Hacker News Reader

Comment by deong | Hacker News Reader

Yes, the stuff he's saying has been known for ages, but there's been a greater trend recently of people hinting otherwise. I worked in a research group that was focused on AGI until I left academia last year, and there are spirited debates going on about just how close we are.

For what it's worth, I completely agree with the sentiment presented here that we're far enough away that we don't know what direction even we need to start down to make the best progress, which in my view makes any sort of prognostication of questionable value. It's also why I think it's not that productive to have people like Stephen Hawking arguing about future AI as a threat that somehow needs to be mitigated. Even if he's right in the general point, we know so little about how that future system would look that it would be like asking Isaac Newton to make cars safer.

View on HN

>> we're far enough away that we don't know what direction even we need to start down to make the best progress

Even worse, we have some strong theoretical results suggesting that none of the techniques we have developed so far can ever lead to AGI, and that refers very much to machine learning primarily.

I'm talking about the work that was done on inductive inference, and then PAC-learning, in decades past (starting in the 1960's). There are many negative results and very few positive ones.

I wouldn't say I've reviewed the bibliography extensively (it's a huge area spanning a couple of fields), but from the little I've read I'm getting the feeling that, if you want to be able to learn any function in polynomial time, with a polynomial number of training examples, and with arbitrary error, you're basically stuck with the easiest classes of concept to learn. And if you want to learn specific distributions or representations (like DFAs or PCFGs) then you can just forget about it.

Which reminds me of something that Andrew Ng has said, about how recent results in machine learning and deep learning only reached the "low hanging fruit" and further progress is going to be much harder. No references to all this- I'll dig 'em up if required.

The results sound interesting, and I'd love to read them if you can find the references, but I'd caution against being too sure as well.

Any result which predicts that human intelligence cannot exist.. has to have an error somewhere. Perhaps the error here is in assuming that a learning algorithm able to fit any function in polynomial time is needed in the first place.

I agree completely.

People are too ready to leap at results that say, "Well, machines can never do X, and I can do X, therefore human-level AI is impossible."

In general, my counter to that would be "you probably can't do X either". It's just that you can do it well enough. You don't encounter the weird cases where your ability fails. But if that's allowed as an argument, then of course machines are more powerful too. I can absolutely solve the halting problem if "solve" can be interpreted as "almost always come up with the correct answer for the kinds of problems people encounter in the real world".

That's basically learning (and more generally, operating) in restricted settings. I agree that much progress can be made in that way, but there are problems that are really, really hard to solve and that both humans and computers are crap at. Then again there are entire classes of problems that humans are very good at, but computers are crap at and the other way around.

Also, don't forget what Terry Pratchett said, about how thousand-to-one odds crop up one time out of ten. The problems with AI in the past started when people tried to attempt to solve anything beyond the trivial, toy-problems they could demonstrate were solvable "in the lab". Like SHRDLU, which solved blocks world, but could not really scale. The same goes with machine learning today: some problems are solvable in restricted settings, or in a lab environment, under controlled conditions and with safe assumptions etc, but once you try to do the same thing in the wild...

You can even look at the big success of Deep nets for examples of that: image processing and speech recognition. Whereas it's possible to get really good results in both fields, image processing still relies very much on having good quality images to work with, and speech recognition is really crap if you have a bit of an accent, or a speech impediment etc. (at least that's what I gather- these are not my specialties).

Some of my references are below. These are mostly online, but the first one from Pitt is behind a paywall and I couldn't find a free version of it, unfortunately. Email me if you need clarifications.

Dana Angluin, Inductive inference of formal languages from positive data. [http://www-personal.umich.edu/~yinw/papers/Angluin80.pdf]

Dana Angluin, Learning regular sets from queries and counterexamples. [https://people.eecs.berkeley.edu/~dawnsong/teaching/s10/pape...]

Mark E. Gold, Complexity of automaton identification from given data. [http://sebastian.doc.gold.ac.uk/papers/Language_Learning/gol...]

Mark E. Gold, Language Identification in the Limit.[http://web.mit.edu/6.863/www/spring2010/readings/gold67limit...]

Pitt, L., & Pitt, L. (1989). Inductive Inference, DFAs and Computational Complexity. 2ND INT. WORKSHOP ON ANALOGICAL AND INDUCTIVE INFERENCE (AII, 18--44. Retrieved from http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.187....

Leonard Pitt & Manfred K. Warmuth. The minimum consistent DFA problem cannot be approximated within any polynomial. [http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.86....]

Also see below for a gentle introduction to PAC-Learning, which actually allows for learning several concepts in restricted (but non-trivial) settings:

http://web.cs.iastate.edu/~honavar/pac.pdf

There is a lot more bibliography on both Inductive Inference tractability and PAC-Learnability. You'll have to follow the trail of references :)

The comment was not that no technique could exist, but that none of the techniques we currently have developed; this could simply argue that a human brain is not like any of our models (which I would personally not find shocking or even surprising).

Yes and no. The results I mention (see reference above) discuss learnability in the context of Computational Theory, so the branch of maths that looks at what is computable with languages and automata (a.k.a. the theoretical part of Computer Science).

Yes, it's entirely possible that we have missed something really important and when we figure that thing out we'll end up with a completely different Theory of Computation than we have now, and that may allow us to progress in learning concept classes that, today, are not learnable. That however, would be a revolution on par with Einstein's in physics.

Another thing that might save our collective ass, in terms of learnability, is that many of those results are assuming that P != NP. If that turns to be wrong, then those results will be off.

However, all this is very unlikely and in any case a) we can hardly expect such a profound re-interpretation of all that we know to happen within the next couple of generations and b) even if it does it remains the case that at this point in time everything we know suggests that there are some concept classes that we can simply not learn efficiently, at least not with computers.

Edit: what I'm really trying to say is that we may not understand human minds, but we're talking about computers here, so basically I agree with what you say, but with an added flourish :)

>> Any result which predicts that human intelligence cannot exist.. has to have an error somewhere.

See deong's comment below. We may not understand how our minds work, but we do understand how computers work and that's where the learnability results come in: because they pertain to computers, not human minds.

Also, it may be that we don't need to learn every single thing we know. See Chomsky's Universal Grammar, etc.

>> we have some strong theoretical results suggesting that none of the techniques we have developed so far can ever lead to AGI

I've been looking for work in this area without much success since publication is dominated by the opposite movement. I'd be grateful for any references you can dig up.

I believe AI is much more of a hardware problem than people realize. It's such a computationally difficult task. The energy efficiency and parallelism of brains is unbelievable. However, I think an improvement of even just two orders of magnitude would be sufficient to achieve something that many people would call rudimentary AGI.

Note that he didn't say anything about the point you made : a possible change in common sense systems being more tractable to build. That'd have been fascinating. He just said something old and techcrunch made it into a headline.

Love your point about hawking. I treat that as pure entertainment but since the general population treats hawking with high regard (well deserved) his opinions about things he doesn't know much about also carry weight in shaping public perception.

As far as speculating on how scary an AI driven future can be, a filmmaker in the 60s or a high school student today can do as compelling a job as Hawking.

Reply on news.ycombinator.com