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

[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.

Do you know where this definition comes from? I have used it since my childhood, I don't remember the origin but I doubt I came up with it myself. A few years ago I tried to look up the source but failed.

Most contemporary definitions seem to include things like lexical knowledge and it's application ability, so very different. And there is this one too: "intelligence is what intelligence tests measure".

[Uberdude]

How about intelligence is knowing there are better things to do than join Mensa

[Bill Spight]

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.

Do you know where this definition comes from? I have used it since my childhood, I don't remember the origin but I doubt I came up with it myself. A few years ago I tried to look up the source but failed.

Most contemporary definitions seem to include things like lexical knowledge and it's application ability, so very different. And there is this one too: "intelligence is what intelligence tests measure".

Well, it is not exactly a definition, but I came to it when I heard a real story about someone doing something well that they had not done before. What primed me to think that that was intelligence, I don't know.

OC, I had long doubted whether IQ tests test anything, factor analysis be damned.

[Kirby]

it makes me question the meaning in life for humans. If it's intelligence, then computers are beginning to have more meaning than humans.

That is still in the future. 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.

Fairly obviously, I think that there are different kinds of intelligence.

@moha: Nice video. That matchbox experiment is pretty cool.

@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.

Best to take a detached Zen-like view, and recall that life, like go, is a game of co-existence.

Good advice, John.

An often overlooked implication of rooting the meaning of human life or the moral status of humans in our special intellectual capacities is that human beings lacking them have less meaningful (or non-meaningful) lives or do not have the moral status of unimpaired human beings. This is an implication that I certainly find problematic, and so it tends to undermine the idea that we should base our understanding of these concepts in the special intellectual capacities of human beings.

You bring up a good point, Monadology. I also find it problematic to equate moral status with particular intellectual capacities, even for unimpaired human beings. My perspective was more reflective in terms of finding self-worth, perhaps (e.g. what makes me valuable as a human being?). But maybe the implications are similar. If I try to equate my value as a human being, I suppose there is still the aspect of being a father to my children, and a husband to my wife. I'd say that I bring some value from that dimension, which can't currently be replaced by machines. Maybe that status could be replaced by another person, but I hope it won't be.

Maybe people are like go stones - no inherent value, but useful at the right place and the right time, under the right circumstances, working in coordination with the rest of the universe...

Or maybe I'm just pushing an analogy too far :-p

[sorin]

I'm wondering why DeepMind choose Shogi out of many board games beside Go

Another usual suspect: Xiangqi ; ...Chaturanga.

For diplomatic reason, choosing Xianqi to conquer by (non-Chinese) AI is a really bad choice.

Really? How about Go then?

[pookpooi]

Really? How about Go then?

See the 'Chinese coverage' part of Future of Go Summit article in Wikipedia

[Uberdude]

Really? How about Go then?

See the 'Chinese coverage' part of Future of Go Summit article in Wikipedia

I thought that was more of the Chinese government not wanting to give free PR to a foreign tech company which doesn't play by their censorship rules than being upset about foreigners doing part of their culture better with computers. I presume the best Chinese chess players are Chinese though (and the bots too? how good are they?) so if you wanted another exhibition match that'd be in China and face the same PR problems.

[Bill Spight]

I rarely do something well the first time I try it. Only after practice can I get any sort of competency.

Well, that's how to get to Carnegie Hall.

[djhbrown]

I've been quite interested in machine learning, and artificial intelligence in general, since college.

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. For sure, Alfie's impressive achievements are remarkable, but could it be that they tell us more about the nature of games like chess and Go than they do about the nature of intelligence?

To examine that question deeper than handwaving, we need to be aware that Alfie's method is basically a statistical number-crunching approach, which requires hardware able to read (probabilistically) all the way to the end of the game and back again gerzillions of times each time she has to make a move.

She can do this because although the game tree of Go is kind of big (sic), it's not that deep - at most 361. To be fair, her probabilistic navigator is sufficiently better than random to weed out enough nonsense that she can find moves good enough to beat people at the game they invented.

Alfie is as impressive at playing Go as a mobile phone is at recording and transmitting information - and considering the state of the telephony art 50 years ago, that's pretty impressive.

But is Alfie as smart as, say, an acacia tree that responds to overgrazing by releasing poison gas to ward off kudu?
https://www.youtube.com/watch?v=Q-4w5xYLwiU
Or a Douglas Fir that nurtures its young?
https://www.youtube.com/watch?v=CrrSAc-vjG4

The robot football world cup maybe gives us a better idea about how smart today's machines are.
https://www.youtube.com/watch?v=R2S9x0gUkpM

PS Needless to say, i think Swim is more intelligent than Alfie, but i suppose you could justifiably argue that i am a bit biased...

PPS Don't fret Kirby, your existence is not as meaningless as that of a Go stone, albeit no more meaningful than that of an acacia tree:
https://www.youtube.com/watch?v=wq3B5prBsK0

[John Fairbairn]

I thought that was more of the Chinese government not wanting to give free PR to a foreign tech company which doesn't play by their censorship rules than being upset about foreigners doing part of their culture better with computers

I think this is right. When I was in China a couple of months ago, talking to politicians, the people who run Chinese go and pros, AlphaGo was mentioned approvingly over and over again (and FineArt not once, I think). DeepMind got some mentions, But Google was never mentioned, except in a couple of conversations where I was reminded (by ordinary punters, not officials) about censorship. But actually it's similar in Japan, except that lack of mentions (though not total) of Google (but plenty of AlphaGo and DeepMind and Hassibis mentions) is due to self-censorship. Although there are far, far fewer people who talk about go in Britain, I'm certain it's close to that situation here, too (plus DeepMind is British, Google isn't). And throughout the world, although billions of people happily use Google/Amazon/Facebook etc, if you did a word association test for each company, I'm sure you'd get lots of words in the "evil" nexus.

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. For sure, Alfie's impressive achievements are remarkable, but could it be that they tell us more about the nature of games like chess and Go than they do about the nature of intelligence?

I have tried to warn against treating AlphaGo as a cargo cult, so welcome this reminder from an AI professional. 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. Inserting a screw in a brick wall by hand is not just hard for humans, but impossible. Yet with the right tools it's a doddle. (And we don't get any Angst about that, do we?)

[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.)