Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

What you wanted to know about AI

fastml.com

Post reply on HN
Print view View original

151–160 of 165 posts

Re: What you wanted to know about AI

#151

Post by rl3 »

maaku wrote: post #117
AndrewKemendo wrote: post #85

Earlier quoted context omitted.

Could you argue against the position directly? No I can't because there isn't a way to argue it. I can't argue that it would be safe because I have nothing to point to that says it will be or even could be. We simply do not know enough about how it would be built to reason on it. It's not within the scope of the quoted works to elaborate a technical roadmap or provide firm dates for the arrival of superhuman machine…

Well, color me disappointed. I think there are many arguments against the viewpoints of MIRI and Bostrom, both discrediting the perceived risk of super-intelligences and the the idea that we could build a "provably friendly" alternative. But I don't think it's at all a fair criticism to say that you need a technical roadmap to engage in that debate. I would rather say that they need to be willing to discuss specific…

>The field of "AI safety" needs more engineering and less philosophy.

An alternate phrasing might state that it needs more philosophy and an even greater amount of engineering.

If the field is in its infancy from a technical standpoint, then it probably is from a philosophical one as well.

Re: What you wanted to know about AI

#152

Post by pauletienney »

NhanH wrote: post #12
pauletienney wrote: post #2

Indeed we are far far away from true AI (to me it implies self-consciousness). The point is even if it happens in 100 or 200 years, it will be a huge change in human history. I guess Gates, Hawking and Stark are talking about a further AI creation, they are kind of long-term thinking guys.

Seems like it's a popular sentiment to think of Elon Musk as Tony Stark ;). I believe you mean to say "Gates, Hawking and Musk". Stark doesn't tend to talk about AI, he build them.

Was just a little joke ;)

Re: What you wanted to know about AI

#153

Post by cousin_it »

eli_gottlieb wrote: post #150
cousin_it wrote: post #112

Earlier quoted context omitted.

Yeah, I agree that reinforcement learning is probably a bad approach to FAI. Most of our toy models involve utility functions encoded directly into the AI, not reinforcement. That said, it's indeed very hard to directly specify a utility function involving paperclips. If our universe were a Game of Life universe and we knew exactly which configuration corresponds to a paperclip, I'd be able to do that right now. But…

>That said, it's indeed very hard to directly specify a utility function involving paperclips. If our universe were a Game of Life universe and we knew exactly which configuration corresponds to a paperclip, I'd be able to do that right now. But since we don't know the true laws of physics, the "hard way" involves encoding some kind of Solomonoff prior over all possible physical universes, and a rule for recognizing…

AIXItl isn't really the kind of AI that I like, because it's reflectively inconsistent. In any case, the time complexity of AIXItl is kind of irrelevant at this stage, because we're trying to figure out what is the right thing to optimize. Only then we should start figuring out how to optimize that thing efficiently, because we really don't want to optimize the wrong thing efficiently. I'm very skeptical that approaches based on "conceptual abstraction" can tell us the right thing to optimize, as opposed to my preferred approach (defining a utility function over mathematical objects directly).

Re: What you wanted to know about AI

#154

Post by renox »

zanny wrote: post #77
dangerlibrary wrote: post #65

Earlier quoted context omitted.

> I suspect that the current google car is already safer than the overall average driver. Based on what? Even Google admits that the car is essentially blind (~30 feet visibility) in light rain. They've done little to no road testing in poor weather conditions. The vast majority of their "total miles driven" are highway miles in good weather, with the tricky city-driving bits at either end taken over by humans. Googl…

A self-driving car that can only do highways and can't drive in bad weather still destroys the trucker industry overnight. And that is today. No, I'm not drinking the AI koolaid here thinking there will be some breakout solution to fuzzy visibility problems these cars have. The difficulty differentiating contextual moving objects correctly, knowing what is "safe" debris and what is not, etc, are all neural net proble…

> A self-driving car that can only do highways and can't drive in bad weather still destroys the trucker industry overnight.

Destroying the trucker industry with a self-driving car which can't drive in bad weather?? I don't think so! Trucker's clients usually care a lot about predictability, and they would NOT be happy to hear "sorry it is raining and the robot car couldn't arrive".

Re: What you wanted to know about AI

#155

Post by foobarqux »

eli_gottlieb wrote: post #147
foobarqux wrote: post #115

Earlier quoted context omitted.

What's absurd is to have a strong conviction about inviolable restrictions on systems that haven't been imagined yet. The examples you cite aren't even compelling on their face. For instance if you include whole-brain emulations in AGI (and we should because the consequences are essentially the same) it would take zero time to train. But even without whole-brain emulations a child learns rapidly from its environment…

>What's absurd is to have a strong conviction about inviolable restrictions on systems that haven't been imagined yet. AGI has already been imagined. And yes, it does have to obey the laws of information theory: only one bit of knowledge can be learned from sensory inputs with one bit of entropy. >An AGI could conceivably do the same but faster, and moreover could take advantage of access to the Internet. Children st…

All the manifestations of AGI have not been imagined, and even ones that have like WBE refute your claims.

Even if you could somehow create a model precise enough to talk about information theory (define a symbol universe, map information to knowledge, itself a not-well defined term) it would not seem to be restrictive. Again, children take in a massive amount of sensory input.

Finally, the singularity hypothesis is that AGI could learn exponentially faster, not merely 10x. And again, WBEs don't need to learn at all.

Re: What you wanted to know about AI

#156

Post by davmre »

themgt wrote: post #128
davmre wrote: post #45

Earlier quoted context omitted.

I'm not sure I understand what you mean by "straw man" here. The usual meaning is that to attack a straw man means to argue against a position that no one actually holds, but which is easier to attack than your opponent's actual position. The concerns about the long-term future of AI are real, actual beliefs held by serious people. At this point there's a respectable literature on the potential dangers of unconstrain…

Really I would describe these people as bikeshedding. They're pontificating about the only abstractions they understand, some hand-wavey idea of AI and then the sci-fi short stories of the imagined dangers. Because if you can't do, fanfic.

I don't pretend that an argument from authority resolves this debate, but the fact that people like Stuart Russell take these arguments seriously implies that there's a bit more substance there than you're acknowledging.

To actually argue the point a little bit, the theory of expected-utility-maximizing agents is pretty much the framework in which all of mainstream AI research is situated. Yes, most current work is focused on tiny special cases, in limited domains, with a whole lot of tricks, hacks, and one-off implementations required to get decent results. You really do need a lot of specialized knowledge to be a successful researcher in deep learning, computer vision, probabilistic inference, robotics, etc. But almost all of that knowledge is ultimately in the service of trying to implement better and better approximations to optimal decision-theoretic agents. It's not an unreasonable question to ask, "what if this project succeeds?" (not at true optimality -- that's obviously excluded by computational hardness results -- but just at approximations that are as good or better than what the human brain does).

Do Nick Bostrom and Eliezer Yudkowsky understand when you would use a tanh vs rectified linear nonlinearity in a deep network? Do they know the relative merits of extended vs unscented Kalman filters, MCMC vs variational inference, gradient descent vs BFGS? I don't know, but I'd guess largely not. Is it relevant to their arguments? Not really. You can do a lot of interesting and clarifying reasoning about the behavior of agents at the decision-theoretic level of abstraction, without cluttering the argument with details of current techniques that may or may not be relevant to the limitations of whatever we eventually build.

Re: What you wanted to know about AI

#157

Post by JoeAltmaier »

davmre wrote: post #156
themgt wrote: post #128

Earlier quoted context omitted.

Really I would describe these people as bikeshedding. They're pontificating about the only abstractions they understand, some hand-wavey idea of AI and then the sci-fi short stories of the imagined dangers. Because if you can't do, fanfic.

I don't pretend that an argument from authority resolves this debate, but the fact that people like Stuart Russell take these arguments seriously implies that there's a bit more substance there than you're acknowledging. To actually argue the point a little bit, the theory of expected-utility-maximizing agents is pretty much the framework in which all of mainstream AI research is situated. Yes, most current work is f…

All that talk about maximizing utility, isn't about intelligence at all. Its about a sensory feedback loop maybe. Intelligence is what lets you say "This isn't working. Maybe I should try a different approach. Maybe I should change the problem. Maybe I should get a different job". Until you're operating at that meta level, you're not talking 'intelligence' at all, just control systems.

Re: What you wanted to know about AI

#158

Post by davmre »

JoeAltmaier wrote: post #157
davmre wrote: post #156

Earlier quoted context omitted.

I don't pretend that an argument from authority resolves this debate, but the fact that people like Stuart Russell take these arguments seriously implies that there's a bit more substance there than you're acknowledging. To actually argue the point a little bit, the theory of expected-utility-maximizing agents is pretty much the framework in which all of mainstream AI research is situated. Yes, most current work is f…

All that talk about maximizing utility, isn't about intelligence at all. Its about a sensory feedback loop maybe. Intelligence is what lets you say "This isn't working. Maybe I should try a different approach. Maybe I should change the problem. Maybe I should get a different job". Until you're operating at that meta level, you're not talking 'intelligence' at all, just control systems.

That's not the definition the mainstream AI community has taken, for what I think are largely good reasons, but you could define intelligence that way if you wanted. It's only a renaming of the debate though - instead of calling the things we're worried about "intelligent machines", you'd now call them "very effective control systems".

The issue is still the same: if a system that doesn't perfectly share your goals is making decisions more effectively than you, it's cold comfort to tell yourself "this is just a control system, it's not really intelligent". As Gary Kasparov can confirm, a system with non-human reasoning patterns is still perfectly capable of beating you.

Re: What you wanted to know about AI

#159

Post by maaku »

rl3 wrote: post #151
maaku wrote: post #117

Earlier quoted context omitted.

Well, color me disappointed. I think there are many arguments against the viewpoints of MIRI and Bostrom, both discrediting the perceived risk of super-intelligences and the the idea that we could build a "provably friendly" alternative. But I don't think it's at all a fair criticism to say that you need a technical roadmap to engage in that debate. I would rather say that they need to be willing to discuss specific…

> The field of "AI safety" needs more engineering and less philosophy. An alternate phrasing might state that it needs more philosophy and an even greater amount of engineering. If the field is in its infancy from a technical standpoint, then it probably is from a philosophical one as well.

No the philosophical musings have proven to be worse than a distraction IMHO.

Re: What you wanted to know about AI

#160

Post by maaku »

eli_gottlieb wrote: post #149
maaku wrote: post #114

Earlier quoted context omitted.

> That said, you put it exactly right. The arguments about the potential long-term risk are persuasive regardless of where we are today. It bothers me to see this argument ignored time and time again. Actually I am almost completely unpersuaded by the arguments of Bostrom, Yudkowsky, et al, at least in their presented strong form. Superintelligent AI is not a magical black box with infinite compute capacity, the two…

I just want to point out: the history of software, just regular software, has been typified by the New Jersey approach and the MIT approach. The former consists in just hacking together something that kinda-mostly works, releasing fast, and trying to ameliorate problems later. The latter consists in thoroughly considering what the software needs to do, designing the code correctly the first time with all necessary fu…

No one is making "world optimizations" engines. The concept doesn't even make sense when he wheels hit the road. No AI research done anytime in the foreseeable future would even be at risk of resulting in a runaway world optimizer, and contrary to exaggerated claims being made there would be plenty of clear signs something was amis if it did happen and planty of time to pull the plug.
Post reply on HN