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 1,301–1,400 of 3,626
DSMC Evaluation Stages: Fostering Robust and Safe Behavior in Deep Reinforcement Learning
Gros, T. P., Höller, D., Hoffmann, J., Klauck, M., Meerkamp, H., & Wolf, V. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-85172-9_11 · Cited by 10
TraceVis: Towards Visualization for Deep Statistical Model Checking
Gros, T. P., Gross, D. C., Gumhold, S., Hoffmann, J., Klauck, M., & Steinmetz, M. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-83723-5_3 · Cited by 14
Effects of Sampling and Prediction Horizon in Reinforcement Learning
Osinenko, P., & Dobriborsci, D. (2021) · IEEE Access · doi:10.1109/access.2021.3112498 · Cited by 10
Robust ASV Navigation Through Ground to Water Cross-Domain Deep Reinforcement Learning
Lambert, R., Li, J., Wu, L., & Mahmoudian, N. (2021) · Frontiers in Robotics and AI · doi:10.3389/frobt.2021.739023 · Cited by 11
“, 2013), playing board games such as Chess and Go (Silver et al., 2018), and robot body simulations (Todorov et al.”
BackgroundReinforcement Learning Policies With Local LQR Guarantees For Nonlinear Discrete-Time Systems
Zoboli, S., Andrieu, V., Astolfi, D., Casadei, G., Dibangoye, J., & Nadri, M. (2021) · IEEE Conference on Decision and Control (CDC) · doi:10.1109/cdc45484.2021.9683721 · Cited by 10
Deep Learning and Computational Chemistry
James, T., & Hristozov, D. (2021) · Methods in molecular biology · doi:10.1007/978-1-0716-1787-8_5 · Cited by 10
Noy, I., & Uher, T. (2022) · Economics of Disasters and Climate Change · doi:10.1007/s41885-022-00105-x · Cited by 14
“Notable progress has also been made towards achieving a more general kind of intelligence, such that may eventually be able to employ multiple cognitive abilities, abstraction, and reasoning, as is typical for humans (Strogatz 2018; Badia et al. 2020; Silver et al. 2018).”
BackgroundHuber, M., & Raidl, G. R. (2022) · Lecture notes in computer science · doi:10.1007/978-3-030-95470-3_22 · Cited by 10
“AlphaGo and its successor AlphaZero gained broader recognition as agents excelling in the games of Go, chess, and shogi [25].”
BackgroundResultDecentralized Safe Multi-agent Stochastic Optimal Control using Deep FBSDEs and ADMM
Pereira, M. A., Saravanos, A. D., So, O., et al. (2022) · doi:10.15607/rss.2022.xviii.055 · Cited by 9
“Deep reinforcement learning has enjoyed significant fame over the past few years [30, 40, 41, 28], although being restricted to simulation settings where safety is not a primary concern.”
BackgroundCockrell, C., Larie, D., & An, G. C. (2022) · Frontiers in Immunology · doi:10.3389/fimmu.2022.995395 · Cited by 13
“Speci fi cally, the latter projects described in Refs (20, 21) utilized the method, Deep Reinforcement Learning (DRL), employed by ML/Arti fi cial Intelligence (AI) systems to successfully play and win a series of games against human experts (22 – 24).”
MethodFerber, P., Geißer, F., Trevizan, F., Helmert, M., & Hoffmann, J. (2022) · Proceedings of the International Conference on Automated Planning and Scheduling · doi:10.1609/icaps.v32i1.19845 · Cited by 11
“Given the success of neural networks (NN) as game-state evaluators (Silver et al. 2016, 2017, 2018; Agostinelli et al. 2019), and the prominence of heuristic search in planning, (e.g., Hoffmann and Nebel 2001; Helmert and Domshlak 2009; Richter and Westphal 2010; Helmert et al. 2014; Domshlak,…”
BackgroundPlanning spatial networks with Monte Carlo tree search
Darvariu, V.-A., Hailes, S. M. V., & Musolesi, M. (2023) · Proceedings of the Royal Society A Mathematical Physical and Engineering Sciences · doi:10.1098/rspa.2022.0383 · Cited by 14
“MCTS has been applied to great success in connectionist games such as Morpion Solitaire (Rosin, 2011), Hex (Nash, 1952; Anthony et al., 2017), and Go, which was previously thought computationally intractable (Silver et al., 2016; 2018).”
MethodFoundations of human spatial problem solving
Zarr, N., & Brown, J. W. (2023) · Scientific Reports · doi:10.1038/s41598-023-28834-3 · Cited by 12
Altamimi, S., Altamimi, B., Côté, D., & Shirmohammadi, S. (2023) · IEEE Access · doi:10.1109/access.2023.3248652 · Cited by 9
“The superintelligence of AI has been demonstrated for games already [5], and we speculate that a growing number of such superintelligent agents will be established in a variety of domains.”
BackgroundMulti-Agent Reinforcement Learning
(2001) · The CRC Press international series on computational intelligence/International series on computational intelligence · doi:10.1201/9781420038835.ch7 · Cited by 9
Muddling-Through and Deep Learning for Managing Large-Scale Uncertain Risks
Cox, T. (2019) · Journal of Benefit-Cost Analysis · doi:10.1017/bca.2019.17 · Cited by 11
“For large problems, recent MCTS algorithms (Munos, 2014; Silver et al., 2016, 2018) sample possible future paths and rewards to avoid enumerating all possibilities.”
Goldstone, R. L. (2019) · Topics in Cognitive Science · doi:10.1111/tops.12463 · Cited by 15
Self-Play for Training General Fighting Game AI
Takano, Y., Inoue, H., Thawonmas, R., et al. (2019) · Nicograph International (NicoInt) · doi:10.1109/nicoint.2019.00034 · Cited by 9
“AlphaZero recently achieved superhuman performance in Go, shogi (Japanese chess), and chess by reinforcement learning from self-play [2].”
BackgroundLearning self-play agents for combinatorial optimization problems
Xu, R., & Lieberherr, K. (2020) · Knowledge engineering review (Print) · doi:10.1017/s026988892000020x · Cited by 7
“…combinatorial optimization problems into games via a process called Zermelo Gamification (ZG), so that a neural Monte Carlo Tree Search (neural MCTS) (Anthony et al., 2017; Silver et al., 2017b, 2018) self-play agent can be leveraged to play the transformed games and solve the original problems.”
MethodWarm-Start AlphaZero Self-play Search Enhancements
Wang, H., Preuß, M., & Plaat, A. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-58115-2_37 · Cited by 15
Self-Learning Molecular Design for High Lithium-Ion Conductive Ionic Liquids using Maze Game
Kikkawa, N., Kajita, S., & Takechi, K. (2020) · Journal of Chemical Information and Modeling · doi:10.1021/acs.jcim.0c00692 · Cited by 9
A novel method for GPR imaging based on neural networks and dictionary learning
Hajipour, S., Namin, F. A., & Shirazi, R. S. (2021) · Waves in Random and Complex Media · doi:10.1080/17455030.2021.1880667 · Cited by 8
Enforcement Heuristics for Argumentation with Deep Reinforcement Learning
Craandijk, D., & Bex, F. (2022) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v36i5.20497 · Cited by 16
“A growing body of literature in this field combines deep learning with reinforcement learning to automatically learn heuristics for symbolic reasoning problems (such as game playing (Silver et al. 2018) and combinatorial optimization (Bengio, Lodi, and Prouvost 2021)).”
BackgroundA Deep Reinforcement Learning Approach for Global RoutingPreprint
Liao, H., Zhang, W., Dong, X., et al. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1906.08809 · Cited by 9
“DRL combines reinforcement learning and deep learning, and has been successfully utilized in a variety of sequential decision making problems [16,17,18,19,20,21,16].”
MethodComputational model discovery with reinforcement learningPreprint
Bassenne, M., & Lozano-Durán, A. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2001.00008 · Cited by 18
“This accomplishment tops the list of sophisticated human tasks at which AI is now performing at a human or superhuman level [19], and enlarges the previous body of achievements in game playing [27, 26] and in other tasks, e.”
BackgroundFast Adaptation via Policy-Dynamics Value FunctionsPreprint
Răileanu, R., Goldstein, M., Szlam, A. D., & Fergus, R. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2007.02879 · Cited by 14
“Deep reinforcement learning (RL) has achieved impressive results on a wide range of complex tasks (Mnih et al., 2015; Silver et al., 2016; 2017; 2018; Jaderberg et al., 2019; Berner et al., 2019; Vinyals et al., 2019).”
Learning Personalized Models of Human Behavior in ChessPreprint
McIlroy-Young, R., Wang, R., Sen, S., Kleinberg, J. M., & Anderson, A. (2020) · arXiv (Cornell University) · Cited by 11
Human-Level Performance in No-Press Diplomacy via Equilibrium SearchPreprint
Gray, J., Lerer, A. K., Bakhtin, A., & Brown, N. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2010.02923 · Cited by 55
“…have achieved expert-level or even superhuman performance in benchmark games such as backgammon (Tesauro, 1994), chess (Campbell et al., 2002), Go (Silver et al., 2016; 2017; 2018), poker (Morav ˇ cík et al., 2017; Brown & Sandholm, 2017; 2019b), and real-time strategy games (Berner et al., 2019;…”
BackgroundMethodScaling up Mean Field Games with Online Mirror DescentPreprint
Pérolat, J., Perrin, S., Élie, R., Laurière, M., Piliouras, G., Geist, M., Tuyls, K., & Pietquin, O. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.00623 · Cited by 47
"Weak AI" is Likely to Never Become "Strong AI", So What is its Greatest Value for us?Preprint
Liu, B. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.15294 · Cited by 44
“The striking point lies in that the computing agent has reached and even surpassed humans in many tasks, e.g., image classification [2], speech recognition [3, 4], games [5–7], protein structure prediction [8].”
BackgroundScaling Instructable Agents Across Many Simulated Worlds
SIMA Team, Raad, M. A., Ahuja, A., Barros, C., Besse, F., Bolt, A., Bolton, A., Brownfield, B., Buttimore, G., Cant, M., et al. (2024) · arXiv · doi:10.48550/arxiv.2404.10179 · Cited by 9
“For example, backgammon provided the initial proving ground for early deep reinforcement learning methods (Tesauro et al., 1995), and later works have achieved superhuman performance even in complex board games like Go (Silver et al., 2016, 2018).”
Unsupervised Active Visual Search With Monte Carlo Planning Under Uncertain Detections
Taioli, F., Giuliari, F., Wang, Y., Berra, R., Castellini, A., Bue, A. D., Farinelli, A., Cristani, M., & Setti, F. (2024) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2024.3451994 · Cited by 8
“Popular MCTS-based approaches have been recently used also for developing agents with superhuman performance in the game of Go [43], [44].”
MethodTesting for Fault Diversity in Reinforcement Learning
Mazouni, Q., Spieker, H., Gotlieb, A., et al. (2024) · doi:10.1145/3644032.3644458 · Cited by 8
“In the last decade, Reinforcement Learning (RL) combined with neural networks (NNs) was shown to be able to effectively solve complex sequential decision problems in various fields, such as planning, game playing or system control [10, 21, 28].”
BackgroundSample-Efficient Multi-Agent DQNs for Scalable Multi-Domain 5G+ Inter-Slice Orchestration
Doanis, P., & Spyropoulos, T. (2024) · IEEE Transactions on Machine Learning in Communications and Networking · doi:10.1109/tmlcn.2024.3420268 · Cited by 11
“Go, Chess [32]) do need to tackle large state and action spaces, they tackle ‘‘planning’’ problems, where the environment and rules are fully known, unlike our problem, where environment variables (the VNF demands) are unknown and time-varying.”
BackgroundEbrahimi, R., Chai, Y., Li, W., Pacheco, J. L., & Chen, H. (2024) · MIS Quarterly · doi:10.25300/misq/2024/17339 · Cited by 11
Feng, X., Xu, M., & Li, Y. (2024) · Advanced Engineering Informatics · doi:10.1016/j.aei.2024.102915 · Cited by 8
Annealing-inspired training of an optical neural network with ternary weights
Skalli, A., Goldmann, M., Haghighi, N., Reitzenstein, S., Lott, J. A., & Brunner, D. (2025) · Communications Physics · doi:10.1038/s42005-025-01972-y · Cited by 25
“Over the past decade, ANNs have revolutionized computing 1 , disrupting fi elds ranging from natural language 2,3 and image processing 4 to self-driving vehicles 5 and game playing 6 .”
Sideris, I., Yan, Y., Duncan, S., Afrasiabi, M., & Bambach, M. (2025) · Additive manufacturing · doi:10.1016/j.addma.2025.104831 · Cited by 8
Improving reservoir water quality by optimizing weir operations with reinforcement learning and SWAT
Pyo, J., Baek, S., Abbas, A., Kim, H. G., Lee, J., Kim, S., Chun, J. A., & Cho, K. H. (2025) · Ecohydrology & Hydrobiology · doi:10.1016/j.ecohyd.2025.100679 · Cited by 12
Introduction of a new dataset and method for location predicting based on deep learning in wargame
Liu, M., Zhang, H., Hao, W., Qi, X., Cheng, K., Jin, D., & Feng, X. (2021) · Journal of Intelligent & Fuzzy Systems · doi:10.3233/jifs-201726 · Cited by 8
HetSeq: Distributed GPU Training on Heterogeneous Infrastructure
Ding, Y., Botzer, N., & Weninger, T. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i17.17813 · Cited by 18
Hong, J.-W. (2021) · Journal of Control and Decision · doi:10.1080/23307706.2021.1978882 · Cited by 11
Rigoli, L. M., Patil, G., Nalepka, P., Kallen, R. W., Hosking, S. G., Best, C. J., & Richardson, M. J. (2022) · Journal of Cognitive Engineering and Decision Making · doi:10.1177/15553434221092930 · Cited by 13
“Although DRL agents can achieve ‘ super-human ’ levels of performance in multiagent tasks (Bansal et al., 2017; Jaderberg et al., 2019; Silver et al., 2018; Vinyals et al., 2019), the behavior of these AAs is often easily distinguishable from human behavior and can also lead to suboptimal…”
BackgroundGawęda, A. E., Lederer, E. D., & Brier, M. E. (2022) · Frontiers in Medicine · doi:10.3389/fmed.2022.807994 · Cited by 8
“Compared to Supervised and Unsupervised Learning which strictly depend on the training data, RL has the unique ability to discover novel information by interactively generating new data (28).”
BackgroundWang, J., Qi, H., Guang, M., Zhang, C., Yan, C., & Jiang, C. (2021) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2021.3084902 · Cited by 9
“For example, AlphaGo has exceeded the level of human world champion [1], the correct recognition rate of the images on ImageNet is higher than that of people [2], [3], and fraud detection methods have been designed based on the recent behavioral profile of cardholders [4], [5].”
MethodInfusing common-sensical prior knowledge into topological representations of learning robots
Ogawa, K., & Hartono, P. (2022) · Artificial Life and Robotics · doi:10.1007/s10015-022-00776-5 · Cited by 9
“In recent years, with the rapid growth of AI as represented by reinforcement learning (RL) [11, 18, 19, 26] and deep learning [9], AI has been applied in a wide range of fields, such as image recognition, speech recognition, text processing, object detection, games, and industrial and home robots.”
BackgroundScore vs. Winrate in Score-Based Games: which Reward for Reinforcement Learning?
Pasqualini, L., Parton, M., Morandin, F., et al. (2022) · IEEE International Conference on Machine Learning and Applications (ICMLA) · doi:10.1109/icmla55696.2022.00099 · Cited by 8
“We validate this hypothesis in a AlphaZero-like software with a human expert.”
MethodInfluentialLearning new attack vectors from misuse cases with deep reinforcement learning
Veith, E. M., Wellßow, A., & Uslar, M. (2023) · Frontiers in Energy Research · doi:10.3389/fenrg.2023.1138446 · Cited by 10
How to control hydrodynamic force on fluidic pinball via deep reinforcement learning
Feng, H., Wang, Y., Xiang, H., et al. (2023) · Physics of Fluids · doi:10.1063/5.0142949 · Cited by 8
Hyperparameter Selection in Reinforcement Learning Using the “Design of Experiments” Method
McKenzie, M. C., & McDonnell, M. D. (2023) · Procedia Computer Science · doi:10.1016/j.procs.2023.08.140 · Cited by 8
Predictive reinforcement learning: map-less navigation method for mobile robot
Dobriborsci, D., Zashchitin, R., Kakanov, M. A., Aumer, W., & Osinenko, P. (2023) · Journal of Intelligent Manufacturing · doi:10.1007/s10845-023-02197-y · Cited by 10
“Over the past decade, RL methods have made remarkable progress in several tasks and competitions such as Go, chess, shogi, StarCraft II, and Rubik’s Cube (Akkaya et al., 2019; Silver et al., 2016, 2018; Vinyals et al., 2019).”
Li, S., Han, C., Yu, P., Edwards, C., Li, M., Wang, X., Fung, Y. R., Yu, C., Tetreault, J., Hovy, E. H., et al. (2023) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2023.findings-emnlp.799 · Cited by 27
FM3Q: Factorized Multi-Agent MiniMax Q-Learning for Two-Team Zero-Sum Markov Game
Hu, G., Zhu, Y., Li, H., & Zhao, D. (2024) · IEEE Transactions on Emerging Topics in Computational Intelligence · doi:10.1109/tetci.2024.3383454 · Cited by 11
“zero-sum games (2p0sMGs) [5], [6], [7], [8].”
Learning Topological Operations on Meshes with Application to Block Decomposition of Polygons
Narayanan, A., Pan, Y., & Persson, P. (2024) · Computer-Aided Design · doi:10.1016/j.cad.2024.103744 · Cited by 7
Tackling Cooperative Incompatibility for Zero-Shot Human-AI Coordination
Li, Y., Zhang, S., Sun, J., Zhang, W., Du, Y., Wen, Y., Wang, X., & Pan, W. (2024) · Journal of Artificial Intelligence Research · doi:10.1613/jair.1.15884 · Cited by 31
Unitary Synthesis of Clifford+T Circuits with Reinforcement Learning
Rietsch, S., Dubey, A. Y., Ufrecht, C., Periyasamy, M., Plinge, A., Mutschler, C., & Scherer, D. D. (2024) · IEEE International Conference on Quantum Computing and Engineering (QCE) · doi:10.1109/qce60285.2024.00102 · Cited by 30
“In recent years, model-based RL utilizing Monte Carlo tree search (MCTS) [14], [15], such as AlphaGo [16] and Alp-haZero [17], have shown great promise in solving complex problems with large search spaces [18]–[20].”
Multiple Policy Value Monte Carlo Tree Search
Lan, L.-C., Li, W., Wei, T.-H., & Wu, I. (2019) · International Joint Conference on Artificial Intelligence · doi:10.24963/ijcai.2019/653 · Cited by 8
A Game Model for Gomoku Based on Deep Learning and Monte Carlo Tree Search
Li, X., He, S., Wu, L., Chen, D., & Zhao, Y. (2019) · Lecture notes in electrical engineering · doi:10.1007/978-981-32-9050-1_10 · Cited by 8
Distance-based classifier on the Quantum Inspire
Wezeman, R. S., Neumann, N. M. P., & Phillipson, F. (2019) · Digitale Welt · doi:10.1007/s42354-019-0240-5 · Cited by 8
Computer Vision and Hybrid Reality for Construction Safety Risks: A Pilot Study
Li, R. Y. M., & Leung, T. H. (2020) · Advances in intelligent systems and computing · doi:10.1007/978-981-32-9343-4_2 · Cited by 9
Deep Reinforcement Learning Applied to Airport Surface Movement Planning
Tien, S.-L. A., Tang, H., Kirk, D. B., Vargo, E. P., & Liu, S. (2019) · IEEE/AIAA Digital Avionics Systems Conference (DASC) · doi:10.1109/dasc43569.2019.9081720 · Cited by 8
“Among all the candidate AI approaches, Deep Reinforcement Learning (DRL) is a breakthrough that has been successfully applied to train an AI agent that can perform at or above the expert human level in complex problems; examples include Google DeepMind applications in chess, shogi, Go and StarCraft [4-6].”
MethodStrength Adjustment and Assessment for MCTS-Based Programs [Research Frontier]
Liu, A.-J., Wu, T.-R., Wu, I., Guei, H., & Wei, T.-H. (2020) · IEEE Computational Intelligence Magazine · doi:10.1109/mci.2020.2998315 · Cited by 8
“In terms of future work, the AlphaZero algorithm has also been successfully applied to other games such as chess and shogi, also achieving super-human level strength [6].”
MethodA Survey of Nash Equilibrium Strategy Solving Based on CFR
Li, H., Wang, X., Jia, F., Li, Y.-F., & Chen, Q. (2020) · Archives of Computational Methods in Engineering · doi:10.1007/s11831-020-09475-5 · Cited by 8
“The game of Go, as a typical PIG, its positions of all chess pieces are visible to both of the players when playing games, which is successfully solved already by the computer programs based on deep reinforcement learning like AlphaGo, AlphaZero, et al. [1–5].”
BackgroundImproving Model and Search for Computer Go
Cazenave, T. (2021) · IEEE Conference on Games (CoG) · doi:10.1109/cog52621.2021.9619078 · Cited by 8
“Deep reinforcement learning combining self play and Monte Carlo Tree Search (MCTS) [1], [2] with the PUCT algorithm [3] is the current state of the art for computer Go [4], [5] and for other board games [6], [7].”
MethodPHASE: PHysically-grounded Abstract Social Events for Machine Social Perception
Netanyahu, A., Shu, T., Katz, B., Barbu, A., & Tenenbaum, J. B. (2021) · AAAI Publications (The Association for the Advancement of Artificial Intelligence (AAAI)) · doi:10.1609/aaai.v35i1.16167 · Cited by 24
Full Gradient DQN Reinforcement Learning: A Provably Convergent Scheme
Avrachenkov, K. E., Borkar, V. S., Dolhare, H., & Patil, K. (2021) · Emergence, complexity and computation · doi:10.1007/978-3-030-76928-4_10 · Cited by 15
Programmatic Strategies for Real-Time Strategy Games
Mariño, J. R. H., Moraes, R. O., Oliveira, T. C., Toledo, C. F. M., & Lelis, L. H. S. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i1.16114 · Cited by 8
“Search and learning-based algorithms represent the current state-of-the-art approaches for playing zero-sum games, e.g., AlphaZero (Silver et al. 2018) and AlphaStar (Vinyals et al. 2019).”
BackgroundA Generalized Stacked Reinforcement Learning Method for Sampled Systems
Osinenko, P., Dobriborsci, D., Yaremenko, G., & Malaniya, G. (2023) · IEEE Transactions on Automatic Control · doi:10.1109/tac.2023.3250032 · Cited by 8
“Reinforcement learning (RL) showed remarkable performance in puzzles and video games [1], [2], [3], while also having demonstrated certain success in industrial tasks [4], [5].”
BackgroundBoosting Intelligent Data Analysis in Smart Sensors by Integrating Knowledge and Machine Learning
Łuczak, P., Kucharski, P., Jaworski, T., Perenc, I., Ślot, K., & Kucharski, J. (2021) · Italian National Conference on Sensors · doi:10.3390/s21186168 · Cited by 8
“Not only are intelligent systems now capable of achieving state-of-the-art performance on multiple complex games, as shown by AlphaZero [2], but they are also capable of solving extremely complex real-world problems such as protein folding.”
BackgroundWatching a Language Model Learning Chess
Stöckl, A. (2021) · International conference Recent advances in natural language processing · doi:10.26615/978-954-452-072-4_153 · Cited by 22
The Era of Immersive Health Technology
Bremner, R., Gibbs, A., & Mitchell, A. R. J. (2020) · EMJ Innovation · doi:10.33590/emjinnov/19-00132 · Cited by 8
Reinforcement-Learning-Based Task Planning for Self-Reconfiguration of Cellular Satellites
Zhang, Y., Wang, W., Zhang, P., & Huang, P. (2021) · IEEE Aerospace and Electronic Systems Magazine · doi:10.1109/maes.2021.3089252 · Cited by 8
Accurate and effective framework for identifying track defects
Yang, H., Bi, Q., Yao, Z., & Wang, Y. (2022) · Measurement · doi:10.1016/j.measurement.2021.110625 · Cited by 8
Barros, P. V. A., & Sciutti, A. (2022) · Neural Networks · doi:10.1016/j.neunet.2022.03.013 · Cited by 9
Stock Price Formation: Precepts from a Multi-Agent Reinforcement Learning Model
Lussange, J., Vrizzi, S., Bourgeois‐Gironde, S., Palminteri, S., & Gutkin, B. S. (2022) · Computational Economics · doi:10.1007/s10614-022-10249-3 · Cited by 16
“In machine learning, multi-agent reinforcement learning (Silver et al. 2018) (RL) has yielded remarkable results with far-reaching applications in domains closely related to quantitative finance, such as decision theory and game theory (Lanctot et al. 2017).”
Yue, Z., Wang, Y., & Lyu, P. (2022) · Physica A Statistical Mechanics and its Applications · doi:10.1016/j.physa.2022.127538 · Cited by 9
On the Road to Perfection? Evaluating Leela Chess Zero Against Endgame Tablebases
Haque, R., Wei, T.-H., & Müller, M. (2021) · Advances in Computer Games · doi:10.1007/978-3-031-11488-5_13 · Cited by 8
“Moreover, instead of waiting for an iteration to complete, and evaluating it against the best previous network, AlphaZero maintains only one network that is updated continually [47].”
BackgroundMethodInfluentialAnalysis of Hyper-Parameters for AlphaZero-Like Deep Reinforcement Learning
Wang, H., Emmerich, M., Preuß, M., & Plaat, A. (2022) · International Journal of Information Technology & Decision Making · doi:10.1142/s0219622022500547 · Cited by 14
“The AlphaGo series of papers 1 , 2 , 3 have sparked much interest of researchers and the general public alike into deep reinforcement learning.”
BackgroundMethodInfluentialSpeedup Training Artificial Intelligence for Mahjong via Reward Variance Reduction
Li, J., Wu, S., Fu, H., Fu, Q., Zhao, E., & Xing, J. (2022) · IEEE Conference on Games (CoG) · doi:10.1109/cog51982.2022.9893584 · Cited by 8
“AlphaZero [4] introduces a general reinforcement learning framework to solve perfect information games without any domain knowledge by self-play, which outperforms human masters in the Go game.”
BackgroundMeta-Optics-Empowered Switchable Integrated Mode Converter Based on the Adjoint Method
Ha, Y., Guo, Y., Pu, M., Xu, M., Li, X., Ma, X., Zou, F., & Luo, X. (2022) · Nanomaterials · doi:10.3390/nano12193395 · Cited by 8
“One of the obstacles on the road to subwavelength PIC application is the fact that static subwavelength PICs cannot solve dynamic information transmission [10,11,15,16].”
BackgroundEnabling deep reinforcement learning autonomous driving by 3D-LiDAR point clouds
Chen, Y., Tse, R. T., Bosello, M., Aguiari, D., Tang, S.-K., & Pau, G. (2022) · International Conference on Digital Image Processing · doi:10.1117/12.2644369 · Cited by 8
“Since its success in several complex tasks, 14-16 RL has been also widely studied in AD tasks.”
BackgroundScalable Model-based Policy Optimization for Decentralized Networked Systems
Du, Y., Ma, C., Liu, Y., Lin, R., Dong, H., Wang, Y.-X., & Yang, Y. (2022) · IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros47612.2022.9982253 · Cited by 19
“Reinforcement learning has achieved remarkable success in many decision making tasks [21], [6], [22], [23].”
BackgroundDebugging a Policy: Automatic Action-Policy Testing in AI Planning
Steinmetz, M., Fišer, D., Enişer, H. F., et al. (2022) · Proceedings of the International Conference on Automated Planning and Scheduling · doi:10.1609/icaps.v32i1.19820 · Cited by 8
“The use of neural networks (NN) to learn action policies π is highly successful in game playing (Mnih et al. 2013; Silver et al. 2018), and is gaining traction in AI Planning (Is-sakkimuthu, Fern, and Tadepalli 2018; Groshev et al. 2018; Garg, Bajpai, and Mausam 2019; Toyer et al. 2020; Karia and…”
BackgroundLiu, P., Zhou, J., & Lv, J. (2022) · Knowledge-Based Systems · doi:10.1016/j.knosys.2022.110207 · Cited by 8
The graph structure of two-player games
Biggar, O., & Shames, I. (2023) · Scientific Reports · doi:10.1038/s41598-023-28627-8 · Cited by 12
Adversarial Sybil attacks against Deep RL based drone trajectory planning
Bhandarkar, A. B., Jayaweera, S. K., & Lane, S. A. (2022) · IEEE Military Communications Conference (MILCOM) · doi:10.1109/milcom55135.2022.10017870 · Cited by 8
“…increasingly popular in solving problems that require making complex decisions [1] including, for example, playing board games such as chess and Go [2], controlling traffic signals in multi-intersection vehicular networks [3], beamforming in intelligent reflecting surface (IRS)-aided wireless…”
BackgroundState of the Art of Adaptive Dynamic Programming and Reinforcement Learning
Liu, D., Ha, M., & Xue, S. (2022) · CAAI Artificial Intelligence Research · doi:10.26599/air.2022.9150007 · Cited by 9
The Morphospace of Consciousness: Three Kinds of Complexity for Minds and Machines
Arsiwalla, X. D., Solé, R. V., Moulin-Frier, C., Herreros, I., Sánchez-Fibla, M., & Verschure, P. F. M. J. (2023) · NeuroSci · doi:10.3390/neurosci4020009 · Cited by 19
“Then we have AlphaZero, which is a single system that taught itself from scratch how to master the games of chess, shogi and Go [66].”
BackgroundAn interpretable machine-learned model for international oil trade network
Xie, W.-J., Wei, N., & Zhou, W. (2023) · Resources Policy · doi:10.1016/j.resourpol.2023.103513 · Cited by 13
Official International Mahjong: A New Playground for AI Research
Lu, Y., Li, W., & Li, W. (2023) · Algorithms · doi:10.3390/a16050235 · Cited by 12
Policy ensemble gradient for continuous control problems in deep reinforcement learning
Liu, G., Chen, G., & Huang, V. (2023) · Neurocomputing · doi:10.1016/j.neucom.2023.126381 · Cited by 10
Korecki, M. (2023) · Entropy · doi:10.3390/e25070982 · Cited by 11
Stendal, J. A., & Bambach, M. (2023) · Journal of Intelligent Manufacturing · doi:10.1007/s10845-023-02173-6 · Cited by 8
“For instance, an RL algorithm can learn to master chess through self-play on a simulated board (Silver et al., 2018).”
A K-means Supported Reinforcement Learning Framework to Multi-dimensional Knapsack
Bushaj, S., & Büyüktahtakın, İ. E. (2024) · Journal of Global Optimization · doi:10.1007/s10898-024-01364-6 · Cited by 16
Reinforcement Learning for PID Gain Tuning in Ball-and-Beam Systems: A Comparative Study
Blake, X. J., & Aphiratsakun, N. (2024) · International Conference on Robotics, Engineering, Science, and Technology (RESTCON) · doi:10.1109/restcon60981.2024.10463594 · Cited by 9
“Applications for RL have been developed across diverse domains, such as mastering board games like Chess, Shogi, and Go [4], recommendation systems [5], energy management [6], healthcare [7], and computer vision [8].”
MethodEvaluating Superhuman Models with Consistency Checks
Fluri, L., Paleka, D., & Tramèr, F. (2024) · IEEE Conference on Secure and Trustworthy Machine Learning (SaTML) · doi:10.1109/satml59370.2024.00017 · Cited by 67
“To illustrate the challenge, consider a model trained to play chess—a canonical setting where models surpass humans [67, 17].”
BackgroundMethodInfluentialSarkhi, S. M. K., & Koyuncu, H. (2024) · AI · doi:10.3390/ai5030057 · Cited by 12
Learning quantum phases via single-qubit disentanglement
An, Z., Cao, C., Xu, C.-Q., & Zhou, D. L. (2024) · Quantum · doi:10.22331/q-2024-07-22-1421 · Cited by 12
What does AI’s success playing complex board games tell brain scientists?
Purves, D. (2019) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1909565116 · Cited by 8
Vision-Based Autonomous Driving: A Model Learning Approach
Baheri, A., Kolmanovsky, I. V., Girard, A., Tseng, H. E., & Filev, D. (2020) · Proceedings of the ... American Control Conference/Proceedings of the American Control Conference · doi:10.23919/acc45564.2020.9147510 · Cited by 12
“Reinforcement learning has shown dramatic success at learning to play Atari games and acting in complex environments to gain human-level control [9]–[11].”
Background
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.