A general reinforcement learning algorithm that masters chess, shogi, and Go through self-play
Open accessThe DOI matches its Crossref record; 30 of the 30 references it cites were found in a registry or an index; 5 of the 16 sources its claims cite were checked for support.
Abstract
One program to rule them all: Computers can beat humans at increasingly complex games, including chess and Go. However, these programs are typically constructed for a particular game, exploiting its properties, such as the symmetries of the board on which it is played. Silver et al. developed a program called AlphaZero, which taught itself to play Go, chess, and shogi (a Japanese version of chess) (see the Editorial, and the Perspective by Campbell). AlphaZero managed to beat state-of-the-art programs specializing in these three games. The ability of AlphaZero to adapt to various game rules is a notable step toward achieving a general game-playing system. Science , this issue p. 1140 ; see also pp. 1087 and 1118
Overview
CoChat Summary
TL;DR
AlphaZero is a single reinforcement-learning and MCTS algorithm that, given only game rules, learned superhuman play in chess, shogi, and Go through self-play. It defeated world-champion programs Stockfish, Elmo, and AlphaGo Zero using the same algorithm and network architecture across all three games.
Key findings
- In chess, AlphaZero defeated Stockfish 8, winning 155 games and losing 6 out of 1,000; it also won matches from common human openings and 2016 TCEC opening positions.
- In shogi, AlphaZero defeated Elmo, winning 98.2% of games as black and 91.2% overall; in Go, it defeated AlphaGo Zero, winning 61% of games.
- AlphaZero first outperformed Stockfish after 4 hours (300,000 steps), Elmo after 2 hours (110,000 steps), and AlphaGo Lee after 30 hours (74,000 steps); training ran for 700,000 steps, taking about 9 hours in chess, 12 hours in shogi, and 13 days in Go.
- AlphaZero searched only about 60,000 positions per second in chess and shogi, versus 60 million for Stockfish and 25 million for Elmo, yet still defeated Stockfish with 1/10 thinking time and won 46% against Elmo with 1/100 thinking time.
Methods
- AlphaZero uses a deep neural network \(f_\theta(s)\) that outputs move probabilities \(p\) and a scalar value \(v\), trained by reinforcement learning from self-play starting from random parameters; MCTS uses these outputs to guide search, with 800 simulations per move during training.
- The same algorithm, network architecture, and hyperparameters were used for chess, shogi, and Go, with no domain knowledge except game rules; no opening book, endgame tablebases, or domain-specific heuristics were used.
- Training used 5,000 first-generation TPUs to generate self-play games and 16 second-generation TPUs to train networks; evaluation matches used 3 hours per game plus 15 seconds per move, with opponents on 44 CPU cores and AlphaZero on four first-generation TPUs and 44 CPU cores.
Limitations
- Stated: AlphaZero did not use opening books, endgame tablebases, or domain-specific heuristics; the authors note that some such techniques could further improve performance but were left for future research.
- Stated: hyperparameters were reused across games without game-specific tuning, except for exploration noise and the learning rate schedule; matches against an opening book were low in diversity, and forced randomization increased losses.
- Evident from the text: results cover only three games with separate trained instances, no cross-game transfer is demonstrated, and evaluation compares AlphaZero on TPU hardware against CPU-based opponents, so hardware and time-control differences qualify direct strength comparisons.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
3 Valid · 2 Not valid · 8 Not checked · 43 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“Apart from the representation of positions and actions described above, AlphaZero uses the same network architecture as AlphaGo Zero ( 9 ), briefly recapitulated here.”
What the sources say
- Supportedcites [9] source
The Neural Network Architecture Comparison (p. 29) identifies AlphaGo Zero’s architecture as a residual tower with policy and value heads. The citation supplies the AlphaGo Zero architecture that the citing authors say AlphaZero uses; the comparison to AlphaZero is their own statement. (LLM, conf=0.88)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“and in Go, AlphaZero first outperformed AlphaGo Lee ( 9 ) after 30 hours (74,000 steps).”
What the sources say
- Partly supportedcites [9] sourceExperimental — accuracy not yet measured
Section 2, p. 7 reports that AlphaGo Zero outperformed AlphaGo Lee, but says this happened after 36 hours, not 30. The passage does not give 74,000 steps, and it names the system AlphaGo Zero rather than AlphaZero. (LLM, conf=0.90)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The training algorithm achieved similar performance in all independent runs (see fig. S3), suggesting that the high performance of AlphaZero's training algorithm is repeatable.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“AlphaZero also defeated Stockfish when given 1 / 10 as much thinking time as its opponent (i.e. searching ~ 1 / 10 , 000 as many positions),”
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“and won 46% of games against Elmo when given 1 / 100 as much time (i.e. searching ~ 1 / 40 , 000 as many positions), see Fig. 2.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
23 more claims
Create a free account to see all 56 claims and to verify the 3 references still open.
- Probably not a substantial claim. 10
- It may state two separate claims as one. 23
- The quote only partly states this claim. 1
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Games have historically served as highly effective metrics for evaluating progress in deep reinforcement learning research”
From BALROG: Benchmarking Agentic LLM and VLM Reasoning On Games (2024) · cites this paper as Silver et al., 2018
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“In 2017, AlphaGo, using the same learning algorithms in TD-Gammon and a billion times more computing power, soundly beat Ke Jie, the world Go champion, in a game that is to chess what chess is to backgammon in complexity (Silver et al., 2018).”
From Large Language Models and the Reverse Turing Test (2023) · cites this paper as Silver et al., 2018
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Most of the attention was on who won, but more important was how it won: AlphaGo was creative and had discovered new moves and board positions that were outside human experience.”
From Large Language Models and the Reverse Turing Test (2023) · cites this paper as Silver et al., 2018
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“The primary method for applying more computation at inference, common in the AlphaZero family of agents2,72, is scaling the tree search and running multiple independent search attempts.”
From Olympiad-level formal mathematical reasoning with reinforcement learning (2026) · cites this paper as [2]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“Modern AI can write computer programs (Li et al., 2022) or play chess at super-human level (Silver et al., 2018), but the ability of AI to perceive and act in the world remains far below human level.”
From Scaling Instructable Agents Across Many Simulated Worlds (2024) · cites this paper as Silver et al., 2018
1 more claim about this paper
Create a free account to see every claim other papers make about this one.
Citations
Cites
Mastering the game of Go with deep neural networks and tree search
Silver, D., Huang, A., Maddison, C. J., Guez, A., Sifre, L., van den Driessche, G., Schrittwieser, J., Antonoglou, I., Panneershelvam, V., Lanctot, M., et al. (2016) · Nature · doi:10.1038/nature16961 · Cited by 15,612
Self-citationAdvances in neural information processing systems 7
(1996) · Computers & Mathematics with Applications · doi:10.1016/s0898-1221(96)90463-0 · Cited by 14,281
Identity mappings in deep residual networks
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-46493-0_38 · Cited by 10,741
Identity Mappings in Deep Residual Networks
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · arXiv · doi:10.48550/arxiv.1603.05027 · Cited by 10,665
Mastering the game of Go without human knowledge
Silver, D., Schrittwieser, J., Simonyan, K., Antonoglou, I., Huang, A., Guez, A., Hubert, T., Baker, L., Lai, M., Bolton, A., et al. (2017) · Nature · doi:10.1038/nature24270 · Cited by 9,084
Self-citationIn-Datacenter Performance Analysis of a Tensor Processing Unit
Jouppi, N. P., Young, C., Patil, N., Patterson, D., Agrawal, G., Bajwa, R., Bates, S., Bhatia, S., Boden, N., Borchers, A., et al. (2017) · Proceedings of the 44th Annual International Symposium on Computer Architecture · doi:10.1145/3079856.3080246 · Cited by 4,537
Encyclopedia of Artificial Intelligence
(2009) · IGI Global eBooks · doi:10.4018/978-1-59904-849-9 · Cited by 1,396
Campbell, M., Hoane, A. J., & Hsu, F.-H. (2002) · Artificial Intelligence · doi:10.1016/s0004-3702(01)00129-1 · Cited by 1,226
Lim, T.-S., Loh, W.-Y., & Shih, Y.-S. (2000) · Machine Learning · doi:10.1023/a:1007608224229 · Cited by 1,133
An analysis of alpha-beta pruning
Knuth, D. E., & Moore, R. W. (1975) · Artificial Intelligence · doi:10.1016/0004-3702(75)90019-3 · Cited by 1,004
Research works printed but not linked: 7
- C. J. Maddison A. Huang I. Sutskever D. Silver paper presented at the International Conference on Learning Representations 2015 San Diego CA 7 to 9 May 2015.
- D. N. L. Levy M. Newborn How Computers Play Chess (Ishi Press 2009).
- W. Steinitz The Modern Chess Instructor (Edition Olms 1990).
- O. Arenz “Monte Carlo chess ” master’s thesis Technische Universität Darmstadt (2012).
- 10.1023/a:1007634325138 · DOI
- T. Kaneko K. Hoki in Advances in Computer Games – 13th International Conference ACG 2011 Revised Selected Papers Tilburg Netherlands 20 to 22 November 2011 (Springer 2012) pp. 158–169.
- R. Ramanujan A. Sabharwal B. Selman in Proceedings of the 26th Conference on Uncertainty in Artificial Intelligence (UAI 2010) Catalina Island CA 8 to 11 July (AUAI Press 2010).
Cited by
Items 2,001–2,100 of 3,633
Pacaud, A., Bechler, A., & Coupechoux, M. (2024) · IEEE Transactions on Games · doi:10.1109/tg.2024.3424246 · Cited by 3
“Restrictions apply. model of AlphaZero [28], exploring approaches, which combine MCTSwithneuralnetworksmightalsobeapromisingdirection.”
More Human-Like Gameplay by Blending Policies From Supervised and Reinforcement Learning
Ogawa, T., Hsueh, C.-H., & Ikeda, K. (2024) · IEEE Transactions on Games · doi:10.1109/tg.2024.3424668 · Cited by 3
“The game has been actively studied for a long time for various purposes, such as creating strong game AI [4], [7], solving smaller boards [14], and teaching human players [15], [16].”
InfluentialUnlocking AI Literacy: The SMaILE-App Gaming Experience
Franco, S., Ballatore, M. G., Bernardini, S., Como, G., Damonte, L., Neacsu, A., & Vanelli, M. (2024) · IEEE Gaming, Entertainment, and Media Conference (GEM) · doi:10.1109/gem61861.2024.10585394 · Cited by 3
“In the Station game, the player learns to fine-tune the Al-phaZero algorithm [16] for creating virtual opponents for the Planting minigame.”
MethodHeuristic Knowledge Transfer for General Game Playing
Jung, J. D. A., & Hoey, J. (2024) · IEEE Conference on Games (CoG) · doi:10.1109/cog60054.2024.10645542 · Cited by 3
“Recent deep learning projects, like AlphaZero [1], have bucked this trend by achieving world-class performance across a variety of games.”
BackgroundHammersborg, P., & Strümke, I. (2024) · Scientific Reports · doi:10.1038/s41598-024-70701-2 · Cited by 13
“The project also produced a family of models 2 using the same architecture as AlphaZero presented in Silver et al. (2018).”
BackgroundMethodInfluentialMarisetty, H. V., Fatima, N., Gupta, M., & Saxena, P. (2024) · Internet of Things · doi:10.1016/j.iot.2024.101375 · Cited by 5
Chan, E. J., & Tuckerman, M. E. (2024) · Acta Crystallographica Section B Structural Science Crystal Engineering and Materials · doi:10.1107/s205252062400132x · Cited by 3
Rao, N., Xu, H., Qi, Z., Wang, D., Zhang, Y., Peng, X., & Jiang, L. (2024) · Computer Communications · doi:10.1016/j.comcom.2024.107993 · Cited by 3
Learning How to Play Bomberman with Deep Reinforcement and Imitation Learning
Goulart, Í., Paes, A., & Clua, E. (2019) · Lecture notes in computer science · doi:10.1007/978-3-030-34644-7_10 · Cited by 4
Receptors P1 and P2 as Targets for Drug Therapy in Humans
Faria, R. (2019) · IntechOpen eBooks · doi:10.5772/intechopen.78115 · Cited by 3
EquiNox: Equivalent NoC Injection Routers for Silicon Interposer-Based Throughput Processors
Li, Y., & Chen, L. (2020) · IEEE International Symposium on High-Performance Computer Architecture (HPCA) · doi:10.1109/hpca47549.2020.00043 · Cited by 3
“MCTS is a classic search algorithm in machine learning [29] and has been recently employed to enhance the AI algorithms in the game of Go, Shogi and StarCraft II [30, 31, 32] to search their huge solution spaces.”
MethodKhalid, M., Aslam, N., & Wang, L. (2020) · International Conference on Communications and Networking (ComNet) · doi:10.1109/comnet47917.2020.9306103 · Cited by 6
“The reinforcement learning model provide promising solution in scenarios where environment is unknown [17].”
BackgroundHas machine learning rendered simple rules obsolete?
Fernández‐Villaverde, J. (2021) · European Journal of Law and Economics · doi:10.1007/s10657-021-09690-w · Cited by 4
“A simple example of this approach is the training of AlphaZero, a computer program that learned to play Go, chess, and shogi through self-play (Silver et al. 2018).”
Emerging CMOS Compatible Magnetic Memories and Logic
Ender, J., Fiorentini, S., de Orio, R. L., et al. (2021) · IEEE Journal of the Electron Devices Society · doi:10.1109/jeds.2021.3066679 · Cited by 3
Eye of the Beholder: Improved Relation Generalization for Text-Based Reinforcement Learning Agents
Murugesan, K., Chaudhury, S., & Talamadupula, K. (2022) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v36i10.21358 · Cited by 4
“Successful applications of RL have included domains such as Chess [27], Go [28], and Atari games [21].”
MethodMakarov, I., Gorbachev, R., Novikov, A., et al. (2020) · doi:10.1109/ent50437.2020.9431284 · Cited by 3
Adaptive random quantum eigensolver
Barraza, N., Pan, C.-Y., Lamata, L., Solano, E., & Albarrán-Arriagada, F. (2022) · Physical Review A · doi:10.1103/physreva.105.052406 · Cited by 3
Machine learning assisted quantum adiabatic algorithm design
Lin, J. H., Ye, M., Zhu, J., & Li, X. (2021) · Acta Physica Sinica · doi:10.7498/aps.70.20210831 · Cited by 3
Knowledge Transfer between Similar Atari Games Using Deep Q-Networks to Improve Performance
Das, A., Shroff, V., Jain, A., & Sharma, G. (2021) · International Conference on Computing, Communication and Networking Technologies (ICCCNT) · doi:10.1109/icccnt51525.2021.9580091 · Cited by 4
“In [7], DeepMind released the AlphaGo Zero’s generalized approach under the “AlphaZero” algorithm, which achieved a superhuman level of play in the games of chess, shogi, and Go in just 24 hours.”
Chess Endgame Compression via Logic Minimization
Gomboc, D., & Shelton, C. R. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-11488-5_14 · Cited by 3
“Chess engines that reach such pre-tabulated positions from within their heuristic searches can propagate back an exact score, which can be used either to improve on-line game play or to improve the accuracy and efficiency of off-line reinforcement learning [29] via endgame table rescoring [20].”
BackgroundHeredia, L. G., & Cazenave, T. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-11488-5_3 · Cited by 3
“This idea was called Predictor+UCB (PUCT) and it is the way in which the AlphaZero and Expert Iteration frameworks combine neural networks and MCTS.”
MethodInfluentialLogic-based AI for Interpretable Board Game Winner Prediction with Tsetlin Machine
Giri, C., Granmo, O., van Hoof, H., & Blakely, C. D. (2022) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn55064.2022.9892796 · Cited by 5
“Board game winner prediction is a critical part of state-of-the-art AIs for board game playing, such as AlphaZero [22].”
Continuous Control of Autonomous Vehicles using Plan-assisted Deep Reinforcement Learning
Dwivedi, T., Betz, T., Sauerbeck, F., Manivannan, P. V., & Lienkamp, M. (2022) · International Conference on Control, Automation and Systems (ICCAS) · doi:10.23919/iccas55662.2022.10003698 · Cited by 9
“While modular software architectures, such as [10], [11], are currently the benchmark in autonomous racing competitions, end-to-end DRL offers a promising alternative as these have demonstrated competence beyond human expertise in many real-world applications [6], [7], [8].”
BackgroundPineda, L. A. (2024) · Cognitive Systems Research · doi:10.1016/j.cogsys.2023.101204 · Cited by 9
On Solving the Problem of 7-Piece Chess Endgames
Zakharov, V. B., Мальковский, М. Г., & Mostyaev, A. I. (2019) · Programming and Computer Software · doi:10.1134/s036176881903006x · Cited by 3
Reverse Engineering of Option Pricing: An AI Application
Herzog, B., & Osamah, S. (2019) · International Journal of Financial Studies · doi:10.3390/ijfs7040068 · Cited by 4
“The most remarkable scientific breakthrough in recent years is the creation of super-human intelligence via deep neural networks in the field of artificial intelligence (AI) (Silver et al. 2016, 2017, 2018; Tian et al. 2019).”
BackgroundAbukhait, J. J., Aljaafreh, A., & Al-Oudat, N. (2019) · International Conference on Social Networks Analysis, Management and Security (SNAMS) · doi:10.1109/snams.2019.8931879 · Cited by 3
“Studies on mastering Chess, Shogi, and Go games, without human knowledge and based on reinforcement learning, have been introduced in [22-24].”
BackgroundChess, Artificial Intelligence, and Epistemic Opacity
Grünke, P. (2020) · Információs Társadalom · doi:10.22503/inftars.xix.2019.4.1 · Cited by 6
Applying Monte-Carlo Tree Search in HTN Planning
Wichlacz, J., Höller, D., Torralba, Á., & Hoffmann, J. (2021) · Proceedings of the International Symposium on Combinatorial Search · doi:10.1609/socs.v11i1.18538 · Cited by 3
“This is an important feature of, e. g., the AlphaGo/Zero system series (Silver et al. 2016; 2018).”
BackgroundParallel Monte Carlo Integration Algorithm Based on GPU
Zong, H., Hua, R., Zhao, J., et al. (2019) · IEEE International Conference on Intelligent Systems and Knowledge Engineering · doi:10.1109/iske47853.2019.9170329 · Cited by 3
Are Strong Policies Also Good Playout Policies? Playout Policy Optimization for RTS Games
Yang, Z., & Ontañón, S. (2020) · Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment · doi:10.1609/aiide.v16i1.7423 · Cited by 3
“In recent Go research, playout policies is some times abandoned and replaced by refined evaluation functions (Silver et al. 2017; 2018).”
BackgroundLearning representation of stock traders and immediate price impacts
Xie, W.-J., Li, M.-Y., & Zhou, W. (2021) · Emerging Markets Review · doi:10.1016/j.ememar.2020.100791 · Cited by 3
Mininni, C. J., & Zanutto, B. S. (2021) · Scientific Reports · doi:10.1038/s41598-021-82964-0 · Cited by 3
Uncertainty maximization in partially observable domains: A cognitive perspective
Ramičić, M., & Bonarini, A. (2023) · Neural Networks · doi:10.1016/j.neunet.2023.02.044 · Cited by 5
“More complex architectures of neural network function approximators coupled with the increase of computational power allowed temporal-difference RL algorithms to achieve super-human level control in problems that were designed for the complexity and scale of human cognition, such as Atari games [36], complex board gam…”
BackgroundA Computer-Based Method for the Investigation of Human Behavior in the Iterative Chicken Game
Kim, S.-P., Kim, M., Lee, J., Cho, Y. S., & Kwon, O.-S. (2021) · Frontiers in Psychology · doi:10.3389/fpsyg.2021.576404 · Cited by 3
“…a number of studies have developed artificial agents that can learn to play the chicken game in a data-driven manner (van den Dries and Wiering, 2012; Silver et al., 2018), an artificial agent developed in this study is different from previous ones since it can play the chicken game with varying…”
BackgroundLearning to Stop: Dynamic Simulation Monte-Carlo Tree Search
Lan, L.-C., Wu, T.-R., Wu, I., & Hsieh, C. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i1.16100 · Cited by 10
“After fine-tuning the thresholds on NoGo, DS-MCTS can speed up more than 2.5x and still maintain a 46.5% win rate under a much larger Nmax.”
BackgroundInfluentialEffectiveness of Transfer Learning in Autonomous Driving using Model Car
Chiba, S., & Sasaoka, H. (2021) · International Conference on Machine Learning and Computing (ICMLC) · doi:10.1145/3457682.3457773 · Cited by 3
“The research of Alpha Zero [4] has introduced their algorithm based solely on re-inforcement learning, without personal data, guidance, or domain knowledge beyond game rules.”
BackgroundReinforcement Learning: An Industrial Perspective
Surana, A. (2021) · Studies in systems, decision and control · doi:10.1007/978-3-030-60990-0_21 · Cited by 3
Walsh, M., Menthe, L., Geist, E., Hastings, E., Kerrigan, J., Léveillé, J., Margolis, J., Martin, N., & Donnelly, B. P. (2021) · RAND Corporation eBooks · doi:10.7249/rr-a263-2 · Cited by 3
Distance-Based Mapping for General Game Playing
Jung, J. D. A., & Hoey, J. (2021) · IEEE Conference on Games (CoG) · doi:10.1109/cog52621.2021.9619036 · Cited by 6
“Among these bots, AlphaZero stands out as a generalist, but even AlphaZero can require days or weeks to train on a specific problem [1].”
BackgroundLearning Curricula for Humans: An Empirical Study with Puzzles from The Witness
Lelis, L. H. S., Nova, J. G. G. V., Chen, E., Sturtevant, N., Epp, C. D., & Bowling, M. (2022) · International Joint Conference on Artificial Intelligence · doi:10.24963/ijcai.2022/538 · Cited by 4
“For example, AlphaZero learns how to play chess and Go by playing the game with itself [Silver et al. , 2018].”
BackgroundWorld-class interpretable poker
Bertsimas, D., & Paskov, A. (2022) · Machine Learning · doi:10.1007/s10994-022-06179-8 · Cited by 3
“Over the past two decades, reinforcement learning has yielded phenomenal successes in the domain of perfect-information games: it has produced world-class bots capable of outperforming even the strongest human competitors in games such as Chess and GO [Silver et al. (2018)].”
BackgroundAzad, A. S., Kim, E., Wu, Q., Lee, K., Stoica, I. G., Abbeel, P., Sangiovanni-Vincentelli, A. L., & Seshia, S. A. (2022) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v36i6.20549 · Cited by 8
Towards Tackling QSAT Problems with Deep Learning and Monte Carlo Tree Search
Xu, R., & Lieberherr, K. J. (2022) · Lecture notes in networks and systems · doi:10.1007/978-3-031-10464-0_4 · Cited by 4
Scalable evolutionary hierarchical reinforcement learning
Abramowitz, S., & Nitschke, G. (2022) · Genetic and Evolutionary Computation Conference Companion (GECCO Companion) · doi:10.1145/3520304.3528937 · Cited by 4
“Reinforcement learning (RL) [18] has been used to create artificially intelligent agents for tasks ranging from robot locomotion [8] to board games such as chess and Go [17].”
MethodDeep learning approaches for de novo drug design: an overview
Wang, M., Li, D., Hou, T., & Kang, Y. (2022) · Scientia Sinica Chimica · doi:10.1360/ssc-2022-0135 · Cited by 4
A behavior fusion method based on inverse reinforcement learning
Shi, H., Li, J., Chen, S., & Hwang, K. (2022) · Information Sciences · doi:10.1016/j.ins.2022.07.100 · Cited by 5
Multi-scene Scheduling of Power System With Renewable Energy Based on DDPG
Han, X., Shi, W., Yuan, X., & Yi, H. (2023) · Asia Conference on Power and Electrical Engineering (ACPEE) · doi:10.1109/acpee56931.2023.10135852 · Cited by 4
“The simulation environment can be applied in various fields[4], including generation planning, load forecasting, real-time operation, power market analysis, and reliability assessment.”
BackgroundAdamczyk, J., Arriojas, A., Tiomkin, S., & Kulkarni, R. V. (2023) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v37i6.25817 · Cited by 14
Deep Q-learning to globally optimize a k-D parameter search for medical imaging
Zhang, H., Liang, S., Matkovic, L. A., Momin, S., Wang, K., Yang, X., & Insana, M. F. (2023) · Quantitative Imaging in Medicine and Surgery · doi:10.21037/qims-22-1147 · Cited by 5
“With the successful application of DL in AlphaGo (14,15), the power of DL has been validated in a variety of applications (16-22).”
Multi-task Learning by Pareto Optimality
Dyankov, D., Riccio, D., Fatta, G. D., & Nicosia, G. (2019) · Lecture notes in computer science · doi:10.1007/978-3-030-37599-7_50 · Cited by 3
A Modified I2A Agent for Learning in a Stochastic Environment
Pal, C.-V., & Leon, F. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-63007-2_30 · Cited by 4
Structure and Randomness in Planning and Reinforcement Learning
Czechowski, K., Januszewski, P., Kozakowski, P., Kuciński, Ł., & Miłoś, P. (2021) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn52387.2021.9533317 · Cited by 3
“Bandit Shooting, presented in Algorithm 4, is a Multi-armed Bandits variant of Random Shooting and uses PUCT [7] rule to improve exploration and thus achieve more reliable evaluations of actions.”
BackgroundInfluentialLiu, Z., Zhou, M., Cao, W., Qu, Q., Yeung, H. W. F., & Chung, Y. Y. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1903.01747 · Cited by 3
Paradox in Deep Neural Networks: Similar yet Different while Different yet SimilarPreprint
Akbarinia, A., & Gegenfurtner, K. R. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1903.04772 · Cited by 4
“…at a great rate, surpassing our very own in various fields, from the famous AlphaZero , a generic model conquering the world of intellectual board games, such as, go, chess and shogi, (Silver et al., 2018), to a 3D-CNN that swiftly reports abnormalities in medical images (Titano et al., 2018).”
BackgroundSchomaker, L. (2019) · arXiv (Cornell University) · doi:10.13140/rg.2.2.11940.53120 · Cited by 6
Pretrain Soft Q-Learning with Imperfect DemonstrationsPreprint
Zhang, X., Li, Y., Ma, H., & Luo, X. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1905.03501 · Cited by 3
Learning Latent State Spaces for Planning through Reward PredictionPreprint
Havens, A., Ouyang, Y., Nagarajan, P., & Fujita, Y. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1912.04201 · Cited by 7
“Deep reinforcement learning (DRL) has demonstrated some of the most impressive results on several complex, high-dimensional sequential decision-making tasks such as video games [1] and board games [2].”
BackgroundFormulaZero: Distributionally Robust Online Adaptation via Offline Population SynthesisPreprint
Sinha, A., O’Kelly, M., Zheng, H., Mangharam, R., Duchi, J. C., & Tedrake, R. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2003.03900 · Cited by 20
“Reinforcement learning (RL) has achieved unprecedented success on classic two-player games (e.g. Silver et al., 2018), leading to new approaches in partially-observable games with continuous action spaces (Arulkumaran et al., 2019; Berner et al., 2019).”
BackgroundResultApproximate exploitability: Learning a best response in large gamesPreprint
Timbers, F., Bard, N., Lockhart, E., Lanctot, M., Schmid, M., Burch, N., Schrittwieser, J., Hubert, T., & Bowling, M. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2004.09677 · Cited by 29
Self-citationDeep reinforcement learning for optical systems: A case study of mode-locked lasersPreprint
Sun, C., Kaiser, E., Brunton, S. L., et al. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.05579 · Cited by 3
“This is accomplished in chess and Go by self-play [14], which is exactly what the mode-locked laser is allowed to do to learn an optimal policy.”
BackgroundContinuous Control for Searching and Planning with a Learned ModelPreprint
Yang, X., Duvaud, W., & Wei, P. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.07430 · Cited by 7
“|A | ≥ , an action a k will be selected according to the stored statistics of node s k 1 , by maximizing over the PUCB score [21, 28] Here we note the difference with the MuZero algorithm is that the prior value here ¯ π ( a | s, θ ) is normalized from the policy probability density function value…”
MethodAgazzi, A., & Lu, J. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1905.10917 · Cited by 7
“…potential have appeared over the past few years, with such algorithms mastering games and tasks of increasing complexity, from playing Atari to learning to walk and beating world grandmasters at the game of Go (Mnih et al., 2013; Mnih et al.; Silver et al., 2016, 2017, 2018; Haarnoja et al., 2018).”
BackgroundContinual Learning: Tackling Catastrophic Forgetting in Deep Neural Networks with Replay ProcessesPreprint
Lesort, T. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2007.00487 · Cited by 37
“Classical benchmarks Reinforcement learning has known recent big success in games such as chess Silver et al. (2018), go Silver et al. (2016) or Dota 2 video games OpenAI et al. (2019).”
BackgroundSingle-partition adaptive Q-learningPreprint
Araújo, J. P., Figueiredo, M. A. T., & Botto, M. A. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2007.06741 · Cited by 3
“This approach has led to many successful applications of RL in areas such as control of autonomous vehicles (Kuutti et al., 2020), game playing (Silver et al., 2018; OpenAI et al., 2019), healthcare (Shortreed et al., 2011), and education (Mu et al., 2018b), to name a few.”
BackgroundComposite Q-learning: Multi-scale Q-function Decomposition and Separable OptimizationPreprint
Kalweit, G., Hügle, M., & Boedecker, J. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1909.13518 · Cited by 9
“…Q-learning (Watkins and Dayan, 1992) has achieved major successes in a broad range of areas by employing deep neural networks (Mnih et al., 2015; Silver et al., 2018; Lillicrap et al., 2016), including environments of higher complexity (Riedmiller et al., 2018) and even in first real world…”
BackgroundLearning Robust Algorithms for Online Allocation Problems Using Adversarial TrainingPreprint
Žužić, G., Wang, D., Mehta, A., & Sivakumar, D. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2010.08418 · Cited by 4
“A long-standing ambition of machine learning is to learn from first principles, that is, an ML system that learns without any human-provided data beyond an appropriate definition of the task at hand [27].”
BackgroundInfluentialControl with adaptive Q-learningPreprint
Araújo, J. P., Figueiredo, M. A. T., & Botto, M. A. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2011.02141 · Cited by 3
“, 2020), game playing (Silver et al., 2018; OpenAI et al., 2019), healthcare (Shortreed et al.”
Continuous Transition: Improving Sample Efficiency for Continuous Control Problems via MixUpPreprint
Lin, J., Huang, Z., Wang, K., et al. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2011.14487 · Cited by 3
“Deep reinforcement learning (RL), with high-capacity deep neural networks, has been applied to solve various complex decision-making tasks, including chess games [1], [2], video games [3], [4], etc.”
Model-based controlled learning of MDP policies with an application to lost-sales inventory control.Preprint
van Jaarsveld, W. L. (2020) · arXiv (Cornell University) · Cited by 3
Solving the Job-Shop Scheduling Problem with Reinforcement Learning
Schlebusch, D. (2020) · Institutional Repository (IHS Vienna) · doi:10.26041/fhnw-3437 · Cited by 3
“In general if the model of the environment, in other words, the “rules”, are known a model-based approach can be used to train the algorithms with known state transitions, for example, the chess computer AlphaZero [11].”
MethodExplaining Natural Language Processing Classifiers with Occlusion and Language ModelingPreprint
Harbecke, D. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2101.11889 · Cited by 6
“Deep neural networks have achieved state-of-the-art performance on a wide variety of tasks, such as • image recognition (Ciregan et al., 2012; Krizhevsky et al., 2012), • text classification (Kim, 2014), • perfect information games (Silver et al., 2016, 2018), 2 Chapter 1.”
BackgroundMulti-Agent Coordination in Adversarial Environments through Signal Mediated StrategiesPreprint
Cacciamani, F., Celli, A., Ciccone, M., et al. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2102.05026 · Cited by 3
“Thanks to the recent successes of deep RL in single-agent environments [42, 52, 53], MARL is recently experiencing a newwave of interest and some old ideas have been adapted to leverage the power of function approximators [44, 45].”
Deep Latent Competition: Learning to Race Using Visual Control Policies in Latent SpacePreprint
Schwarting, W., Seyde, T., Gilitschenski, I., Liebenwein, L., Sander, R., Karaman, S., & Rus, D. L. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2102.09812 · Cited by 40
“Multi-agent RL Recently, MARL agents have surpassed human-level performance in many multiagent environments including complex board and card games such as Go [19], chess, shogi [20], and Poker [21, 22].”
BackgroundAgazzi, A., & Lu, J. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2010.11858 · Cited by 15
“…have appeared over the last few years, with reinforcement learning algorithms mastering games and tasks of increasing complexity, from learning to walk to the games of Go and Starcraft (Mnih et al., 2013; 2015; Silver et al., 2016; 2017; 2018; Haarnoja et al., 2018a; Vinyals et al., 2019).”
BackgroundThe Autodidactic UniversePreprint
Alexander, S., Cunningham, W. J., Lanier, J., Smolin, L., Stanojevic, S., Toomey, M. W., & Wecker, D. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2104.03902 · Cited by 39
“• The look-ahead protocol is somewhat related to algorithms in reinforcement learning such as AlphaZero [58].”
MethodA Study of the Mathematics of Deep LearningPreprint
Mukherjee, A. (2021) · SSRN Electronic Journal · doi:10.2139/ssrn.3833584 · Cited by 3
The best laid plans: Computational principles of ACCPreprint
Holroyd, C. B., & Verguts, T. (2020) · PsyArXiv (OSF Preprints) · doi:10.31234/osf.io/3df8y · Cited by 6
Evading Malware Classifiers via Monte Carlo Mutant Feature DiscoveryPreprint
Boutsikas, J., Eren, M. E., Varga, C. K., et al. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.07860 · Cited by 3
“MCTS has gained popularity based on its application for the game Go as described by (Coulom, 2009), as well as its latest implementation in Google’s AlphaZero (Silver et al., 2018).”
MethodScenic4RL: Programmatic Modeling and Generation of Reinforcement Learning EnvironmentsPreprint
Azad, A. S., Kim, E., Wu, Q., Lee, K., Stoica, I. G., Abbeel, P., & Seshia, S. A. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.10365 · Cited by 5
Backprop-Free Reinforcement Learning with Active Neural Generative CodingPreprint
Ororbia, A. G., & Mali, A. (2021) · doi:10.31219/osf.io/3nfcm · Cited by 3
“The emergence of this framework is timely – in reinforcement learning (RL) research, despite the recent successes afforded by artificial neural networks (ANNs) [8, 9], most models require exorbitant quantities of data to train well, struggling to learn tasks as efficiently as humans and animals [10].”
BackgroundModEL: A Modularized End-to-end Reinforcement Learning Framework for Autonomous Driving.Preprint
Wang, G., Niu, H., Zhu, D., Hu, J., Zhan, X., & Zhou, G. (2021) · arXiv (Cornell University) · Cited by 3
AI in Games: Techniques, Challenges and OpportunitiesPreprint
Yin, Q., Yang, J. J., Ni, W., Liang, B., & Huang, K. (2021) · arXiv (Cornell University) · Cited by 4
Human learning follows the dynamics of gradient descentPreprint
Barry, D. N., & Love, B. C. (2021) · PsyArXiv (OSF Preprints) · doi:10.31234/osf.io/v6bt9 · Cited by 3
Fabrication of a Tactile Sensor for Artificial Skin Based on Electrical Impedance TomographyPreprint
Liu, K., Qin, Z., Wu, Y., et al. (2021) · SSRN Electronic Journal · doi:10.2139/ssrn.3990028 · Cited by 3
On the impact of MDP design for Reinforcement Learning agents in Resource ManagementPreprint
Cunha, R. L. F., & Chaimowicz, L. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2109.03202 · Cited by 3
“Popular function approximators include linear combinations of features [6] and neural networks [13, 15].”
BackgroundMachine Learning in German Official Statistics 1
Dumpert, F. (2023) · doi:10.1002/9781119672333.ch23 · Cited by 3
DeepMind AI creates algorithms that sort data faster than those built by people
Hutson, M. (2023) · Nature · doi:10.1038/d41586-023-01883-4 · Cited by 2
Deep Watermarking for Deep Intellectual Property Protection: A Comprehensive SurveyPreprint
Sun, Y., Liu, L., Yu, N., Liu, Y., Tian, Q. C., & Guo, D. (2024) · SSRN Electronic Journal · doi:10.2139/ssrn.4697020 · Cited by 7
Predicting Optimal Moves in Chess Board Using Artificial Intelligence
Thangaramya, K., Logeswari, G., Sudhakaran, G., Aadharsh, R., Bhuvaneshwar, S., Dheepakraaj, R., & Sunny, P. (2024) · Wiley eBooks · doi:10.1002/9781394214068.ch4 · Cited by 5
Cockrell, C., Vodovotz, Y., Zamora, R., & An, G. C. (2024) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/2024.06.04.595972 · Cited by 5
Khodadadi, Z., Trpevski, D., Lindroos, R., & Kotaleski, J. H. (2025) · eLife · doi:10.7554/elife.97274.2 · Cited by 3
MAESTRO: Open-Ended Environment Design for Multi-Agent Reinforcement Learning
Samvelyan, M., Khan, A., Dennis, M., Jiang, M., Parker-Holder, J., Foerster, J., Raileanu, R., & Rocktäschel, T. (2023) · arXiv · doi:10.48550/arxiv.2303.03376 · Cited by 8
“…co-players in competitive games (Leibo et al., 2019; Garnelo et al., 2021; Baker et al., 2019; Bansal et al., 2018; Feng et al., 2021), where it is crucial to play against increasingly stronger opponents (Silver et al., 2018) and avoid being exploited by other agents (Vinyals et al., 2019).”
BackgroundMethodSWE-Search: Enhancing Software Agents with Monte Carlo Tree Search and Iterative Refinement
Antoniades, A., Örwall, A., Zhang, K., Xie, Y., Goyal, A., & Wang, W. (2024) · arXiv · doi:10.48550/arxiv.2410.20285 · Cited by 3
“Instead, we adopt a more straightforward heuristic-based selection function, similar to the approach used in AlphaZero Silver et al. (2016a; 2018).”
Safe Search for Stackelberg Equilibria in Extensive-Form Games
Ling, C. K., & Brown, N. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i6.16697 · Cited by 4
“…been a key component for AI in single-agent settings (Lin 1965; Hart, Nilsson, and Raphael 1968), perfect-information games (Tesauro 1995; Campbell, Hoane Jr, and Hsu 2002; Silver et al. 2016, 2018), and zero-sum imperfect-information games (Moravčı́k et al. 2017; Brown and Sandholm 2017b, 2019).”
BackgroundOffline Policy Comparison Under Limited Historical Agent-Environment Interactions
Dereventsov, A., Daws, J., & Webster, C. (2024) · Lecture notes in computational science and engineering · doi:10.1007/978-3-031-55060-7_5 · Cited by 5
Reinforcement Learning for Weighted p-median Problem
Matis, D., & Tarábek, P. (2023) · International Conference on Information and Digital Technologies (IDT) · doi:10.1109/idt59031.2023.10194404 · Cited by 4
“Model-based methods (e.g. AlphaZero [10]) use a model of the environment to plan and make decisions, while model-free methods do not rely on an explicit model of the environment and instead learn directly from experience.”
MethodFrom Principles to Practices: The Risks of Being Unethical
Floridi, L. (2023) · Oxford University Press eBooks · doi:10.1093/oso/9780198883098.003.0005 · Cited by 2
Estimation and Inference in Distributional Reinforcement Learning
Zhang, L.-Y., Peng, Y., Liang, J.-D., Yang, W., & Zhang, Z.-H. (2023) · Annals of Statistics · doi:10.1214/25-aos2527 · Cited by 10
Collaborative robots can augment human cognition in regret-sensitive tasks
Schlafly, M., Prabhakar, A., Popovic, K., Schlafly, G., Kim, C., & Murphey, T. D. (2024) · PNAS Nexus · doi:10.1093/pnasnexus/pgae016 · Cited by 3
“Similar techniques have been used to beat international human champions in Chess and Go (32) as well as model animal (33, 34) and human behavior (35 – 40).”
Method
Metadata
- Authors
- Silver, D., Hubert, T., Schrittwieser, J., Antonoglou, I., Lai, M., Guez, A., Lanctot, M., Sifre, L., Kumaran, D., Graepel, T., Lillicrap, T., Simonyan, K., & Hassabis, D.
- Affiliations
- Google DeepMind (United Kingdom) (GB)
- Corresponding author
- Silver, David
- Hubert, Thomas
- Schrittwieser, Julian
- Hassabis, Demis
- ORCID
- Venue
- Science
- Published
- 2018-12-07
- Volume
- 362
- Issue
- 6419
- Pages
- 1140-1144
- ISSN
- 0036-8075 (ISSN-L)
- 1095-9203
- OpenAlex
- W2902907165
- Cited by
- 3,562
- References
- 30
- Citations by year
- 2026: 257 · 2025: 427 · 2024: 512 · 2023: 514 · 2022: 569 · 2021: 546 · 2020: 462 · 2019: 256 · 2018: 13
- FWCI
- 179.89
- Type
- journal article
- Language
- en
- Open access
- bronze
- doi.org
- Locations
- Science (publishedVersion)
- PubMed (publishedVersion)
- UCL Discovery (University College London) (submittedVersion)
- Funders
- DeepMind
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Artificial Intelligence in Games
- Physical Sciences › Computer Science › Artificial Intelligence › Reinforcement Learning in Robotics
- Physical Sciences › Computer Science › Artificial Intelligence › Artificial Intelligence Applications
- Keywords
- AlphaZero, chess, shogi, Go, evaluation function learning, game tree search
- MeSH terms
- Algorithms, Artificial Intelligence, Humans, Reinforcement, Psychology, Software, Video Games
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.