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 2,301–2,400 of 3,633
Comparison of Search Behaviors in Chess, Shogi, and the game of Go
Takeuchi, S. (2022) · Conference on Technologies and Applications of Artificial Intelligence (TAAI) · doi:10.1109/taai57707.2022.00041 · Cited by 2
“…the next visited node is selected based on the previous score and visited counts: where, Q ( s, a ) is the mean value and P ( s, a ) is the prior probability, N ( s, a ) is the visit count of selecting action a in state s , and a constant C ( s ) = log ((1+ N ( s )+ c base ) /c base )+ c init [10].”
BackgroundTordesillas, J., & Arbelaiz, J. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1904.01200 · Cited by 2
“stories of using RL for control [5], [6], we use model-free Deep-Reinforcement Learning (DRL) algorithms to design a personalized cancer chemotherapy schedule.”
MethodStructured agents for physical constructionPreprint
Bapst, V., Sánchez‐González, Á., Doersch, C., Stachenfeld, K., Kohli, P., Battaglia, P., & Hamrick, J. B. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1904.03177 · Cited by 99
“The efficiency of Monte-Carlo Tree Search (MCTS) (Coulom, 2006) planning in RL has recently been highlighted in (Guo et al., 2014; Silver et al., 2016; 2017; 2018).”
BackgroundMethodInfluentialDirect Policy Gradients: Direct Optimization of Policies in Discrete Action SpacesPreprint
Lorberbom, G., Maddison, C. J., Heess, N., Hazan, T., & Tarlow, D. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1906.06062 · Cited by 14
“MCTS is quite different because—unlike DirPG—it uses search and a simulator at test time, but it becomes closer when search results are distilled into a policy as in [36].”
BackgroundExample Based Hebbian Learning may be sufficient to support Human IntelligencePreprint
Wong, E. C. (2019) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/758375 · Cited by 2
“In addition to networks that are thought to provide a time base, there are also cells known as place cells, that together provide a spatial basis that appears to be stable (Muller, 1996;Smith and Mizumori, 2006;Moser et al., 2015).”
BackgroundZhan, H., & Cao, Y. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1909.12268 · Cited by 10
“In recent years, the application of RL in tasks with highdimensional sensory inputs has shown the potential of creating artificial agents that can learn to accomplish a number of challenging tasks, including the Atari games [1], [2], [3], [4], [5], [6], [7], self-driving cars [8], and Go [9], [10], [11].”
BackgroundZhan, H., & Cao, Y. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1910.01919 · Cited by 2
“In recent years, the application of reinforcement learning in missions with high-dimensional sensory inputs has shown the potential of creating artificial agents that can learn to accomplish a number of challenging tasks, including the Atari games [1]–[7], self-driving cars [8], and Go [9]–[11].”
BackgroundRare Rewards Amplify Dopamine Learning ResponsesPreprint
Rothenhoefer, K. M., Hong, T., Alikaya, A., & Stauffer, W. R. (2019) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.1101/851709 · Cited by 2
“Biological learning signals have inspired deep reinforcement learning algorithms with performance that exceeds expert human performance on Atari games, chess, and Go ( 26, 27 ).”
BackgroundD'Orazio, R., Morrill, D., Wright, J. R., et al. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1912.02967 · Cited by 2
“Function approximation has facilitated many recent successes in game playing more broadly [24, 25, 29].”
BackgroundWorld Programs for Model-Based Learning and Planning in Compositional State and Action SpacesPreprint
Segler, M. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1912.13007 · Cited by 6
“To conclude, we would also like to pose a small challenge: Given only a small set of historically played Go and Chess games as grid graphs, can you first learn the rules of the games to learn a simulator, and then compete with an AlphaZero-like agent (Silver et al., 2018) using an agent trained in the learned simulato…”
BackgroundMethodOn the Existence of Characterization Logics and Fundamental Properties of Argumentation SemanticsPreprint
Baumann, R. (2019) · arXiv (Cornell University) · doi:10.13140/rg.2.2.21831.24483 · Cited by 2
Analysis of Hyper-Parameters for Small Games: Iterations or Epochs in Self-Play?Preprint
Wang, H., Emmerich, M., Preuß, M., et al. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2003.05988 · Cited by 2
“Following the works by Silver et al. [2,3] the fundamental structure of AlphaZero-like Self-play is an iteration over three different stages (see Algorithm 1).”
BackgroundMethodMimicking Evolution with Reinforcement LearningPreprint
Abrantes, J. P., Abrantes, A. J., & Oliehoek, F. A. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2004.00048 · Cited by 5
“Arms-race with gradient methods With regards to gradient methods (deep learning methods in specific), impressive results have been recently achieved by training NN, through back-propagation, to compete against each other in simulated games (OpenFive 6 , AlphaZero 42 , GAN 19 ).”
Method6G: Connecting Everything by 1000 Times Price ReductionPreprint
Zhang, S., Xiang, C., & Xu, S. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2004.00853 · Cited by 2
“deep reinforcement learning [24], has demonstrated their potential to deal with huge scale optimization problems, such as playing chess or Go games, and the famous AlphaZero can beat top Go player even without the prior human experts’ knowledge [25].”
BackgroundA Brief Look at Generalization in Visual Meta-Reinforcement LearningPreprint
Alver, S., & Precup, D. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.07262 · Cited by 6
“In recent years, deep reinforcement learning (RL) algorithms have achieved significant success in a wide variety of challenging tasks, ranging from board games (Silver et al., 2017; 2018) to video games (Mnih et al., 2015; Vinyals et al., 2017).”
BackgroundApplication of variational policy gradient to atomic-scale materials synthesisPreprint
Liu, S., Borodinov, N., Vlček, L., et al. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.15644 · Cited by 2
“Reinforcement learning (RL) in recent years has achieved impressive results in an array of problems in continuous and discrete action spaces, including in games such as Chess, Go, Atari (Silver et al., 2018; Mnih et al., 2015), as well as in robotics (Devin et al.”
BackgroundLearning to Play Two-Player Perfect-Information Games without KnowledgePreprint
Cohen-Solal, Q. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2008.01188 · Cited by 18
“This knowledge was first generated by supervised learning [14, 15, 16, 17, 18] then by supervised learning followed by reinforcement learning [19], and finally by only reinforcement learning [2, 20, 21].”
Deep Replication of a Runoff PortfolioPreprint
Krabichler, T., & Teichmann, J. (2020) · RePEc: Research Papers in Economics · doi:10.48550/arxiv.2009.05034 · Cited by 2
“, see [31] by a research team of Google’s DeepMind), robotics (e.”
BackgroundEntropy, Computing and RationalityPreprint
Pineda, L. A. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2009.10224 · Cited by 4
Beyond Tabula-Rasa: a Modular Reinforcement Learning Approach for Physically Embedded 3D SokobanPreprint
Karkus, P., Mirza, M., Guez, A., Jaegle, A., Lillicrap, T., Buesing, L., Heess, N., & Weber, T. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2010.01298 · Cited by 8
“For example, consider the line of work leading from AlphaGo to MuZero [5, 57, 58, 59]: the earliest models in this family included domain-specific engineering, such as data augmentation by board rotation and perfect knowledge of game rules.”
BackgroundSelf-citationEvidence of Accelerating Mismeasurement of Growth and Inflation in the U.S. in the 21st Century
Nakamura, L. I. (2020) · Working paper · doi:10.21799/frbp.wp.2020.41 · Cited by 2
Affordance as general value function: A computational modelPreprint
Graves, D., Günther, J., & Luo, J. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2010.14289 · Cited by 2
“More recent successes using this approach range from video games (Mnih et al., 2013; Vinyals et al., 2019) to board games (Silver et al., 2016, 2018) to physically manipulating Rubik’s cube (OpenAI et al., 2019).”
A* Search Without Expansions: Learning Heuristic Functions with Deep Q-NetworksPreprint
Agostinelli, F., Shperberg, S. S., Shmakov, A., McAleer, S., Fox, R., & Baldi, P. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2102.04518 · Cited by 26
Learned Belief Search: Efficiently Improving Policies in Partially Observable SettingsPreprint
Hu, H., Lerer, A. K., Brown, N., & Foerster, J. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.09086 · Cited by 13
“Most of these successes were achieved in fully observable games such as Backgammon (Tesauro, 1994), Chess (Campbell et al., 2002) and Go (Silver et al., 2016; 2017; 2018).”
BackgroundOLIVAW: Mastering Othello with neither Humans nor a Penny.Preprint
Norelli, A., & Panconesi, A. (2021) · arXiv (Cornell University) · Cited by 2
Towards Teachable Autonomous AgentsPreprint
Sigaud, O., Caselles-Dupré, H., Colas, C., Akakzia, A., Oudeyer, P., & Chétouani, M. (2021) · HAL · Cited by 8
Ernst, E., & Mishra, S. (2021) · SSRN Electronic Journal · doi:10.2139/ssrn.3800783 · Cited by 2
Sample Efficient Reinforcement Learning via Model-Ensemble Exploration and ExploitationPreprint
Yao, Y., Li, X., An, Z., et al. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2107.01825 · Cited by 2
“Reinforcement learning (RL) has achieved many remarkable results in recent years, including outperforming human performances on video games [1]–[3], mastering the game of GO [4], playing card games [5], and even learning manipulation skills for robotics [6], [7].”
BackgroundCARL: Conditional-value-at-risk adversarial reinforcement learningPreprint
Godbout, M., Heuillet, M., Chandra, S., Bhati, R., & Durand, A. (2021) · arXiv (Cornell University) · doi:10.5281/zenodo.23112617 · Cited by 3
A deep learning guided memetic framework for graph coloring problemsPreprint
Goudet, O., Grelier, C., & Hao, J. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2109.05948 · Cited by 2
“These attempts have been given a new lease of life, with the emergence of deep learning techniques for combinatorial optimization problems [7, 46], inspired from the great success of the AlphaZero al-gorithm for combinatorial games [40].”
Wu, J., Huang, Z., Huang, W., & Lv, C. (2021) · arXiv (Cornell University) · Cited by 2
Motivating Physical Activity via Competitive Human-Robot InteractionPreprint
Yang, B., Habibi, G., Lancaster, P., Boots, B., & Smith, J. R. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2202.07068 · Cited by 9
Human learning follows the dynamics of gradient descentPreprint
Barry, D. N., & Love, B. C. (2021) · PsyArXiv (OSF Preprints) · doi:10.31234/osf.io/75e4t · Cited by 2
The Nature of Information, Semantics, and Effectiveness for Artificial Intelligence and CognitionPreprint
Brown, K. (2021) · OSF Preprints (OSF Preprints) · doi:10.31219/osf.io/dehkj · Cited by 4
Artificial Intelligence Guided De Novo Molecular Design Targeting COVID-19Preprint
Srinivasan, S., Batra, R., Chan, H., et al. (2020) · ChemRxiv · doi:10.26434/chemrxiv.12581075 · Cited by 2
On CVaR-Based Reinforcement Learning in Quantitative InvestmentPreprint
Alameer, A., & Alshehri, K. A. (2022) · doi:10.36227/techrxiv.19961525.v1 · Cited by 2
Deep Reinforcement Learning-Based Long-Range Autonomous Valet Parking for Smart CitiesPreprint
Khalid, M., Wang, L., Wang, K., Pan, C., Aslam, N., & Cao, Y. (2022) · SSRN Electronic Journal · doi:10.2139/ssrn.4136570 · Cited by 2
“The reinforcement learning being an emerging technology playing a beneficial role in scenarios where environment is dynamic and keep changing frequently [22].”
BackgroundHigher Cognition: A Mechanical Perspective
Friedman, R. (2022) · Encyclopedia · doi:10.3390/encyclopedia2030102 · Cited by 1
“The visual and symbolic puzzles, such as in the case of chess and related boardgames [52], are solvable by deep learning approaches, but the machine reasoning is not generalized across a space of abstract environments and objects.”
BackgroundJulia, E. S. (2022) · doi:10.14393/ufu.di.2022.562 · Cited by 2
Cognitive Escape Reinforcement Learning for Complex Decision MakingPreprint
Wu, Q., Zhao, S., Zhou, F., Zhang, H., Huang, Y., & Ma, K. (2023) · Research Square · doi:10.21203/rs.3.rs-2661516/v1 · Cited by 2
Learning state-action features for general game playing
Soemers, D. J. N. J. (2023) · Research Publications (Maastricht University) · doi:10.26481/dis.20230425ds · Cited by 4
Long short-term prediction guides human metacognitive reinforcement learningPreprint
Lee, S. W., Kim, D., Lee, J. H., Jung, W., & Kim, S. H. (2023) · Research Square · doi:10.21203/rs.3.rs-3080402/v1 · Cited by 2
Reverse-engineering the selfPreprint
Paul, L. A., Ullman, T. D., Freitas, J. D., & Tenenbaum, J. B. (2023) · PsyArXiv (OSF Preprints) · doi:10.31234/osf.io/vzwrn · Cited by 6
Iterative improvements from feedback for language modelsPreprint
Li, Y. (2023) · doi:10.14293/pr2199.000220.v1 · Cited by 5
Emergence of Cooperation in Two-Agent Repeated Games with Reinforcement LearningPreprint
Zhang, J., Cai, W.-R., Zheng, G.-Z., et al. (2023) · SSRN Electronic Journal · doi:10.2139/ssrn.4523815 · Cited by 2
MARLIM: Multi-Agent Reinforcement Learning for Inventory ManagementPreprint
Leluc, R., Kadoche, E., Bertoncello, A., & Gourvénec, S. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2308.01649 · Cited by 14
“Such paradigm has reached tremendous success in games [35, 48, 56], but the early applications of reinforcement learning to real-world tasks, e.”
BackgroundHippocampus shapes cortical sensory output and novelty coding through a direct feedback circuitPreprint
Butola, T., Hernández‐Frausto, M., Blankvoort, S., Flatset, M. S., Peng, L., Elmaleh, M., Hairston, A., Hussain, F., Claudia Clopath, A., Kentros, C. G., et al. (2023) · Research Square · doi:10.21203/rs.3.rs-3270016/v1 · Cited by 1
Patel, V., Shrivastava, A., Nagar, S. K., & GUPTA, N. (2023) · doi:10.61909/isbn.978-81-966743-8-0.amkedtb122304 · Cited by 5
End-to-End Path Planning for Homogeneous Temperature Fields in Additive ManufacturingPreprint
Sideris, I., Duncan, S., Fabbri, M., et al. (2023) · SSRN Electronic Journal · doi:10.2139/ssrn.4656481 · Cited by 2
Human Divergent Exploration Capacity for Material Design: A Comparison with Artificial IntelligencePreprint
Sakai, H., Matsuda, K., Kikkawa, N., et al. (2023) · doi:10.31234/osf.io/y69px · Cited by 2
Application of Reinforcement Learning to Deduce Nuclear Power Plant Severe Accident ScenarioPreprint
Song, S. H., Lee, Y., Bae, J. Y., et al. (2023) · SSRN Electronic Journal · doi:10.2139/ssrn.4676198 · Cited by 2
Learning from rewards and social information in naturalistic strategic behaviorPreprint
Kuperwajs, I., van Opheusden, B., Russek, E. M., & Griffiths, T. L. (2024) · PsyArXiv (OSF Preprints) · doi:10.31234/osf.io/d8zje · Cited by 4
Data Quality Evaluation and Improvement for Machine Learning
Chen, H. (2022) · University of North Texas Digital Library (University of North Texas) · doi:10.12794/metadc1944318 · Cited by 10
Pseudo-Consciousness in AI: Bridging the Gap Between Narrow AI and True AGIPreprint
Prestes, J. A. D. L. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5147424 · Cited by 5
Diagnostic and Surgical Uses of Explainable AI (XAI)
Budha, R. R., Khan, S. W. A. M., Lokhande, T., Rao, G. S. N. K., & Aaghaz, S. (2025) · Wiley eBooks · doi:10.1002/9781394302444.ch3 · Cited by 2
Müller, V. C. (2025) · Wiley eBooks · doi:10.1002/9781394238651.ch6 · Cited by 4
Symphony: Edge-powered Decentralized Multi-agent Framework for Autonomous Co-evolving IntelligencePreprint
Wang, J., Feng, Y., Ai, L., & Shi, B. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5361139 · Cited by 2
Rosenbloom, P. S., Laird, J. E., & Lebiere, C. J. (2022) · The Conversation · doi:10.64628/aai.gs9hyen9g · Cited by 3
Non-adiabatic molecular dynamics of PAH-related complexes
Posenitskiy, E. (2020) · theses.fr (ABES) · doi:10.70675/bfb35103z00bbz4bf9z80fdze96b19b060bb · Cited by 3
Schuster, T., Kalyan, A., Polozov, O., & Kalai, A. T. (2021) · arXiv · doi:10.48550/arxiv.2106.05784 · Cited by 2
“Several neighboring fields that have made substantial progress in reasoning include theorem prov357 ing [6], two-player game playing [53], and SAT-solving [7].”
BackgroundMethodBALROG: Benchmarking Agentic LLM and VLM Reasoning On Games
Paglieri, D., Cupiał, B., Coward, S., Piterbarg, U., Wolczyk, M., Khan, A., Pignatelli, E., Kuciński, Ł., Pinto, L., Fergus, R., et al. (2024) · arXiv · doi:10.48550/arxiv.2411.13543 · Cited by 2
“…game environments into a unified testbed for research on long-context LLMs. Games have historically served as highly effective metrics for evaluating progress in deep reinforcement learning research (Bellemare et al., 2013; Silver et al., 2018; Schrittwieser et al., 2020; Vinyals et al., 2019).”
Models of learning in games: An overview
Chernov, G., & Susin, I. S. (2019) · Journal of the New Economic Association · doi:10.31737/2221-2264-2019-44-4-3 · Cited by 1
Oliveira, T. G., & Oliveira, A. L. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-16474-3_35 · Cited by 1
Rabaey, J. M. (2022) · Foundations and Trends® in Integrated Circuits and Systems · doi:10.1561/3500000006 · Cited by 1
Load Balancing in Compute Clusters With Delayed Feedback
Tahir, A., Alt, B., Rizk, A., & Koeppl, H. (2022) · IEEE Transactions on Computers · doi:10.1109/tc.2022.3215907 · Cited by 2
An AlphaZero-Inspired Approach to Solving Search Problems
Dantsin, E. Y., Крейнович, В., & Wolpert, A. (2023) · Studies in systems, decision and control · doi:10.1007/978-3-031-16415-6_20 · Cited by 1
Summerfield, C. (2022) · Oxford University Press eBooks · doi:10.1093/oso/9780192843883.003.0006 · Cited by 1
Reward Delay Attacks on Deep Reinforcement Learning
Sarkar, A., Feng, J., Vorobeychik, Y., Gill, C. D., & Zhang, N. (2023) · Lecture notes in computer science · doi:10.1007/978-3-031-26369-9_11 · Cited by 5
Valeri, J. A., Krishnan, A., & Collins, J. J. (2023) · Nature Microbiology · doi:10.1038/s41564-023-01390-2 · Cited by 1
A Deep Reinforcement Learning Technique for PNPSC Net Player Strategies
Bearss, E. M., & Petty, M. D. (2023) · ACM Southeast Conference (ACMSE) · doi:10.1145/3564746.3587011 · Cited by 1
“MCTS is then used to search this game tree [10].”
MethodDu, K., Gemp, I., Wu, Y., et al. (2023) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v37i13.26962 · Cited by 1
Model-Based Reinforcement Learning with Multinomial Logistic Function Approximation
Hwang, T., & Oh, M.-H. (2023) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v37i7.25964 · Cited by 15
Bandit Approach to Conflict-Free Parallel Q-Learning in View of Photonic Implementation
Shinkawa, H., Chauvet, N., Röhm, A., Mihana, T., Horisaki, R., Bachelier, G., & Naruse, M. (2023) · Intelligent Computing · doi:10.34133/icomputing.0046 · Cited by 3
Xu, Y., Pu, Z., Cai, Q., Li, F., & Chai, X. (2023) · Lecture notes in computer science · doi:10.1007/978-3-031-44223-0_5 · Cited by 1
Overview of Game Decision Intelligence
Wang, K., & Bai, D. (2023) · IEEE International Conference on Information, Communication and Networks (ICICN) · doi:10.1109/icicn59530.2023.10392489 · Cited by 1
“DeepMind then launched an upgraded version of AlphaGo Zero called AlphaZero, which achieved excellent results in the game of Go, chess, and Japanese shogi and was published in Science [18], causing widespread attention.”
ResultPrior-Information Enhanced Reinforcement Learning for Energy Management Systems
Zangato, T., Osmani, A., & Alizadeh, P. (2024) · International Conference on AI, Machine Learning and Applications (AIMLA) · doi:10.5121/csit.2024.140207 · Cited by 1
“Recently, RL has gained attention, particularly for its success in the gaming field [12], resulting in the emergence of RL-based EMS as a vibrant research area [13].”
BackgroundMEPE: A Minimalist Ensemble Policy Evaluation Operator for Deep Reinforcement Learning
He, Q., & Hou, X. (2024) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp48485.2024.10448009 · Cited by 1
“Deep Reinforcement Learning (DRL) has achieved remarkable success in a wide range of domains such as board games [1], video games [2], robot manipulation [3, 4], etc.”
BackgroundIsmail, M. S. (2024) · Mathematical Social Sciences · doi:10.1016/j.mathsocsci.2024.03.004 · Cited by 7
“Silver et al. [32] introduced AlphaZero, which achieved great success in not only one game, but in three di ff erent games: chess, shogi, and Go.”
BackgroundFast and Knowledge-Free Deep Learning for General Game Playing (Student Abstract)
Maras, M., Kępa, M., Kowalski, J., et al. (2024) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v38i21.30480 · Cited by 1
GoFish: A customizable open-source chess engine in Go
Jubair, F., Salem, A., Assaf, Y. A., et al. (2024) · Expert Systems with Applications · doi:10.1016/j.eswa.2024.123855 · Cited by 1
New Fuzzing Biases for Action Policy Testing
Eisenhut, J., Schuler, X., Fišer, D., Höller, D., Christakis, M., & Hoffmann, J. (2024) · Proceedings of the International Conference on Automated Planning and Scheduling · doi:10.1609/icaps.v34i1.31472 · Cited by 1
“Learned action policies, in particular ones represented by neural networks, are gaining traction in AI (e.g., Mnih et al. 2013; Silver et al. 2016, 2018), and are being intensively explored in AI planning (Issakkimuthu, Fern, and Tadepalli 2018; Groshev et al. 2018; Garg, Bajpai, and Mausam 2019;…”
BackgroundIncorporating Domain Knowledge Into Monte Carlo Tree Search in Dark Chess
Yen, S.-J., Huang, C.-W., Yen, Y.-F., & Chen, J.-C. (2024) · International Conference on Applied System Innovation · doi:10.1109/icasi60819.2024.10547962 · Cited by 1
“This method uses multiple simulations to collect and count the simulation performance of each move to evaluate the quality of moves and search directions [5][6][7].”
MethodOn Necessity of Conscious Learning: From Robots to Humans
Weng, J. (2024) · Mechanisms and machine science · doi:10.1007/978-3-031-62664-7_4 · Cited by 1
Cheney-Lippold, J. (2024) · Public Culture · doi:10.1215/08992363-11158965 · Cited by 1
Generalization and Abstraction: Human Memory as a Magic Library
Rogers, T. T. (2024) · Oxford University Press eBooks · doi:10.1093/oxfordhb/9780190917982.013.7 · Cited by 3
GameTable Working Group 1 meeting report on search, planning, learning, and explainability
Soemers, D. J. N. J., Kowalski, J., Piette, É., Morenville, A., & Crist, W. (2024) · ICGA Journal · doi:10.3233/icg-240251 · Cited by 4
“…with a focus on using concept-based approaches (Kim et al., 2018; Achtibat et al., 2023) to discover human-understandable concepts learned using AlphaZero-like combinations of deep learning and search (Silver et al., 2018; McGrath et al., 2022; Hammersborg and Strümke, 2023; Schut et al., 2023).”
Yet More Optimistic Temporal Difference Learning for Game Mini2048
Matsuzaki, K., & Terauchi, S. (2024) · IEEE Conference on Games (CoG) · doi:10.1109/cog60054.2024.10645595 · Cited by 1
“For example, a classic application of temporal difference learning goes back to TD-Gammon [1], and recent ones include Deep Q Learning applied to Atari games [2] and AlphaZero and its successors [3], [4] for many games.”
XQSV: A Structurally Variable Network to Imitate Human Play in Xiangqi
Zhou, C. (2024) · IEEE Conference on Games (CoG) · doi:10.1109/cog60054.2024.10645635 · Cited by 1
“More recently, deep reinforcement learning has emerged as a powerful method including DeepMind’s AlphaGo [2] and AlphaZero [3]–[5] demonstrating superhuman performance in chess and Shogi, learning solely from game rules without supervised learning.”
MethodLearning to Play 7 Wonders Duel Without Human Supervision
Paolini, G., Moreschini, L., Veneziano, F., et al. (2024) · IEEE Conference on Games (CoG) · doi:10.1109/cog60054.2024.10645555 · Cited by 1
“Our approach is similar to AlphaZero [1], the algorithm developed by DeepMind to achieve super-human performance in games such as Go, Chess, and Shogi.”
MethodHarnessing RRAM Technology for Efficient AI Implementation
Gu, F.-Y. (2024) · Artificial intelligence · doi:10.5772/intechopen.1006094 · Cited by 1
Unsupervised Replay Strategies for Continual Learning with Limited Data
Bazhenov, A., Dewasurendra, P., Krishnan, G. P., & Delanois, J. E. (2024) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn60899.2024.10650116 · Cited by 5
“Although artificial neural networks (ANNs) have recently begun to approach human performance in various tasks, from intricate games [33] to image classification [34] to language processing [35], they exhibit several limitations.”
BackgroundLerousseau, J. P., & Summerfield, C. (2024) · Behavioral and Brain Sciences · doi:10.1017/s0140525x24000190 · Cited by 1
An Asymmetric Game Theoretic Learning Model
Yin, Q., Yu, T., Feng, X., Yang, J. J., & Huang, K. (2024) · Lecture notes in computer science · doi:10.1007/978-981-97-8502-5_10 · Cited by 1
Analysis of Medical Image Processing using Machine Learning Applications - A Review
(2024) · International Journal of Advanced Trends in Computer Science and Engineering · doi:10.30534/ijatcse/2024/061362024 · Cited by 1
On Generating Explanations for Reinforcement Learning Policies: An Empirical Study
Yuasa, M., Tran, H., & Sreenivas, R. S. (2024) · IEEE Control Systems Letters · doi:10.1109/lcsys.2024.3519301 · Cited by 1
“Whether it is triumphing over world champions in games such as Chess and Go or enhancing recommendation systems and robotics, RL has unmistakably demonstrated its vast potential [1]–[3].”
BackgroundReinforcement Learning for Optimizing Routing in the Production Supply of Matrix Production Systems
Ried, F., Niederdränk, S., & Fottner, J. (2025) · Communications in computer and information science · doi:10.1007/978-3-031-80760-2_17 · Cited by 1
Role-Specific Reward Design with Large Language Model for StarCraft II
Li, S., Lou, H., Zhang, X., Zeng, X., Shen, Z., & Li, T. (2025) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp49660.2025.10890857 · Cited by 2
“On the other hand, Reinforcement Learning (RL) has shown excellent results in ATARI games [6], Chess [7], and StarCraft [8].”
Confluence of Photonics and Artificial Intelligence
Chen, Y.-K. (2025) · IEEE Journal of Selected Topics in Quantum Electronics · doi:10.1109/jstqe.2025.3552430 · Cited by 3
“AI systems have achieved exceptional performance in complex games like Go and poker, surpassing human players [5].”
Classification Using Hybrid Classical-Quantum Machine Learning
Bala, R., Kher, G., Singh, V., & Singh, R. (2025) · Communications in computer and information science · doi:10.1007/978-3-031-84059-3_14 · Cited by 1
Convolutional Neural Networks with Specific Kernels for Computer Chess
Goudet, O., Joshi, B., & Cazenave, T. (2025) · Lecture notes in computer science · doi:10.1007/978-3-031-86585-5_2 · Cited by 1
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.