AlphaZero paper discussion (Mastering Go, Chess, and Shogi)

[Uberdude]

I have tried to warn against treating AlphaGo as a cargo cult, so welcome this reminder from an AI professional.

His recent output is not of professional quality.

I'm also constantly reminded of something that Michael Redmond said: go is a very simple game. It may be hard for humans but change the tool and Redmond's proven right.

In some ways I think Go is actually relatively easy for humans given its game tree/state complexity compared to chess and shogi: because the pieces don't move it makes it easier to visualise and chunk, and we have excellent neural networks for visual patterns. One of the points Demis made in the Q&A after a recent AlphaGo film showing was that he is pretty confident (but weren't the last failures, hah!) this latest generation of AI research with neural networks was on the right track to general AI is that it is informed and inspired by a system we know works, namely our brains. He reckoned now we were on maybe the 2nd rung of several dozens.

[moha]

IMO intelligence means ability to solve previously unseen tasks, so I'm not sure if these examples qualify, even as early birds. Unless matchboxes are intelligent .

Yes, I regard intelligence as the ability to do something well that you have never done before.

@Bill: I'd say that I'm not really intelligent at anything, then. I rarely do something well the first time I try it. Only after practice can I get any sort of competency.

This would still fit, at least by my definition. I didn't mean right at first try, just reasonably soon.

This is the biggest problem with RL today, which is why it doesn't seem intelligence - yet. Basically it is usable only where the environment is known and can be simulated, because of the insane amount of experimental data (trial and error) required. Humans - maybe animals in general - always needed to learn and to adapt fast, to stay alive. But this may change for AI in time, neural networks are still in their infancy. Perhaps real cognitive abilities will also be developed.

BTW, it's amazing Deepmind never tried anything that would be out-of-the box. The first time a programmer tries NNs, the first thoughts are often about potential changes (new activation functions / network structures). But they just took the state of the art (from the visual field), and applied it vanilla to go/chess. Perhaps as a demonstration that these things are already somewhat mature.

[dfan]

Me too; in my case, that's since 1971; but i don't think that during that time AI has advanced much compared to advances in computer hardware, despite the current hullabaloo.

I have tried to warn against treating AlphaGo as a cargo cult, so welcome this reminder from an AI professional.

As another data point, I am an AI professional and think that AI has advanced hugely since 1971. Of course computer hardware has advanced a ridiculous amount too! Most (all?) current AI applications would not have been feasible on 1971 hardware but that doesn't cancel out the huge amount of progress that has been made.

To put it a different way: if you handed a bunch of 1971-era AI researchers 2017-level hardware, they wouldn't be able to create AlphaGo (until they had duplicated decades of research).

Meanwhile, on the subject of centipawns vs. winning percentage: one thing to keep in mind is that chess has three results. You can just treat draws as 1/2 (as tournaments do) and calculate expected value, but there really is a difference between positions that are 40%W 20%D 40%L and 5%W 90%D 5%L. And this difference is meaningful not just on the board but off of it; if you had the choice between those two positions, your choice would depend a lot on your position in the tournament (if you're half a point behind the leader and it's the last round, not much matters besides the %W bin). I haven't looked carefully at the AlphaZero paper yet so I don't know what they do here, though I would expect (no pun intended) that they just try to maximize expected value.

[moha]

if you handed a bunch of 1971-era AI researchers 2017-level hardware, they wouldn't be able to create AlphaGo

They weren't hopeless though.

[dfan]

if you handed a bunch of 1971-era AI researchers 2017-level hardware, they wouldn't be able to create AlphaGo

They weren't hopeless though.

Indeed, though note that TD-Gammon was developed 21 years after 1971 and benefited from (and extended!) the theories of reinforcement learning developed during those decades. (I saw Tesauro talk at a conference last year about his experience creating TD-Gammon and it was quite interesting.)

[Uberdude]

I don't understand much of backgammon, bur found it interesting that TD-Gammon upended some opening theory and made humans start playing moves they previously thought were bad, rather like AlphaGo and the early 3-3 invasions.

[djhbrown]

His recent output is not of professional quality.

Please can someone who is competent to judge explain why Uberdude who has never manifested any evidence of professional competence in AI has licence to make repeated personal abuse attacks upon me in this forum over several years without ever receiving a reprimand from a moderator?

Being good at Go does not make one good at being human.

[Kirby]

if you handed a bunch of 1971-era AI researchers 2017-level hardware, they wouldn't be able to create AlphaGo

They weren't hopeless though.

Indeed, though note that TD-Gammon was developed 21 years after 1971 and benefited from (and extended!) the theories of reinforcement learning developed during those decades. (I saw Tesauro talk at a conference last year about his experience creating TD-Gammon and it was quite interesting.)

I agree that advancements have been made from AI outside of hardware. But it's also hard to separate the two. For example, a number of research advancements may have been made possible with increased computing power and availability.

It's somewhat of a hypothetical argument, similar to how people talk about who would be stronger between a "modern day Shusaku" and, for example, Ke Jie. It's nice to think about as a thought experiment and make various reasoning, but the idea can't extend the limits of its hypothetical nature.

As to whether 1971 researchers could replicate AlphaGo, I'd say that dfan is probably right, but who really knows?

[Kirby]

His recent output is not of professional quality.

Please can someone who is competent to judge explain why Uberdude who has never manifested any evidence of professional competence in AI has licence to make repeated personal abuse attacks upon me in this forum over several years without ever receiving a reprimand from a moderator?

Being good at Go does not make one good at being human.

OK. As an admin, I'd like not to take sides on the matter. Let's remain cordial with one another. For all in the thread, please refrain from personal attacks, whether that means questioning an individual's professional credentials, or questioning their qualities as human beings.

Let's keep the discussion related to topics related to AlphaZero, AI, Go, and the like - not about individuals.

[dfan]

I agree that advancements have been made from AI outside of hardware. But it's also hard to separate the two. For example, a number of research advancements may have been made possible with increased computing power and availability.

Oh, this is 100% the case. There was a lot of neural network research in the 1980s that seemed like a dead end because it would require too much computing power to be useful. Thirty years later...

As to whether 1971 researchers could replicate AlphaGo, I'd say that dfan is probably right, but who really knows?

Maybe they'd get there eventually! But a lot of AI research has been done in the last 46 years, and I'm pretty reluctant to believe that it all could have been reproduced in a couple of years or something if only computers were faster. A lot of very smart people did a lot of very hard thinking for a long time to get where we are now.

[Kirby]

Probably correct, dfan.

[pookpooi]

In a rare occasion American Go Association post an article from chess.com on their E-Journal but IMHO they should add details on AZ outclass AGZ 20 blocks as well, otherwise the article is not related to Go at all.

[djhbrown]

There was a lot of neural network research in the 1980s that seemed like a dead end because it would require too much computing power to be useful.

That's an overstatement and a misattribution of cause, since it was not lack of computing power but lack of functionality that held NNs up - the breakthrough was Le Cann's idea of convolutions. Go-wise, the idea of trying out Monte-Carlo search was also a quantum leap, so putting 2 and 2 together was a smart move.

So far, so good...

Where to next? A Go program that can talk? Or one that can see eyes? NNs can separate images containing a cat from those that don't, but can they draw a line around the cat?

[Gomoto]

I for one do not care if AI is not there yet.

If it plays go and wins, I like it.

[Gomoto]

I learned some things about life, while playing go.

Therefore if it plays go, it knows some things about life too.