Reid McIlroy-Young, Siddhartha Sen, Jon Kleinberg, Ashton Anderson
2020 · arXiv
As artificial intelligence becomes increasingly intelligent---in some cases, achieving superhuman performance---there is growing potential for humans to learn from and collaborate with algorithms. However, the ways in which AI systems approach problems are often different from the ways people do, and thus may be uninterpretable and hard to learn from. A crucial step in bridging this gap between human and artificial intelligence is modeling the granular actions that constitute human behavior, rather than simply matching aggregate human performance. We pursue this goal in a model system with a long history in artificial intelligence: chess. The aggregate performance of a chess player unfolds as they make decisions over the course of a game. The hundreds of millions of games played online by players at every skill level form a rich source of data in which these decisions, and their exact context, are recorded in minute detail. Applying existing chess engines to this data, including an open-source implementation of AlphaZero, we find that they do not predict human moves well. We develop and introduce Maia, a customized version of Alpha-Zero trained on human chess games, that predicts human moves at a much higher accuracy than existing engines, and can achieve maximum accuracy when predicting decisions made by players at a specific skill level in a tuneable way. For a dual task of predicting whether a human will make a large mistake on the next move, we develop a deep neural network that significantly outperforms competitive baselines. Taken together, our results suggest that there is substantial promise in designing artificial intelligence systems with human collaboration in mind by first accurately modeling granular human decision-making.
Metadata: arXiv. Abstracts retain their original authors’ rights.
Original HN comments, with labels from simple text rules. A label describes the whole comment and may miss context or irony. Read the discussion to make up your own mind. About the labels
4 selected excerpts · 4 total mentions
https://arxiv.org/pdf/2006.01855 If you check figure 4, SF at depth 15 matches about 47% of the time which is lower than their NN. So quantitatively, SF isn’t amazing at move matching, and lower than the best NN at its relevant rating range (1900 -> 54%). Qualitatively, SF is pretty unsatisfying as a model of a pro human because a lot of its moves are counterintuitive.
Wow I played a single game but the prediction was very good, it predicted 1636 on lichess while my best elo was ~1600 (though I haven't played for about a year) But what's even more impressive is that the game felt great to play, I didn't get this feeling that the AI would pick on every of my smallest mistakes like you have with Stockfish. I know there has been several research work [0,1] recently to improve the human-ness of chess AIs, but it's even more impressive that you are able to do it "online", in the course of a single game, bravo. I'd be interested in a blog post with details on the underlying algorithm [0] https://arxiv.org/abs/2006.01855 [1] https://openreview.net/forum?id=fJY2iCssvIs
It turns out that taking a superhuman player and making it play like a weak human is surprisingly tricky. It's not so hard to make a weak player - you just take suboptimal moves instead of the best moves. But often these suboptimal moves are bizarre. A weak human chess player will lose their pieces as they fall prey to forks and skewers and so on - tricks that are hard to see coming for new player. But they will rarely actively throw a piece away by moving it into danger. Even a novice human is pretty decent at looking one move ahead. But to a chess engine, or an MuZero agent, a move that loses the queen immediately and a move that leads to a sequence that loses the queen…
in five turns are basically equal. And so an artificially-weak MuZero agent, or an artificially-weak Stockfish agent will tend to make 'mistakes' that not even a weak human would make. This makes them a little difficult to learn from. There does exist research on how to make a human-like weak player: https://arxiv.org/abs/2006.01855 The basic idea is to look at weak human games and try to predict when a mistake will be made. But I don't know if there's any approach that can do that without access to a corpus of human errors.
Hi everyone, We're happy to announce a research project that has been in the works for almost two years! Please meet Maia, a human-like neural network chess engine. Maia is a Leela-style framework that learns from human play instead of self-play, with the goal of making human-like moves instead of optimal moves. Maia predicts the exact moves humans play in real online games over 50% of the time. We intend Maia to power data-driven learning tools and teaching aids, as well as be a fun sparring partner to play against. We trained 9 different versions on 12M Lichess games each, one for each rating level between 1100 and 1900. Each version captures human style at its targeted…
level, meaning that Maia 1500's play is most similar to 1500-rated players, etc. You can play different versions of Maia yourself on Lichess: Maia 1100 (https://lichess.org/@/maia1), Maia 1500 (https://lichess.org/@/maia5) and Maia 190 (https://lichess.org/@/maia9) This is an ongoing research project using chess as a model system for understanding how to design machine learning models for better human-AI interaction. For more information about the project, check out http://maiachess.com. We published a research paper (https://arxiv.org/abs/2006.01855) and blog post (http://csslab.cs.toronto.edu/blog/2020/08/24/maia_chess_kdd/) on Maia, and the Microsoft Research blog (https://www.microsoft.com/en-us/research/blog/the-human-side...) covered the project here. All of our code is available on our [GitHub repo](https://github.com/CSSLab/maia-chess). We are super grateful to Lichess.org for making this project possible with their open data policy. In current work, we are developing Maia models that are personalized to individual players. It turns out that personalized Maia can predict a particular player's moves up to 75% of the time. You can read a preprint about this work https://arxiv.org/abs/2008.10086. We'd love to hear your feedback! You can contact us at maiachess@cs.toronto.edu or on our new Twitter account @maiachess.
More from the same topics.
Sébastien Bubeck, Varun Chandrasekaran, Ronen Eldan +11
Artificial intelligence (AI) researchers have been developing and refining large language models (LLMs) that exhibit remarkable capabilities across a variety of domains and tasks, challenging our understanding of learning and cognition.
Tom B. Brown, Benjamin Mann, Nick Ryder +28
Recent work has demonstrated substantial gains on many NLP tasks and benchmarks by pre-training on a large corpus of text followed by fine-tuning on a specific task.
Chiyuan Zhang, Samy Bengio, Moritz Hardt +2
Despite their massive size, successful deep artificial neural networks can exhibit a remarkably small difference between training and test performance.
A. Pluchino. A. E. Biondo, A. Rapisarda
The largely dominant meritocratic paradigm of highly competitive Western cultures is rooted on the belief that success is due mainly, if not exclusively, to personal qualities such as talent, intelligence, skills, efforts or risk taking.