- cross-posted to:
- technology@lemmy.ml
- cross-posted to:
- technology@lemmy.ml
cross-posted from: https://lemmy.ml/post/20858435
Will AI soon surpass the human brain? If you ask employees at OpenAI, Google DeepMind and other large tech companies, it is inevitable. However, researchers at Radboud University and other institutes show new proof that those claims are overblown and unlikely to ever come to fruition. Their findings are published in Computational Brain & Behavior today.
You must log in or register to comment.
Meh. It’s not a problem of scale. It’s a problem of we have no idea how the fuck to do that. Scaling up existing techniques is neither necessary nor sufficient.
Right on the money. One of the big things with AI safety is “we have no fucking clue how AGI can originate so we are constantly in the dark.” If we ever did create it, we likely would not immediately know it was AGI, and that creation could go very terribly in a number of ways.
Sounds really counterintuitive to say that it’s impossible.
The article says that we would run out of computing power, and that’s definitely true for current hardware and software. It’s just that they are being developed all the time, so I think we need to leave that door open. Who knows how efficient things can get within the next decade or century. The article didn’t even mention any fundamental obstacle that would make AGI completely impossible. It’s not like AGI would be violating the laws of physics.
Whenever I hear someone say that something is impossible with current technology, I think about my grandma. When she was a kid, only some important people had telephones. Doctors, police, etc.
In her lifetime we went from that to today, and, since she’s still alive, even further into the future.
Whenever someone calls something impossible, I think about how far technology will progress in my own lifetime and I know that they’ve got no idea what they’re talking about. (Unless, like you said, it’s against the laws of physics. But sometimes even then I’m not so sure, cause it’s not like we understand those entirely. )
The thing is, we have no idea where technological progress is taking us. So far, most predictions have been wrong. 50 to 60 years ago, people thought we would already be colonizing other planets by now. Barely anyone was able to predict the Internet, smartphones, social media, etc. - the kind of technology that is actually shaping our civilization’s future right now.
Another aspect that I feel is often neglected is the assumption that technological progress will continue forever or at least continue at this current rapid pace. This wasn’t true in the past and we might simply be experiencing a historical anomaly right now, one that could correct itself very soon in the future, either towards stagnation or even regression.
The space example is extremely apt. Its possible we could have had tons of space stations, a moon colony, maybe even some other stuff going on around the solar system, asteroid mining, etc. But thay would have at least required the space race to continue longer and for spending to grow to create a big enoigh industry to ensure thay outcome, assuming no capacity or time issue. Alas, we took another path.
Something that seems important to us might not matter in even 10 years, or at least, not have a monetary and/or societal incentive to keep advancing.
I was also based on the assumption that the rapid progress of aerospace technology that happened in the 1920s to 1960s would continue onward at the same pace, whereas what actually happened was that barriers emerged that nobody was able to circumvent, like for example engineering things to withstand incredibly abrasive Moon dust (or really do anything productive on that lifeless rock), how to deal with the endless pitfalls of a long Mars journey, how to bring down the cost of launch vehicles so that grand projects like giant space stations would even be remotely possible (von Braun was already thinking about huge space stations all the way back in 1945). Many of these issues couldn’t simply be solved by throwing more money at them, which is important. Deciders, both in Washington and Moscow, were smart enough to realize this in the 1970s, for the most part at least (the Space Shuttle and its Soviet clone, each a gigantic waste of money, are major counter example from this era).
The point I’m making here is that everyone assumed linear progress in this area, just like there are people currently making many billion dollar bets on linear progress in regards to computer technology in general and AI in particular, but at least, with the benefit of hindsight given past examples, there’s a reasonable amount of doubt this time around.
This wasn’t true in the past and we might simply be experiencing a historical anomaly right now
While our exact pacing might be slightly different from the pure extrapolation, human history has been a long, steady increase in the rate of invention. Access to education has meant that more people are making things, and then the next generations build on top of their work to make even bigger things.
In addition, technological development can take unexpected twists and turns. For a while, it looked like analogue technology involving gears was going to solve every problem… until transistors were developed and mechanical calculators were soon forgotten. Also, the development of fertilizers revolutionized farming and and food production, which changed the world more than anyone even realized.
That’s not an apt comparison.
More like “we’ll have flying cars 50 years from now.”
I love the flying car example because it reveals a huge issue with the whole “tech will get better” idea. People are still trying to make flying cars happen but it’s running in to the same fundamental issues; large things that are mechanically complex, energy intensive, and moving at high speeds in a crowded urban environments are just too expensive and dangerous.
There is no way around the physical realities, no clever trick or efficiency that will push it over some threshold of practicality.
Let’s put it this way: If in our lifetime we can simulate the intelligence of a vinegar fly as general intelligence, that would be a monumental landmark in AGI. And we’re far, far, far away from it.
As far as the iron age was from the metal alloys used in the Space Shuttle.
Talking about AGI simulating higher intelligence at the level of a dog or a cat, dear I say a pigeon or a crow is as far fetched as expecting ancient Egyptians to harness the power of the atom.
Let’s put it this way: If in our lifetime we can simulate the intelligence of a vinegar fly as general intelligence, that would be a monumental landmark in AGI. And we’re far, far, far away from it.
I get what you mean here and I agree with it, if we’re talking about current “AI”, which isn’t anywhere close. I know, because I’ve programmed some simple “AIs” (Mainly ML models) myself.
But your comparison to ancient egypt is somewhat lacking, considering we had the aptly named dark ages between then and now.
Lot’s of knowledge got lost all the time during humanity’s history, but ever since the printing press, and more recently the internet, came into existence, this problem has all but disappeared. As long as humanity doesn’t nuke itself back to said dark ages, I recon we aren’t that far away from AGI, or at least something close to it. Maybe not in my lifetime, but another ~2000 years seems a little extreme.
I recommend this thread btw https://www.reddit.com/r/AskHistory/comments/18ydzok/has_the_term_dark_ages_truly_become_an_obsolete/
Nothing to do with the rest of your comment.
Could take a while, but how long? Progress tends to be non-linear, so things can slow down and speed up suddenly. I’m pretty sure we’ll get there sooner or later unless we nuke ourselves to oblivion before that.
If AI development isn’t prioritized, it could take centuries. Maybe we’re still missing some crucial corner stores we haven’t even thought of yet. Just imagine what it was like to build an airplane in an age when the internal combustion engine hadn’t been invented yet. Maybe we’re still missing something that big. On the other hand, it could also be just around the corner, but I find it unlikely.
Actually, we do already know that we’re close to a theoretical limit of increasing computing power as we currently know it. The transistor can’t really get that much smaller, before it stops working.
Also, if you’re talking about the article as linked, that is a mere introduction to a much longer paper.
The fact that human brain is capable of general intelligence tells us everything we need to know about the processing power needed to run one.
Well it sets an upper bound on compute requirements at ‘simulate 10^27 atoms for thirty years’ remains to be seen if what we can optimize away ever converges with what’s feasible to build.
The article did mention a fundamental obstacle. It said quite clearly that we would run out of resources before we had enough computing power. I suppose you could counter that by arguing that we could discover magic, or magical technology, or a lot of new resources through space exploration.
Of course things get more efficient. But in the past few decades they’ve gotten efficient in predictable, and mostly predicted, ways. It’s certainly possible that totally unexpected things can happen. I could win the lottery next week. Is that the standard? Are you pushing the stance that says AGI is somewhat less likely than winning the lottery or getting struck by lightning, but by golly it’s more than zero, how dare you suggest that it’s anywhere close to zero?
It really depends on your assumptions. If you assume that software and hardware will stay at the current level, then the article does present a valid point. I would argue that those assumptions are only reasonable in the short term. AGI development does depend on some big technological changes we haven’t seen yet, so it could take decades or even a century, but I wouldn’t call it impossible.
If you assumed that 1950s style vacuum tube computers were the best thing ever, you could safely say that playing a game like fortnite with your buddies living in different countries is completely impossible. Modern semiconductors and integrated circuits would have seemed pretty magical in that context.
If we assume that we’re going to be stuck with silicon, you can safely say that AGI just isn’t going to happen with these tools and methods. Since quantum computers aren’t quite useful just yet and optical computers aren’t even in the news in any meaningful way, it seems that we will be stuck with silicon for quite some time. However, in the long term, you can’t really say that for sure. Technological developments have taken sudden and unpredictable jumps from time to time.
The actual paper is an interesting read. They present an actual computational proof, stating that even if you have essentially infinite memory, a computer that’s a billion times faster than what we have now, perfect training data that you can sample without bias and you’re only aiming for an AGI that performs slightly better than chance, it’s still completely infeasible to do within the next few millenia. Ergo, it’s definitely not “right around the corner”. We’re lightyears off still.
They prove this by proving that if you could train an AI in a tractable amount of time, you would have proven P=NP. And thus, training an AI is NP-hard. Given the minimum data that needs to be learned to be better than chance, this results in a ridiculously long training time well beyond the realm of what’s even remotely feasible. And that’s provided you don’t even have to deal with all the constraints that exist in the real world.
We perhaps need some breakthrough in quantum computing in order to get closer. That is not to say that AI won’t improve or anything, it’ll get a bit better. But there is a computationally proven ceiling here, and breaking through that is exceptionally hard.
It also raises (imo) the question of whether or not we can truly consider humans to have general intelligence or not. Perhaps we’re not as smart as we think we are either.
The paper’s scope is to prove that AI cannot feasibly be trained, using training data and learning algorithms, into something that approximates human cognition.
The limits of that finding are important here: it’s not that creating an AGI is impossible, it’s just that however it will be made, it will need to be made some other way, not by training alone.
Our squishy brains (or perhaps more accurately, our nervous systems contained within a biochemical organism influenced by a microbiome) arose out of evolutionary selection algorithms, so general intelligence is clearly possible.
So it may still be the case that AGI via computation alone is possible, and that creating such an AGI will not require solution of an NP-hard problem. But this paper closes one potential pathway that many believe is a viable pathway (if the paper’s proof is actually correct, I definitely am not the person to make that evaluation). That doesn’t mean they’ve proven there’s no pathway at all.
Our squishy brains (or perhaps more accurately, our nervous systems contained within a biochemical organism influenced by a microbiome) arose out of evolutionary selection algorithms, so general intelligence is clearly possible.
That’s assuming that we are a general intelligence. I’m actually unsure if that’s even true.
That doesn’t mean they’ve proven there’s no pathway at all.
True, they’ve only calculated it’d take perhaps millions of years. Which might be accurate, I’m not sure to what kind of computer global evolution over trillions of organisms over millions of years adds up to. And yes, perhaps some breakthrough happens, but it’s still very unlikely and definitely not “right around the corner” as the AI-bros claim (and that near-future thing is what the paper set out to disprove).
That’s assuming that we are a general intelligence.
But it’s easy to just define general intelligence as something approximating what humans already do. The paper itself only analyzed whether it was feasible to have a computational system that produces outputs approximately similar to humans, whatever that is.
True, they’ve only calculated it’d take perhaps millions of years.
No, you’re missing my point, at least how I read the paper. They’re saying that the method of using training data to computationally develop a neural network is a conceptual dead end. Throwing more resources at the NP-hard problem isn’t going to solve it.
What they didn’t prove, at least by my reading of this paper, is that achieving general intelligence itself is an NP-hard problem. It’s just that this particular method of inferential training, what they call “AI-by-Learning,” is an NP-hard computational problem.
What they didn’t prove, at least by my reading of this paper, is that achieving general intelligence itself is an NP-hard problem. It’s just that this particular method of inferential training, what they call “AI-by-Learning,” is an NP-hard computational problem.
This is exactly what they’ve proven. They found that if you can solve AI-by-Learning in polynomial time, you can also solve random-vs-chance (or whatever it was called) in a tractable time, which is a known NP-Hard problem. Ergo, the current learning techniques which are tractable will never result in AGI, and any technique that could must necessarily be considerably slower (otherwise you can use the exact same proof presented in the paper again).
They merely mentioned these methods to show that it doesn’t matter which method you pick. The explicit point is to show that it doesn’t matter if you use LLMs or RNNs or whatever; it will never be able to turn into a true AGI. It could be a good AI of course, but that G is pretty important here.
But it’s easy to just define general intelligence as something approximating what humans already do.
No, General Intelligence has a set definition that the paper’s authors stick with. It’s not as simple as “it’s a human-like intelligence” or something that merely approximates it.
This isn’t my field, and some undergraduate philosophy classes I took more than 20 years ago might not be leaving me well equipped to understand this paper. So I’ll admit I’m probably out of my element, and want to understand.
That being said, I’m not reading this paper with your interpretation.
This is exactly what they’ve proven. They found that if you can solve AI-by-Learning in polynomial time, you can also solve random-vs-chance (or whatever it was called) in a tractable time, which is a known NP-Hard problem. Ergo, the current learning techniques which are tractable will never result in AGI, and any technique that could must necessarily be considerably slower (otherwise you can use the exact same proof presented in the paper again).
But they’ve defined the AI-by-Learning problem in a specific way (here’s the informal definition):
Given: A way of sampling from a distribution D.
Task: Find an algorithm A (i.e., ‘an AI’) that, when run for different possible situations as input, outputs behaviours that are human-like (i.e., approximately like D for some meaning of ‘approximate’).
I read this definition of the problem to be defined by needing to sample from D, that is, to “learn.”
The explicit point is to show that it doesn’t matter if you use LLMs or RNNs or whatever; it will never be able to turn into a true AGI
But the caveat I’m reading, implicit in the paper’s definition of the AI-by-Learning problem, is that it’s about an entire class of methods, of learning from a perfect sample of intelligent outputs to itself be able to mimic intelligent outputs.
General Intelligence has a set definition that the paper’s authors stick with. It’s not as simple as “it’s a human-like intelligence” or something that merely approximates it.
The paper defines it:
Specifically, in our formalisation of AI-by-Learning, we will make the simplifying assumption that there is a finite set of possible behaviours and that for each situation s there is a fixed number of behaviours Bs that humans may display in situation s.
It’s just defining an approximation of human behavior, and saying that achieving that formalized approximation is intractable, using inferences from training data. So I’m still seeing the definition of human-like behavior, which would by definition be satisfied by human behavior. So that’s the circular reasoning here, and whether human behavior fits another definition of AGI doesn’t actually affect the proof here. They’re proving that learning to be human-like is intractable, not that achieving AGI is itself intractable.
I think it’s an important distinction, if I’m reading it correctly. But if I’m not, I’m also happy to be proven wrong.
A breakthrough in quantum computing wouldn’t necessarily help. QC isn’t faster than classical computing in the general case, it just happens to be for a few specific algorithms (e.g. factoring numbers). It’s not impossible that a QC breakthrough might speed up training AI models (although to my knowledge we don’t have any reason to believe that it would) and maybe that’s what you’re referring to, but there’s a widespread misconception that Quantum computers are essentially non-deterministic turing machines that “evaluate all possible states at the same time” which isn’t the case.
I was more hinting at that through conventional computational means we’re just not getting there, and that some completely hypothetical breakthrough somewhere is required. QC is the best guess I have for where it might be but it’s still far-fetched.
But yes, you’re absolutely right that QC in general isn’t a magic bullet here.
Yeah thought that might be the case! It’s just a thing that a lot of people have misconceptions about so it’s something that I have a bit of a knee jerk reaction to.
Haha it’s good that you do though, because now there’s a helpful comment providing more context :)
the limitation is specifically using the primary machine learning technique, same one all chatbots use at places claiming to pursue agi, which is statistical imitation, is np-hard.
Not just that, they’ve proven it’s not possible using any tractable algorithm. If it were you’d run into a contradiction. Their example uses basically any machine learning algorithm we know, but the proof generalizes.
via statistical imitation. other methods, such as solving and implementing by first principles analytically, has not been shown to be np hard. the difference is important but the end result is still no agigpt in the foreseeable and unforeseeable future.
Nitpick: a lightyear is a measure of distance, not of time :)
Yes, hence we’re not “right around the corner”, it’s a figure of speech that uses spatial distance to metaphorically show we’re very far away from something.
AGI is inevitable unless:
-
General intelligence is substrate independent and what the brain does cannot be replicated in silica. However, since both are made of matter, and matter obeys the laws of physics, I see no reason to assume this.
-
We destroy ourselves before we reach AGI.
Other than that, we will keep incrementally improving our technology and it’s only a matter of time untill we get there. May take 5 years, 50 or 500 but it seems pretty inevitable to me.
@ContrarianTrail @JRepin well I guess somebody would first need to clearly define what “AGI” is. Currently it’s just “whatever the techbro hypers want it to be”.
And then there’s the matter (ha!) of your assumption that we understand all laws of physics necessary that “matter obeys”, or that we can reasonably understand them. That’s a pretty strong assumption: individual human minds are pretty limited and communication adds overhead, and we might reach a point where we’re stuck.
A chess engine is intelligent in one thing: playing chess. That narrow intelligence doesn’t translate to any other skill, even if it’s sometimes superhuman at that one task, like a calculator.
Humans, on the other hand, are generally intelligent. We can perform a variety of cognitive tasks that are unrelated to each other, with our only limitations being the physical ones of our “meat computer.”
Artificial General Intelligence (AGI) is the artificial version of human cognitive capabilities, but without the brain’s limitations. It should be noted that AGI is not synonymous with AI. AGI is a type of AI, but not all AI is generally intelligent. The next step from AGI would be Artificial Super Intelligence (ASI), which would not only be generally intelligent but also superhumanly so. This is what the “AI doomers” are concerned about.
> A chess engine is intelligent in one thing: playing chess
No. That’s not how the adjective “intelligent” works, outside of marketing drivel of course (“intelligent washing machine” etc).
> Artificial General Intelligence (AGI) is the artificial version of human cognitive capabilities
Can you give a definition of “intelligence” or “human cognitive abilities” that would allow us to somehow unequivocably establish that “X is intelligent” or “X has human cognitive abilities”?
IIRC, within computer science, which is the field most heavily driving AI design and research forward, an ‘intelligent agent’ is essentially defined as any ‘agent’ which takes external stimulai from a collection of sensors in some form of environment, processes that stimulai in a dynamic fashion (one of the criteria IIRC is a branching decision tree based on the stimulai), and then applies that processing to a collection of affectors in the environment.
Yes, this definition is an extremely low bar and includes a massive amount of code, software and scripts. It also includes basic natural intelligences such as worms, ants, amoeba, and even viruses. One example of mechanical AI are some of Theo Jansen’s StrandBeasts
“Artificial intelligence” is for the marketing department’s benefit. At least mainly so. What people envision with the term AI is because of preconceived notions based science fiction not what it actually is.
@JayDee so two things.
First: sure, we can redefine words in any way we want, but then:
-
talking about “AI” becomes much less interesting if it merely means “walking a decision tree based on data coming from external sensors”
-
the whole talk about “intelligence” becomes a bait-and-switch, as the conversation started with the term “intelligence” being used in the general sense we tend to apply to people and some animals.
I am not bait-and-switching here. The switchers were the business-minded grifters which made the term synonymous with LLMs and eventually destroyed its meaning completely.
The definition I gave is from the most popular and widely used CS textbook on AI and has been the meaning used in the field since the early 90s. It’s why videogame NPCs are always called AI, because they fit the conventional CS definition, and were one of the major things it was about the most.
As for your ‘1’, AI is a wide-but-very-specialized field and pertains from everything from robots to text autocomplete. If you want the most out of it, you need to get down into the nitty gritty and really research the field.
On a Seperate note, while AI safety, AGI, and the risk of the intelligence explosion are somewhat related to computer science’s pursuit of AI systems, they are much more philosophical currently, and adhere to much vaguer definitions of AI, Such as Alan Turing’s.
@JayDee I didn’t say you are, I clarified in my later post. Sorry, should have been clearer.
I am vehemently agreeing with you here, in fact.
The context is the conversation above in the thread, where it was claimed that “AGI” is “pretty inevitable”.
And the point I’ve been making is:
-
we don’t have a good definition of what “intelligence” is, in the sense presumably used above;
-
if we decide to use a somewhat simplistic definition, the whole “AI” issue stops being all that exciting.
-
-
You’re right that we need a clear definition of intelligence if we are to make any predictions about achieving AGI. The researchers behind this article appear to mean “human-level cognition” which doesn’t seem to be a particularly objective or useful yardstick. To begin with, which human are we talking about? If they’re talking about an idealised maximally intelligent human, then I don’t think we should be surprised that we aren’t about to achieve that. The goal is not to recreate human cognition as if that’s some kind of holy grail. The goal is to make intelligent systems which can give results which are at least as good as what would be produced by a skilled and well-trained human working on the same problem.
Can I ask you how you would define intelligence? And in particular, how would you - if you would at all - differentiate intelligence from being clever, or from being well educated?
@lightstream I wouldn’t, because I am not the one making claims about “AGI” being just around the corner.
That’s the thing, OpenAI and others benefiting from the hype make extraordinary claims – along the lines of “human-level AGI is just around the corner” – so they are the ones that need to define their terms.
You are asking all the right questions here (“which human are we talking about”), the point is that these questions should be answered by those who make such extraordinary claims.
I certainly am not surprised that OpenAI, Google and so on are overstating the capabilities of the products they are developing and currently selling. Obviously it’s important for the public at large to be aware that you can’t trust a company to accurately describe products it’s trying to sell you, regardless of what the product is.
I am more interested in what academics have to say though. I expect them to be more objective and have more altruistic motivations than your typical marketeer. The reason I asked how you would define intelligence was really just because I find it an interesting area of thought which fascinates me and has done long before this new wave of LLMs hit the scene. It’s also one which does not have clear answers, and different people will have different insights and perspectives. There are different concepts which are often blurred together: intelligence, being clever, being well educated, and consciousness. I personally consider all of these to be separate concepts, and while they may have some overlap, they nevertheless are all very different things. I have met many people who have very little formal education but are nonetheless very intelligent. And in terms of AI and LLMs, I believe that an LLM does encapsulate some degree of genuine intelligence - they appear to somehow encode a model of the universe in their billions of parameters and they are able to meaningfully respond to natural language questions on almost any subject - however an LLM is unquestionably not a conscious being.
@ContrarianTrail @JRepin and finally, there’s a question of whether we actually decide to pursue it.
Nuclear power was supposed to be the “inevitable” power source for all of humanity mere 50 years ago. But at some point we decided not to pursue that goal.
Cryptocurrencies were supposed to be “inevitable” replacement for the banking system.
And we *have* cryptocurrencies and nuclear power. These exist. As opposed to whatever nebulous concept hides beneath “AGI”.
Since they still exist, only time will tell if the promise of nuclear power and/or cryptocurrencies come to be.
AGI and even IMHO AI do not exist. Whatever product is being marketed as AI isn’t what I would consider AI. “AI” can have its uses but I really do not think they will be what people expect because it fundamentally lacks what I would consider crucial aspects of human intelligence.
AI makes for a very good grammar checker. It is good at producing filler content for SEO. And it is good at producing “stuff” that looks like it could be right. Probably will have some uses in creative work since it doesn’t have to be “correct” so as a tool to aid an artist, that’s seems pretty cool - I’m sure that is already happening. It will have its uses and a lot of companies will find out the hard way, it is not that they think. That’s my prediction.
incremental improvements on a dead end, still gets you to the dead end.
Then you need to give me an explanation for why it’s a dead end
because, having coded them myself, I am under no illusions as to their capabilities. They are not magic. “just” some matrix multiplications that generate a probability distribution for the next token, which is then randomly sampled.
You seem to be talking about LLMs now and I’m not. LLMs being a dead end is perfectly compatible with what I just said. We’ll just try a different approach next then. Even the fact of realising they’re a dead end is yet another step towards AGI.
yeah, so that means that it’s not incremental improvement on what we have that we need. That will get us nowhere. We need a (as yet unknown) completely different approach. Which is the opposite of incremental improvement.
I didn’t say we need to improve on what we have. We just need to keep making better technology which we will keep doing unless we destroy ourselves first.
Did you read the article, or the actual research paper? They present a mathematical proof that any hypothetical method of training an AI that produces an algorithm that performs better than random chance could also be used to solve a known intractible problem, which is impossible with all known current methods. This means that any algorithm we can produce that works by training an AI would run in exponential time or worse.
The paper authors point out that this also has severe implications for current AI, too–since the current AI-by-learning method that underpins all LLMs is fundamentally NP-hard and can’t run in polynomial time, “the sample-and-time requirements grow non-polynomially (e.g. exponentially or worse) in n.” They present a thought experiment of an AI that handles a 15-minute conversation, assuming 60 words are spoken per minute (keep in mind the average is roughly 160). The resources this AI would require to process this would be 60*15 = 900. The authors then conclude:
“Now the AI needs to learn to respond appropriately to conversations of this size (and not just to short prompts). Since resource requirements for AI-by-Learning grow exponentially or worse, let us take a simple exponential function O(2n ) as our proxy of the order of magnitude of resources needed as a function of n. 2^900 ∼ 10^270 is already unimaginably larger than the number of atoms in the universe (∼10^81 ). Imagine us sampling this super-astronomical space of possible situations using so-called ‘Big Data’. Even if we grant that billions of trillions (10 21 ) of relevant data samples could be generated (or scraped) and stored, then this is still but a miniscule proportion of the order of magnitude of samples needed to solve the learning problem for even moderate size n.”
That’s why LLMs are a dead end.
But I wasn’t talking about LLMs
You literally were LMAO
Other than that, we will keep incrementally improving our technology and it’s only a matter of time untill we get there. May take 5 years, 50 or 500 but it seems pretty inevitable to me.
Literally a direct quote. In what world is this not talking about LLMs?
There’s not a single mention of LLM’s in my entire post. The argument I’m making there isn’t even mine. I heard it from Sam Harris way before LLMs were even a thing.
Yeah, suuuuure you weren’t.
Note that the proof also generalizes to any form of creating an AI by training it on a dataset, not just LLMs. But sure, we’ll absolutely develop an entirely new approach to cognitive science in a few years, we’re definitely not boiling the planet and funneling enough money to end world poverty several times over into a scientific dead end!
Another possibility is that humans just aren’t smart enough to figure out AGI. While I’m sure that we will continue incrementally improving technology in some form, it’s not at all self-evident that these improvements will eventually add up to AGI.
I get what you’re saying but to me, that still just sounds like a timescale issue. I can’t think of a scenario where we’ve improved something so much that there’s just absolutely nothing we could improve on further. With AI we only need to reach the point of making it have human-level cognitive capabilities and from there on it can improve itself.
There are a couple of reasons that might not work:
- Maybe we’ll asymptotically approach a point that is lower than human-level cognitive capabilities
- Gradual improvements are susceptible to getting stuck in a local maxima. This is a problem in evolution as well. A lot of animals could in theory evolve, say, human level intelligence in principle, but to reach that point they’d have to go through a bunch of intermediate steps that lead to worse fitness. Gradual scientific improvements are a bit like evolution in this way.
- We also lose knowledge over time. Something as dramatic as a nuclear war would significantly set back the progress in developing AGI, but something less dramatic might also lead to us forgetting things that we’ve already learned.
To be clear, most of the arguments I’m making aren’t really about AGI specifically but about humanities capability to develop arbitrary in principle feasible technologies in general.
I can’t think of a scenario where we’ve improved something so much that there’s just absolutely nothing we could improve on further.
Progress itself isn’t inevitable. Just because it’s possible doesn’t mean that we’ll get there, because the history of human development shows that societies can and do stall, reverse, etc.
And even if all human societies tends towards progress, it could still hit dead ends and stop there. Conceptually, it’s like climbing a mountain through the algorithm of “if there is a higher elevation near you, go towards that, and avoid stepping downward in elevation.” Eventually that algorithm brings you to a local peak. But the local peak might not be the highest point on the mountain, and while it is theoretically possible to have gotten to the other true peak from the beginning, the person who is insistent on never stepping downward is now stuck. Or, it’s possible to get to the true peak but it requires climbing downward for a time and climbing up past elevations we’ve already been to, on paths we hadn’t been on. One can imagine a society that refuses to step downward, breaking the inevitability of progress.
This paper identifies a specific dead end and advocates against hoping for general AI through computational training. It is, in effect, arguing that even though we can still see plenty of places that are higher elevation than where we are standing, we’re headed towards a dead end, and should climb back down. I suspect that not a lot of the actual climbers will heed that advice.
-
Will AI soon surpass the human brain?
If you ask employees at OpenAI, Google DeepMind and other large tech companies, it is inevitable.That doesn’t answer the question.
If it will happen is unrelated to When it will happen.
I’d expect we’ll see AGI some time between the next 20 and 200 years. I think that’s pretty soon. You may not.If there were a giant asteroid hurling toward Earth, set to impact sometime in the next 20 to 200 years, I’d say there’s definitely a need for urgency. A true AGI is somewhat of an asteroidal impact in itself.
A single AGI would not be to different from a human. But it may not take long for AGI to develop ASI, superior to human intelligence.
Thats not an astronaut impact but alien contact
A single AGI could be copied into a million copies near-instantly. That would be significant
None of those companies are suggesting 20 years. They’re suggesting much less than 10, and selling investors on that promise.
The steam engine won’t replace John Henry!!!
Not really a good comparison. The steam engine was an extant technology at that point. AGI is not, and we really no idea if/when it will be. One thing is clear though, it is not as close on the horizon as tech bros want us to think it is.
Possible or not I don’t think we’ll get to the point of AGI. I’m pretty sure at some point someone will do something monumentally stupid with AI that will wipe out humanity.
Like wrecking the biosphere in its persuit.
Maybe. But I have a feeling it’ll be a dumb single mistake that’ll make someone say “ah, shit” just before we’re wiped out.
When the Soviets trained anti-tank dogs in WW2 they did so on tanks that weren’t running to save fuel: “Their deployment revealed some serious problems… In the field, the dogs refused to dive under moving tanks.” https://en.m.wikipedia.org/wiki/Anti-tank_dog
History is littered with these kinds of mistakes. It would only take one military AI with access to autonomous weapons to have a similar issue in it’s training data to potentially kill us all.
Why in God’s name would we put weapons that pose a legitimate threat to the whole of humanity under the control of an ai? I just don’t think this one sounds plausible.
- I’m so glad they weren’t “kamikaze” dogs.
- I very much expected that story to end with the dogs targeting Soviet tanks.
I like SCUMM but AGI is okay I just don’t like typing commands
