Go master Lee Se-dol says he quits, unable to win over AI Go players • Hacker News Board

Re: Go master Lee Se-dol says he quits, unable to win over AI Go players

#491

Post by 1123581321 » Thu, Nov 28, 2019, 5:51 AM UTC

dragontamer wrote: post #284

CydeWeys wrote: post #276

Earlier quoted context omitted.

Even if it is 100% deterministic (which I'm not convinced of, especially seeing as how distributed and thus ordering-dependent it is), if it's the best in the world, how does that help you compete against it? In order to take advantage of its determinism you'd need to be better than it, and nothing else is.

> if it's the best in the world, how does that help you compete against it? In order to take advantage of its determinism you'd need to be better than it, and nothing else is. Play AlphaGo against itself. Go rarely has draws (requires Triple-Ko, a very, very rare position). Almost every game you play with AlphaZero vs AlphaZero will result in a winner-and-loser. You will quickly be able to characterize the positions…

Go engine training surpasses this basic level of self-play effectively instantly.

The strongest early moves create the most potential for winning (maximizing potential winning paths, sort of); they do not push the game towards one best end state. They do not have counters. I saw elsewhere you have some understanding of the game (15kyu) so you should be able to demonstrate this to yourself by playing some of AlphaGo’s openings on a board and trying to write deterministic counters to them. You will not be able to push the AI into a situation where it has too few options to avoid loss. You will also find you need to create a book much larger than a few moves to meaningfully predict play and so will exceed the number of states that can be stored (referencing your 16TB comment elsewhere.)

Please actually try this as I think it is a key to improving your skill in addition to understanding the challenges in automating play.

Re: Go master Lee Se-dol says he quits, unable to win over AI Go players

#492

Post by MiddleEndian » Thu, Nov 28, 2019, 6:04 AM UTC

SubiculumCode wrote: post #489

55555 wrote: post #474

Earlier quoted context omitted.

> Humans do not run to beat horses. We never compete. That would be silly I hate to be that guy, but: https://en.m.wikipedia.org/wiki/Man_versus_Horse_Marathon It is very silly indeed.

2004 2007 a human won. Wow.

The race's setup is horse-friendly as well: Shorter than a regular marathon by a few miles and on flat terrain. A longer distance on hillier terrain would be won by humans much more often.

Re: Go master Lee Se-dol says he quits, unable to win over AI Go players

#493

Post by UnbugMe » Thu, Nov 28, 2019, 6:18 AM UTC

needToCrust wrote: post #445

Zaak wrote: post #430

Earlier quoted context omitted.

So you're defining "AI" to be anything that we can currently program a computer to do, and "I" to be anything we can't yet? That doesn't seem like a useful distinction to me. Unless you're using "I" to mean general (artificial) intelligence, in which case you should probably use the more well-known term.

No, please don’t straw man my point. I’ll assume you know what ai is and that you understand there is a huge difference between that and human intelligence. I am arguing that ai will never be able to DM a D&D game. For that, a computer will need human intelligence.

But AI definition is still target-in-motion, very blurred.

Re: Go master Lee Se-dol says he quits, unable to win over AI Go players

#494

Post by p1esk » Thu, Nov 28, 2019, 6:20 AM UTC

amelius wrote: post #135

neohaven wrote: post #93

Earlier quoted context omitted.

As a person who likes music, making it, listening to it, breaking it down and hacking it... Making a classical arrangement that evokes a particular expression in the listener is the job of the musician. If an AI system helps you explore the possibilities there, it's more like a studio musician that's able to improvise. You're still the person, the human, the emotional filter, that picks "This sounds right" or "This d…

The point is: what if the tool becomes so great that practically anybody can use it? Anybody could be that "filter" and "be" a great musician.

http://aiva.ai - anybody could be that "filter"

Re: Go master Lee Se-dol says he quits, unable to win over AI Go players

#495

Post by EsssM7QVMehFPAs » Thu, Nov 28, 2019, 6:20 AM UTC

babesh wrote: post #226

avs733 wrote: post #180

Earlier quoted context omitted.

Another way to look at your metaphor is that research on exercise physiology has shown the enormous importance of rest and proper nutrition during training. Prior to cars, getting from place to place and access to optimal nutrition were both mediated by transport over long distances. AI is the future of Go because it enables those new perspectives and new processes by which human players can learn. AI is smart-dumb,…

Do these AIs actually have much of a strategy? Is the strategy mostly correct evaluations of positions and optimized search?

The search space of Go is way too large for dumb traverse of the tree, even with high end optimizations.

What makes recent breakthroughs in AI agents playing adversarial games possible is the fact that deep neural networks are able to develop patterns that yield short- and long-term strategic planning. And the ability to self train without human intervention to reach unprecedented training levels.

Re: Go master Lee Se-dol says he quits, unable to win over AI Go players

#496

Post by EsssM7QVMehFPAs » Thu, Nov 28, 2019, 6:32 AM UTC

dragontamer wrote: post #233

btilly wrote: post #230

Earlier quoted context omitted.

This comment sounds reasonable, but shows that you don't understand the challenge of Go on computers very well. Go does not have openings. It has reasonable choices to make, with an insane branching factor, with few moves making much of a difference by themselves. Therefore your database will only extend a few moves, and all of the positions that it winds up with will still be very close to even. So your database con…

> Go does not have openings. There are 381 opening moves in Go, but really only 96 because of symmetry. 96 (opening moves) x 380 responses x 379 x 378 x 377 == ~2 Trillion positions after 5 ply. These 2-trillion positions will easily fit in a 16TB hard drive for $400. That's 8-bytes per position, so you probably can get there with more symmetries and some compression applied. ---------- You're thinking too much like…

> Alpha-Go doesn't seem to implement much randomness at all into the moves it plays. The source of randomness is in time-controls (AlphaGo may choose MoveX before 30 seconds of analysis, or MoveY after 30 seconds of analysis), but this is a fairly constrained number of moves.

The basis of reinforcement learning algorithms is the exploratory nature of learning due to the initial application of largely random moves.

Only after some time the agent is given confidence into his learned ability and grafually moved into a more deterministic behaviour mode.

This is the exact opposite of your statement. Star Craft players have noted that the fleet of different AlphaStar instances training in ensemble observed very different behaviour due to this property of RL.

Re: Go master Lee Se-dol says he quits, unable to win over AI Go players

#497

Post by kerkeslager » Thu, Nov 28, 2019, 6:43 AM UTC

dragontamer wrote: post #233

btilly wrote: post #230

Earlier quoted context omitted.

This comment sounds reasonable, but shows that you don't understand the challenge of Go on computers very well. Go does not have openings. It has reasonable choices to make, with an insane branching factor, with few moves making much of a difference by themselves. Therefore your database will only extend a few moves, and all of the positions that it winds up with will still be very close to even. So your database con…

> Go does not have openings. There are 381 opening moves in Go, but really only 96 because of symmetry. 96 (opening moves) x 380 responses x 379 x 378 x 377 == ~2 Trillion positions after 5 ply. These 2-trillion positions will easily fit in a 16TB hard drive for $400. That's 8-bytes per position, so you probably can get there with more symmetries and some compression applied. ---------- You're thinking too much like…

Gosh, it's sure clear that you don't know what you're talking about.

At 5 ply, the complexity of the game hasn't started in any meaningful way. In a typical game, that's 4 corner moves, and then one of a: an approach to a corner (kakari), b: an enclosure (shimari), c: a wedge (waruichi) or d: creating a side framework (such as the Chinese fuseki or Sanrensei). There are some odd opening such as tengen, or corner-corner-corner-kakari which typically turns into a sente fight, but 99% of games will fall into the aforementioned pattern. The database you describe is about as useful as a database of amateur games, since most games, including AlphaGo's games, follow just a few basic openings that early, and even amateurs can play these first few moves "correctly".

Even if you get out to 10 ply you're still only getting partway into a single joseki sequence, often leaving three whole corners of the board which haven't even been approached, so this database still isn't very useful.

Incidentally, your numbers are also wrong. Symmetry reduces the first move to 55 possibilities, not 96, and there are 361 points on the board, not 381.

Re: Go master Lee Se-dol says he quits, unable to win over AI Go players

#498

Post by s_Hogg » Thu, Nov 28, 2019, 7:01 AM UTC

cmonnow wrote: post #477

braythwayt wrote: post #460

Earlier quoted context omitted.

> It used to be that you might be able to believe that there is some kind of art behind go, some sort of abstract beauty to it, and that the pursuit of this beauty is the path to being good at go ... All but the last phrase are still true. Pursuit of Go for beauty is still pursuit of Go for beauty. I will never play any instrument as well as a sequencer. I believe algorithms will make beautiful Jazz improvisations in…

'AI' is not an adversary. It is simply a tool created by hundreds of humans coming together. It's like watching 1-v-100 boxing match. Of course the 100 are going to win. Enjoyable sport has always been about ~similarly matched opponents. When we have DeepMind AI Go vs. MindDeep AI Go, that's when things get interesting.

That's kind of what we already have though, given that these things are often trained by playing against themselves these days

Re: Go master Lee Se-dol says he quits, unable to win over AI Go players

#499

Post by TheGallopedHigh » Thu, Nov 28, 2019, 7:34 AM UTC

2bitencryption wrote: post #372

hindsightRegret wrote: post #254

Earlier quoted context omitted.

I think achieving superiority over humans is practically solving the problem though. Solving chess or go by going through a complete search space seems more like a hardware/computational goal than a practical ml/ai goal.

It all hinges on your definition of "solved". "Solved" in the AI/game theory has a very strict definition. It indicates that you have formally proven that one of the players can guarantee an outcome from the very beginning of the game. The less-strict definition being thrown around here in the comments is more like "This AI can always beat this human because it is much stronger."

I think most people discussing this mean the later, less pedantic option. I mean, that’s the spirit of AI. Can we make it think like a human, or even more so. We are the yardstick.

Re: Go master Lee Se-dol says he quits, unable to win over AI Go players

#500

Post by bsaul » Thu, Nov 28, 2019, 7:44 AM UTC

braythwayt wrote: post #460

breatheoften wrote: post #374

I sympathize. It used to be that you might be able to believe that there is some kind of art behind go, some sort of abstract beauty to it, and that the pursuit of this beauty is the path to being good at go ... But the defeat of the tactics born from this mindset by MCTS at least for now lay bare the fact that the path to being good at go is actually to probabilistically sample the phase space of the game and perfor…

> It used to be that you might be able to believe that there is some kind of art behind go, some sort of abstract beauty to it, and that the pursuit of this beauty is the path to being good at go ... All but the last phrase are still true. Pursuit of Go for beauty is still pursuit of Go for beauty. I will never play any instrument as well as a sequencer. I believe algorithms will make beautiful Jazz improvisations in…

About jazz improvisation: I knew people at the creative labs research center back in 2003 that told me a researcher once showed them a program that did just that. It was able to improvise « the way player x would » by just listening to it, and it would continue the impro in the same style.