The whole Enlightenment idea that begat modernity was that man could learn from the universe. We could use science to learn, improve, grow & progress. And that is what we expanded the future with, jointly, with Royal Academies of Science & newly begat intellectual peerships.
AI defies these roots we have sprung from. Mankind will not have access to this artificially created universe. The future now happening is safeguarded deep inside corporate keeps. Answers will be shared but understanding will be withheld. We will not be able to study or learn about this universe.
When we say this is The Dark Timeline, it's very much because the light from this era will not expand. The progress made here is not shared, is not free to spread, and will inevitably be lost, likely replaced with newer also corporate controlled automated-answering-machines. This is an entirely asymmetric future, a springboard into the dark, replacing humans and human knowledge with insubstantial easily manipulated highly controlled unobservable non-participating un-democratic artifice. If you care about humanity & the human cause, you should fear what's happening here & detest it.
it sounds like your concept of the fear isn't AI comes and destroys us all like a lot of people fear, but rather huge corporate walls only increasing the disparity between the weak and the powerful, hoarding this stuff to themselves. I agree with this take and to me that is the most realistic. We desperately need OSS LLMs and democratic access to all of this goodness that AI will bring. I dont want black box subscriptions
My fear is that they actually succeed in creating AGI and us plebes are no longer required. Do people actually think that AGI will herald a new era for humanity like in Star Trek? No, the wealthy and powerful will just eliminate the worker class.
its definitely the darker timeline. i do think its unrealistic unless there are certain ways to protect themselves from uprisings/revolutions. unless they build a castle in the sky hovering in LEO or AGI builds sentries, that will be unlikely. i imagine the more likely scenario is systemic injustices that aren't as obvious but carefully controlled to keep the workers "happy enough." kind of similar to today's age, except there is some tiny percentage of chance to become rich, whereas in the future it will be literally zero with a thin veil of hope
To add to your point, surveillance is a key component in such a system, since "undesireble" behavior can quickly be detected and punished. The current zero-privacy interfaces are a huge step in the wrong direction.
This is my fan theory of The Terminator: that the machines never took over. They worked for billionaires and the billionaires wanted the rest of humanity to THINK the machines had taken over.
It's unlikely that OSS LLMs will ever be able to compete with corporate LLMs. I can only think of a few scenarios where this could work:
1. Someone develops a procedure for training models with distributed computing resources, including consumer CPUs and GPUs. Even this is not really guaranteed to work, since corporations will probably just buy up all the consumer GPUs, since they are cheaper on a FLOPS/$ metric.
2. One or more governments provide a lot of funding for OSS models, including being willing to pay competitive salaries for the best talent (potentially millions of dollars per year). This is unlikely for a lot of reasons. The only thing that could speed up the process enough to compete with private organizations is a major war that required AI to win. In that case, though, open source would be the least of their concern.
3. Scaling laws stop working and Moore's law catches up. In that case adding more compute won't really help and eventually even organizations with small budgets can afford to train a SOTA LLM. We can't know until we find the limit, but we haven't hit the limit of scaling laws so far, despite scaling up massively over the last few years, so I doubt we will find the limit any time soon.
4. A bunch of corporations that have no hope of reaching first place decide to combine their resources to beat OpenAI and thus prevent a monopoly. I'm not sure if there is precedent, but even if there is, that would still require a lot of coordination and resources for little direct monetary gain.
We'll see what happens, but I'm not really confident about any of these possibilities.
I think this is inherent to the technology itself. It's a black box the owners can tweak but the users cannot and the users are expected to trust it at face value.
It's the idea of going to a corporate entity's black box in order to know how to live. It's a blind trust that is truly blind.
> The future now happening is safeguarded deep inside corporate keeps.
Is it though? It seems to me the current AI revolution is much as corporate enterprise as it is hackers running models on their own computers. In that sense, it's no different from any other technology.
This doom and gloom prediction paints past scientific achievements with a rosy patina that is unjustified. Major scientific endeavours have always required funding from less than palatable sources, and while understanding may initially be withheld behind corporate moats, it cannot be contained there forever.
Previously the world was material & it was extremely hard to make something good & useful & leverage it well while denying access to it, while keeping it secret & unobservable.
But this virtual construct does not actually live in the real world. No one can poke or prod the systems, no one can see the models it was trained on.
These are dark by default, are so immaterial as to be unlike anything else mankind has experienced. And their impact & use, how they change the world is almost unlimited; global connectivity gives them unlimited access & breadth, from within the iron mountain security of their corporate keeps.
Even if the superhuman corporate bodies did decide they want to share, theres such unusual barriers to discocery & learning here. Assuming the petabytes of training models are made available, who is there that can accurately review them? Who can train their own systems? How can we experiment with other trainings, other systems? This is a megaproject, one that produces unique distinct gems. Watson's quip come back again, in a different form, "I think there is a world market that can maybe afford five AI."
The ratio of how leveragable it is versus what the world can see or ape has gone quickly towards infinity. Nothing like this exists, this is unprecedented entirely.
Oh I don't buy this "dark and immaterial" argument. There are many physical processes that seem magical to those unaware, and having access to the final product may reveal nothing about the processes used to make them.
As an example consider gunpowder. Having the finished product doesn't provide any insights into how it was made. It requires, what is now basic, but what was once, a pretty strong grasp of chemistry.
Another example is antibiotics. These life-saving drugs are almost exclusively created by for-profit corporations, yet despite over a hundred years since their invention, they remain easily accessible and some would argue too plentiful.
> Nothing like this exists, this is unprecedented entirely.
The above examples illustrate that life-changing inventions are par for course in human history and represent the cornerstones of the betterment of our species.
The early groups that had gunpowder used that power to inflict a lot of harm on groups who didn't. Same with many other technologies, and I expect AI will be the same. Except that unlike gunpowder, which was often controlled at the nation-state level, successful AI will largely be controlled by a handful of billionaires.
Exactly my point. I was trying to illustrate for the OP's benefit that this is not the first time in human history that game-changing tech was invented.
Gunpowder was controlled by kings for the longest time. No different from modern billionaires or oligarchs, heck monarchs had greater control and inflicted extensive misery. But despite all that, despite the control, the misery and the harm, do you seriously consider that we should not have invented gunpowder?
Gunpowder is an interesting case to suggest. It was already well known around 0AD that saltpeter emitted a dangerous poisonous smoke, already was clearly an interesting element anyone could go mine & put to use hurting people with. It's already overtly a mineral that, like, doesn't conflagrate, but does pretty weird interesting shit when on fire, and lots of folks figured that out.
It seems like by the time 13th century medeival Christiandom rolled around there were two main sources, Aragon & Montpellier. The limited supply indeed did probably greatly slow down the understanding & use.
And indeed, you can make saltpeter... by collecting & refining excrement. Queen Elizabeth paid 300 pounds of gold to the Germans after going to war with Spain & thereby losing access to Spain's natural extract.
The Chinese developing actual black powder is another interesting case. Known for centuries, it wasn't until 904 we had the first recorded use of it for combat. By the 11th century the Song dynasty started greatly controlling access & trade.
In general I think there's probably a ton of primitive chemistry/discover even much earlier ages could have done with access. The actual material being somewhat not easy to scare up is definitely an issue here.
I think the situations you're talking about don't apply very well to the moderm world. The Enlightenment project of building instruments & understanding natural science have built up, and today it would not be hard to figure out how to reproduce black powder. There is definitely more murkiness still with biology, with figuring out how to concot some chemical compounds, or even in material science with figuring out what conditions a certain combination of elements were brought together in. But these historical challenges rely primarily on a darker time in human history to contain the not-so-secret. Enlightenment observation won, it brought us out of that age, it gave us tools to figure things out with.
Im interested in being challenged more, but holy heck AI still feels enormously forbodingly dark & immaterial. Yes, bring able to observe a thing counts. Trying to persuade me that having zero access is OK or normal in the world doesn't sit with me well at all.
These models aren't tools that are built once and then released to be used freely. They require immense resources to stay at the cutting edge.
This is a very different scenario than something like the first PCs. Early computers could be taken apart, experimented with, and explored freely. This allowed countless creative uses to emerge and evolve.
The headline LLMs are the exact opposite of that. Every single use is checked and anything the owner dislikes is blocked. Many types of exploration are actively discouraged. Usage requires continuous payment and connectivity.
I'd say it's mostly ignorance and misunderstanding (or the result of being conned). Laypeople equate AI with some kind of sentient and powerful adversary, or get caught up in "took our jobs" rhetoric, or whatever.
And stories like this just capitalize on it to keep promoting the hype. There's nothing to see here.
My concern is stagnation. Everything about AI currently is derivative, by definition. Arguably much human endeavour is derivative also, but some is not. Why create something original for it to be sucked up by AI?
Look at music. There was a fleeting moment between the invention of recording and Napster where distributors could rent seek by controlling how recordings are used. The creative music industry has not suffered at all now that copying and sharing is free, and people still make stuff. Why would it be different for "AI" scraping content for training data? Some people may lose out but it won't be creators.
Music as a medium seems to be pretty stagnant to me, the major technological innovations re: music have enabled hobbyists to distribute their music across the planet, but by and large the artform itself hasn't seemingly changed much in 20 years even with the invention of all these music creation tools, it's now just a lot easier to make a song sound good.
I think we have essentially perfected music. We can generate any kind of sound and we can record it and play it back at will. As you said, the tools to do this are now available to everyone. The physics and the psychology of music processing, interpretation, and enjoyment are well known.
I think there still is evolution happening in music but there will be no more revolutions.
People in music are doing things that weren't even possible 20 years ago. Just off the top of my head: Live in-game concerts, twitch collaborations, music "meme-ing" with youtube and tiktok. A musician can push a button and broadcast a live performance to the world now, and with a few more button pushes incorporate other artists or their works.
Those appear to be marketing innovations to me, but I don't understand everything you listed. I claim the music itself as a medium is stagnant. I.e. the art form hasn't really changed.
I think that's a valid concern. I have a friend who's very bullish on ChatGPT and has encouraged me to use it more for programming tasks. I was working on parsing binary data stored as sound in a WAV file and ChatGPT4 could not figure out the task despite having a detailed description of the problem and input and output examples. However, it did a good job of just parsing the WAV file format so I could concentrate on the rest. In the end it generated probably 70% of the code and I did the remaining 30%. But that 70% was relatively uninteresting to the task; the remaining 30% was the interesting part.
It’s easy to dismiss those fears as the probability of creating a humanity-threatening technology is quite low, but on the other hand, the pre-1945 experts who feared that nuclear power could cause total annihilation kinda had a point...
The author seems confused about what "alignment" means in the context of AI. Either that or they think our future robot overloads are the same thing as grandchildren (culturally different)?
Some history you may not be aware of: the lesswrong sequences were originally an extended discussion between Yudkowsky and Hanson, held via blog post on overcomingbias.org.
Which isn't to say I disagree; if anything I find it a bit sad that he says things like “But even ignoring AI, our default future was never going to be aligned. […] These assumptions imply that our descendants will likely soon have very different cultures, including different values, beliefs, and behaviors. That is, they will be unaligned.” as not squelching humanity's choices was pretty explicitly an important part of alignment.
Note that the author also has a "fear of the future": A fear of a future where other people "...vote to stop this change. And then also the next big one. And so on until progress grinds to a halt."
There are many possible futures, and humanity, collectively, has some say in which come about. When someone claims we are impeding progress, we also have the right to ask "progress towards what"? The author seems to think that "progress" towards the total elimination of humanity is totally cool, so I'm not really bought into his notion of progress and I'm more than happy to see it grind to a halt.
This is pretty naive. My fear isn't that AI is going to suddenly turn into Skynet(although I think it is possible) but that AI is going to:
1. Enable more sophisticated scams and exploitation.
2. Create even more distrust in people through the use of Deepfakes and AI generated voice.
3. Put a lot of people out of work without the required safety nets to give them time to find their way through the turmoil.
4. Lead to a giant deluge of content spam that is just regurgitated AI text and art that makes the discoverability problem even harder to solve than it is now.
And from what I can tell, all 4 of those are currently happening.
You’re generalizing the particular. This is a fear of a very specific variety of future. Nobody is afraid of the a future that has a vaccine for HIV or the elimination of extreme poverty. The one where we are inundated with AI is special.
Sure, if speaking generally. I mainly take issue with the author passing it off as fear of change or fear of the worst case scenario as if it were just a phobia. The things I listed I feel are very real and actively happening.
These sound no different to me than any other transformative tech during other time frames in history. More scams [some people eventually learned, those who didn't lost out], less of the same jobs we have now [different jobs were created, those who didn't adapt or refused lost out], more noise [people learned to filter differently, those who didnt or refused lost out].
It's already happening. Numerous deepfakes have spread around online, and possibly many more that nobody even realizes are deepfakes. There are numerous stories of people losing their jobs because ChatGPT. One story comes to mind, a freelance writer who lost their biggest client because they started using ChatGPT to generate everything.
It's already happening now. It's only going to get worse as these AIs improve.
I think the bigger idea here that hanson maybe doesn't do a good job of highlighting, using your example, is that if you look at the historical examples (ice cores, old rocks, etc) of big shifts in the world climate, it's going to happen regardless of humans.
This sounds awfully handwavy. The fact that the climate changes over millennea doesn't do anything to justify climate destruction. For a small scale experiment, try cutting down all of the trees in your neighbor's yard and justifying it with "Well all trees fall down eventually".
In your example, the likely outcome depending on the jurisdiction would be some kind of fine (neglecting some kind of physical confrontation). Just because something negative would happen to me (maybe more parties as well if one includes the trees, the neighbor, anyone/thing else that tries to be involved in such), doesn't really change the longer scale. We kill plenty of things in our daily lives without a second thought with every breath we take, to favor some killing over others is arbitrary. Some will manage to survive in some form (maybe even unrecognizable to humans today) is such an end state where all trees are gone, most who are dogmatic about others trying to do so, will not.
Many eukaryote's from billions of years ago died out, and some went on to become humans. If one could imagine that those eukaryote's that died out then could ponder the future, no doubt plenty would have sought to be comforted by some kind of pause in their development for some kind of eukaryote safety net to avoid such far future couplings in the form of humans.
> If one could imagine that those eukaryote's that died out then could ponder the future, no doubt plenty would have sought to be comforted by some kind of pause in their development for some kind of eukaryote safety net to avoid such far future couplings in the form of humans.
That's the kind of short term thinking that gets eukaryotes in trouble. The heat death of the universe in 1.7e106 years means none of that ultimately matters, right?
> The heat death of the universe in 1.7e106 years means none of that ultimately matters, right?
Makes no sense for an actor that can perceive such a theory among many others that they cannot currently test, to weight that higher for any other (esp ones that they can test today despite others telling them not to, who are without any feasible means at scale to stop such).
I don't see how that's a useful discussion though. It's not really lofty futurist fearmongering, it's uncertainty about a very real, currently happening event in history. It would be like dismissing someone in the US for worrying about WW2 in 1939. Yeah, it might not impact the US, but also it's not immaterial daydreaming either. It is something you can reason about and prepare some plans for.
Cool, so any consideration of the downsides of technology is just "fear of the future"? We're just suppose to terminate all thinking and mindlessly charge ahead?
This is a small tweak in a search engine's algo that /detects/ AI and downranks it. Only organic human-crafted content should be allowed to surface in a search engine.
You’re eliding how hard that detection bit is. As of this moment, it’s effectively not possible to do accurately, and there’s nothing visible on the horizon.
It will be a constant arms race though and I don't see how the detection side could possibly keep up. The generative models will continue evolving and propogating. Any detectable pattern in one model won't be reliable for other cases.
About the AI art aspect, having just experienced this week what's probably the greatest hour of TV ever filmed until now, certainly on a personal level, in Succession's episode 3, season 4, I am fairly certain AI tools without agency will never be able to achieve such levels because art is fundamentally about humans, the manner of depiction is secondary: art has been impressive from Lascaux to Giotto to Stable Diffusion, the point however is the transference of feelings, from the artist to the viewer, and for transference to be achieved the artist must have feelings to begin with.
This is not to say that great artists of the future won't use AI tools for shaping their vision. And if AI goes beyond being a mere transformer of input and starts having agency, well, we're not there yet at least.
Sure, "feeling" is rather ineffable and plenty of ink and bits have flown trying to fixate it. Taking it down a notch, you can probably put it as a very strong dichotomy like-dislike. I know my dog has some proto-feelings because he really likes belly rubs and he really dislikes pulling his tail. Nevertheless, there is probably something akin to the Weber–Fechner law [1] through which "feelings" will be disambiguated. And when that happens, we might discover it was us who have been artificial from the start: biochemistry being merely ridiculously advanced technology, indistinguishable from magic as the saying goes, by comparison with our rusty transistors.
>3. Put a lot of people out of work without the required safety nets to give them time to find their way through the turmoil.
I agree with this, but I also think the risk is less that lots of people will be "successfully" replaced by AI systems, and more that management will force the adoption of models as cost cutting measures. I wrote at some point that people shouldn't be afraid that AI will do their job, but that managers will think that it can because they don't understand the exigencies of the job or the limitations of the model.
Yeah I think we'll see a lot of Good Enough (to cut salary expense) replacements rolled out, which is even more disturbing. The middle class will shrink and in exchange we'll get crappier products.
To your point #1. The first conviction for printing counterfeit money happened only 3 months after the introduction of the US dollar. Scams always follow new technology. I see no reason why now should be different.
You can find stories online about content writers and what not being replaced but ones that I personally know of are a lot of minor art roles in game development being replaced. Things like icon art for example that would usually be outsourced to a contracting studio. I will give the disclaimer that I don't work in game development but I'm in the PNW so I know a lot of people who work at studios.
To add to this, the fact that AI tools are black boxes that not even the creators can fully understand. This leads to a "trust us" mentality where actual answers about why something is can never actually be answered.
The TV show Humans is what, 2 years old now, and one of the main characters was worried they were not needed by the future because robots would take all the jobs.
Now IRL, there already is a steady stream of anxiety & angst, regular questions of "why would I bother becoming a coder anymore, since all the jobs will be wiped out by AI?"
To me it's an indicator we overly mythologize these things we cannot know. We not just believe the hype, but we amplify it enormously.
After Zuboff's Age of Survelliance Capitalism, Cory Doctorow's put out a response position that while the book is great, it 100% accepts every capability these folks say they have to completely reprogram human behavior, and it takes it further.
This tendency to assume technologies have high oracular powers, to- even when we are critiquing (Zuboff)- assume the system is at least if not vastly impactful & powerful than the claims... it's a savage problem for society, a nature of people & how we view the unobservable unexplained world as mystical & overaweing. How we dampen the hysteria is beyond my guessing, atm.
I fear AI because, in our cargo cult industry, people will start optimizing for AI as opposed to optimizing AI for us. Example: chatGPT can write semi-functional Python code, but chatGPT cannot write semi-functional ReasonML code. Prediction - ReasonML will be dead in 6 months or so.
I'm not personally on the Eliezer end of being an AI hawk, but I dislike the type of arguments Hanson uses here and find them boring & shallow. Humanity has a long history of inventing & then subsequently regulating very dangerous things- nuclear weapons, explosives in general, military-grade rifles (well more in other countries....), a vast array of narcotics (the morphine/opium/heroin/opioids/fentanyl family, the meth/amphetamine family, etc.), thalomide, antibiotics.... Boring things we don't think about are regulated for safety (automobiles, stoves, building construction, bridges, phone chargers, knives, bedframes, toasters). I mean there are safety regulations & strict requirements around, I don't know, leasing a crane? An 18-wheeler truck? Why would it be reasonable to regulate cranes and trucks but not AI?
Anyways I think it's uncharitable on Hanson's part (and numerous others online) to just call people with reasonable AI concerns Luddites. Sometimes they try to present their argument as being historically savvy ("people were afraid of the radio and video games when they came out!"), but a historical look at Dangerous Things Humanity Invented sure shows a lot of safety regulation too. I think that's just as reasonable a historical frame to view this new field
It’s easy to say this, but I think AI is very different from nuclear weapons.
It’s hard to make nuclear weapons and there little profitability for businesses to get involved.
AI on the other hand has huge profitability potential so even if governments wanted to regulate things, it’s a losing game. Hundreds of countries with different governments and different incentives. Businesses can easily move between countries.
christ on his throne ive never been gaslit so hard from a single author.
"Most people have always feared change. And if they had really understood what changes were coming, most probably would have voted against most changes we’ve seen."
what we constitute as AI these days is functionally indistinguishable from the digitized autism of a tiktok and reddit binge. The real concerns --from people who really understand what changes are coming-- include the closed and questionless nature of a dichotomy of "thought leading" products that have been ingested at lightspeed by the largest FAANG players with seemingly no review process in a desperate bid to remain relevant in 2023. it includes serious concerns that this could be used to impersonate legitimate people, wreak havock in courts, indict or frame someone for crimes, and bypass flimsy articles of biometrics like voice.
the issue isnt that we dont know whats coming, the issue is that whats coming doesnt seem to have any serious safeguard or checkvalves because its billed as an AI that is "friendly" despite having already offered solutions to problems that are either irreverant or suicidal at best.
In general I'm extremely optimistic about the near future. I think practical fusion power is likely to happen by the early 2030s, SpaceX is going to drop the cost to LEO below $100/kg within the decade, all sorts of fantastic medical advances will arrive soon including printed organs and the beginnings of workable anti-aging, and that all this will have overwhelmingly positive effects overall.
But I still pay attention to AI safety arguments and find them worrisome, and I'm not sanguine about being replaced by "descendants made of metal and math."
I don't think it's fair to be dismissive of AI fears by equating it with fear of future because what matters isn't change but the RATE of change
The rate of change is the difference between slowing for a red light and crashing into a brick wall. In both cases there is a force acting on the front of your skull but only in the latter is it accelerated into the back of your skull.
Many serious, pro-technology scientists, philosophers, and science fiction writers have been publicly worried about AI for decades and people are now acting like they're all suddenly fools or have nefarious motivations. Here's a thought: those who are uncritical of AI have far more of an incentive to ignore and downplay the risks!
You would think 3 years into the fallout of a possibly lab-created virus (or, if you don't believe that, certainly a poorly-managed response to a natural one) people would be a little, I don't know, concerned about human institutions' ability to handle anything at all that is massive and consequential in scope?
I would understand if AI proponents were putting out material that demonstrated they take critics' concerns seriously. But all I'm seeing in the opposite: a flippant disregard, as if we're all stupid for caring. Forgive me if my confidence is not through the roof right now.
A sensible approach, which is disappearing from LLM-related HN threads. (Going away to a designated area, I hope.)
But don't you see that you are preaching to the same set of outdated "human institutions"? HN is also a part of it (if you have perused it in the early 2020).
For me my biggest worry is about jobs. Taking the current trajectory of trends in AI it seems we need bigger compute and better data. Only big players seem to be able to train and host such models. So as far as new jobs being because of AI, I don't think many will be. With the AI systems that we have now I don't think it's possible to automate a job completely but AI seems to be only getting better
The Skynet Terminator future seems unlikely at this point. The fear shouldn't be some future in which AI takes over and enslaves humanity. The real fear is how these AIs will be used by nefarious actors. Phishing attacks. Misinformation. Deepfakes. These are the threats of AI that exist today. There's literally another thread on the Hacker News frontpage right now titled "Joe Rogan Issues Warning After AI-Generated Version of His Podcast Surfaces." Or take this story:
The real fear of future should really be that AI is only going to improve and get better at running phishing attacks, and generating deepfakes and misinformation. It's already happening today, and these problems are only going to get worse.
Let's talk about optimism of future. To all of you who keep talking about "hindering progress", what utopian future do you hope we're racing towards? How is AI really going to improve our lives? And is it really worth the risk? Or maybe it's just people hoping to get rich off the hype train, in the same way as cryptocurrency...
I'm not afraid of the future in general, I'm afraid of a specific vision of the future. One that holds that workers are not entitled to the ease or value gained from automation, that requires everyone work to live with dignity and also demands the right to decide how many may work, and who.
It's not the AI it's the consequences of owners applying it to labor. It will be used to further disempower workers, increase their precarity and desperation, continue and accelerate the concentration of wealth and power to a small coterie of owners.
I'm also worried about the impact of inevitably applying it to high impact decisions that affect individuals' lives. Sentencing & parole, school & job applications, loans approval & rates, police surveillance, drone assassinations. These are already decisions that disproportionately create and reproduce inequality, violence, and immiseration in our society, we don't do them well. AI will not do better, but it will ossify our divisions and grant the same old results a veneer of quantitative legitimacy. The small daily atrocities will become even harder to stop, appeal, or even see.
That's what I'm afraid of. Framing it as a general abstract fear of "the future" paints us as immature, irrational and cowardly. But what worries me about AI is that it won't be the future, it will just be the same mistakes and horrors we're making now, but even more efficient and impersonal.
I've never felt more "conservative" than when probing my feelings of fear around the future in light of AI. It makes me so aware of how often my visions of the future were largely based around managing the known uncertainties I can see today as opposed to true radical change. The latter, uniquely, implies a threat to that which I've invested my life in to so far, a threat to the payout I've delayed for so long.
It's an odd and uncomfortable feeling.
My synthesis has lately been that these changes are, in some sense, inevitable. It's not that we couldn't stop this particular one, but instead that the practice implied by that, which Hanson abridges to "progress grinds to a halt" doesn't seem practical. Even if we align AI, it's existence is potent indication that we are not yet masters of our universe and can still, even in all of our modern enlightenment, be blindsided and surprised.
So I begin to wonder what it's like to manage future unknown unknowns as a whole human. Not just a founder or a company, but as someone who values family, culture, and legacy and their place in the both immediate and more distant future.
It leads me to see AI as something more like our children than as this exogenous threat. It doesn't hurt that it speaks the way we do, too, of course. We're all nervous parents trying to read the tea leaves, but the actual activities that help might be better oriented around acceptance and guidance as opposed to resistance and conservatism.
I feel like Hanson is also speaking to this to a degree.
Most of the people I talk to directly about this don't seem to be too concerned about AI replacing their jobs. But, when I look online, I see a bunch of journalists talking about it as if it's going to replace all of our jobs.
Most of those stories lack a concern about AI replacing the journalist's job, yet that is the most likely common job I see where AI might be able to do quite well, given people were willing to talk to the AI about a particular topic.
One thing to consider is that journalism is research of documents + discussions with people + the ability for the story to be interesting and viral in nature (so it does well in the "news" ecosystem).
I can see where someone might build a system that ingests document related to topics of interest (which are pulled from discussions online) and then have a way for people to come in and discuss those documents interactively (like in a forum) with the end product then produced that is a "report" on the topic of interest.
All fear is fear of some future, so this argument is empty calories.
What I fear about AI is mostly how efficiently and effectively it will be deployed to disable people and society. It is a dream tool for grifters, despots, and capitalists who want to spread oil and sharp tacks on the road behind them.
As a tester, I stand for responsible engineering. Now there are people who say no need for testing, because AI itself will be responsible for that. Baloney!
My fear is not some vague fear of change, anymore than not wanting to drive off a cliff is a fear of change.
85 comments
[ 2.9 ms ] story [ 163 ms ] threadAI defies these roots we have sprung from. Mankind will not have access to this artificially created universe. The future now happening is safeguarded deep inside corporate keeps. Answers will be shared but understanding will be withheld. We will not be able to study or learn about this universe.
When we say this is The Dark Timeline, it's very much because the light from this era will not expand. The progress made here is not shared, is not free to spread, and will inevitably be lost, likely replaced with newer also corporate controlled automated-answering-machines. This is an entirely asymmetric future, a springboard into the dark, replacing humans and human knowledge with insubstantial easily manipulated highly controlled unobservable non-participating un-democratic artifice. If you care about humanity & the human cause, you should fear what's happening here & detest it.
1. Someone develops a procedure for training models with distributed computing resources, including consumer CPUs and GPUs. Even this is not really guaranteed to work, since corporations will probably just buy up all the consumer GPUs, since they are cheaper on a FLOPS/$ metric.
2. One or more governments provide a lot of funding for OSS models, including being willing to pay competitive salaries for the best talent (potentially millions of dollars per year). This is unlikely for a lot of reasons. The only thing that could speed up the process enough to compete with private organizations is a major war that required AI to win. In that case, though, open source would be the least of their concern.
3. Scaling laws stop working and Moore's law catches up. In that case adding more compute won't really help and eventually even organizations with small budgets can afford to train a SOTA LLM. We can't know until we find the limit, but we haven't hit the limit of scaling laws so far, despite scaling up massively over the last few years, so I doubt we will find the limit any time soon.
4. A bunch of corporations that have no hope of reaching first place decide to combine their resources to beat OpenAI and thus prevent a monopoly. I'm not sure if there is precedent, but even if there is, that would still require a lot of coordination and resources for little direct monetary gain.
We'll see what happens, but I'm not really confident about any of these possibilities.
It's the idea of going to a corporate entity's black box in order to know how to live. It's a blind trust that is truly blind.
Is it though? It seems to me the current AI revolution is much as corporate enterprise as it is hackers running models on their own computers. In that sense, it's no different from any other technology.
But this virtual construct does not actually live in the real world. No one can poke or prod the systems, no one can see the models it was trained on.
These are dark by default, are so immaterial as to be unlike anything else mankind has experienced. And their impact & use, how they change the world is almost unlimited; global connectivity gives them unlimited access & breadth, from within the iron mountain security of their corporate keeps.
Even if the superhuman corporate bodies did decide they want to share, theres such unusual barriers to discocery & learning here. Assuming the petabytes of training models are made available, who is there that can accurately review them? Who can train their own systems? How can we experiment with other trainings, other systems? This is a megaproject, one that produces unique distinct gems. Watson's quip come back again, in a different form, "I think there is a world market that can maybe afford five AI."
The ratio of how leveragable it is versus what the world can see or ape has gone quickly towards infinity. Nothing like this exists, this is unprecedented entirely.
As an example consider gunpowder. Having the finished product doesn't provide any insights into how it was made. It requires, what is now basic, but what was once, a pretty strong grasp of chemistry.
Another example is antibiotics. These life-saving drugs are almost exclusively created by for-profit corporations, yet despite over a hundred years since their invention, they remain easily accessible and some would argue too plentiful.
> Nothing like this exists, this is unprecedented entirely.
The above examples illustrate that life-changing inventions are par for course in human history and represent the cornerstones of the betterment of our species.
Gunpowder was controlled by kings for the longest time. No different from modern billionaires or oligarchs, heck monarchs had greater control and inflicted extensive misery. But despite all that, despite the control, the misery and the harm, do you seriously consider that we should not have invented gunpowder?
It seems like by the time 13th century medeival Christiandom rolled around there were two main sources, Aragon & Montpellier. The limited supply indeed did probably greatly slow down the understanding & use.
And indeed, you can make saltpeter... by collecting & refining excrement. Queen Elizabeth paid 300 pounds of gold to the Germans after going to war with Spain & thereby losing access to Spain's natural extract.
The Chinese developing actual black powder is another interesting case. Known for centuries, it wasn't until 904 we had the first recorded use of it for combat. By the 11th century the Song dynasty started greatly controlling access & trade.
In general I think there's probably a ton of primitive chemistry/discover even much earlier ages could have done with access. The actual material being somewhat not easy to scare up is definitely an issue here.
I think the situations you're talking about don't apply very well to the moderm world. The Enlightenment project of building instruments & understanding natural science have built up, and today it would not be hard to figure out how to reproduce black powder. There is definitely more murkiness still with biology, with figuring out how to concot some chemical compounds, or even in material science with figuring out what conditions a certain combination of elements were brought together in. But these historical challenges rely primarily on a darker time in human history to contain the not-so-secret. Enlightenment observation won, it brought us out of that age, it gave us tools to figure things out with.
Im interested in being challenged more, but holy heck AI still feels enormously forbodingly dark & immaterial. Yes, bring able to observe a thing counts. Trying to persuade me that having zero access is OK or normal in the world doesn't sit with me well at all.
This is a very different scenario than something like the first PCs. Early computers could be taken apart, experimented with, and explored freely. This allowed countless creative uses to emerge and evolve.
The headline LLMs are the exact opposite of that. Every single use is checked and anything the owner dislikes is blocked. Many types of exploration are actively discouraged. Usage requires continuous payment and connectivity.
And stories like this just capitalize on it to keep promoting the hype. There's nothing to see here.
Except it's not rhetoric or "whatever" to them. The threat is quite real.
I think there still is evolution happening in music but there will be no more revolutions.
Which isn't to say I disagree; if anything I find it a bit sad that he says things like “But even ignoring AI, our default future was never going to be aligned. […] These assumptions imply that our descendants will likely soon have very different cultures, including different values, beliefs, and behaviors. That is, they will be unaligned.” as not squelching humanity's choices was pretty explicitly an important part of alignment.
See https://www.lesswrong.com/posts/HawFh7RvDM4RyoJ2d/three-worl..., which was originally posted on overcomingbias (see https://web.archive.org/web/20090608073156/http://robinhanso... ) many years ago now.
There are many possible futures, and humanity, collectively, has some say in which come about. When someone claims we are impeding progress, we also have the right to ask "progress towards what"? The author seems to think that "progress" towards the total elimination of humanity is totally cool, so I'm not really bought into his notion of progress and I'm more than happy to see it grind to a halt.
1. Enable more sophisticated scams and exploitation.
2. Create even more distrust in people through the use of Deepfakes and AI generated voice.
3. Put a lot of people out of work without the required safety nets to give them time to find their way through the turmoil.
4. Lead to a giant deluge of content spam that is just regurgitated AI text and art that makes the discoverability problem even harder to solve than it is now.
And from what I can tell, all 4 of those are currently happening.
all fears are fear of the future when you apply enough abstraction.
It's already happening now. It's only going to get worse as these AIs improve.
Many eukaryote's from billions of years ago died out, and some went on to become humans. If one could imagine that those eukaryote's that died out then could ponder the future, no doubt plenty would have sought to be comforted by some kind of pause in their development for some kind of eukaryote safety net to avoid such far future couplings in the form of humans.
That's the kind of short term thinking that gets eukaryotes in trouble. The heat death of the universe in 1.7e106 years means none of that ultimately matters, right?
Makes no sense for an actor that can perceive such a theory among many others that they cannot currently test, to weight that higher for any other (esp ones that they can test today despite others telling them not to, who are without any feasible means at scale to stop such).
This is a small tweak in a search engine's algo that /detects/ AI and downranks it. Only organic human-crafted content should be allowed to surface in a search engine.
This is not to say that great artists of the future won't use AI tools for shaping their vision. And if AI goes beyond being a mere transformer of input and starts having agency, well, we're not there yet at least.
This is an interesting concept since researchers really aren’t sure how consciousness in humans and animals even happens. We just don’t know.
Feelings could be theoretically be possible in artificial intelligence. We really don’t know enough about the subject to say either way.
[1] https://en.wikipedia.org/wiki/Weber%E2%80%93Fechner_law
I agree with this, but I also think the risk is less that lots of people will be "successfully" replaced by AI systems, and more that management will force the adoption of models as cost cutting measures. I wrote at some point that people shouldn't be afraid that AI will do their job, but that managers will think that it can because they don't understand the exigencies of the job or the limitations of the model.
Not about to be but currently are…
Now IRL, there already is a steady stream of anxiety & angst, regular questions of "why would I bother becoming a coder anymore, since all the jobs will be wiped out by AI?"
To me it's an indicator we overly mythologize these things we cannot know. We not just believe the hype, but we amplify it enormously.
After Zuboff's Age of Survelliance Capitalism, Cory Doctorow's put out a response position that while the book is great, it 100% accepts every capability these folks say they have to completely reprogram human behavior, and it takes it further.
This tendency to assume technologies have high oracular powers, to- even when we are critiquing (Zuboff)- assume the system is at least if not vastly impactful & powerful than the claims... it's a savage problem for society, a nature of people & how we view the unobservable unexplained world as mystical & overaweing. How we dampen the hysteria is beyond my guessing, atm.
Anyways I think it's uncharitable on Hanson's part (and numerous others online) to just call people with reasonable AI concerns Luddites. Sometimes they try to present their argument as being historically savvy ("people were afraid of the radio and video games when they came out!"), but a historical look at Dangerous Things Humanity Invented sure shows a lot of safety regulation too. I think that's just as reasonable a historical frame to view this new field
It’s hard to make nuclear weapons and there little profitability for businesses to get involved.
AI on the other hand has huge profitability potential so even if governments wanted to regulate things, it’s a losing game. Hundreds of countries with different governments and different incentives. Businesses can easily move between countries.
"Most people have always feared change. And if they had really understood what changes were coming, most probably would have voted against most changes we’ve seen."
what we constitute as AI these days is functionally indistinguishable from the digitized autism of a tiktok and reddit binge. The real concerns --from people who really understand what changes are coming-- include the closed and questionless nature of a dichotomy of "thought leading" products that have been ingested at lightspeed by the largest FAANG players with seemingly no review process in a desperate bid to remain relevant in 2023. it includes serious concerns that this could be used to impersonate legitimate people, wreak havock in courts, indict or frame someone for crimes, and bypass flimsy articles of biometrics like voice.
the issue isnt that we dont know whats coming, the issue is that whats coming doesnt seem to have any serious safeguard or checkvalves because its billed as an AI that is "friendly" despite having already offered solutions to problems that are either irreverant or suicidal at best.
But I still pay attention to AI safety arguments and find them worrisome, and I'm not sanguine about being replaced by "descendants made of metal and math."
The rate of change is the difference between slowing for a red light and crashing into a brick wall. In both cases there is a force acting on the front of your skull but only in the latter is it accelerated into the back of your skull.
And also direction of change. Not all change is for the better.
There is not one possible future, there are many; but only one will happen. We have some agency in determining its shape.
Some of the possible futures are dystopian, and we should be afraid of them, and that fear should prompt us to turn away from them.
You would think 3 years into the fallout of a possibly lab-created virus (or, if you don't believe that, certainly a poorly-managed response to a natural one) people would be a little, I don't know, concerned about human institutions' ability to handle anything at all that is massive and consequential in scope?
I would understand if AI proponents were putting out material that demonstrated they take critics' concerns seriously. But all I'm seeing in the opposite: a flippant disregard, as if we're all stupid for caring. Forgive me if my confidence is not through the roof right now.
But don't you see that you are preaching to the same set of outdated "human institutions"? HN is also a part of it (if you have perused it in the early 2020).
https://www.wired.com/story/large-language-model-phishing-sc...
The real fear of future should really be that AI is only going to improve and get better at running phishing attacks, and generating deepfakes and misinformation. It's already happening today, and these problems are only going to get worse.
Let's talk about optimism of future. To all of you who keep talking about "hindering progress", what utopian future do you hope we're racing towards? How is AI really going to improve our lives? And is it really worth the risk? Or maybe it's just people hoping to get rich off the hype train, in the same way as cryptocurrency...
It's not the AI it's the consequences of owners applying it to labor. It will be used to further disempower workers, increase their precarity and desperation, continue and accelerate the concentration of wealth and power to a small coterie of owners.
I'm also worried about the impact of inevitably applying it to high impact decisions that affect individuals' lives. Sentencing & parole, school & job applications, loans approval & rates, police surveillance, drone assassinations. These are already decisions that disproportionately create and reproduce inequality, violence, and immiseration in our society, we don't do them well. AI will not do better, but it will ossify our divisions and grant the same old results a veneer of quantitative legitimacy. The small daily atrocities will become even harder to stop, appeal, or even see.
That's what I'm afraid of. Framing it as a general abstract fear of "the future" paints us as immature, irrational and cowardly. But what worries me about AI is that it won't be the future, it will just be the same mistakes and horrors we're making now, but even more efficient and impersonal.
It's an odd and uncomfortable feeling.
My synthesis has lately been that these changes are, in some sense, inevitable. It's not that we couldn't stop this particular one, but instead that the practice implied by that, which Hanson abridges to "progress grinds to a halt" doesn't seem practical. Even if we align AI, it's existence is potent indication that we are not yet masters of our universe and can still, even in all of our modern enlightenment, be blindsided and surprised.
So I begin to wonder what it's like to manage future unknown unknowns as a whole human. Not just a founder or a company, but as someone who values family, culture, and legacy and their place in the both immediate and more distant future.
It leads me to see AI as something more like our children than as this exogenous threat. It doesn't hurt that it speaks the way we do, too, of course. We're all nervous parents trying to read the tea leaves, but the actual activities that help might be better oriented around acceptance and guidance as opposed to resistance and conservatism.
I feel like Hanson is also speaking to this to a degree.
Most of those stories lack a concern about AI replacing the journalist's job, yet that is the most likely common job I see where AI might be able to do quite well, given people were willing to talk to the AI about a particular topic.
One thing to consider is that journalism is research of documents + discussions with people + the ability for the story to be interesting and viral in nature (so it does well in the "news" ecosystem).
I can see where someone might build a system that ingests document related to topics of interest (which are pulled from discussions online) and then have a way for people to come in and discuss those documents interactively (like in a forum) with the end product then produced that is a "report" on the topic of interest.
What I fear about AI is mostly how efficiently and effectively it will be deployed to disable people and society. It is a dream tool for grifters, despots, and capitalists who want to spread oil and sharp tacks on the road behind them.
As a tester, I stand for responsible engineering. Now there are people who say no need for testing, because AI itself will be responsible for that. Baloney!
My fear is not some vague fear of change, anymore than not wanting to drive off a cliff is a fear of change.
prompt: Begin.
The enabling of generative fakes, automated actions, and replication of human-like works in little time are the main concerns.