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

Geoffrey Hinton and Demis Hassabis: AGI is nowhere close to being a reality

venturebeat.com

Post reply on HN
Print view View original

91–100 of 181 posts

Re: Geoffrey Hinton and Demis Hassabis: AGI is nowhere close to being a reality

#91

Post by jacquesm »

roenxi wrote: post #30

The computational power of the hardware is getting really close to what a human brain is capable of (on an exponential scale, anyway). If "nowhere close" means not in the next 5 years then sure. Over the medium term I'm not sure AI researchers are the best people to ask. They are completely dependent on how much power the electrical engineers give them - I doubt there is a deeper understanding what a doubling or quad…

> I doubt there is a deeper understanding what a doubling or quadrupling of computer power will do than any programmer learning about neural networks.

Sure they do. They just hook up four times as much compute power or simulate whatever they want to do in four times as much time. A slow AGI would still be an AGI. But we do not see anything like that if we use four times as much power as in the control. It is still nowhere near.

Re: Geoffrey Hinton and Demis Hassabis: AGI is nowhere close to being a reality

#92

Post by taneq »

kahoon wrote: post #51
karmasimida wrote: post #17

And no one should be surprised by this. The NN advancement of late doesn't help addressing human-style symbolic reasoning at all. All we have is a much more powerful function approximator with a drastic increased capacity (very deep networks with billions of parameters) and scalable training scheme (SGD and its variants). Such architecture works great for differentiable data, such's images/audios, but the improvement…

I don't understand this fixation on symbolic reasoning. Do any other animals practice this? If the answer is no, then it is probably not the most important milestone to AGI or at least not the one we should be currently aiming for. Right now we can not replicate the cognition of a mouse. Feels like we want to go to Mars before figuring out how to build a rocket.

The thing is, whatever the hell it is that human brains actually do in the background to produce our 'understanding' of the world and our ability to synthesize new ways to manipulate it, we're also very good at back-fitting explanations based on symbolic reasoning. So it looks like machines need symbolic reasoning to replicate human abilities, whereas I'd bet a dollar that actually, we're doing something quite different (and messy and Bayesian and statistical) in the background and then, using the same process, coming up with a story to explain our outcome semantically. It's not insight so much as parallel construction.

Re: Geoffrey Hinton and Demis Hassabis: AGI is nowhere close to being a reality

#93

Post by barrkel »

joe_the_user wrote: post #46
sdenton4 wrote: post #36

Earlier quoted context omitted.

Maaaaybe. I tend to think that symbolic reasoning is a learning tool, rather than a goalpost for general intelligence. For example, we use symbolic reasoning quite extensively when learning to read a new language, but once fluent can rely on something closer to raw processing - no more reading and sounding out character sequences. Similarly with chess - eventually we have good mnemonics for what make good plays, and…

Once someone is fluent in a language, the logical operations and judgements involved stop being overt and highly visible to the conscious mind. But that doesn't mean that one stops getting the benefits and results of logical operations. What you might see as logical operations "not mattering", I would see as logical operations integrated so deeply into reflexive operations that it's hard to see where one ends and the…

There's a parallel between something being logical and it "feeling right" without a necessary connection at the "implementation level" between the two, just like there may be a parallel between an artificial NN recognizer recognizing something unambiguously and not caught awkwardly with multiple weak or conflicting activations, and a logical system using rules to determine a contradiction, without ever needing to embed the second in the first, however deep - it's just that illogical inputs didn't get good training because they either don't happen or have no meaningful training data.

I, personally, just know I don't use logical rules very often at all. Usually I apply them retroactively as a post-hoc justification, or narrative, to explain a sense of discomfort or internal conflict or dissonance, but I have no way of knowing if my rationale is true other than how it makes me feel - I'm simply relying on the same mechanism, with an extra set of pattern recognition learned specifically to identify fallacies and incorrect logical constructs. If I didn't have that extra training, my explanations could be illogical and I'd be none the wiser.

I think humans are very bad at logical reasoning and very inefficient at it. Only a small % of the population ever does it and they usually do it incorrectly with biases, constructively to justify an already held conclusion. They're great at pattern recognition though. I don't think logical reasoning is anywhere on the critical path to human level AGI at a deep level. It could very well be a parallel system though to help train recognition if we don't figure out better ways of doing that.

Re: Geoffrey Hinton and Demis Hassabis: AGI is nowhere close to being a reality

#94

Post by taneq »

mindgam3 wrote: post #54

Demis Hassabis (true) statements here would be much more credible if DeepMind wasn't currently making a mint by promoting AlphaZero to the masses as a "general purpose artificial intelligence system". Don't believe me? Check out this series of marketing videos on YouTube by GM Matthew Sadler. 1. “Hi, I’m GM Matthew Sadler, and in this series of videos we’re taking a look at new games between AlphaZero, DeepMind’s gen…

I don't think they mean the two in the same way. AlphaZero is "general purpose artificial intelligence" because if you formulate a problem in the right way and then throw a server cluster at it for a few weeks, it often comes back with pretty good performance at solving that problem. It's probably our current best crack at creating AGI, but it's a long way from a machine that can take a very high level goal and figure out the rest for itself, which is what we usually mean by "AGI" - not just a machine that answers multiple questions, but a thing analogous to a human mind which can analyse new things, infer properties and mechanics, generalise those to new contexts, and apply that knowledge to achieve new outcomes.

Re: Geoffrey Hinton and Demis Hassabis: AGI is nowhere close to being a reality

#95

Post by aerodude »

mortivore wrote: post #28
izzydata wrote: post #24

I'm not even convinced that a real AI is possible with conventional computer hardware or anything remotely similar to it. Not even considering software I get the impression there is a fundamental limitation of hardware.

I'm not convinced we've even defined the problem space well enough to solve it. Like what is the concrete measure(something to target) for intelligence? If we develop general intelligence is it going to be human, dog, or fish?

I think this is the real answer. When we developed flight, the measure wasn't "can we fly like birds?" We still haven't achieved that even today, but we fly in otherwise unimagined, but equally powerful ways.

We seem to be looking at intelligence in humans and thinking we need to develop that, without first defining what intelligence actually is. We don't exist in isolation, and it's likely that the components of intelligence exist to varying degrees in other organisms. In the same way that birds, bats, gliders and insects all have wings that generate lift, what are the things that we have in common with other animals?

Re: Geoffrey Hinton and Demis Hassabis: AGI is nowhere close to being a reality

#96

Post by mindgam3 »

taneq wrote: post #94
mindgam3 wrote: post #54

Demis Hassabis (true) statements here would be much more credible if DeepMind wasn't currently making a mint by promoting AlphaZero to the masses as a "general purpose artificial intelligence system". Don't believe me? Check out this series of marketing videos on YouTube by GM Matthew Sadler. 1. “Hi, I’m GM Matthew Sadler, and in this series of videos we’re taking a look at new games between AlphaZero, DeepMind’s gen…

I don't think they mean the two in the same way. AlphaZero is "general purpose artificial intelligence" because if you formulate a problem in the right way and then throw a server cluster at it for a few weeks, it often comes back with pretty good performance at solving that problem. It's probably our current best crack at creating AGI, but it's a long way from a machine that can take a very high level goal and figur…

“if you formulate a [very specific type of problem involving perfect information games] in the right way”

Re: Geoffrey Hinton and Demis Hassabis: AGI is nowhere close to being a reality

#97

Post by barrkel »

taneq wrote: post #92
kahoon wrote: post #51

Earlier quoted context omitted.

I don't understand this fixation on symbolic reasoning. Do any other animals practice this? If the answer is no, then it is probably not the most important milestone to AGI or at least not the one we should be currently aiming for. Right now we can not replicate the cognition of a mouse. Feels like we want to go to Mars before figuring out how to build a rocket.

The thing is, whatever the hell it is that human brains actually do in the background to produce our 'understanding' of the world and our ability to synthesize new ways to manipulate it, we're also very good at back-fitting explanations based on symbolic reasoning. So it looks like machines need symbolic reasoning to replicate human abilities, whereas I'd bet a dollar that actually, we're doing something quite differ…

I fully agree, as I wrote in my other comment in here. Logical symbolic reasoning is usually post-hoc rationalisation built constructively to come to an already held conclusion that "feels right". It's rare that someone changes their mind due to logic, especially if the topic isn't abstract and has real-world consequences and emotional engagement.

Re: Geoffrey Hinton and Demis Hassabis: AGI is nowhere close to being a reality

#98

Post by karmasimida »

sdenton4 wrote: post #36
karmasimida wrote: post #17

And no one should be surprised by this. The NN advancement of late doesn't help addressing human-style symbolic reasoning at all. All we have is a much more powerful function approximator with a drastic increased capacity (very deep networks with billions of parameters) and scalable training scheme (SGD and its variants). Such architecture works great for differentiable data, such's images/audios, but the improvement…

Maaaaybe. I tend to think that symbolic reasoning is a learning tool, rather than a goalpost for general intelligence. For example, we use symbolic reasoning quite extensively when learning to read a new language, but once fluent can rely on something closer to raw processing - no more reading and sounding out character sequences. Similarly with chess - eventually we have good mnemonics for what make good plays, and…

I put symbolic reasoning at the spotlight for it is something that NN is particularly bad at: discrete data, hard to design, often approximate and non-differentiable measurement.

The problem is so inherently hard that we are struggling even to come up with a meaningful task, telling us how bad we are doing. That comes to your first point, I think finding the right loss function is a like a chicken-and-egg situation here. When you have the loss function at hand, you already what task and problem you are going to solve, then it becomes easier. But that is apparently not our current situation.

That is why I think DeepMind has a good reason to go after reinforcement learning, after all, that is how we human are trained, through exams and the feedbacks.

As to your point about LSTM, I am not very passionate to qualitatively claim it whether it can/can't handle short/long term memory. That is apparently task dependent, and all the concepts involved are ill-defined.

Re: Geoffrey Hinton and Demis Hassabis: AGI is nowhere close to being a reality

#99

Post by aerodude »

roenxi wrote: post #75
anonytrary wrote: post #43

Earlier quoted context omitted.

> The computational power of the hardware is getting really close to what a human brain is capable of Why do you say that? AFAIK computing architecture and brain architecture are completely different. How would you even begin to compare their power?

Well, Wikipedia was my source [0] and it links http://hplusmagazine.com/2009/04/07/brain-chip/ as its source. Google has TPU that are off from the estimated power required to simulate a brain by a factor of 3, so technology is reaching the ballpark. Given that brains were evolved, the part that does symbolic thinking is probably "easy to stumble on" in some practical sense. [0] https://en.wikipedia.org/wiki/Computer_…

Faster hardware will help, but I'm not convinced that it's the answer. OpenAI Five used on the order of 2000 years of experience to train their agent. There are clearly still huge algorithmic gains to be had.

Given how we've managed to improve on nature in other domains (see solar cell efficiency, for example), I think that if we can figure out how intelligent organisms manage to learn so quickly we can likely beat nature's efficiency.

Re: Geoffrey Hinton and Demis Hassabis: AGI is nowhere close to being a reality

#100

Post by barrkel »

joe_the_user wrote: post #86
narag wrote: post #72

Earlier quoted context omitted.

Cada vez que intento usar un sitio de traducción para traducir más de un párrafo (Facebook o Google), sale un lío confuso - no significa que algunas oraciones sean aparentemente claras y significativas. Lo importante es que se atraganta con los modismos, no entendiéndolos, no dejándolos como están, sino adivinando algún significado claramente erróneo. Ocasionalmente encuentro que las publicaciones de una sola oración…

Oh, that just means I happen to write in clear, unidiomatic English ;-). Add just a smidgen of irregular usage, contemporary metaphors and such and things can go South pretty quickly.

That'll just be a training problem (often translation is driven from example texts that have lots of translations, and we have many of these for multi-national orgs like EU, which necessarily don't include a lot of colloquialisms), and the inconsistency isn't found because the models built don't extend all the way out to real world experiences (training) and recognizing text as a real world experience narrative. I think a much deeper network could do better but we don't know how to train them, we take way too long as it is.
Post reply on HN