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

EURISKO Lives

blog.funcall.org

Post reply on HN
Print view View original

81–90 of 102 posts

Re: EURISKO Lives

#81

Post by YeGoblynQueenne »

lisper wrote: post #72
avmich wrote: post #70

Earlier quoted context omitted.

It's great to read conversation of towering HN experts in the field. Lisper, as I understand this part - > In order to have kinematic skills, something in your brain must be doing something that can be equated by some mapping to solving kinematic equations you're talking about an equivalent of YeGoblynQueenne's > that humans ... do not find solutions to kinematic equations, but instead use simple heuristics that expl…

Let's go back to the original formulation so we don't lose the plot here: Me: As an analogy, consider a professional tennis or baseball player. YeGoblynQueenne: humans e.g. playing baseball do not find solutions to kinematic equations, but instead use simple heuristics that exploit our senses and body configuration, like placing their hands in front of their eyes so that they line up with the ball etc. At the risk of…

>> But even just moving your hand in front of your eyes is nowhere near as trivial as YeGoblynQueenne implies.

Yeah, I was actually thinking of kinematics as in classical mechanics. I think you were speaking about kinematic equations as in robotics. My bad, I misunderstood.

I agree that moving your hand in the right place is not a simple problem, and I don't actually have an insight into that, but I think it's easier than calculating the trajectory of an object, let alone many at once (think juggling). Maybe that's another source of our disagreement- but see my comment about having multiple models for a process.

Re: EURISKO Lives

#82

Post by YeGoblynQueenne »

avmich wrote: post #73
lisper wrote: post #72

Earlier quoted context omitted.

Let's go back to the original formulation so we don't lose the plot here: Me: As an analogy, consider a professional tennis or baseball player. YeGoblynQueenne: humans e.g. playing baseball do not find solutions to kinematic equations, but instead use simple heuristics that exploit our senses and body configuration, like placing their hands in front of their eyes so that they line up with the ball etc. At the risk of…

I suspect there's a terminological difference. > being a professional tennis or baseball player involves a lot more than "simple heuristics ... Mmm, a combination of simple heuristics, all of which are of course learned, but still simple heuristics, could in itself be a simple heuristic. Yet it could allow performing pretty complex-looking actions, including those you described. Simple heuristic here could be a linea…

What's meant by heuristics is some times unclear. I wonder if by heuristics you mean a shortcut. In CS and AI our model of a shortcut is the heuristic cost functions in heuristic search algorithms like A* and its variants.

It's interesting because I've thought along the following lines. A* is a pathfinding algorithm so it's a natural choice for path planning -the process of planning a path through some environment for an autonomous agent to follow. The funny thing is, as it turns out, pathfinding can be abstracted as finding a "path" through a graph: a set of nodes connected by edges; and that's a great abstraction for general task planning - the task of achieving any arbitrary objective - so A* is also widely used for task planning.

Well, isn't path planning an almost universal ability of intelligent animals? Most animals are motile for some part of their lives and they seem to use their intelligence at the very least to navigate their environment. So is it that far-fetched to think that an ancestral ability for path planning, essentially identical to a heuristic search algorithm like A*, evolved into general intelligence? And wouldn't that mean that general intelligence can be, ultimately, modeled as some kind of heuristic search?

The answer I think is: no, and that's a dangerous way to think. A model is a model, it's not the process it models. And I think that's my fundamental disagreement with lisper, disregarding my confusion about the meaning of "kinematics".

Re: EURISKO Lives

#83

Post by YeGoblynQueenne »

gwern wrote: post #68
YeGoblynQueenne wrote: post #42

Earlier quoted context omitted.

That's the "bitter lesson", right? Which is really a sour lesson- as in sour grapes. See, Rich Sutton's point with his Bitter Lesson is that encoding expert knowledge only improves performance temporarily, which is eventually surpassed by more data and compute. There are only two problems with this: One, statistical machine learning systems have an extremely limited ability to encode expert knowledge. The language of…

Vapnik: https://www.cs.princeton.edu/courses/archive/spring13/cos511... https://engineering.columbia.edu/files/engineering/vapnik.pd... https://www.learningtheory.org/learning-has-just-started-an-... https://nautil.us/teaching-me-softly-234576/ The main paper: https://gwern.net/doc/reinforcement-learning/exploration/act... It sounds kinda crazy (is there really that much far transfer?), but you know, I think it would…

Yeah, that's one of the papers in that line of research by Vapnik. He's got a few with similar content. Visually, it's not the paper I remember, I'll have to read it again to be sure.

If I remember correctly, Vapnik's point is, we know that Big Data Deep Learning works; now, try to do the same thing with small data. Very much like my point that capabilities of models are not improving, only the scale increasing.

Re: EURISKO Lives

#84

Post by YeGoblynQueenne »

adw wrote: post #55
YeGoblynQueenne wrote: post #42

Earlier quoted context omitted.

That's the "bitter lesson", right? Which is really a sour lesson- as in sour grapes. See, Rich Sutton's point with his Bitter Lesson is that encoding expert knowledge only improves performance temporarily, which is eventually surpassed by more data and compute. There are only two problems with this: One, statistical machine learning systems have an extremely limited ability to encode expert knowledge. The language of…

> The language of continuous functions is alien to most humans and it's very difficult to encode one's intuitive, common sense knowledge into a system using that language In other words; machine learned models are octopus brains ( https://www.scientificamerican.com/article/the-mind-of-an-oc... ) and that creeps you out. Fair enough, it creeps me out too, and we should honour our emotions — I'm no rationalist – but we…

Please don't god mode me? Machine learning doesn't creep me out. I'm sorry it creeps you out. In my culture, octopus is a prized delicacy, my dad used to fish them out of the sea with his bare hands when I was a kid. If you wanna creep me out, you should try snake, not octopus.

Re: EURISKO Lives

#85

Post by YeGoblynQueenne »

adw wrote: post #66
YeGoblynQueenne wrote: post #45

Earlier quoted context omitted.

>> Inefficient is a whole lot better than can't even play the game, the story of GOFAI for the last few decades. See e.g. my link above where GOFAI plays the game (Atari) very well indeed. Also see Watson winning Jeopardy (a hybrid system, but mainly GOFAI - using frames and Prolog for knowledge extraction, encoding and retrieval). And Deep Blue beating Kasparov. And MCTS still the SOTA search algo in Go etc. And EUR…

> And MCTS still the SOTA search algo in Go etc It's often forgotten that Rich Sutton said the two things which work are learning (the AlphaGo/Leela Zero policy network) and search (MCTS). (I think the most interesting research in ML is around the circumstances in which large models wind up performing implicit search.)

Well, gradient optimisation is a form of search.

Re: EURISKO Lives

#86

Post by YeGoblynQueenne »

adw wrote: post #65
YeGoblynQueenne wrote: post #54

Earlier quoted context omitted.

They haven't (eaten the world etc). They just happen to be the models that trend hard right now. I bet if you could compare like for like you'd be able to see some improvement in performance from Transformers, but that 'd be extremely hard to separate from the expected improvement from the constantly increasing amounts of data and compute. For example, you could, today, train a much bigger and deeper Multi-Layered Pe…

> much more inefficient The tricky part here is that "efficiency" is not a single dimension! Transformers are much more "efficient" in one sense, in that they appear to be able to absorb much more data before they saturate; they're in general less computationally efficient in that you can't exploit symmetries as hard, for example, at implementation time. Let's talk about that in terms of a concrete example: the big i…

>> Let's talk about that in terms of a concrete example: the big inductive bias of CNNs for vision problems is that CNNs essentially presuppose that the model should be translation-invariant. This works great — speeds up training and makes it more stable – until it doesn't and that inductive bias starts limiting your performance, which is in the large-data limit.

I don't know about that, I'll be honest. Do you have a reference? I suspect it won't disagree with what I'm saying, that neural nets just can't use strong enough bias to avoid overfitting. I didn't say that in so many words, above, but that's the point of having a good inductive bias, that you're not left, as a learner, to the mercy of the data.

>> Someone who comes at things from the perspective of mathematical logic is going to find that worldview very weird, I suspect.

No that's absolutely a standard assumption in logic :) Think of grammars; like Chomsky likes to say, human language "makes infinite use of finite means" (quoting Wilhelm von Humboldt). Chomsky of course believes that human language is the result of a simple set of rules, very much like logical theories. Personally, I have no idea, but Chomsky consistently and even today pisses off all the linguists and all the machine learning people, so he must be doing something right.

Btw, I'm not coming from the perspective of mathematical logic, only. It's complicated, but, e.g. my MSc was in data science and my PhD in a symbolic form of machine learning. See, learning and logic, or learning and reasoning, are not incompatible, they're fundamentally the same.

Re: EURISKO Lives

#87

Post by YeGoblynQueenne »

adw wrote: post #67
adw wrote: post #65

Earlier quoted context omitted.

> much more inefficient The tricky part here is that "efficiency" is not a single dimension! Transformers are much more "efficient" in one sense, in that they appear to be able to absorb much more data before they saturate; they're in general less computationally efficient in that you can't exploit symmetries as hard, for example, at implementation time. Let's talk about that in terms of a concrete example: the big i…

> They haven't (eaten the world etc). To clarify what I mean on this specific bit: the SOTA results in 2D and 3D vision, audio, translation, NLP, etc are all transformers. Past results do not necessarily predict future performance, and it would be absurd to claim that an immutable state of affairs, but it's certainly interesting that all of the domain-specific architectures have been flattened in a very short period…

Thanks for clarifying. Well, my argument is that the state of the art is more the result of trends in research than of the true capabilities of different approaches.

Take my little rant about Rich Sutton's (a god, btw) Bitter Lesson with respect to RL. So, there's AlphaGo, AlphaZero and μZero, yes? AlphaGo knows the rules of Go and starts with some expert knowledge, and beats very human Go player. AlphaZero knows the rules of Go but has no expert knowledge and it beats AlphaGo. And μZero neither knows the rules of Go, nor has expert knowledge, and it beats AlphaZero, and can also plays chess, shoggi and Atari games, with one hand while eating a banana. Do you know how hard it is to eat a banana with one hand? Unpeeled!

Easy to draw a conclusion from that. Except all those systems were developed and used by DeepMind, and there are very few entities besides DeepMind that can even train them, so all we know is what DeepMind claims and we have no way to check their claims. For example, can I test different configurations of μZero, with and without knowledge of the rules of the game and expert knowledge? Not really. And it's clear to me that DeepMind are pushing very, very hard a form of AI that relies on having gigantic resources, like the ones the just completely coincidentally happen to be among the few entities to have access to. So I remain unconvinced.

(I need to re-read the μZero paper, it's in my pdf buffer. I didn't get it the first time I read it, and it might well be that they did make sufficient ablation studies to convince even me and I just don't remember it).

Re: EURISKO Lives

#88

Post by YeGoblynQueenne »

YeGoblynQueenne wrote: post #58
famouswaffles wrote: post #51

Earlier quoted context omitted.

NNs can do the things GOFAI is good at a whole lot better than GOFAI can do the things NNs are good at.

That's wishful thinking not supported by empirical results.

Hey, og_kalu, I vouched for your comment but it stays dead. It's not you, it was me who was out of line, with my comment: "wishful thinking"; that's not a very polite thing to say. And my original comment was a bit prissy, too.

To be honest, I'm always a bit jumpy around your comments because I've noticed them all over the place and they're often grayed-out. You kind of tend to go for the jugular. I don't mean that as a good thing. I think others have noticed it too and you get more reaction than you should. That's a shame, because it's clear there's lots of interesting conversations to be had, given you have such strong views and you seem to have done quite a bit of reading; though only on one side of things.

Anyway sorry for starting it this time around and that you got dead'ed, I hope we get to disagree more in the future.

Re: EURISKO Lives

#89

Post by lisper »

YeGoblynQueenne wrote: post #81
lisper wrote: post #72

Earlier quoted context omitted.

Let's go back to the original formulation so we don't lose the plot here: Me: As an analogy, consider a professional tennis or baseball player. YeGoblynQueenne: humans e.g. playing baseball do not find solutions to kinematic equations, but instead use simple heuristics that exploit our senses and body configuration, like placing their hands in front of their eyes so that they line up with the ball etc. At the risk of…

>> But even just moving your hand in front of your eyes is nowhere near as trivial as YeGoblynQueenne implies. Yeah, I was actually thinking of kinematics as in classical mechanics. I think you were speaking about kinematic equations as in robotics. My bad, I misunderstood. I agree that moving your hand in the right place is not a simple problem, and I don't actually have an insight into that, but I think it's easier…

> Yeah, I was actually thinking of kinematics as in classical mechanics. I think you were speaking about kinematic equations as in robotics.

What do you see as the relevant difference?

> I think it's easier than calculating the trajectory of an object, let alone many at once

Well, yeah, but there isn't anything fundamentally more difficult about juggling. It all boils down to Newton's laws.

My point is that there are two different ways that human brains can apply Newton's laws. We can do it intuitively, without even being consciously aware of Newton's laws, which is why humans were able to throw and catch objects before 1687. Or we can do it consciously by manipulating symbolic representations of the equations of motion. Those two activities are in some sense equivalent because they both involve producing a model of a physical system in our brains and using that model to make accurate predictions about that system. But they are also obviously radically different in other ways, and being skilled at one in now way implies being skilled at the other.

Re: EURISKO Lives

#90

Post by lisper »

YeGoblynQueenne wrote: post #80
lisper wrote: post #63

Earlier quoted context omitted.

Sorry about that, I'm dealing with a troll on another thread so I'm on a bit of a hair trigger. I think we have a fundamental disconnect somewhere, so let's try to diagnose it. Where do you start to disagree in the following series of claims: 1. People can have kinematic skills, like throwing and catching balls, without having math or physics skills, like solving kinematic equations. 2. In order to have kinematic ski…

>> Sorry about that, I'm dealing with a troll on another thread so I'm on a bit of a hair trigger. Hey, no worries. Thanks for being a gentleman and I'm sorry you're being harassed. Btw, just to be clear: I'm perfectly fine with robust disagreement, I just don't deal well with personal attacks; which you didn't do, I was just worried that's where this conversation was going. So, thanks for the very detailed analysis…

> it's entirely possible for humans to be using a completely different process to catch balls etc, than kinematic equations.

This all turns on what you mean by "completely different". Yes, obviously when you learn to actually catch a ball your brain is not doing anything that maps straightforwardly onto the kinds of symbolic manipulations that happen when you do math. On the other hand, it has to map onto doing math somehow even if that mapping is not straightforward. The only other possibility is that your brain is actually doing something that doesn't map onto math in any way, but still somehow produces the same results that math does by sheer coincidence. If you could actually demonstrate that, it would be one of the biggest breakthroughs in the history of science because it would refute the Church-Turing thesis.

Post reply on HN