If AlphaGo wins all 5 matches, what do you think DeepMind will do with it? My intuition is that they won't continue development, and instead focus on other applications.
I think people in the Go community wants to see AlphaGo play against the top ranked player. Otherwise I hope they'd at the very least release it as an AI for people to play against.
The probability that any human player will learn to beat state of the art chess engines is zero (at least until we have humans with biologically or electronically augmented brains). There was a very small time window in which an expert player could beat chess engines by 'adapting their style'. Do you have reason to believe Go will be any different?
Lee is widely considered the greatest player of the last decade, winning 18 world titles, but has been surpassed in the last few years by younger players, with this unofficial ELO rating system ranking him at No.4: http://www.goratings.org/. Go is really a young man's game right now.
They'll keep using the technology. They probably won't want to play Go again - they'll have little to gain and everything to lose - and I suspect most of the possible audience will lose interest.
Putting on a conspiracy hat for pretend, the whole thing is clearly all a big PR scheme to get Google the brand better name recognition and improve their position in Asian markets.
That's not a conspiracy theory - I think everyone agrees this is as much a publicity stunt as a true test. Unless you're suggesting they bribed Sodol to lose?
Ke Jie believes he is slightly better than AlphaGo given the play he has seen, although he thinks AlphaGo will outpower him in a few months.
The belief that every good player that has so far played against AlphaGo would win highlights a common misunderstanding of the nature of the human mind. There seems to be in fact a very small difference between the aptitudes of the average players and those of the top players, such that it is very easy for a machine to go from average to superhuman. Getting to average is by far the harder part.
> There seems to be in fact a very small difference between the aptitudes of the average players and those of the top players
This is an interesting observation across many domains that involve "intelligence". It's actually pretty easy to see why, if you reconsider the scales for "good" vs "bad": The "bad" end of the scale is not at "the village idiot". The "bad" end of the scale is at "a rock". So, yes, once you get from "rock" to "village idiot", "superhuman" is a comparatively tiny step away.
What's even more exciting is that there weren't direct mistakes by Lee Sedol in this game, like there were in Game 1. So does that mean that AlphaGo is just playing on a level beyond him?
That might be the only strategy to learn from AlphaGo: If she makes an unexpected bold move, like she did in the center a couple of times, she's onto something and you certainly miscalculated that area. Don't gamble, defend that.
This game was largely played extremely well by both sides. There were a a few peculiar-seeming moves made by AlphaGo that the commentator found very atypical. These moves ended up playing a very important role in the end game.
I should also say that it's somewhat clear that Sedol made one suboptimal move, and AlphaGo capitalized on it. Interestingly, the English commentator made the same mistake as he was predicting lines of play. This involved play in the center of the board, in a very complicated position. Prior to this set of moves, the game was almost a tie. Afterwards, it was very heavily in AlphaGo's favor.
Just that in that time AlphaGo will have played millions of more matches against itself, learning at a much faster rate than him. Not sure, he might still beat the machine though. That needs to be seen.
Impossible for anyone of us to know. The Google people might know, but it's in there interest to not present AlphaGo as clearly superior because an even game attracts more views.
Seeing the way chess computers have evolved, this won't be far into the future. There are several tournaments just for chess computers, and it is very hard even for grandmasters to follow every more. Somehow, after a 20-move sequence that appears to accomplish nothing, one side is slightly up, and the rest of the game is decided.
There's been a big annual Go computer cup for a while now, it's been fun to read about: http://jsb.cs.uec.ac.jp/~igo/eng/index.html Of interest this year is Facebook's new bot is competing, which employs the same strategy as AlphaGo (Deep nets with MCTS), but from what I remember they've said (they've been a lot more open about its performance, you could even play a version on KGS, though I did really like AlphaGo's paper) it's quite a bit weaker. However I suspect it's also running on a lot less computer power, and that it will still be a while before you can get AlphaGo or better performance from a mere modest gaming rig let alone a standard laptop.
One benefit of programs like AlphaGo eventually becoming available to individuals is that we'll see a rise in some incredibly strong young players that leave their ancestors in the dust, as I think has happened with Chess via Magnus. For the amount of training that can only be done by playing more and more games, being able to do that against a computer will be a lot more efficient.
> Seeing the way chess computers have evolved, this won't be far into the future.
With chess, there were two breakthroughs. First, there was Deep Blue, which threw massive hardware resources at the problem and achieved world champion level play.
That was interesting, of course, but didn't really do anything for human chess, because most humans did not have access to the necessary hardware.
The second breakthrough was when the developers of chess programs that ran on commodity desktop computers improved their algorithms to the point that they could play at (and far beyond) Deep Blue's level even though they were only able to search about 1/100th as many positions per second.
That was when humans started being able to really use computers to help the humans understand chess.
The breakthroughs in chess algorithms on commodity computers had little, if anything, to do with the Deep Blue breakthrough. The two are just too different.
Can AlphaGo be made available on hardware that top human Go players have access to, or is AlphaGo to Go as Deep Blue was to chess?
Hey, that wetware is ten time as powerful as what AlphaGo has to work with. Give or take a few orders of magnitude. And given that Lee's brain only uses a portion of that on Go.
> Hey, that wetware is ten time as powerful as what AlphaGo has to work with.
I don't think that's actually true. The hardware that AlphaGo is on is probably a lot more powerful than the one that is available in a single human brain, the big difference is in the software.
See the difference between the very best chess programs of a decade ago versus the ones now.
Seems like somewhere in-between. They created a novel approach that is scalable and improves as you throw more hardware at it. And Google is throwing a lot of hardware at it based on their past matches with hundreds of CPUs and GPUs. I think the fact that they have been so mum about what hardware they're using suggests it's quite extreme, but hopefully they release more details soon.
Its somewhat interesting to think about the differences in marketing between IBM and Google; IBM was marketing hardware and HPC with deep blue, but Google is marketing AI when so much of their advances in AlphaGo are enabled by distributed systems and HPC running billions of games training deep neural networks. It feels a little smoke and mirrors which is probably why they won't release much until after they get enough marketing value from this tournament :)
Fan Hui has already learned from AlphaGo. He's been playing matches against her regularly, and (perhaps as a result of that) won _all_ his games in the last European championship.
Myungwan Kim said he feels like it plays like a she. Personally I think its informal gender was determined by the nigiri of the first match, as it's common to refer to black as he and white as she absent of player names. And I'm expecting to see at least one really cute AlphaGo-tan drawing any day now.
I watched the live stream commentated by Myungwan Kim so I know what the parent is referring to (later in the game, Myungwan Kim referred back to that possibly being a mistake), although not being a Go player myself most of it was over my head.
It was fairly early in the game and about 4-5 lines down from the top and towards the center, center-left. Apparently, Lee Sedol played a little conservatively and "didn't take a ko" (?) when he could have.
Hope that helps you figure out what they're referring to, or maybe someone else can chime in, but that's what I remember.
Lee Sedol seemed to be doing well before he went into extra time (as far as I could follow from the commentators). How is it ensured that this is a fair game given the time constraints? I'm guessing adding more computing power to the AlphaGo program should definitely help it in this regard.
The human's strength is intuition and insight. They can look at a move and have a good understanding of strengths and weaknesses of positions by "feel" developed by long practice of the game. More time doesn't really help this much.
Another part of the game is "reading" -- playing out scenarios of response and counter-response to evaluate how strong a move is. The computer excels at this, because it can play out as many moves as its computation time allows and remember all the results with complete accuracy. Humans are slower and prone to mistakes when they do lots of reading.
So adding clock time lets the computer increase its advantage over the human in reading depth, but doesn't so much increase the human's advantage of intuition and insight.
Well, I think the last time computers played the world chess champion (in 2006), they didn't allow the computer to think on the human time!
And you can always give the computer less time than the human, but this just shows that it's stronger than you and you need to handicap it to have a chance.
Increasing the human time is not an option, since no one wants to watch 8 hour games, and fatigue could also come into play, so the only real option for a winning chance is reducing computer time/power
Correspondence go, similarly, would see fewer errors. Holding a world championship would take quite a bit longer than in chess, though (rough guess: 300-ish half-moves per game versus 100-ish half moves, the latter, I guess, with a bit more variation). That could be problematic, as a world championship in chess already takes years (curiously, some championships finished before the one started a year earlier did)
Irrelevant in this case, there was the latency of the operator reading the move on the screen and physically picking up a stone and placing it on the board.
Just as arbitrary. Unless you're a self-sustaining vegan, you're costing a lot of energy even to just get you the few thousands of kcal your metabolism consumes.
Unless you're not counting support systems, in case it becomes very complicated to calculate and decide exactly which energy expenses are for support systems and which are directly integral for function.
We have two computational substrates, human brains, and CPU/GPU clusters. Forget what it takes to support them, just consider what they consume while computing, that is, the energy consumed while they are playing the game.
Lee Sedol is vastly more efficient than the entire AlphaGo cluster. However, while AlphaGo gains a predictable amount of power as its computing power is increased, it's not clear that one could do the same with humans. Our Go players optimize individual play, not multi-brain distributed play. What would the match look like if we trained up a bunch of humans to play Go as a team, and pitted AlphaGo against a team of humans that consume the same number of joules over the course of the match as it does?
Let’s not forget that aside from being vastly more energy efficient as a Go player, Lee Sedol is additionally capable of taking on a virtually unlimited list of other, equally machine-challenging tasks – while AlphaGo can only do one thing. In fact, Lee can lift himself off the chair to a standing position, pace around the table, lift a glass to his mouth, keep it there while emptying some of it, and think about his next move – all at the same time. (And on the same energy budget.) And far beyond all that, he decides whether to do these things – or something else instead.
I admit my first thought on fairness did go in the same direction of limiting energy budgets. But after reflecting on it just long enough to realise the above, I am finding myself surprisingly uninterested. It now seems to me that nothing particularly insightful would be revealed: limiting energy budget is no less arbitrary than limiting time unless the artificial opponent is expected to be capable of a range of things comparable to that expectable of an average human. Or if expectations are much lower, the artificial opponent would need to contend with drastically tighter limits to approach “fairness” – though at this time it would be guesswork how much tighter they ought to be. Either way, it is glaringly obvious that no computer would come within miles of competing.
So ultimately the fact that Go has been “broken” (in a particular sense) at all is far more interesting to me than whether the machine is competitive with the human in any more general sense. “It’s not” as the universal answer is boring.
And to digress a bit from there: From that perspective, this was ultimately a very human achievement. It was humans who chose Go as a problem to attack and it was them who picked MCTS and deep learning as the way to go. (Uh, no pun intended.) That’s not just reassuring. It’s also a framing we should keep in mind as computers become more entangled with the physical world and more autonomous.
I find it interesting that AlphaGo seems to take a long-ish time to play the move every other pro would play instantly. These pauses probably help balance any time oddities. I think Sedol managed his time well, and even AlphaGo went into byoyomi near the end. Also the fact that it's on even time is more than fair, since the standard until now has been "go on, add more computing power, take more time to play a move, you'll still lose to a pro". As for AlphaGo, I kind of remember reading that doubling the computing resources at this point gave an increase in 60 ELO points. (So if they solidly win against all the Sedol matches they may need to double once or twice or find enough software optimizations to take down Ke Jie using standard time control, but it's not out of reach..)
The next person that will beat alphaGo may not be a top go player.
In particular, I'm wondering if a computer scientist with access to the alphaGo source code and all the weights of the network could trick alphaGo in order to win games automatically (cf. the papers that show a neural net can be tricked to classify a plane as any other class).
If a human with the knowledge of the source code and the weights can do this, it is scary. Imagine a similar algorithm runs your car. An attacker that knows the source code and the weights may trick the algorithm to send your car in a wall!
You cannot do something unpredictable in a game with strict rules and borders. You had to trick this algorithm the billion times it trained for this game.
I had this conversation earlier with a friend wondering if any of Sedol's Korean pro buddies have noticed any systematic biases that could be exploited. I think it would be possible to make the neural net relatively useless by playing strange sequences that hack the weights, but you're still left with a monte-carlo-tree-search bot which alone depending on its implementation is between 2d and 6d amateur (so on the far weaker side of the professional scale). Whether you could make those strange sequences to trick the neural net while also not dying to MCTS, I'm skeptical.
I'm kind of hoping for an unconventional opening in game #3. Come on tengen or 5,5... Additionally I think that might be one way of weakening the bot in that it will find its net for suggesting candidate moves less useful and lean more on unguided MCTS, but that's just a wild guess.
Sure, almost any HnG fan will play them at least once. ;) But there's so few of them in general I would think any training data derived from them would be low-value.
I sense a change in the announcer's attitude towards AlphaGo. Yesterday there were a few strange moves from AlphaGo that were called mistakes; today, similar moves were called "interesting".
fun fact: when TD-Gammon hit the backgammon scene in the 90's, it didn't just defeat the top level human pros, it shattered the whole metagame and changed how humans play the game. It could be that human vs human go will look very different in the future due to what AlphaGo has learned and can teach us.
AlphaGo maximises the probability of winning, and not the margin by which it does. So those "mistakes" yesterday turned out to be fortifying moves because AlphaGo was confident of a win. And similarly today the weird moves were interesting because they perhaps indicated that AlphaGo thought it was ahead.
Yeah, that explanation from the DeepMind team member today put a whole new spin on some of the 'odd' late game moves. It doesn't 'care' about about margins so it will shore up its odds of a win in preference to increasing the margin if it wins.
I had a feeling after yesterday's win that people might be tempted to call it a close win by AlphaGo and predict that Lee could overtake it with a bit better play, but that we'd find no matter how good Lee played, AlphaGo would adjust and continue to come out just a bit more on top. It's actually probably really hard to tell how much better AlphaGo is because it probably plays quite conservatively overall and hides a lot of potential strength.
Does anyone know what DeepMind's software stack looks like? Just based on past work of some of the people working there, I'm guessing most of the code is C++ with some Lua. Anyone know for sure?
At the same time, only now are people seeing how its hard for people to play Go against computers. Until AlphaGo I didn't even think computers were close.
Go is much harder than chess. The tree of available moves is massive and difficult to prune, as it's hard to get a reasonable heuristic about the "strength" of a given Go position.
Actually computers were already in the top 10 percentile or so. I mean top go playing bots ranked something like 5dan amateur which is very hard to achieve, so it could already defeat most human players.
To address this, has anyone considered pitting AlphaGo against a team of 9p players consulting with each other, and perhaps taking charge of different conflicts on the board?
I can't find the source now, but I remember reading a while ago that a team of go players is, perhaps contrary to intuition, not significantly better than their strongest individual alone.
There's just no useful way to "pool" human thinking power and redistribute it to where it's needed the most – all the players will simultaneously consider the same branches. At best you reduce the risk of making a silly mistake, but sitting pros are already pretty good at that.
> all the players will simultaneously consider the same branches
A computer-assisted pool of humans might work. Feed each human a board state advanced by 1/2/3/N moves down the decision tree in some direction, and have them evaluate that particular sub-tree. It's a map-reduce problem!
For some reason this made me think of the Focused in A Deepness in the Sky - where real general AI isn't possible [at least where we are] so human minds are harnessed to solve problems in a deeply unpleasant way.
AFAIK Mechanical Turk is just getting random people to do stuff for you - the concept of Focus is something else entirely and genuinely quite terrifying.
If you haven't read A Fire Upon the Deep I'd recommend reading that first. It's not vital but there are some subtle links between the two that are rather cool (not the shared character, more the shared "physics" and tech).
Is there a non-dystopian way we can do this? The immediate problems seem to be bandwidth of communication and ability to quickly generate shared culture and jargon.
Are Bridgewater Capital's employees Ray Dalio's focused?
That's actually a really interesting idea. At this point you're basically replacing the "policy" neural network in AlphaGo with biological human neural networks.
Is there a meaningful response that each one could give that would be optimal? I figure this would only stand a chance of working if the pool of humans were cloned from a 9 dan :-)
Each one needn't give a 9 dan response. Each person gives one response to the best of their ability, and scores the situation on a scale of like 1–100. Collect all the responses and distribute the new board situations over the same – or a different – group of people. The score is mostly used to value the quality of previous moves. If a branch leads from a situation most people rated as 60 to a situation where most people rate 10 it's not a good branch.
Of course, Go has way too many eligible possible moves for random people to vote on, but a large enough group of top pros might be able to do well just by voting.
Four to five expert chess players suggested moves for the world team. I feel that for any reasonable non-expert it comes down to "choose between these suggested moves" rather than "pick a move". It's also not really random human beings, since the selection of participants is self-selected and therefore much more likely to contain very good chess players.
This is mostly semantics, but anyone who doesn't read your link might get the wrong idea, so I felt like clarifying it a bit.
Fair points. Thanks for pointing this out. I used "random" to really mean "not-top-pros." Moves suggested by experts being voted on by a large sum of people are kind of similar to how democracy and capitalism work.
Part of the reason go is a difficult computational problem is that local conflicts can have global relevance. If it was the kind of thing that could be decomposed easily to different humans then it would be an easier game for a computer to crack.
Yes, particularly because of the sente/gote dynamic. However, it takes time to read individual situations through, and this might be chunked and distributed. The problem, as usual, is efficiently communicating analysis in a useful way. There would probably need to be a hierarchy where "specialists" report to the captain about the battle, and then the captain has to prioritize and decide on the order of action for each region. But the minute detailed analysis for each region would be entirely delegated. Note that leiutenents would be responsible for understanding their region's relationship to the whole, but only the captain is truly responsible for understanding it all.
I'm totally uninformed about Go, but by now it seems that unless you're clearly in the lead by the end of the midgame, AlphaGo is going to win, simply because at that point its Monte Carlo Tree Search is going to our-compute you in examining all the tactical variations in the endgame. Lee Sedol really seemed to be under a lot of time pressure at the end.
EDIT: clarified to what I originally meant: "end of midgame"
It really depends on how good AlphaGo is at avoiding mistakes. If you get critical points on the board, it's possible to create a resilient structure that is defensible against comebacks. But when I play, there are times when I can make an overwhelming comeback because my opponent missed one critical move. I've likewise had it done to me. This is of course probably less common at higher and higher levels of play.
I'm reminded of the time when Deep Blue made a buggy move that totally threw Kasparov off because it was so counterintuitive. But it turned out to be a real mistake. If Kasparov had kept his composure, who knows what could have happened?
Of course, as been stated here, AlphaGo doesn't have the issue of having to keep composure, while Lee does have that issue and pressure. But who's to say that AlphaGo won't make a similar significant mistake, and that Lee won't be able to capitalize on it to make a comeback in that particular game?
edit: admittedly, as pshc as noted, the search space probably grows smaller as the game continues.
edit 2: suddenly, his comment disappeared for some reason?
In the mid-game – and hell, even in the early end-game – the search space is still too big to practically exhaust the useful branches with MCTS. Keep in mind "mid-game" in Go means there's still over 100 moves left in the game – well longer than a full chess game.
I'm not saying you're wrong, I'm just saying I don't agree with your motivation.
I don't think so. Pros are already pretty good at the endgame. Yeah, they make blunders now and then, so it's possible that AlphaGo would gain a few points in the endgame in some games, but not enough to overcome a significant lead and not every time.
In any case, I personally find it more interesting to see what we can learn from AlphaGo about the opening and the midgame.
My writing was sloppy, let me rephrase it: in a pro match, whatever is the estimated score when the game goes into the endgame, that will likely be the final score.
Where I wrote that a pro will blunder "now and then" I should have written "rarely" -- I don't have data to back this up but I'd guess once in every n games for some n > 30.
The thing I find amazing about this is how soon this has happened. We all were expecting this to eventually happen but if you asked anyone who played go and was across the computer go scene when it would happen, say a year ago, they would say it was "10 years out". AlphaGo is one incredible feat of engineering.
Is there a complete recording of the commentary? They had one for game 1. The current live stream only goes back two hours and doesn't include the beginning of the game.
I tried to watch commentary on the DeepMind channel for game #1 and found it to be very un-informative: I recommend the AGA channel (although it's been linked a few times): https://www.youtube.com/watch?v=6ZugVil2v4w, still waiting on the game 2 static video.
There's a good reason for that: there is no game on Friday. Next game is Saturday, and they may or may not broadcast that. Andrew is ready, but he needs a pro or otherwise really skilled player to provide high-level commentary. Hyungwan who did it the last two games will not be available.
Last night I found that on the DeepMind channel they were explaining the game a lot for beginners. This was a bit tedious for me but then this is a historic event so there'll be noobs tuning in so I can understand that, and I appreciate that the TV people are thinking of them.
Also, the AGA channel you just seem to have a split screen showing two people with headphones on talking over Skype? (Granted, one of them is Myungwan Kim, 9p, a super likeable guy - but Michael Redmond, 9p, is also super likeable.) The DeepMind channel switches between the game board and the large commentary board, with occasional shots of Lee Sedol and occasional shots of the DeepMind terminal.
Michael Redmond was reading out lots of variations and repeatedly trying to count the board and evaluate the position. I think he was trying to be as informative as he could. Occasionally he would get lost in thought as he calculated owing to the complexity. You could call it many things but uninformative is not one of them.
> Last night I found that on the DeepMind channel they were explaining the game a lot for beginners.
Yeah, I think I caught some 15 minutes of this and stopped watching, maybe it got better later? I would have expected them to the introduction for beginners before the start of the real game, and they ended up skipping commentary on what seemed like pretty key choice in the early game to do so.
> Also, the AGA channel you just seem to have a split screen showing two people with headphones on talking over Skype? (Granted, one of them is Myungwan Kim, 9p, a super likeable guy - but Michael Redmond, 9p, is also super likeable.)
There were even some technical difficulties with the skype feed on a few occasions and I wouldn't call it great quality, but on the other hand I rather prefer the KGS virtual board - on which it's quicker to show the possible alternatives and easier to point out the key stones.
On this I do agree with you :) I found it annoying that there was not more analysis happening in the fuseki. I was like, "is it not important to be evaluating these moves?" I guess because the DeepMind crew didn't it meant that beyond pointing out that Lee Sedol tried to break out of opening patterns quickly and that the order of moves in the top-right joseki were considered sub-optimal for AlphaGo there was not much that urgently needed commenting on.
I switched to KGS at intervals to see what people were saying on the mirrored board kibitz. And the DeepMind commentary did begin to irritate at times but I feel they had to cater for noobs. Once the middle game got going it was fine though. Agreed on the KGS virtual board.
I was wondering why they didn't keep a running total of the score after each move, and show it somewhere on the screen. Isn't that pretty easy to automate?
> I was wondering why they didn't keep a running total of the score after each move, and show it somewhere on the screen. Isn't that pretty easy to automate?
Until you are close to late-game the scoring of many of the positions is quite flexible (depending on sente, and how certain local situations are resolved) so although you would may have a semi-reasonable approximations for total points, the games have been close enough that a +/- 5 point error in estimating will make it impossible to tell who is actually ahead.
That said, it would be very curious if we could get a virtual-board with AlphaGo's evaluation function used to score each position.
They put a version of the broadcast up after it is finished, if last time is anything to go by. Took a little while because the stream is still going, with the press conference.
I find it very interesting that to a layperson, the idea of a computer being able to beat a human at a logic game is pretty much expected and uninteresting.
You try and share this story with a non-technical person and they will likely say "Well, duh..it's a computer".
I encountered this many times over the past couple of days haha. I then have to explain that previous AIs, like those in chess largely used brute force computation to simulate each move, while these AIs "actually learn, similar to our brain". Probably not the most scientific explanation, but I feel your sentiment.
They are using Monte Carlo Methods for looking around, more like a tiny sampling, it's an incomprehensibly large space. They have both value and policy (deep) neural nets. Train one to get the sense of good/bad individual board states (value) and one to get the sense of good/bad trajectories (policy, evaluates chains of moves).
Chess has a fairly straight forward ranking of pieces, fairly well established ranking of piece power. A knight is more or less always a knight.
Go, you kinda make up your "pieces" from scratch and there isn't any single common ranking. This makes the problem of even evaluating the board difficult (value), which is a sub step of making your higher plans (policy).
The output of the two dNN help cut off, or prevent the sampling of the bad game states and encourage looking into the fruitful areas.
Well, it's a bit more complicated than that. Neither chess engines nor go engines use brute-force (that is, exhaustive search). Although go does have a much higher branching factor and that does affect the search algorithm used, the biggest challenge is being able to write good evaluation and move-guessing heuristics.
It's a creative game only to the degree your logic (reading) is not strong enough :)
I mean, deep down, Go is a perfect-information-zero-sum-discrete-2-player game.
I think a more successful pitch is showing that the strategies and skills coming from Go (human or robot) can be useful outside the game, as I argue here: http://rare-technologies.com/go_games_life/
To me, "creative" is not the right word. Maybe try "intuitive".
As Radim says, of course, the intuition is only relevant when logic fails. However, no computer, including AlphaGo, has sufficient processing power to take the logic-only approach to the, uh, logical extreme (more board states than atoms in the universe, etc etc).
So both humans and computers play by a combination of logic and intuition. Surely Lee Sedol must be incredibly good at both. Perhaps AlphaGo is better than Lee Sedol at the logic part (or perhaps not, but just suppose for the sake of argument).
In this light, what is interesting is that AlphaGo is sufficiently good at intuition, a domain we might have considered uniquely human[0], to complement its ferocious logical power.
[0] I find it foolish to consider anything uniquely human, but a humanist essentialist might make such a claim.
The other interesting bit is the combination, the intuition guiding (and evaluating the results of) the logical searching.
I'm still holding out for the discovery of a proof of winning strategy like other combinatorial games. That to me is the logical extreme, not magic evaluation of all states. There are two interesting books on analyzing Go with combinatorial game theory, they found a neat system of scoring in the endgame and can create example boards with one best move that can be found by mechanically applying their system but stumps 9p players (who use whatever systems).
That's why task based notions for general intelligence are rubbish.
Here is what humans can do: When presented with pretty much any task, specified however poorly, they can design hardware and algorithms that can beat themselve at that task.
That's a decent measure of intelligence. A decent measure of creativity is coming up with tasks that make the intelligence part as interesting as possible.
Optimizing compilers do this already and they aren't 'intelligent'. Iirc alpha go is an extension of the same algo that could beat all atari games sight unseen..
Optimizing compilers do this already and they aren't 'intelligent'. Iirc alpha go is based on the same algo that could beat all atari games sight unseen.. satisfying the 'novel situation' requirement you set forth..
Optimizing compilers do this already and they aren't 'intelligent'. Iirc alpha go is based on the same algo that could beat all atari games sight unseen.. satisfying the 'novel situation' requirement you set forth..
That's a point of concern for using the Turing Test to assess the intelligence of a computer, but that's only because the computer has to also be trained to expose human-like features that are considered to be unintelligent.
But on one hand, in order to have a shot at passing the Turing Test, a computer has to first be able to understand and speak human language. And that's pretty damn hard actually.
Yes we can deploy statistical methods, deep learning or what have you for classification, feature extraction and so on, but natural language is context rich and ambiguous. NLP isn't a conscious process for us either, but our brains are capable to disambiguate effortlessly. And a big part of that is also the rich (human) experience we gain while growing up. Being able to speak natural language is the one big trait that separates us from animals. It's why we can cheat death, because our children can learn from the acquired knowledge of all of their ancestors.
This is different from playing Chess at least. With chess you can deploy smart algorithms, but in the end it's still a raw search for good moves in the space of all possible moves, while giving up on branches with a bad score. Raw search is what computers have always been good at. That's not really possible with NLP, because at least with chess your vocabulary is very limited and for calculating good moves you don't need extensive knowledge about the world we live in.
Computers will surely be able to do that in the future and that will be a major milestone towards "true AI", but for now computers cannot do it.
On the Turing Test, some features considered as being unintelligent, like the tendency for lying, or being sensitive to insults are actually evolutionary traits, that have arguably helped humans to survive. So while emulating some human traits will be counter productive, a "true AI" will be concerned with survival and as a consequence will end up doing whatever it takes.
So while passing the Turing Test may not be enough, not passing the Turing Test is a sign that the computer is unintelligent.
I know this is a semi-joke, but it's worth discussing. While such people are less smart than average or have a deficiency in their education or both, unless you're talking about mentally handicapped people with a medical diagnosis, all normal humans have a conscience, are capable of reasoning, can understand complex symbols, can entertain complex thought, are very good pattern matching machines, can speak natural language and can learn and acquire new skills, which makes them intelligent.
I know the kind of people you're talking about. I have a family member like that. She's not that smart, she failed her baccalaureate, she's semi-illiterate and she probably suffers from ADHD, though a lot has to do with her upbringing. But getting answers like that from such a person means that you're not asking the right questions or the incentive to answer is not there. Give a cash reward to a cash-strapped person and you will never get an answer like "im so bored this test sux".
Yes, but you also have a database of answers from people, so even if you ask things like "what color is the sky" you're still getting the right answer.
it jokes with you, it gives vulgar answers sometimes, etc.
it's NOT good enough to fool someone, but what if you just made it sound like it's a person with a disability? Is that still beating the Turing test? Is talking to a "five year old" still beating the Turing test?
Or does it have to be a 100 IQ adult person fluent in English? In which case that's just improving the bot a bit. To make it a funny and charming bot it would require even more effort. But it's just slowly improving the state of the art with some kind of techniques like reinforcement learning or neural networks or whatnot.
When a bot actually beats the Turing test it wouldn't be big news because it would just be a slightly better jabberwacky.
> So while passing the Turing Test may not be enough, not passing
> the Turing Test is a sign that the computer is unintelligent.
You could have an intelligence that's just not smart in the human
sense. Consider running into an alien intelligence evolved from our
equivalent of octopuses, you ask it a questions but it only
communicates via color changes on its body.
Similarly you can conceive of an AI that's smart, self-aware and
intelligent just hasn't been developed to talk to humans.
The Turing test is a fine test to figure out if your AI is
conversational with humans, but the OP I was replying to was
suggesting it as a general AI intelligence test, it's not meant for
that, and will give you both false positives & negatives.
I'd say if anything the average person's perception of what AI can do in the opposite direction. In the mid twentieth century the idea of a near future involving computers/robots that thought and interacted much like humans but never made mistakes was pretty mainstream. People have rather dialled back their expectations of feasible computers since then, to the point the average layman thinks a hardcoded easter egg humorous response in Siri is impressive because although talking to Siri is just like a more error-prone alternative to using the keypad, the response sort of seems like how a human would handle the question.
The average person isn't impressed with computers winning at Go because they vastly underestimate the complexity and open-endedness of Go and wonder why it's really all that much more complex than chasing Pac Man through a maze like their computers were doing quite happily, and even with apparent personality, in the 1980s.
Go is a lot more psychological and emotional than playing tic-tac-toe. Its a strategy game, which undecided outcomes, where ones style and vision changes how you play and see the game.
The computer was able to overcome the computational difficulty humans compensated with abstractions and strategic concepts. We still use those to understand whats going on in the game, but the computer is oblivious to them and only uses a tactical view of it.
Michael Redmond. I thought he was excellent too, and I really liked his willingness to jump into hypothetical situations and play it out. Really, really good commentator.
As the only 9p player in the western world, Michael Redmond is of course very impressive.
Chris Garlock, on the other hand, doesn't add that much value to the broadcast. Maybe somebody will start a "Left Commentator" meme, just like Left Shark.
As I understand it, the value network takes the place of the heuristic for scoring a given board layout, and the policy network takes the place of the heuristic for ordering moves from most to least promising.
When searching the game tree, at each ply the most promising N moves are examined (as determined by the policy network) and leaves of the game tree are scored by the value network.
I mean it's classic sports broadcasting. You have the expert and the guys who gives the expert someone to back and forth with. E-sports has been using that formula for ever.
I don't think that's fair at all to Chris. As someone who has never watched GO before, he kept the topic often to very beginner level questions, and often kept Michael on topic a little bit. I would say Chris added a lot of value to my experience, although this is the first time watching either of them.
I was also super impressed. I know him as マイケルレドモンド (Michael Redmond in Katakana), and it was the first time I saw him speak English. He's fluent in Japanese and frequently commentates on matches in Japan. This is a video of him commentating on the Meijin title matches.
I may be glad no one took my bet offer of me paying $19 if AlphaGo won 3/5 vs them paying $1 otherwise... I had a prediction at 90% confidence that nothing would show up before the end of this year that would be capable of beating the top players (though since I first heard about MCTS's success the idea of coupling it with deep learning seemed obvious, so I had an unfortunately non-recorded prediction that if a company ever bothered to devote about 8-12 months of research and manpower into combining those two algorithms with a very custom supercomputer or tons of GPUs then they would have something that could beat the best), then AlphaGo was announced. But the top pros weren't too impressed with its defeat of Fan Hui, and Ke Jie estimated something like "less than 5%" chance of it beating Sedol so I updated to 5% for this match of it winning 3/5...
Tonight's game was beautiful. Last night's was a fighting game way too high level for me to really grasp (I have no idea how to play like that, all those straight and thin groups would make me nervous). I'm expecting Sedol to win Friday since I imagine he's going to have a great study session today, but I'm no longer confident he'll win the last two.. Still rooting for him though. :) (I also want to see AlphaGo play Ke Jie (ed: sounds like from the other submission on Ke's thoughts that may happen if Sedol is soundly defeated), and for kicks play Fan Hui again and see whether it now crushes weaker pros or is strangely biased to adopt a style just slightly stronger than who it's facing.)
Re last paragraph: "just slightly stronger" is expected. AlphaGo is designed to maximise its probability of winning, not its margin of victory. You can expect solid not-very-flashy plays that definitely maintain an advantage, rather than even slightly risky plays that probably increase the advantage.
Let's compare Go and Chess. We all know that Go is more complex that Chess, but how much more?
There's 10^50 atoms in the planet Earth. That's a lot.
Let's put a chess board in each of them. We'll count each possible permutation of each of the chess boards as a separate position. That's a lot, right? There's 10^50 atoms, and 10^40 positions in each chess board so that gives us 10^90 total positions.
That's a lot of positions, but we're not quite there yet.
What we do now is we shrink this planet Earth full of chess board atoms down to the size of an atom itself, and make a whole universe out of these atoms.
So each atom in the universe is a planet Earth, and each atom in this planet Earth is a separate chess board. There's 10^80 atoms in the universe, and 10^90 positions in each of these atoms.
That makes 10^170 positions in total, which is the same as a single Go board.
I am not sure that calculating the raw number of positions is a good indication of complexity at a given point. What if most positions are obviously junk in go while they are more difficult to assess in chess? Not saying this is the case in this particular example but thats a possibility in theory.
> What if most positions are obviously junk in go while they are more difficult to assess in chess?
I wouldn't go with most (because I don't know about that), but many of these boards would also be either impossible to achieve (in a normal game) or illegal.
To illustrate your point: you can just add rows to a game of Nim (https://en.wikipedia.org/wiki/Nim) to get a truly enormous state space, without changing the simple winning strategy.
This doesn't seem to be the main reason why Go is harder than chess for computers. It was noted that even in 9x9 Go, with a comparable branching factor to Chess, traditional Go programs are still no stronger than on big boards. The main difficulty for Go is that it is much harder to evaluate board positions. So in Chess the depth of the search can be significantly reduced by using a reasonable evaluation function, whereas in Go no such function seems to exist.
Well, I guess it was more true before the advent of Monte Carlo Tree Search. Even so, note that even in the case of MoGoTW in 2011, it played blind Go (this helps the computer), and out of 4 games, won two games against a 9p player, and lost 1 game to a 5p player. Though it is perhaps better than MoGo's performance on 19x19, it still isn't very good, doesn't seem much better than MoGo on 13x13, and performs much worse than computer Chess, despite a similar branching factor.
The branching factor of 9x9 Go isn't 75. 75 could be the factor in early game, but the average factor is somewhere between 40 and 50, versus 35 in chess. State-space complexity is also considerably higher in Chess than in 9x9 Go.
Not sure what you meant regarding MCTS, I never said anything about MCTS not being able to beat pros.
This evaluation function does exist, and it's better than the super-simple chess evaluation function.
See, a chess program needs to find a lot of valid moves (see Deep Blue which won because it had stupid but extremely fast HW move generators), evaluate the moves and do a very deep search, up to 14, out of the very few alternatives. Russian chess programmers were better those times. They came up with AVL trees e.g. But hardware won.
In Go it's completely different. A move generator makes no sense at all, and a depth search of 14 neither. There are not a few alternatives, there are too many.
What you need is a good overall pattern matching of areas of interest and an evaluation of those areas. And we saw that this feature outplayed Lee Sedol. Sedol couldn't quite follow in the recalculation of the areas.
Same as in chess AlphaGo learned the easy thing, that the center is more important than the corners, something Lee forgot during the game.
But it's not a deep search, it's a very broad search, and very complicated evaluation function. A neural net is perfect for this function.
> whereas in Go no such function seems to exist.
It does exist. It's the neural net. It's a simple pattern recognizer, which learns over time more and more.
On the other hand, 2^565 is already slightly larger than 10^170. In other words, a couple of hydrogen atoms as quantum bits can perfectly well encode every possible position.
575 comments
[ 2.9 ms ] story [ 324 ms ] threadGreat game btw, a pleasure to watch.
In a sense this is unfair: alphaGo was trained with a lot of human data, but AFAIK Lee Sedol is playing AI for the first time.
I would assume the development will continue.
However, they won't make it such a big PR focus. (And even if they stop, other people will implement the same ideas.)
http://www.shanghaidaily.com/national/AlphaGo-cant-beat-me-s...
Ke Jie believes he is slightly better than AlphaGo given the play he has seen, although he thinks AlphaGo will outpower him in a few months.
The belief that every good player that has so far played against AlphaGo would win highlights a common misunderstanding of the nature of the human mind. There seems to be in fact a very small difference between the aptitudes of the average players and those of the top players, such that it is very easy for a machine to go from average to superhuman. Getting to average is by far the harder part.
This is an interesting observation across many domains that involve "intelligence". It's actually pretty easy to see why, if you reconsider the scales for "good" vs "bad": The "bad" end of the scale is not at "the village idiot". The "bad" end of the scale is at "a rock". So, yes, once you get from "rock" to "village idiot", "superhuman" is a comparatively tiny step away.
http://lesswrong.com/lw/ql/my_childhood_role_model/
good to know they'll play all 5 games no matter what the result is though
People seem to think Lee knew he lost and was just playing to learn more. Hope he learned enough to take the overlord down in the next three games
I should also say that it's somewhat clear that Sedol made one suboptimal move, and AlphaGo capitalized on it. Interestingly, the English commentator made the same mistake as he was predicting lines of play. This involved play in the center of the board, in a very complicated position. Prior to this set of moves, the game was almost a tie. Afterwards, it was very heavily in AlphaGo's favor.
Does this mean AlphaGo was playing at a higher level or is it just a coincident?
Myungwan Kim 9-dan professional (comments on match #1): "She knows everything [...] Unthinkable move for human [...] AlphaGo plays like the god of go"
https://youtu.be/6ZugVil2v4w?t=5054
I want to see AlphaGo vs AlphaGo :-)
One benefit of programs like AlphaGo eventually becoming available to individuals is that we'll see a rise in some incredibly strong young players that leave their ancestors in the dust, as I think has happened with Chess via Magnus. For the amount of training that can only be done by playing more and more games, being able to do that against a computer will be a lot more efficient.
With chess, there were two breakthroughs. First, there was Deep Blue, which threw massive hardware resources at the problem and achieved world champion level play.
That was interesting, of course, but didn't really do anything for human chess, because most humans did not have access to the necessary hardware.
The second breakthrough was when the developers of chess programs that ran on commodity desktop computers improved their algorithms to the point that they could play at (and far beyond) Deep Blue's level even though they were only able to search about 1/100th as many positions per second.
That was when humans started being able to really use computers to help the humans understand chess.
The breakthroughs in chess algorithms on commodity computers had little, if anything, to do with the Deep Blue breakthrough. The two are just too different.
Can AlphaGo be made available on hardware that top human Go players have access to, or is AlphaGo to Go as Deep Blue was to chess?
Whether those insights will come soon or not is the big question.
I don't think that's actually true. The hardware that AlphaGo is on is probably a lot more powerful than the one that is available in a single human brain, the big difference is in the software.
See the difference between the very best chess programs of a decade ago versus the ones now.
Its somewhat interesting to think about the differences in marketing between IBM and Google; IBM was marketing hardware and HPC with deep blue, but Google is marketing AI when so much of their advances in AlphaGo are enabled by distributed systems and HPC running billions of games training deep neural networks. It feels a little smoke and mirrors which is probably why they won't release much until after they get enough marketing value from this tournament :)
https://youtu.be/4fjmnOQuqao?t=2881
https://youtu.be/4fjmnOQuqao?t=2881
https://twitter.com/zpdldlfrkwl/status/707858454317178880
Can you please indicate this time in the video of the game? Thanks a lot.
It was fairly early in the game and about 4-5 lines down from the top and towards the center, center-left. Apparently, Lee Sedol played a little conservatively and "didn't take a ko" (?) when he could have.
Hope that helps you figure out what they're referring to, or maybe someone else can chime in, but that's what I remember.
Both players get the same time controls, seems fair to me
But if you're saying humans might fare better against computers in a game with a longer time control, I suspect that's true
It seems that additional time might actually work _against_ humans and for AlphaGo.
The human's strength is intuition and insight. They can look at a move and have a good understanding of strengths and weaknesses of positions by "feel" developed by long practice of the game. More time doesn't really help this much.
Another part of the game is "reading" -- playing out scenarios of response and counter-response to evaluate how strong a move is. The computer excels at this, because it can play out as many moves as its computation time allows and remember all the results with complete accuracy. Humans are slower and prone to mistakes when they do lots of reading.
So adding clock time lets the computer increase its advantage over the human in reading depth, but doesn't so much increase the human's advantage of intuition and insight.
And you can always give the computer less time than the human, but this just shows that it's stronger than you and you need to handicap it to have a chance.
Increasing the human time is not an option, since no one wants to watch 8 hour games, and fatigue could also come into play, so the only real option for a winning chance is reducing computer time/power
Correspondence go, similarly, would see fewer errors. Holding a world championship would take quite a bit longer than in chess, though (rough guess: 300-ish half-moves per game versus 100-ish half moves, the latter, I guess, with a bit more variation). That could be problematic, as a world championship in chess already takes years (curiously, some championships finished before the one started a year earlier did)
FWIW regular tournament Go games go up to 6 hours, and title games can take more than 16h and span over two days.
Unless you're not counting support systems, in case it becomes very complicated to calculate and decide exactly which energy expenses are for support systems and which are directly integral for function.
Lee Sedol is vastly more efficient than the entire AlphaGo cluster. However, while AlphaGo gains a predictable amount of power as its computing power is increased, it's not clear that one could do the same with humans. Our Go players optimize individual play, not multi-brain distributed play. What would the match look like if we trained up a bunch of humans to play Go as a team, and pitted AlphaGo against a team of humans that consume the same number of joules over the course of the match as it does?
I admit my first thought on fairness did go in the same direction of limiting energy budgets. But after reflecting on it just long enough to realise the above, I am finding myself surprisingly uninterested. It now seems to me that nothing particularly insightful would be revealed: limiting energy budget is no less arbitrary than limiting time unless the artificial opponent is expected to be capable of a range of things comparable to that expectable of an average human. Or if expectations are much lower, the artificial opponent would need to contend with drastically tighter limits to approach “fairness” – though at this time it would be guesswork how much tighter they ought to be. Either way, it is glaringly obvious that no computer would come within miles of competing.
So ultimately the fact that Go has been “broken” (in a particular sense) at all is far more interesting to me than whether the machine is competitive with the human in any more general sense. “It’s not” as the universal answer is boring.
And to digress a bit from there: From that perspective, this was ultimately a very human achievement. It was humans who chose Go as a problem to attack and it was them who picked MCTS and deep learning as the way to go. (Uh, no pun intended.) That’s not just reassuring. It’s also a framing we should keep in mind as computers become more entangled with the physical world and more autonomous.
I wonder if Lee Sedol will have an interest in studying deep learning after this =)
Especially when AlphaGo capitalized on just one suboptimal move of Lee Sedol.
In particular, I'm wondering if a computer scientist with access to the alphaGo source code and all the weights of the network could trick alphaGo in order to win games automatically (cf. the papers that show a neural net can be tricked to classify a plane as any other class).
If a human with the knowledge of the source code and the weights can do this, it is scary. Imagine a similar algorithm runs your car. An attacker that knows the source code and the weights may trick the algorithm to send your car in a wall!
Tricking the NN works in a no noise situation, also an image has many more parameters than a Go board
I'm kind of hoping for an unconventional opening in game #3. Come on tengen or 5,5... Additionally I think that might be one way of weakening the bot in that it will find its net for suggesting candidate moves less useful and lean more on unguided MCTS, but that's just a wild guess.
Though I wonder if something emerges. For example, black always wins, or, black always wins from a certain opening position.
https://youtu.be/l-GsfyVCBu0?t=2510
Especially for Lee, the whole world is looking at him. An "ordinary" human like me won't be able to make the right decisions under this pressure.
A great respect to Lee and the Developers of AlphaGo. Good Game!
There's just no useful way to "pool" human thinking power and redistribute it to where it's needed the most – all the players will simultaneously consider the same branches. At best you reduce the risk of making a silly mistake, but sitting pros are already pretty good at that.
A computer-assisted pool of humans might work. Feed each human a board state advanced by 1/2/3/N moves down the decision tree in some direction, and have them evaluate that particular sub-tree. It's a map-reduce problem!
Are Bridgewater Capital's employees Ray Dalio's focused?
https://en.wikipedia.org/wiki/Quarantine_%28Greg_Egan_novel%...
Is there a meaningful response that each one could give that would be optimal? I figure this would only stand a chance of working if the pool of humans were cloned from a 9 dan :-)
Of course, Go has way too many eligible possible moves for random people to vote on, but a large enough group of top pros might be able to do well just by voting.
Four to five expert chess players suggested moves for the world team. I feel that for any reasonable non-expert it comes down to "choose between these suggested moves" rather than "pick a move". It's also not really random human beings, since the selection of participants is self-selected and therefore much more likely to contain very good chess players.
This is mostly semantics, but anyone who doesn't read your link might get the wrong idea, so I felt like clarifying it a bit.
EDIT: clarified to what I originally meant: "end of midgame"
I'm reminded of the time when Deep Blue made a buggy move that totally threw Kasparov off because it was so counterintuitive. But it turned out to be a real mistake. If Kasparov had kept his composure, who knows what could have happened?
http://www.wired.com/2012/09/deep-blue-computer-bug/
Of course, as been stated here, AlphaGo doesn't have the issue of having to keep composure, while Lee does have that issue and pressure. But who's to say that AlphaGo won't make a similar significant mistake, and that Lee won't be able to capitalize on it to make a comeback in that particular game?
edit: admittedly, as pshc as noted, the search space probably grows smaller as the game continues.
edit 2: suddenly, his comment disappeared for some reason?
I'm not saying you're wrong, I'm just saying I don't agree with your motivation.
In any case, I personally find it more interesting to see what we can learn from AlphaGo about the opening and the midgame.
Where I wrote that a pro will blunder "now and then" I should have written "rarely" -- I don't have data to back this up but I'd guess once in every n games for some n > 30.
I'm looking at the DeepMind channel on Youtube: https://www.youtube.com/channel/UCP7jMXSY2xbc3KCAE0MHQ-A
Last night I found that on the DeepMind channel they were explaining the game a lot for beginners. This was a bit tedious for me but then this is a historic event so there'll be noobs tuning in so I can understand that, and I appreciate that the TV people are thinking of them.
Also, the AGA channel you just seem to have a split screen showing two people with headphones on talking over Skype? (Granted, one of them is Myungwan Kim, 9p, a super likeable guy - but Michael Redmond, 9p, is also super likeable.) The DeepMind channel switches between the game board and the large commentary board, with occasional shots of Lee Sedol and occasional shots of the DeepMind terminal.
Michael Redmond was reading out lots of variations and repeatedly trying to count the board and evaluate the position. I think he was trying to be as informative as he could. Occasionally he would get lost in thought as he calculated owing to the complexity. You could call it many things but uninformative is not one of them.
edit: added pro commentators names...
Yeah, I think I caught some 15 minutes of this and stopped watching, maybe it got better later? I would have expected them to the introduction for beginners before the start of the real game, and they ended up skipping commentary on what seemed like pretty key choice in the early game to do so.
> Also, the AGA channel you just seem to have a split screen showing two people with headphones on talking over Skype? (Granted, one of them is Myungwan Kim, 9p, a super likeable guy - but Michael Redmond, 9p, is also super likeable.)
There were even some technical difficulties with the skype feed on a few occasions and I wouldn't call it great quality, but on the other hand I rather prefer the KGS virtual board - on which it's quicker to show the possible alternatives and easier to point out the key stones.
I switched to KGS at intervals to see what people were saying on the mirrored board kibitz. And the DeepMind commentary did begin to irritate at times but I feel they had to cater for noobs. Once the middle game got going it was fine though. Agreed on the KGS virtual board.
Until you are close to late-game the scoring of many of the positions is quite flexible (depending on sente, and how certain local situations are resolved) so although you would may have a semi-reasonable approximations for total points, the games have been close enough that a +/- 5 point error in estimating will make it impossible to tell who is actually ahead.
That said, it would be very curious if we could get a virtual-board with AlphaGo's evaluation function used to score each position.
In future, it will be interesting to see AlphaGo playing against itself!
You try and share this story with a non-technical person and they will likely say "Well, duh..it's a computer".
Is it that the deep net works mainly as an evaluation function of the current position? I guess it does more than that right?
Chess has a fairly straight forward ranking of pieces, fairly well established ranking of piece power. A knight is more or less always a knight.
Go, you kinda make up your "pieces" from scratch and there isn't any single common ranking. This makes the problem of even evaluating the board difficult (value), which is a sub step of making your higher plans (policy).
The output of the two dNN help cut off, or prevent the sampling of the bad game states and encourage looking into the fruitful areas.
Here is a (edit: not old) new paper, https://gogameguru.com/i/2016/03/deepmind-mastering-go.pdf
I mean, deep down, Go is a perfect-information-zero-sum-discrete-2-player game.
I think a more successful pitch is showing that the strategies and skills coming from Go (human or robot) can be useful outside the game, as I argue here: http://rare-technologies.com/go_games_life/
As Radim says, of course, the intuition is only relevant when logic fails. However, no computer, including AlphaGo, has sufficient processing power to take the logic-only approach to the, uh, logical extreme (more board states than atoms in the universe, etc etc).
So both humans and computers play by a combination of logic and intuition. Surely Lee Sedol must be incredibly good at both. Perhaps AlphaGo is better than Lee Sedol at the logic part (or perhaps not, but just suppose for the sake of argument).
In this light, what is interesting is that AlphaGo is sufficiently good at intuition, a domain we might have considered uniquely human[0], to complement its ferocious logical power.
[0] I find it foolish to consider anything uniquely human, but a humanist essentialist might make such a claim.
I'm still holding out for the discovery of a proof of winning strategy like other combinatorial games. That to me is the logical extreme, not magic evaluation of all states. There are two interesting books on analyzing Go with combinatorial game theory, they found a neat system of scoring in the endgame and can create example boards with one best move that can be found by mechanically applying their system but stumps 9p players (who use whatever systems).
If software started writing bestselling novels, it would soon become a "duh that's just what computers do" matter of fact.
Here is what humans can do: When presented with pretty much any task, specified however poorly, they can design hardware and algorithms that can beat themselve at that task.
That's a decent measure of intelligence. A decent measure of creativity is coming up with tasks that make the intelligence part as interesting as possible.
But on one hand, in order to have a shot at passing the Turing Test, a computer has to first be able to understand and speak human language. And that's pretty damn hard actually.
Yes we can deploy statistical methods, deep learning or what have you for classification, feature extraction and so on, but natural language is context rich and ambiguous. NLP isn't a conscious process for us either, but our brains are capable to disambiguate effortlessly. And a big part of that is also the rich (human) experience we gain while growing up. Being able to speak natural language is the one big trait that separates us from animals. It's why we can cheat death, because our children can learn from the acquired knowledge of all of their ancestors.
This is different from playing Chess at least. With chess you can deploy smart algorithms, but in the end it's still a raw search for good moves in the space of all possible moves, while giving up on branches with a bad score. Raw search is what computers have always been good at. That's not really possible with NLP, because at least with chess your vocabulary is very limited and for calculating good moves you don't need extensive knowledge about the world we live in.
Computers will surely be able to do that in the future and that will be a major milestone towards "true AI", but for now computers cannot do it.
On the Turing Test, some features considered as being unintelligent, like the tendency for lying, or being sensitive to insults are actually evolutionary traits, that have arguably helped humans to survive. So while emulating some human traits will be counter productive, a "true AI" will be concerned with survival and as a consequence will end up doing whatever it takes.
So while passing the Turing Test may not be enough, not passing the Turing Test is a sign that the computer is unintelligent.
"im so bored this test sux", "i dunno wat are you asking me", "what if ur a bot lol"
AND THEN I WOULDN'T ACTUALLY KNOW because some people ARE this stupid
I know the kind of people you're talking about. I have a family member like that. She's not that smart, she failed her baccalaureate, she's semi-illiterate and she probably suffers from ADHD, though a lot has to do with her upbringing. But getting answers like that from such a person means that you're not asking the right questions or the incentive to answer is not there. Give a cash reward to a cash-strapped person and you will never get an answer like "im so bored this test sux".
Check out http://www.jabberwacky.com/
it jokes with you, it gives vulgar answers sometimes, etc.
it's NOT good enough to fool someone, but what if you just made it sound like it's a person with a disability? Is that still beating the Turing test? Is talking to a "five year old" still beating the Turing test?
Or does it have to be a 100 IQ adult person fluent in English? In which case that's just improving the bot a bit. To make it a funny and charming bot it would require even more effort. But it's just slowly improving the state of the art with some kind of techniques like reinforcement learning or neural networks or whatnot.
When a bot actually beats the Turing test it wouldn't be big news because it would just be a slightly better jabberwacky.
Similarly you can conceive of an AI that's smart, self-aware and intelligent just hasn't been developed to talk to humans.
The Turing test is a fine test to figure out if your AI is conversational with humans, but the OP I was replying to was suggesting it as a general AI intelligence test, it's not meant for that, and will give you both false positives & negatives.
The average person isn't impressed with computers winning at Go because they vastly underestimate the complexity and open-endedness of Go and wonder why it's really all that much more complex than chasing Pac Man through a maze like their computers were doing quite happily, and even with apparent personality, in the 1980s.
The computer was able to overcome the computational difficulty humans compensated with abstractions and strategic concepts. We still use those to understand whats going on in the game, but the computer is oblivious to them and only uses a tactical view of it.
http://xkcd.com/1425/
Closing in on nearly 600 career championship wins.
https://en.wikipedia.org/wiki/Michael_Redmond_%28Go_player%2...
http://gobase.org/information/players/?pp=Michael%20Redmond
Chris Garlock, on the other hand, doesn't add that much value to the broadcast. Maybe somebody will start a "Left Commentator" meme, just like Left Shark.
When searching the game tree, at each ply the most promising N moves are examined (as determined by the policy network) and leaves of the game tree are scored by the value network.
https://www.youtube.com/watch?v=oSbN2AveJuc
Tonight's game was beautiful. Last night's was a fighting game way too high level for me to really grasp (I have no idea how to play like that, all those straight and thin groups would make me nervous). I'm expecting Sedol to win Friday since I imagine he's going to have a great study session today, but I'm no longer confident he'll win the last two.. Still rooting for him though. :) (I also want to see AlphaGo play Ke Jie (ed: sounds like from the other submission on Ke's thoughts that may happen if Sedol is soundly defeated), and for kicks play Fan Hui again and see whether it now crushes weaker pros or is strangely biased to adopt a style just slightly stronger than who it's facing.)
Not sure if any of the games he played are available, though.
There's 10^50 atoms in the planet Earth. That's a lot.
Let's put a chess board in each of them. We'll count each possible permutation of each of the chess boards as a separate position. That's a lot, right? There's 10^50 atoms, and 10^40 positions in each chess board so that gives us 10^90 total positions.
That's a lot of positions, but we're not quite there yet.
What we do now is we shrink this planet Earth full of chess board atoms down to the size of an atom itself, and make a whole universe out of these atoms.
So each atom in the universe is a planet Earth, and each atom in this planet Earth is a separate chess board. There's 10^80 atoms in the universe, and 10^90 positions in each of these atoms.
That makes 10^170 positions in total, which is the same as a single Go board.
Chess positions: 10^40 (https://en.wikipedia.org/wiki/Shannon_number) Go positions: 10^170 (https://en.wikipedia.org/wiki/Go_and_mathematics) Atoms in the universe: 10^80 (https://en.wikipedia.org/wiki/Observable_universe#Matter_con...) Atoms in the world: 10^50 (http://education.jlab.org/qa/mathatom_05.html)
I wouldn't go with most (because I don't know about that), but many of these boards would also be either impossible to achieve (in a normal game) or illegal.
Evaluation function exists but it is not as simple as it can be for chess.
Are they not? MoGo beat pros of 9 Dan on 9x9 in 2011: https://www.lri.fr/~teytaud/mogo.html
Fuego beat a pro in 2008 using MCTS actually.
Not sure what you meant regarding MCTS, I never said anything about MCTS not being able to beat pros.
See, a chess program needs to find a lot of valid moves (see Deep Blue which won because it had stupid but extremely fast HW move generators), evaluate the moves and do a very deep search, up to 14, out of the very few alternatives. Russian chess programmers were better those times. They came up with AVL trees e.g. But hardware won.
In Go it's completely different. A move generator makes no sense at all, and a depth search of 14 neither. There are not a few alternatives, there are too many. What you need is a good overall pattern matching of areas of interest and an evaluation of those areas. And we saw that this feature outplayed Lee Sedol. Sedol couldn't quite follow in the recalculation of the areas.
Same as in chess AlphaGo learned the easy thing, that the center is more important than the corners, something Lee forgot during the game. But it's not a deep search, it's a very broad search, and very complicated evaluation function. A neural net is perfect for this function.
> whereas in Go no such function seems to exist.
It does exist. It's the neural net. It's a simple pattern recognizer, which learns over time more and more.