Acquisition of chess knowledge in AlphaZero
en.chessbase.com
en.chessbase.com
To me, this is a great illustration that superhuman intelligence is not magic. When a superhuman AI plays chess, the moves usually make sense to an expert. And even if they don't immediately make sense, they usually make sense once the expert plays out some lines to see what happens. Superhuman AIs have crushed humans at chess not by discovering new, wildly counterintuitive openings that dumb humans missed (it actually plays many of the same openings). Rather, it just does what humans do - win pieces, keep your king safe, attack, gain space, restrict your opponent's moves - but better and more consistently.
This paper builds on that concept and finds that not only are a superhuman AI's moves usually understandable by experts, but even some of the superhuman AI's internal representations are understandable by experts (!).
In the 5th century BCE, Greek philosophers thought the Earth was a sphere. Although they were eventually improved upon by Newton's ellipsoid, they still had arrived at the right concept. And here I think we see the same thing: although a superhuman AI 'understands' chess better than human experts, its understanding still builds upon the same concepts.
Thinking broadly, like, suppose we invent some superhuman AGI for $100B in 2100. And then we ask it how to optimize our paperclip production. I don't think its recommendations will be magic. Paperclip production is a fairly well understood problem: you bring in some metal, you reshape it, you send it out. It's probably gonna have much of the same advice we've already figured out: invest in good equipment, take advantage of economies of scale, choose cheap but effective metals, put factories near sources of energy and labor and raw materials, ship out paperclips in a hierarchical network with caches along the way to supply unanticipated demand, etc. Like, an AGI's suggestions may well be better than a management consultant's, but it's not like it's going to wave a magic wand and flip the laws of physics. Rather, I expect it to win within the same frameworks we've developed, but by doing things better and more consistently than humans.
Do you actually play chess? Half the moves made by an AI at 3500+ ELO (e.g. Stockfish 13+) is completely alien to even a human super GM like Nakamura or Carlsen. It's actually the most reliable and used way to know whether someone is cheating or not, if the moves don't make sense to high level players, then it's very probably someone using an engine.
> And here I think we see the same thing: although a superhuman AI 'understands' chess better than human experts, its understanding is still leverages the same concepts.
That is very far from being a truth that lots of chess experts would agree on. Most chess players at high level reason with similar, well established concepts, such as tempo, positions, baits, color dominance, etc.
Super AIs don't use these concepts, and it's obvious when you see their moves. They often break tempo abruptly, disregard color dominance, enter what looks like dangerous positions,etc.
Yes. I have a chess set on my kitchen counter from playing blindfold last night. I led my high school to the state championship. I am rated ~2000 in blitz and crazyhouse on Lichess. I watched 10+ analysis videos of AlphaZero when it came out.
If you have a criticism of my claim, just state the criticism. Even if I was a grandmaster, I would not be immune from being wrong.
And to be clear, my claim is that human experts usually (not always) understand the pros and cons of a move made by AlphaZero. My claim is not that human experts usually know why one move, with its set of pros and cons, was selected over another move, with a different set of pros and cons.
And of course, there are exceptions. Super long endgame sequences, for example. Even with a lifetime I doubt any human can understand these sequences beyond the tautological "it's winning because it wins": https://en.chessbase.com/post/just-one-of-17-823-400-766-pos...
It’s more than human experts apply a posteriori the tools they typically use to analyse chess games to games played by AlphaZero. These tools are specifically crafted to analyse chess so they obviously shed some lights on what’s happening but even then some plays were seen as counterintuitive.
An easy example is how AlphaZero can sacrifice pieces or tempos to cripple the mobility of the opposition pieces. If you carefully analyse the full game, you can see that it’s probably why the move was made but it’s completely counterintuitive when it’s played as the relative value of some winning tradeoffs made by AlphaZero are impossible to evaluate for a human player.
I largely agree with this and I was about to make the same point before reading your comment. However, as GP said upthread:
> This paper builds on that concept and finds that not only are a superhuman AI's moves usually understandable by experts, but even some of the superhuman AI's internal representations are understandable by experts (!).
Internal structures mapping to concepts high level players already use is a good indicator that the AI is "thinking" about chess in the same way as humans do.
Even if most moves are normal, clearly AlphaZero and other bots are doing something different or they wouldn't be world's best.
The analysis in the paper draws the exact opposite conclusion. Alpha zero’s play agrees remarkably with humans despite not seeing a human game (e.g. it found the Berlin defense to the Ruy Lopez).
Also your point is trivially not true for other engines because they use books from human-played games.
Engines don’t win by playing novel moves all the time that humans can’t comprehend. They win by playing the most accurate move every time (never making a mistake).
Books are only used in openings, and in (some) endings, in the form endgame tables. Throughout the middle game - where most fireworks happen - an engine is on its own.
And yes, engine's choices are often incomprehensible even for an elite human player (although it isn't literally every second move).
There was this Polish player, Patrycja Waszczuk, accused of cheating (she got banned for 2 years but she appealed, there's an ongoing legal battle).
The case got quite a bit of attention, and parts of her suspicious games ended up analyzed by Nakamura (a world class player, for those unfamiliar with the field).
There've been several moves which left him dumbfounded, as in "no human would even consider that in this position". I don't have a link, but it can be found on YT.
Incomprehensible moves are going to overwhelmingly be from deep searches while fast polynomial time heuristics such as that would come from neural networks are much more likely to be human discernible.
Sure, but realizing what the reasoning could be in retrospect (sort of reverse engineering) is by definition easier than coming up with a (paradoxical) move in the first place, so if you can't even do the former...
"polynomial time heuristics such as that would come from neural networks are much more likely to be human discernible."
AlphaZero's playing style definitely feels more human-like, as it has a "speculative" quality to it.
To quote no other than Kasparov:
"I admit that I was pleased to see that AlphaZero had a dynamic, open style like my own. The conventional wisdom was that machines would approach perfection with endless dry maneuvering, usually leading to drawn games. But in my observation, AlphaZero prioritizes piece activity over material, preferring positions that to my eye looked risky and aggressive. Programs usually reflect priorities and prejudices of programmers, but because AlphaZero programs itself, I would say that its style reflects the truth."
It's probably a bit deceptive, because what looks risky to us, probably isn't really perceived as "risky" by the neural network, which can "feel" the deeper correctness of the decision, much unlike traditional engines, whose conservative cautiousness stems from their inherent limitations. (We know that now - before AZ it was assumed this is simply the "right" way to play, just like Kasparov says).
Not in this situation. While the chess engine performed a network guided non-deterministic search and returned a result, its neural network did not. With tools like in the paper, we can now map the network's internal states to something close to human relatable concepts.
Even without such tools we could observe what lines were productive by analyzing a few ply deep and comparing with other moves we could have made. It's closer to something becoming obvious with hindsight, where seeing what works compared to what you thought of helps to highlight the flaw in your plan or something you missed, while reinforcing your understanding of the engine's moves. That scenario is a common enough one.
Wikipedia says (about AlphaGo):
"AlphaGo's playing style strongly favours greater probability of winning by fewer points over lesser probability of winning by more points. Its strategy of maximizing its probability of winning is distinct from what human players tend to do which is to maximize territorial gains, and explains some of its odd-looking moves."
In other words optimizing the likelihood of winning looks strange to experts who use other heuristics to assess positions.
Any decent player will prioritise win probability over loss minimisation, even to the point of sacrificing multiple Pokémon to get a single turn of setup with your main sweeper.
While it seems confusing to newbies, experienced players know exactly why they're making the choices that they are.
My guess would be that for chess it's more difficult to gauge the difference in real time.
There are some amazing games of Alpha Zero playing Stockfish (the other best chess program) where it seems Alpha Zero just doesn’t give a fuck. It’s just sacrificing it’s own pieces and not clear (to a human… or Stockfish) what the plan is. Later it achieves a zugzwang against Stockfish, which is the most humiliating thing it could do (if either program had emotions).
[1]https://www.cs.toronto.edu/~ashton/pubs/maia-kdd2020.pdf (scroll down for the graphs)
The method you're referring to is based on detecting how often a human picks the top move out of 2-3 similarly rated ones and most of the time understanding which one of the 56.5%, 57.1% and 57.3% moves is the best is difficult but understanding why 55% move is better than 51% one is usually pretty easy. I am using AlphaZero/Lc0 evals here btw as I find them more intuitive than traditional engines ones.
I think your opinion might have been formed many years ago. There was indeed a time where engines like traditional Stockfish (before NNUE) were already much stronger than humans but their strategic play made little sense. It turned out that they were in fact weak strategically and modern crop of engines showed it. Try running Lc0 with modern network vs say Stockfish 7 and you will see games which could be described as humiliating defeats one after another.
It turned out that we are actually very good at figuring chess out. Not very precise, slow and inconsistent but we've got most of the high level concepts right.
It doesn't mean engines don't come up with occasional move that is just too deep to get without long analysis and too counterintuitive to consider during play. It's just rare. Most of the moves are intuitive, and make a lot of sense to a human.
The usual way to detect cheating is detecting incredible accuracy over a whole game, not individual engine like moves.
No they're not. GMs get thrown out on certain moves, because we tend to play chess strategically, that is, we make a plan and try to find ways to make it work. AIs look like they have a plan and then immediately change course when they compute it's beneficial to do so.
As GP says, it takes a minute or two, but those same GMs will (Nakamura does this constantly in some of his streams) reevaluate the strategy, play some lines to see what is tripping the AI and then figure it out. It's that rapid change of strategy from, as an example, extremely defensive to wildly aggressive, which makes them suspicious that they're playing a computer.
> And here I think we see the same thing: although a superhuman AI 'understands' chess better than human experts, its understanding is still leverages the same concepts.
In that I'd agree with you that GP is wrong. For the AI there's no concepts, it's evaluating the chance of success, the perceived "weights" are irrelevant, you could map any number of concepts and get a correlation between them.
Reminds me of comparisons between Renaissance and other quant funds. If you get enough better it looks like what you're doing is magic until you analyse what was done and see they have better systems and people.
Sometimes, magic is just someone spending more time on something than anyone else might reasonably expect - Teller
"Renaissance Clients Exit After Firm’s Anemic Run of Results"
https://www.bloomberg.com/news/articles/2021-02-08/renaissan...
See GothamChess' very entertaining analysis: https://www.youtube.com/watch?v=8dT6CR9_6l4
It's really interesting to look at very high level players analyzing specific AI moves. Many they understand after a short while, others take longer - and some they don't understand directly, but might have an inkling as to why it's good.
My feeling is that many high level analysts would rather just automatically defer to the AI's analysis rather than challenging it - so there is in some sense a lack of feedback to correct such situations. Even in the 1997 Deep Blue vs Kasparov match, game #1 Deep Blue apparently was a bugged position so it moved randomly (a legal random move) - but because Kasparov believed it was correct he felt the analysis was way beyond him so he conceded.
We designed chess to be easily understood by humans. There's a cap on the conceptual complexity that fits within our capabilities. So it's not surprising that we can understand AI solutions that merely asymptote to that fixed cap.
But based on we actually know about the sort of AI systems humans actually know how to build and obtain useful results from, we'd end up with a computer which absorbed time series of paperclip sales figures, supply chain costs and marketing campaign data which recommends small paperclips for the Kazakh market, and there's no reason to believe a model trained to optimise paperclip production would know anything at all about how to generate fusion power. This article suggests we might even be able to analyse why the algorithm prioritised that obscure market after applying a lot more thought to the relevant data ourselves. (We can also reason that the AI model is more likely to be wrong about the small paperclips for the Kazakh market, because the permutations for the Kazakh market aren't as perfectly encapsulated in the underlying model and simulated by training on data as the permutations of chess)
> suppose we invent some superhuman AGI for $100B in 2100
not
> AI systems humans actually know how to build and obtain useful results from
You're talking about specialized expert system that is possible today.
BTW: AGI don't need to know anything about fusion power, just like Alpha Zero doesn't know anything about chess when it starts. Its supposed to learn the rules as it goes, and in case of AGI its chessboard is the universe.
Chess are a pretty well balanced and stable environment, so there's no way to cheat, but when you give less well-defined environment to AI it routinely cheats. See for example AlphaStar working around the APM limit by "saving" APM for crucial fights and completely demolishing human players in these fights microing at 1000s actions per minute. Or evolutionary systems learning to walk by abusing numerical instability bugs in the physics engine.
We know there are cheatcodes to reality (see industrial revolution), we know we haven't yet discovered all of them, and the assumption was AGI will be smarter than us. What you're doing is wishful thinking AGI won't find them cause it will be too busy with accounting.
The "cheat" argument suppports my argument, not yours. Rather than do something incredibly difficult and indirect like master advanced physics outside the scope of the simulations it can run to invent new power sources to legitimately produce very low cost paperclips, a system to maximise paperclip sales indicators is probably going to exploit gaps in the simulations it can run or data its already supplied with, like data entry errors in a region or logical holes in the supply chain pricing model, or the marketing simulation not including indications of languages spoken. So it might actually be worse than the management consultants at optimising paperclip revenues. On a related note, if you did want your AGI to invent fusion power but your fusion power emulation engine isn't up to scratch, it's likely to design reactors that are very performant inside the simulation but don't actually work, much like those evolutionary systems exploiting physics engine bugs when faced with the comparatively simple task of learning how to walk.
Perhaps it's illustrative to consider the chess variant "hand and brain" played by teams of two people, one ("brain") picking the piece to be moved, and the other ("hand") choosing where to move it; the relationship is complex, but in general having a strong "brain" limit your possible moves to just a single piece would improve the chances of a weaker "hand" above what they could do on their own.
Human players try to maximize score difference, Alpha Go tries to maximize the probability of winning. This means, between playing a risky move that makes more territory, and a safe move that makes less territory, Alpha Go prefers the safer move.
To experts, those move came across as "slow", inefficient or suboptimal. But they turn out to be the superior moves.
GAI/NN doesn't 'fear' variability and so it's OK to ride a razor thin margin all the way to the victory.
[edit: rephrase]
It's not spelled out directly, but it is a pretty low-hanging fruit.
After all, the rules dictate that the number of pieces on the board never increases. The rules also dictate the different kinds of moves allowed by each piece. Together it seems to me those rules invariably lead to pieces not having the same value, aka the concept of material.
I suppose if AGI gets achieved then some of this will seem predictable in hindsight. But we don't yet know if it's achievable.
On the other side, AlphaZero and all the previous Deep Mind's AIs played some very different go than human players did. Every single move was not unheard of but they were dismissed by human masters as not good. The most noticeable contribution was the total change of how to play corner invasions. By looking at game records we can draw a line between pre and post AlphaGo. Maybe go is a little bit more wide than chess and AIs have more space to find novel plays.
What I think would be cool is trying to distill or revise better guidelines from what AlphaZero does.
Take a much simpler game like paper, rock scissors, where certain strategies exist as well (I hear). This should be much easier to analyze. Can somebody apply alphaZero to rock, paper, scissors, please?
If the aliens live in our universe, evolved on a planet with limited resources, need food, breathe, etc., then they'll likely have a lot in common with Earth animals psychologically. If the AI is tasked with doing something that humans also do, then its tactics will be at least recognizable to an expert in that field. Moreso in well-understood systems such as chess.
Is it really plausible that an advanced chess AI's tactics would have no concept of threats or king safety? The same tactics apply to it as do to us.
Does having an evaluation system that takes into account king safety only as a consequence of having taken the entire game state into account, mean that it understands king safety? Can you extract these tactics from the AI? Can it explain the moves?
I would be interested in seeing work here that can provide some comprehensibility to the correlations, as that's how we can get to actual novel intuition. "The computer is good with this neural network due to pattern matching" is less actionable than "the computer is good and found this kind of pattern, and we think it relates to this".
EDIT: I think that if anything Chess is an interesting example of humans emulating computers. People sitting around thinking about variations, trying to snuff out what is good/bad. Granted, the fact that you're playing against other humans is pretty important at a higher level, but it's centuries of people sitting down and running through the search space, using our approximations for what are good and bad.
Swarms of agressive insects are also known to be quite defensive and alert the troops.
Take the human out of the loop and consider mating rituals. It amazes me to no end how simple bugs manage to procreate ...
This does of course not meat the common definition of communication, if it involves pheromones and such like. Nevertheless, when another philosopher says that Pain is impossible tondefine, I will show them pain right there.
Interdimensional aliens might have a more varied purpose I suppose, but then their actions wouldn't be intelligible to us at all. They might decide to convert half all life on earth into the opposite chiral form, because reasons. Or rather because there is some unknown mapping between chirality in this plane to the goals within another plain. The chirality of these biological forms within these dimensions being but the lower part of higher dimensional things which are subject to the motivating forces of higher dimensions. Although it's much more likely that any changes would not be coherent at all in our perspective of reality. The purposeful actions of other beings within other perspective frames of totality would more likely look like random thermal/quantum fluctuations within our perspective. There I solved quantum randomness for you.
I think chess strategy is a flawed way to be thinking about this because we designed the rules and the solution space to be comprehensible to us. It's not surprising that we can understand AI strategy. That simply follows from the solution space itself being small and comprehensible.
The term "psychologically" here is interesting because it skips the "biologically" part. What if they don't have neurons like us or AZ? Could one even talk about psychology?
Sure, non-sapient fungus-like extraterrestrial life-forms will definitely not think like us, even if they have neurons. But "aliens" in most stories implicitly refers to mobile intelligent beings capable of cooperation, communication, developing technology to a level more advanced than our own, and are animal-like and thus need to breathe, eat, drink, maintain homeostasis, etc. Often they'll even have the same body plan as us.
Also, things like the prisoner's dilemma, the tragedy of the commons, and that lying may be beneficial in some circumstances aren't mere artifacts of human psychology, but are straightforward results of game theory.
If a creature is that similar to a human physically, can it really be totally incomprehensible psychologically?
Seeing people analyze AlphaZero games on YouTube, it plays a more open and aggressive game of chess and goes for the win instead of the very defensive play of most of the top players today. Some people hope that if people study how AlphaZero plays chess (the best chess player in the world) to improve their game it might make chess more interesting as most games at the high level now end in draws.
They will learn the best opening move from your first move with white, the best reply from your response to it when they play it, white’s best second move from playing that against your best opening move, etc.
Once they know that, playing against them, all games will be the same and they’ll _look_ as good at chess as you.
Especially if the optimal move isn’t far better than the next best move(s) varying your moves can be used to show that isn’t true, and delay the time it takes your opponents to learn that optimal game.
Stockfish or AlphaZero don't care one bit about what happens in the next game they play.
Chessbase is notoriously sketchy and there is a lawsuit against them for blatantly stealing GPL code.
https://stockfishchess.org/blog/2021/our-lawsuit-against-che...
Because their article is easier to read, your link isn't even directly to the paper.
And unlike the fake news media who gate-keeps information to ensure their monopoly and to stop you Gell-Mann Amnesia them, Chessbase has included a direct link to the PDF and the abstract at the bottom of their article. That's awesome.
Bike-shedding GPL violations over quality and honest articles doesn't interest lots of people. They can read about the OT stuff in proper HN discussions on it.
I find it fascinating that we can create something that can "learn" in some sense in a way we don't fully understand. Or am I misunderstanding the meaning of this passage?
Non-NN chess engines are really all about doing a good job at balancing these heuristics (like it says above). There's fundamentally an upper bound in how good the overall approach can be, simply because the underlying heuristics are defined and selected by humans: there could be many more heuristics that haven't been considered. This version of "learning" is really just removing human forced constraints which were introduced to help learning, by starting from a position without bias.
The graph they have seems to suggest the opposite - nobody playing e4, just d4 early on (and AZ also gravitating most towards d4).