The AI explosion might never happen
amistrongeryet.substack.com
amistrongeryet.substack.com
I dunno, I mean, humanity has made some serious changes in a very short time. We're pretty lucky that the climate's sensitivity to CO2 isn't worse than it is, or we could have taken ourselves out already. Same as if we had had a nuclear war.
And AGI doesn't need to completely destroy the earth to be really bad for humans. Just taking over a lot of the resources we need would do the trick.
> It’s conceivable that some sort of complexity principle makes it increasingly difficult to increase raw intelligence much beyond the human level, as the number of facts to keep in mind and the subtlety of the connections to be made increases.
There's a whole lot of "it's conceivable" in here, which seems to me to be a bit of a coping strategy. For humans, the problem with our biology is that our heads kinda have a physical limit on their size.
The idea that not only is it possible to make machines that are stronger than us, tougher than us, more precise than us, and faster than us, but also just straight up smarter than us in the general sense (instead of just at math or chess or whatever), is not outlandish at all. They might be less efficient and take a whole lot more power, but that sort of thing hasn't stopped us before. We just find ways to get them more power. I accused the author of trying to cope with fear of the development of super-intelligence by figuring out happy scenarios where it will be harder than we think to create. I have done the same thing, but I think we do ourselves a disservice by not facing the potential for this to be a real problem head-on.
> And AGI doesn't need to completely destroy the earth to be really bad for humans. Just taking over a lot of the resources we need would do the trick.
Agreed. I am not arguing that AI could not become superhuman, or could not overwhelm us. I'm merely arguing that:
1. It is not guaranteed that mildly superhuman AGI would inexorably lead to a runaway feedback loop of capabilities increase. I'd agree that it's possible, but I often see statements that it is inevitable, because an AI smarter than us would be able to create an AI smarter than we can. Such statements fail to take into account that the sequence of successive AIs might converge (at mildly superhuman) rather than diverging (to a singularity).
2. Even if AI capabilities diverge, there's no guarantee that would happen quickly ("foom"), because even as AI capabilities increase, the effort needed to achieve each further increment in capability will almost certainly also be increasing.
I do take the potential for superintelligence very seriously; see, for instance, https://amistrongeryet.substack.com/p/get-ready-for-ai-to-ou.... But I also think that some current discussion is going overboard in the other direction, and waving away likely hurdles on the path to superintelligence (see https://amistrongeryet.substack.com/p/the-ai-progress-parado...).
All self-contained with very little human oversight, similar to how humans can reproduce, pass on genes, and then do it all over again until we evolve new traits, AI systems could do the very same thing.
I can see this scenario where language models are evolving weekly or even doubling in abilities, is possible this decade.
Running out of materials? AI drones can setup bases on the moon (less gravity issues) to go out and mine asteroids, bring it back to the moon to build more processing power. The entire moon could be turned into one mega structure super computer that could house an ai that helps manage all the things humanity needs it for. Traversing into scifi territory, but I thought where we are now was nearly impossible by now. LLM's and Stable Diffusion, and coming multi-modal modals are going to affect every aspect of life, one way or another whether it gets to full-super-intelligence doesn't matter, it will still change everything.
The moon lacks a lot of rare earth minerals required to make modern transistors, for example. The AI would have to set up shop in asteroid belt or Kuiper belt or Oort cloud, and except for the last one it limits available energy by quite a lot.
We don't currently have completely automated computer factories. Large chunks of manufacturing are automated, sure, but a) those automated segments are monitored and controlled by humans, and b) there are also significant parts of the assembly process that are fully manual.
Does that mean that fully automated construction is impossible? Of course not. But it does mean it's not something we are certain we can do today, without additional genuine breakthroughs.
Furthermore, what you're proposing is not merely automated construction, but automated improvements to the construction process. That means you need to be able to reconfigure the entire manufacturing process without human intervention. I feel reasonably confident in saying this is not feasible with our current technology.
Beyond that, your very first comment—about needing the means—is very much nontrivial. I don't recall the exact figures, but I was definitely seeing stories about ChatGPT being massively expensive to run, both in terms of money and in terms of energy. And that's just a current-generation LLM. Attempting to bootstrap from there to....the singularity, I guess? is likely to take more energy than is feasible to dedicate to any such project even if it were possible.
And finally, you say the AI will "train future generations"—train on what? An LLM's quality is always going to be largely dictated by its training data. How is an LLM going to be able to train a next-generation LLM any better than it was trained, especially, again, without human intervention? It's nearly impossible for what you're describing to result in anything that can do useful work, simply because training like that requires humans in the loop giving feedback at every step.
There don't seem to be any arguments beyond the trivial observation that things could slow down. But they haven't slowed down.
New models keep surpassing us (in some cases the whole human race) dramatically in new areas of greater generality, while their "weaknesses" also improve dramatically.
It's hard to imagine where a fully multi-modal model, with long context, and a sense of information confidence, and ability to manage its own notes/whiteboard information, will not exceed us. On top of improving in all the areas it already outdoes us.
People are working on each of those improvements right now. (By fully multi-modal I mean text, audio, image, video, simulated and real physics, touch, motor control, team communication, software/internet access, and whatever other senses or decision forms are helpful.)
I don't see a general AI vastly smarter than any one of us, or all of us together, taking longer than 2030-2033 time frame.
But...they have.
How many stories have there been about how GPT-3 and GPT-4 have gotten worse over the period they've been out?
And it's not like we got GPT-3, and then GPT-4, and then, within the same time frame, GPT-5 with a similar increase in quality.
Sure, people are working on more improvements, but they're not here yet, and while that doesn't mean they will never come, it does mean that, compared to what appeared to be a rapid rush of "AI" progress, things have slowed down.
"Progress is still being made" and "progress has slowed down compared to the speed that generated all the hype" are not incompatible statements.
They are experimenting with trade offs. In this case, things like appropriateness, less overconfidence, etc.
That isn’t indicative of any slowdown.
> Sure, people are working on more improvements, but they're not here yet, and while that doesn't mean they will never come, it does mean that, compared to what appeared to be a rapid rush of "AI" progress, things have slowed down.
I don’t follow.
People and teams keep publishing and releasing new techniques but big projects don’t release major updates as quickly.
How are your arguments more relevant to GPT4 than they would have been for GPT3? Earlier models?
Yes, people are still working on stuff. I never said they weren't. I never said they were working any less hard than they had been.
But progress has slowed down from the leaps that it took over the past couple of years to get us to GPT-3 and GPT-4. Whatever "new models" are doing in September 2023, they are absolutely not making the same clear advances that we saw previously. They are incremental improvements. Which is good! It's important! It's progress! But it's unquestionably slower than the breakthrough that led to this generation of LLMs.
Now, it may be that that wasn't what you were thinking of when you said things haven't slowed down, but that was absolutely not clear from your post. And there have been enough people trumpeting loudly that the pace of progress would continue exactly as fast as it was before—breakthrough after breakthrough, leading rapidly to GPT-5 and beyond, and causing millions of white-collar jobs to be automated—that to simply say, without qualification, "things haven't slowed down" is, at best, ignorant of the way it will be taken by many.
[emphasis mine]
GPT-4 was only released March 14, 2023 [0], so until March 14, 2025 you don’t have any data to support your claim.
Elsewhere models continue to update quickly. While we wait for GPT-5, much smaller models keep getting better.
There have been few times in history where for over a decade such a widely impactful tech has kept improving in stunning steps.
There is no evidence of any slow down.
Neither nuclear war nor a climate apocalypse is existential. Some thousands of humans would survive the worst case of either or both.
That said I agree. Human brains arose through evolution: Evolution operates within iron laws of energy efficiency, on the material it has to hand; and it only does enough to provide an advantage in production of grandchildren.
There is no a priori reason to suppose that human-level intelligence is anywhere near being a limit, absent those mechanisms.
We often talk about innovation driving innovation in discussions about AI explosions. However, we must honestly ask ourselves, "Even though it looks like innovation is the limiting factor in AI now, what other factors will be limiting in the future?"
In other words is our model wrong because it could be too simplistic? If innovation was uncapped, what would be our next limiting factor, and how much of a difference would this represent?
1. https://en.wikipedia.org/wiki/Liebig's_law_of_the_minimum
Algorithm limit? Progress is accelerating noticeably, year-to-year.
Demand? No bottleneck there. Tech demos have become accidental products, with dozens of users rising to hundreds of millions in months.
Which, for those of us who grew up in the 17th century, is pretty damn fast!! Let me tell you!
New applications? All kinds of brilliant (and pathetic!) uses springing up. Just on my Mac alone.
Commercial deep learning (neural network) software hit the road in the early 90's. People wondering when the explosion will happen apparently missed the long burning fuse phase. It's happening.
It's quite possible "good enough" AI will trap us in a local maximum of mediocrity where the near-term ROI on thinking hard is too low.
AI will also enable the collection of more data. For example, robots could do the labwork for science experiments in far greater numbers than humans could.
A massive amount of gibberish isn't _useful_ data.
Current training systems extract information on everything from every input. We don't read a cookbook to learn about space, but these models are twiddling weights all over these massive spaghetti balls of logic.
Looking at the data we have, the literal collection of all human knowledge and lamenting not having more is like standing on a ladder sideways.
But it's really starting to feel like pile of ladders is getting tall enough to bootstrap.
And these massive inefficient systems filled with data and churning matrixs will be set in order. And all the ladders laid end to end.
Could AI usher in a new era of wealth where nobody has to work? Sure, if you ignore the fact that the most likely scenario is that the value is captured by the current owning class, which is already well underway.
On the flip side of possible scenarios, could AI start improving itself and lead to an upward spiral of improvement? I can entertain that idea, but I think we'll find that most types of AI innovation won't engineer themselves outside the box that is market need and product-market fit. History has proven over and over again that innovation for innovations' sake gets a few moments in the sun before being quickly forgotten if it doesn't have actual utility.
there’s more than money and the elimination of money.
Cheaper and unconstrained scale of labor for certain types of tasks is going to mean a lot more to smaller and leaner firms than behemoths.
Within the decade I suspect that companies currently or soon cutting talent that have extensive familiarity with domain knowledge in that space will turn out to have made a very poor move for long term viability.
In general, I have a hard time seeing a number of sectors continuing to be dominated by large corporations as they are currently over the next decade as this technology improves.
----
What "fact"? The average working hours per worker have been declining (https://ourworldindata.org/grapher/annual-working-hours-per-...). I assume you have compelling evidence that this trend will reverse?
Working hours will continue to fall as you say, but inequality will only get worse. Working 30 hours instead of 40 hours won’t change how poor everyone is.
In particular, I found one of Cory Doctorow's recent pieces[0] both informative and very hopeful.
[0] https://pluralistic.net/2023/09/14/prop-22-never-again/#norm...
This is an incestuous process that does not produce useful mutations. However it may produce some non-useful ones, "defects".
NB. The eventual "derivative" works in this hypothetical are not derived from use of human language, they are derived from use of a model of human language that represents a point in time that has long passed. Under this hypothetical "future", the source of all work becomes either works created before the computer system existed or works generated with the use of the computer system. Neither the system nor its user can continue to "learn", take inspiration, or, most importantly, deviate unpredictably, from works created without the use of the system because, aside from the works that pre-date the system, such works no longer exist. The birth of new language and ideas arising from sources not created using the system, such as all the works that pre-date the system, is prevented. This is inbreeding.
Humanity isn’t much different. Countless creative works are derivative.
2) Evolution is also incremental in a way engineering doesn't need to be.
3) One could argue we're in the middle of the (biological) singularity already. It just depends on timescales we look at. As life, we've developed more technology in the past 200 years than than in the previous 3B years.
Interpolation works better than extrapolation. I don't really think we know what will happen or where we're going. The Bible, Quoran, or Vedas are as good guideposts as our own extrapolations.
It's a discussion we should be having, but the levels of confidence expressed in predictions, constraints, or dynamics are unrealistic.
We have seen an exponential rate of improvement. And this is the whole point of why people talk about runaway AI; it's relative to the exponential growth of complexity observed in software and computer hardware scaling.
Yes it's slowing now but it many years of exponential growth transforming a lot in tbe process
The assumption in the article that humans haven't been on a "runaway" improvement process could be stronger. Human history is relatively short, but process improvement has been strong.
Oral history led to written history, led to more instant communication, and in most fundamental disciplines there have been systematic process improvements. Humans are messy, but vastly improved over evolution for adapting to varied environments.
I mean, if we've really reached a point where the tools are this self-aware enough to improve, we will probably struggle to get enough power and silicon to run more, faster. I get that this is fantasy stuff, but seems reasonable on a longer timeline (say 150-200 years), no?
I'll give that a "maybe" and "to some degree".
However we had a path of positive feedback loops up to today. Moore's Law is part of this path of positive feedback loops. We have a look-behind bias. We tend to say that such a thing was destined to happen exactly as it happened.
What exactly happens does not really matter.
It is enough that something is found which is better than the rest. We see this in evolution. If a species has a mutation giving it a better chance then this species will thrive and probably displace other species.
I think that there's a chance that the same will happen with AI.
The driving force for evolution is the need to survive and the driving force for human products is the market demand. A company might stumble upon something better they can sell for cheap. Of course there are cartels, rent seeking and market barriers to entry, and if these forces are strong enough world-wide then AI explosion will be delayed for a long time or if humanity goes extinct not establish at all.
But exilees could produce something in the backwaters which can overcome the broken main markets because of sheer ingenuity and because there's demand they will be successful. Think Tesla twenty years ago.
tldr: Something is bound to happen driven by external forces like demand and competition.
That you will never succeed, because the AIs are already better. Because you are too simple. And they can calculate what you will do for all eternity. Because the AIs can simply hold out a false carrot, and then all humanity falls toward the event horizon. Wealth for the wealthy, celebrity for the celebrity. Run toward the carrot and backstab those in front of you. That you were born into Westworld.
That the botnets integrated generative AI the minute it became available. That all other beliefs are foolish. That all the corporations already had AI, and already integrated those ideas a long time ago. 30 years of "maybe AI someday"? Then they all release in a couple months? Right as everyone realizes being squared/boxed/cubed/hypercubed in offices is horrible. Pfft.
More orphans for the orphan grinder.
Locally, you are standing at the precipice of humanity bringing forth intelligence decoupled from physical embodiment. That humanity has brought forth modeling off of ourselves and has put to the task of digitally extending and reproducing ourselves.
We are advancing towards creating virtual universes of greater and greater fidelity where resources are allocated based on what's observed and interacted with and continuously determined geometry is quantified in order to track state changes by free agents.
But when we look at our own universe its macro continuous behavior collapses to quantified parts, and does so based on observation and interaction.
And buried within our collective lore is a document nearly two thousand years old called "the good news of the twin" claiming in it and its associated tradition that there was a spontaneous humanity that brought forth intelligence in light before dying out and that still living intelligence recreated the earlier cosmos and humanity within its light to effectively resurrect them independent of a body and that we're that recreation. And that the proof for this is in the study of motion and rest, and that the ability to find an indivisible point within bodies would only be possible in the non-physical.
A specificity that takes on particular eyebrow raising detail given the increasing likelihood over the past few years that the next stage for AI hardware is optoelectronic (i.e. literally light).
So yeah, maybe all this stuff happened long ago. But maybe it happened REALLY long ago and the nature of our existence isn't what it appears to be at face value.