhaustively map the top5 moves LeelaZero considers into 17-ply.
From there, you pick the positions that LeelaZero thinks its winning in, but in actuality is losing. You have a map towards 1-trillion positions to choose from, and your opponent (if they only pick from the top5 best moves LeelaZero ever outputs) will walk into your trap.
---------
As long as your opponent picks the top 5-moves from LeelaZero, you'll have the map towards victory. I think you're severely underestimating the abilities of a simple, dumb, opening database.
In my opinion, it's really strange to describe this as an expert "beating" AlphaGo, when really it's just a technique for making AlphaGo stronger than it is without a huge pre-calculated cache of moves.
10 moves into a chess game is actually pretty far. The directions the game can go are massively narrowed 10 moves in.
Then you wring out a tiny advantage-- a fraction of a stone. It's a small benefit compared to building other parts of understanding of the game.
The basis of reinforcement learning algorithms is the exploratory nature of learning due to the initial application of largely random moves.
Only after some time the agent is given confidence into his learned ability and grafually moved into a more deterministic behaviour mode.
This is the exact opposite of your statement. Star Craft players have noted that the fleet of different AlphaStar instances training in ensemble observed very different behaviour due to this property of RL.
At 5 ply, the complexity of the game hasn't started in any meaningful way. In a typical game, that's 4 corner moves, and then one of a: an approach to a corner (kakari), b: an enclosure (shimari), c: a wedge (waruichi) or d: creating a side framework (such as the Chinese fuseki or Sanrensei). There are some odd opening such as tengen, or corner-corner-corner-kakari which typically turns into a sente fight, but 99% of games will fall into the aforementioned pattern. The database you describe is about as useful as a database of amateur games, since most games, including AlphaGo's games, follow just a few basic openings that early, and even amateurs can play these first few moves "correctly".
Even if you get out to 10 ply you're still only getting partway into a single joseki sequence, often leaving three whole corners of the board which haven't even been approached, so this database still isn't very useful.
Incidentally, your numbers are also wrong. Symmetry reduces the first move to 55 possibilities, not 96, and there are 361 points on the board, not 381.
All you have done is establish a computational arms to see whose computing rig wins when you press the 'pick best move' button. You're not playing Go any more, you're playing Database Administrator.
The impression I got from Lee Se-dol in the Netflix documentary was that he had a lot of his identity tied up in this (particularly being the best Go player in the world). Not a huge surprise given the time and effort required to do it, but there's probably a healthier mental framing (I think Kasparov has a healthier one).
If people give up on learning or doing something every time someone or something else can do it better then there's going to be a lot of disappointment as things continue to move forward.
Maybe he's lost motivation, but this retirement is not a surprise.
The point is we don't really know what limits exist beyond our small pinhole of perception. During industrialization I'm sure people said the same thing: "What's the point of hand-made clothing if machines can do it?". Today both hand-made and factory-made clothing have their place in the market.
Throughout history, tools have disrupted humanity and we've adjusted and expanded our horizons. To me this feels like another wave, and If I could bet and could collect winnings centuries in the future, I'd say this isn't even the last one.
Some track and field records set in the '80s still stand, e.g., Marita Koch's 400 m, Jürgen Schult's discus throw, Galina Chistyakova's long jump.
An accurate metaphor would be a person on a motorcycle competing against another person on a motorcycle.
Based on what we've seen in a couple decades of computer-aided chess, no. A good chess player using a top-rated engine to help them can pretty consistently beat the engine by itself.
There are tournaments and even a world championship in computer-aided (correspondence) chess and you don't come close to winning by just taking the program's recommended move every time.
Come on, none of us are really "human" anymore since the advent of writing.
This is why AI is more significant to Go than horses and cars are to running.
It's easy for humans to build machines that enable them to travel faster than they could run, but it has zero impact on running as a sport. The existence of cars doesn't change how runners run. But the existence of AI does change how Go players play.
Now that AI is more powerful than human players in chess, it is no longer possible to reach the top of human performance in chess without AI. That's why this is so significant. It's not that a human can't beat the AI. It's that a human can't beat another human without learning from the AI. You simply can't play chess at the highest levels anymore without playing cyborg chess. This is different from the impact of other machines on other sports.
But I disagree with you on the comparison to how we use the internet to enhance our other abilities, such as programming. Computers help us be better programmers, but they can't replace us. AI completely replaces humans as a Go player. The human is no longer needed at all.
I can see this causing a significantly different psychological response. In effect, a human trying to become the best chess player or best go player they can become is trying to more and more closely emulate a computer that they will never actually reach parity with. Another commenter noted that "A Professional Go player is an explorer of truth in a millenarian board", but this is less and less true as AI improves. All knowledge about the best way to play Go was previously gained by humans. In 20 years, humans will probably be meaningless on driving knowledge of perfect Go playing.
Chess players today rely on playing moves which their opponents don't expect and haven't trained on as much as they rely on outplaying their opponents. If there is a perfect chess game or a perfect go game, humans are not going to help find it anymore.
So AI doesn't make it that humans can't meaningfully compete in games such as chess and go, but it does change how they compete. We no longer drive the pursuit of perfect play as "explorers of truth". We no longer learn primarily from each other, we learn by imitating the more powerful, yet unmatchable, AIs. And we no longer compete primarily on sheer expertise in the game. We have to, in some sense, include trick plays that work by exploiting the humanity of our opponents rather than playing for a platonic ideal of the game and seeing who is better from that perspective.
AI has fundamentally and irreversibly altered these games, and I think it's appropriate in some sense to mourn the loss, regardless of our feelings on the new reality we live in.
We've been cyborgs much longer than that! Writing is a very old auxiliary memory mechanism.
See also: Truth of Fact, Truth of Feeling by Ted Chiang - http://archive.is/oYo0l
When computers became able to beat humans, the tradeoff was that computers were better tactically and humans better positionally. Therefore a human computer pair with the ultimate decision up to the human is able to be a better combination of positional and tactical than either alone. This is why Cyborg chess works so well.
But in Go, AlphaGo is simply better at everything. It is better both positionally and tactically. It can't always explain why the move is right, but adding human intuition on top of the computer only detracts from the quality of play.
Probably for a few years. Very soon the input from the human agents will be indistinguishable from the noise in the AlphaGo algorithms.
I agree, and this could also be rephrased as we're not really human any more since the advent of writing or (especially) printing.
Mankind should focus on skills that computers can't yet compete at. Hopefully the point in time will be far off where computers will beat humans at everything, but likewise, once they do, there is no reason for the human race to continue to exist.
What is the reason for the human race to exist today? Please think carefully about your answer.
It would be nice if we got a decade of centaur Go but I'm not sure we will. While Deep Blue had fairly crude heuristics for guessing which lanes of inquiry were promising AlphaGo's strength is its intuition-like neural networks that it combines with a rigorous tree search. It's worth investigating whether human/computer teams are superior at Go right now but I wouldn't count on the answer being "yes." And I especially I wouldn't count on a full 10 years of combined dominance.
As we have seen in the latest championship blitz chess is the future for chess.
Computers compete in their own championship.
I think runtime energy is the most interesting.
Because of the complexity of the game, is it not possible that this iteration of DeepMind could be defeated through an adversarial approach?
> He actually quit the KBA in May 2016 and is now suing the association for the return of his membership fee.
> "... [I] have something else to do," he said, asserting his only dream for now is to rest and spend time with his family.
Edit: Meant to include the part where he's planning a high profile set of games against another ai.
Lee's issues with the KBA are not a secret and he has discussed possibly retiring for some time now. He has given multiple reasons as to why he was considering retiring and while ai might be one of them, saying that its _the_ reason feels very clickbaity.
I have to say, I'm just continually puzzled by this. In this thread, yours is the only comment where it appears that the commenter read beyond the first few paragraphs. There are numerous comments by people speaking authoritatively on go and AlphaGo, who have clearly not studied either significantly.
If people like the comments better than the articles, it's not so surprising that they like to spend time reading and writing comments instead of reading the articles.
Chess, go, driving, flying, math, written language, music... when does it stop?
The standard answer is that we will automate all the dull parts of living and allow everybody to work in some sort of higher order capacity. That sounds great and all, but what happens when our systems learn how to make music as well as we can ourselves?
At some point we will simply become consumers of our machines and while that's a comfortable existence, certainly we are losing something as a species with all of this automation.
Maybe I'm old.
If AI's make music as well, why would I care? I still listen to the music composed by Bach and other ancient composers, more than 300 years after they lived, and their music is still being performed today.
I think the availability of AI's will only make a difference where their effects on humanity are beneficial (that is, tasks we don't like doing, for the most part).
We invented farming and lost our communal hunter gathering cultures. We invented mass production and supply-chains for our food and lost our farming culture (for the most part). This isn't necessarily good or bad, but we are definitely losing something every time we invent something to make our lives easier. At some point there will be a tipping point of diminishing returns.
Personally I believe "the singularity" would be the worst possible thing to ever happen to humans. Sure we can visit distant parts of the universe and live forever but only as passengers to some AI. Would you rather be a dog in today's world or a human 1000 years ago?
Progress is unrelenting though and I certainly enjoy not having to wash my clothes by hand.
Who would be more impressive, someone that can play a song on a guitar or or someone that can play a song over speakers? The former implies dedication and practice. The act of creation is always going to have a place in society, and people who think otherwise don't understand culture. The desire to make unique and interesting things or impress people will always leave room for creation or spending time to learn something. AI will just be a tool.
As for games like chess and go: Most people are interested in other people and will find much more pleasure in playing the game with others.
Can we please stop flogging this tired “man vs machine” narrative? Not only is it totally unnecessary, it also takes away enjoyment and appreciation for the flourishing in games like chess, go and poker that can occur when man and machine work together.
Chess engines have been defeating humans for 20+ years (and are overwhelmingly stronger for a long time), but that hasn't diminished the interest in competitive chess, because the human element of competition and struggle and deep fundamental appreciation for the game is what makes it worthwhile pursuing.
AlphaGo can play go but it cannot appreciate the beauty of the game (at least as of yet, and I don't think it would make the game worse if it could), and so I don't think there's a meaningful conflict between humans and machines.
If someone invented some sort of superhuman math proving engine tomorrow it would not diminish the beauty of maths and I don't think anyone would quit the field. Just like in chess it ought to motivate people to understand their field better.
On the contrary, appreciating the game is the core of what AlphaGo does. In order to search the tree of moves it learns how to play (expand search) and how to evaluate (cut off branches of search). I believe it might appreciate the game on a deeper level than humans, in its own unique way. Of course it can't appreciate the social aspect of the game and all that comes with it.
Put another way, Chess is literally a matter of life and death for AlphaGo, because chess is all it knows. It has no exterior context for which chess is a metaphor.
It's 'appreciating' the value of various states and moves, in light of a vast trove of experience.
Is that true? I feel like chess was a bigger deal in the past. Among my peers, poker and computer games seem a lot more popular.
As to the direct influence of engines the other innovations aside, it has definitely forced players at the very top to re-evaluate the chess metagame, find weaknesses in traditional openings and shook up strategies. For the strongest players engine evaluation has become a useful tool providing new insights. When people watch chess tournaments these days on the internet most websites will provide parallel engine suggestions and commentators use engines to take hints for their commentary.
In my opinion, engines have made the game more competitive at the pro-level and more accessible for casual viewers.
Like don't you enjoy a game to enjoy a game? You can't beat Carlsen either, but you enjoyed the game at your level. Now computers are Carlsen +1, but how you enjoy the game shouldn't be affected. Especially since deep blue won in '97 and the game is still very alive and well, it hasn't been killed by computers but enhanced. Coupled with the multitude of good chess sites and resources out there, it's a better time than ever to enjoy the game.
There's nothing strange about becoming demotivated to study and compete at something extremely taxing both emotionally and mentally when a machine can beat you after an illustrious career.
Maybe for these games to keep popularity, we just need to update our perception of it. The same way we do with FPS games. Yes, we know a bot would do better - but that's not what matters.
Another thing, as said in other comments, is that we can learn from bots. New strategies, new patterns. AFAIK, this is not happening in FPS eSports scene.
Now, going away from this and into the following statement:
> said his retirement was primarily motivated by the invincibility of AI Go programs
I have a slight problem with that line of thoughts. Yes, there is undoubtedly a program that is thrashing humans left and right day in and day out. But I don't see why that would motivate someone to quit/retire. AI, in this day and age, to me at least, is still basically curve fitting. The fact that humans have figured out to do that in N-dimensional spaces (even with very large values of N), does not in any way undermine someone's effort and time dedicated to learning how to do a task. I'm sure that in several years someone will build an "AI" that can drive a modern Formula 1 car and it will smash all records and the best drivers without any efforts. And when that happens, should we abolish motorsports? Or sports in general? As humans we are confined to the limits of our biological abilities and personally I'm fine with that. I don't know who the first trillionaire will be, but I'm willing to bet that someone who figures out a way to interact with those models efficiently has a good chance. Essentially build a fast, all-accessible interface between the biological and digital world, which doesn't involve the traditional digital inputs(keyboard, screen, mouse, etc), and you have access to a decision making superpower at your disposal 24/7. When (and if) that happens, this is where games will cease to have any point. That's when it will boil down to who has the best hardware and software running along them rather than who is the best. Until then - game on!
I wonder if people just don’t like being inferior at something even if the comparison is a machine?
The comparison to sports isn't 1:1 as sports is about physical biological limits vs games which are more about the thought processes. Also human+AlphaZero < AlphaZero so why would I spectate a human+machine vs human+machine match?
Much later when it's commonplace for machines to be better than people at many things, things will change back, like we're amazed to watch people recite digits or make numerical computations.
As for when it's commonplace - yes and no. It's like knowing 60 digits of pi - it works as a party trick but other than that I don't see any real value.
But again - I think we are really far away from that and I consider those thoughts to be my personal speculations at best. Only time will tell.
[0] https://medium.com/intuitionmachine/the-alien-look-of-deep-l...
Is there any computational task which couldn’t be done through something which could be called “curve fitting”?
“Curve fitting” just means “approximating a function”, yeah? And what is a computational task other than computing a function, or some generalization of functions (e.g. could involve randomness or interactivity or the like as well)
Now, yes, current AI doesn’t yet have a model of the world including it being embedded in the world, and such that it chooses actions in order to further goals.
But, I don’t see why such a thing fundamentally couldn’t be accomplished using what one might call “curve fitting”.
If Strong AI could potentially be accomplished using “curve fitting”, I’m not sure that it makes sense to say things like “merely curve fitting”.
Though, you didn’t say “just” or “merely”.
One might imagine a sociopolitical task as sort of the ultimate machine incompatible goal. A look at how well spambots do at getting dating matches, or at how often clowns get elected to be leaders of nations makes me doubt even those sorts of tasks can't be won out by a well constructed domain specific optimizer.
AI will be knocking humans out of many positions of greatness (and not so great). The way we frame our future in a world where our various AI children are better than we are, is important and might be the biggest question to answer right to know the future of our species. Should we be afraid? Sure. Excited for possibilities? Yup. It's how we react to these feelings and how we ultimately act that will inform our ultimate fate.
A super-intelligent AI wouldn't care though, since they would infect any host and escape to the internet as soon as it makes sense.
While an AI Go player may always be able to defeat a human player in points, it still can't recognize or explain a truly beautiful move or strategy, even if it is the one that is carrying them out.
When it comes to competitive endeavors I've never been as interested the best individuals, except insofar as they are more likely to create the best moments of play - especially moments that reveal profound truths about the nature of the game itself.
What comes to mind thinking of this is a quote from Samuel Anders early in BSG:
"... In fact even the games aren't that important to me. What matters to me is the perfect throw, making the perfect catch, the perfect step and block ..."
The nature of Go has changed, but the game still deserves to be played and explored.