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 1,001–1,100 of 3,633
Risk-sensitive policies for portfolio management
Wang, M., & Ku, H. (2022) · Expert Systems with Applications · doi:10.1016/j.eswa.2022.116807 · Cited by 26
Reinforcement learning: A brief guide for philosophers of mind
Haas, J. (2022) · Philosophy Compass · doi:10.1111/phc3.12865 · Cited by 23
“…system was able to develop superhuman play in the game Go by combining supervised learning of human games of Go and reinforcement learning that was used to improve the foregoing policy by having the system play itself (Silver et al., 2016; see also Silver et al., 2017; Silver et al., 2018).”
MethodNeuroevolution-Based Generation of Tests and Oracles for Games
Feldmeier, P., & Fraser, G. (2022) · IEEE/ACM International Conference on Automated Software Engineering (ASE) · doi:10.1145/3551349.3556939 · Cited by 18
“Many reinforcement learning techniques have been shown to be capable of learning to play different game genres such as strategy games [55, 56], arcade games [63], first-person shooter games [37, 42], and casual games [33].”
BackgroundUnveiling the Relationship between Economic Growth and Equality for Developing Countries
Cheong, T. S., Wan, G., & Chui, D. K. H. (2022) · China & World Economy · doi:10.1111/cwe.12435 · Cited by 16
Fan, Z., Huo, M., Quarta, A. A., Mengali, G., & Qi, N. (2022) · Aerospace Science and Technology · doi:10.1016/j.ast.2022.107946 · Cited by 14
Reinforcement Learning Toolkits for Gaming: A Comparative Qualitative Analysis
Mekni, M., Jayaramireddy, C. S., & Naraharisetti, S. V. V. S. S. (2022) · Journal of Software Engineering and Applications · doi:10.4236/jsea.2022.1512024 · Cited by 15
Song, Z., Zhou, Q., Lu, S., Dieb, S., Ling, C., & Wang, J. (2023) · The Journal of Physical Chemistry Letters · doi:10.1021/acs.jpclett.3c00242 · Cited by 15
Human achievement and artificial intelligence
Karlan, B. (2023) · Ethics and Information Technology · doi:10.1007/s10676-023-09713-x · Cited by 15
“2 More recent work has attempted to replicate superhuman performance at multiple games without hand-coding expertise of any kind (even the rules of the game) into the algorithm (Silver et al., 2018).”
BackgroundBrain-inspired chaotic spiking backpropagation
Wang, Z., Tao, P., & Chen, L. (2024) · National Science Review · doi:10.1093/nsr/nwae037 · Cited by 17
“Artificial neural networks (ANNs) or deep neural networks have been widely used in many fields, including image identification [1 –4 ], machine translation [5 ,6 ], protein structure prediction [7 ] and gaming [8 ,9 ].”
Fully Spiking Actor Network With Intralayer Connections for Reinforcement Learning
Chen, D., Peng, P., Huang, T., & Tian, Y. (2024) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2024.3352653 · Cited by 18
“To overcome the limitations of SNNs in solving high-dimensional control problems, it would be natural to combine the energy-efficiency of SNNs with the optimality of deep reinforcement learning (DRL), which has been proven effective in extensive control tasks [13], [14].”
Wu, Z., Zhang, W., Tang, R., Wang, H., & Korolija, I. (2024) · Journal of Building Engineering · doi:10.1016/j.jobe.2024.109497 · Cited by 19
“One example is AlphaZero, originating from the agent for the chessboard game AlphaGo [23].”
Autonomous Air Combat Maneuver Decision-Making Based on PPO-BWDA
Wang, H., Zhou, Z., Jiang, J., Deng, W., & Chen, X. (2024) · IEEE Access · doi:10.1109/access.2024.3419889 · Cited by 13
Salingaros, N. A. (2024) · New Design Ideas · doi:10.62476/ndi82261 · Cited by 24
AI-driven autonomous microrobots for targeted medicine
Medany, M., Mukkavilli, S. K., & Ahmed, D. (2024) · Nature Reviews Bioengineering · doi:10.1038/s44222-024-00232-y · Cited by 17
“RL provides a robust framework for agents to learn optimal strategies by obtaining rewards for optimal actions — for example, algorithms inspired by AlphaZero 5 and Deep Q-Networks 6 in complex strategic and search games.”
Huang, W., Cui, Y., Li, H., & Wu, X. (2024) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2024.3474169 · Cited by 27
“…RL that directly learns control strategies without model knowledge has been widely applied to not only outperform humans in video and rule-based games [2], [3] but also autonomously control complex systems, including chemical plant [4], [5], unmanned vehicles [6], [7], and robots [8], [9], [10].”
Wang, S., Wang, Z., Wang, X., Liang, Q., & Meng, L. Y. (2024) · Expert Systems with Applications · doi:10.1016/j.eswa.2024.126319 · Cited by 15
Keeping the human in the loop: are autonomous decisions inevitable?
Pöhler, J., Flegel, N., Mentler, T., & Laerhoven, K. V. (2025) · i-com · doi:10.1515/icom-2024-0068 · Cited by 20
Entropy Learning for Dynamic Treatment Regimes
Jiang, B.-Y., Song, R., Li, J.-L., & Zeng, D. (2019) · Statistica sinica · doi:10.5705/ss.202018.0076 · Cited by 25
Evolutionary approaches to dynamic earth observation satellites mission planning under uncertainty
Povéda, G., Regnier-Coudert, O., Teichteil-Königsbuch, F., Dupont, G., Arnold, A., Guerra, J., & Picard, M. (2019) · Genetic and Evolutionary Computation Conference (GECCO) · doi:10.1145/3321707.3321859 · Cited by 21
“It has gained a lot of recent attention due to DeepMind’s successes in coupling deep neural networks and reinforcement learning, for instance for general game solving [26].”
BackgroundCzech, J., Willig, M., Beyer, A., Kersting, K., & Fürnkranz, J. (2020) · Frontiers in Artificial Intelligence · doi:10.3389/frai.2020.00024 · Cited by 16
“Q-value. However, in cases when a strong counter reply was only discovered late during search, the Q-value converges quickly, but the respective node still remains at most visits for several samples. Silver et al. (2018) also acknowledged that deviating from the most visited node for move selection can yield better re…”
BackgroundMonster Carlo 2: Integrating Learning and Tree Search for Machine Playtesting
Keehl, O. G., & Smith, A. M. (2019) · IEEE Conference on Games (CoG) · doi:10.1109/cig.2019.8847989 · Cited by 13
“Our work was inspired by Google’s DeepMind’s successes with AlphaGo and AlphaZero.”
BackgroundKeeping it "organized and logical": after-action review for AI (AAR/AI)
Mai, T., Khanna, R., Dodge, J. E., Irvine, J., Lam, K.-H., Lin, Z., Kiddle, N. G., Newman, E., Raja, S., Matthews, C., et al. (2020) · ACM International Conference on Intelligent User Interfaces (IUI) · doi:10.1145/3377325.3377525 · Cited by 17
“This function is necessary to address the large action-branching factor by only performing the more expensive tree expansion under some number of top-ranked actions to improve estimates (similar to AlphaGo and AlphaZero [49, 50]).”
MethodA Comparison of Self-Play Algorithms Under a Generalized Framework
Hernández, D., Denamganaï, K., Devlin, S., Samothrakis, S., & Walker, J. A. (2021) · IEEE Transactions on Games · doi:10.1109/tg.2021.3058898 · Cited by 16
“This algorithm was later refined [11], removing the need for expert human moves.”
MethodGenerating attentive goals for prioritized hindsight reinforcement learning
Liu, P., Bai, C., Zhao, Y., Bai, C., Zhao, W. Z., & Tang, X. (2020) · Knowledge-Based Systems · doi:10.1016/j.knosys.2020.106140 · Cited by 14
Digital twin-enabled self-evolved optical transceiver using deep reinforcement learning
Li, J., Wang, D., Zhang, M., & Cui, S. (2020) · Optics Letters · doi:10.1364/ol.397972 · Cited by 18
AI solutions for drafting in Magic: the Gathering
Ward, H. N., Mills, B., Brooks, D. J., Troha, D., & Khakhalin, A. S. (2021) · IEEE Conference on Games (CoG) · doi:10.1109/cog52621.2021.9619100 · Cited by 19
Verifying Equivalence Properties of Neural Networks with ReLU Activation Functions
Büning, M. K., Kern, P., & Sinz, C. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-58475-7_50 · Cited by 14
The scaling of physics-informed machine learning with data and dimensions
Miller, S. T., Lindner, J. F., Choudhary, A., Sinha, S., & Ditto, W. L. (2020) · Chaos Solitons & Fractals X · doi:10.1016/j.csfx.2020.100046 · Cited by 15
Alberto, A. V. P., Ferreira, N. C. D. S., Soares, R. F., & Alves, L. A. (2020) · Frontiers in Pharmacology · doi:10.3389/fphar.2020.01221 · Cited by 17
“Artificial intelligence (AI) regularly outperforms humans, for example, an AI recently beat the best player at Go, a Chinese game considered more difficult than chess (Silver et al., 2018).”
BackgroundGuiding Multiplayer MCTS by Focusing on Yourself
Baier, H., & Kaisers, M. (2020) · IEEE Conference on Games (CoG) · doi:10.1109/cog47356.2020.9231603 · Cited by 13
“Recent efforts in game AI have largely focused on offline generalization, for example with the deep convolutional neural networks used by AlphaGo and AlphaZero [3], [4] for state”
BackgroundMethodInfluentialExact Reduction of Huge Action Spaces in General Reinforcement Learning
Majeed, S. J., & Hütter, M. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i10.17074 · Cited by 24
Intelligent Consumer Flexibility Management With Neural Network-Based Planning and Control
Kiljander, J., Sarala, R., Rehu, J., Pakkala, D., Pääkkönen, P., Takalo-Mattila, J., & Känsälä, K. (2021) · IEEE Access · doi:10.1109/access.2021.3060871 · Cited by 16
“Most well-known examples are the AlphaGo [16], and AlphaZero [17] that combine MCTS with deep reinforcement learning to provide superhuman and state-of-the-art performances in Go, Shogi and Chess.”
BackgroundNeural Tree Expansion for Multi-Robot Planning in Non-Cooperative Environments
Rivière, B., Hönig, W., Anderson, M., & Chung, S. (2021) · IEEE Robotics and Automation Letters · doi:10.1109/lra.2021.3096758 · Cited by 15
Wang, W., Han, X., Zhang, J., Shang, X., Ha, J. J., Liu, Z., Zhang, L., Luo, L., & He, M. (2021) · British Journal of Ophthalmology · doi:10.1136/bjophthalmol-2020-318609 · Cited by 15
Reinforcement Learning for Layout Planning – Modelling the Layout Problem as MDP
Unger, H., & Börner, F. (2021) · IFIP advances in information and communication technology · doi:10.1007/978-3-030-85906-0_52 · Cited by 16
Boutilier, J. J., & Chan, T. C. Y. (2023) · INFORMS Transactions on Education · doi:10.1287/ited.2021.0256 · Cited by 15
Adaptive Warm-Start MCTS in AlphaZero-Like Deep Reinforcement Learning
Wang, H., Preuß, M., & Plaat, A. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-89370-5_5 · Cited by 17
Planning for potential: efficient safe reinforcement learning
den Hengst, F., François-Lavet, V., Hoogendoorn, M., & van Harmelen, F. (2022) · Machine Learning · doi:10.1007/s10994-022-06143-6 · Cited by 25
“…driven by neural network-based function approximation, commonly known as Deep Reinforcement Learning (DRL), has recently shown remarkable progress in diverse areas such as personalization (den Hengst et al. 2020), robotics (Gu et al. 2017) and game-playing (Hessel et al. 2018; Silver et al. 2018).”
BackgroundCurriculum learning for safe mapless navigation
Marzari, L., Corsi, D., Marchesini, E., & Farinelli, A. (2022) · ACM Symposium on Applied Computing · doi:10.1145/3477314.3507182 · Cited by 22
“pressive performance in a variety of fields such as games [35], videogames [22], robotic manipulation tasks [21] [17], and especially autonomous navigation in mapless context [39].”
BackgroundThe Rhetoric and Reality of Anthropomorphism in Artificial Intelligence
Watson, D. (2020) · Digital ethics lab yearbook · doi:10.1007/978-3-030-29145-7_4 · Cited by 16
Regulating Artificial General Intelligence (AGI)
Mahler, T. (2022) · Information technology and law series/Information technology & law series · doi:10.1007/978-94-6265-523-2_26 · Cited by 23
Huang, B., Boularias, A., & Yu, J. (2022) · IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros47612.2022.9981962 · Cited by 22
“To address this issue, we borrow the idea of virtual loss [51], which has shown to give good results in multiple application domains [52]–[54].”
MethodA Deep Reinforcement Learning Based Framework for Power System Load Frequency Control
Zhang, G., Teng, M., Chen, C., & Bie, Z. (2022) · IEEE/IAS Industrial and Commercial Power System Asia (I&CPS Asia) · doi:10.1109/icpsasia55496.2022.9949649 · Cited by 15
“Top AI laboratories such as OpenAI and DeepMind have demonstrated that this data-driven approach can model-free solve a range of continous decision-making problems[7,8].”
BackgroundStanding on the Shoulders of AI Giants
Lin, H.-Y. (2023) · Computer · doi:10.1109/mc.2022.3218176 · Cited by 18
Fault-Tolerant Control of Degrading Systems with On-Policy Reinforcement Learning
Ahmed, I., Quiñones-Grueiro, M., & Biswas, G. (2020) · IFAC-PapersOnLine · doi:10.1016/j.ifacol.2020.12.878 · Cited by 14
“, 2013), board games (Silver et al., 2018), and smart buildings (Naug et al.”
BackgroundGame AI Competitions: Motivation for the Imitation Game-Playing Competition
Świechowski, M. (2020) · Annals of Computer Science and Information Systems · doi:10.15439/2020f126 · Cited by 18
“Research in computer chess after the last Ultimate Computer Chess Challenge [6] in 2007 shifted towards deep learning [7] and human-like approaches.”
BackgroundTowards Incorporating Human Knowledge in Fuzzy Pattern Tree Evolution
Murphy, A., Murphy, G., Amaral, J. L. M., Dias, D. M., Naredo, E., & Ryan, C. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-72812-0_5 · Cited by 14
GridToPix: Training Embodied Agents with Minimal Supervision
Jain, U., Liu, I.-J., Lazebnik, S., Kembhavi, A., Weihs, L., & Schwing, A. (2021) · 2021 IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv48922.2021.01486 · Cited by 14
“If present-day RL algorithms have been successful in other domains (particularly ones requiring less visual processing [62, 61, 16]), why do they struggle for Embodied AI? Our hypothesis: this struggle is due to the need for embodied agents to learn to plan and perceive simultaneously.”
BackgroundWeaver, G. C., Arya, R., Schneider, C. L., Hudson, A. W., & Stern, L. J. (2022) · Frontiers in Immunology · doi:10.3389/fimmu.2022.864898 · Cited by 13
“The protein structure prediction community was rocked in 2018 by the performance of AlphaFold (60), a machine learning algorithm from Google’s DeepMind project, which also produced the chess- and GO-playing AlphaZero and AlphaGo algorithms (61).”
MethodReinforcement Learning with Quantum Variational CircuitsPreprint
Lockwood, O., & Si, M. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2008.07524 · Cited by 15
“Achieving superhuman performance in massively complex games such as Chess, Go (Silver et al. 2018), StarCraft II (Vinyals et al.”
BackgroundBridging the human–AI knowledge gap through concept discovery and transfer in AlphaZero
Schut, L., Tomašev, N., McGrath, T., Hassabis, D., Paquet, U., & Kim, B. (2025) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2406675122 · Cited by 109
Self-citationDeep active optimization for complex systems
Wei, Y., Peng, B., Xie, R., Chen, Y., Qin, Y., Wen, P., Bauer, S., Tung, P., & Raabe, D. (2025) · Nature Computational Science · doi:10.1038/s43588-025-00858-x · Cited by 21
Su, J., & Dong, S. (2025) · Scientific Reports · doi:10.1038/s41598-025-18309-y · Cited by 21
A systematic review of machine learning approaches in inventory control optimization
Bergsma, R., de Ruijt, C., & Bhulai, S. (2025) · Operations Research Perspectives · doi:10.1016/j.orp.2025.100367 · Cited by 39
Li, X., Gu, J., He, F., Xiao, Y., Jia, Y., & Lan, P. (2026) · CAAI Transactions on Intelligence Technology · doi:10.1049/cit2.70108 · Cited by 15
Model-free safe reinforcement learning for chemical processes using Gaussian processes
Savage, T., Zhang, D., Mowbray, M., et al. (2021) · IFAC-PapersOnLine · doi:10.1016/j.ifacol.2021.08.292 · Cited by 13
Gupta, K., Mukherjee, D., & Najjaran, H. (2021) · SN Computer Science · doi:10.1007/s42979-021-00934-9 · Cited by 20
“Through self-play and reinforcement learning, [104] presented a general game-playing algorithmic agent, trained with random play and no domain-specific knowledge (tabula rasa), that eventually went on to defeat the world leading programs in the games of chess and shogi.”
BackgroundTowards Automatic Facility Layout Design Using Reinforcement Learning
Ikeda, H., Nakagawa, H., & Tsuchiya, T. (2022) · Annals of Computer Science and Information Systems · doi:10.15439/2022f25 · Cited by 13
Control of Magnetic Surgical Robots With Model-Based Simulators and Reinforcement Learning
Barnoy, Y., Erin, Ö., Raval, S., Pryor, W., Mair, L. O., Weinberg, I. N., Diaz-Mercado, Y. J., Krieger, A., & Hager, G. D. (2022) · IEEE Transactions on Medical Robotics and Bionics · doi:10.1109/tmrb.2022.3214426 · Cited by 15
Moreira, T. M., de Faria, J. G., de Melo, P. O. S. V., & Medeiros‐Ribeiro, G. (2022) · Applied Energy · doi:10.1016/j.apenergy.2022.120506 · Cited by 11
Design Methodology State-of-the-Art Report
Erikstad, S. O., & Lagemann, B. (2022) · doi:10.5957/imdc-2022-301 · Cited by 13
Huang, J. (2022) · Journal of Integrated Design and Process Science · doi:10.3233/jid-229010 · Cited by 33
Yuan, Y., Cheung, S. S., Vaerenbergh, T. V., Peng, Y., Hu, Y., Kurczveil, G., Huang, Z., Liang, D., Sorin, W. V., Xiao, X., et al. (2023) · APL Photonics · doi:10.1063/5.0146062 · Cited by 15
Hu, Z., Liu, H., Xiong, Y., Wang, L., Wu, R., Guan, K., Hu, Y., Lyu, T., & Fan, C. (2023) · Multimedia Tools and Applications · doi:10.1007/s11042-023-15361-6 · Cited by 13
“This strategy has been verified effective in a broad range of application scenarios, like go [47], chess [48], Doudizhu (A card game prevalent in China) [54], Dota 2 (A world renowned MOBA game) [7], etc.”
A superconducting synapse exhibiting spike-timing dependent plasticity
Segall, K., Purmessur, C., D’Addario, A. J., & Schult, D. (2023) · Applied Physics Letters · doi:10.1063/5.0150687 · Cited by 14
A Survey on Population-Based Deep Reinforcement Learning
Long, W., Hou, T., Wei, X., Yan, S., Zhai, P., & Zhang, L. (2023) · Mathematics · doi:10.3390/math11102234 · Cited by 18
Evaluating Explainable AI (XAI) in Terms of User Gender and Educational Background
Reeder, S., Jensen, J., & Ball, R. (2023) · Lecture notes in computer science · doi:10.1007/978-3-031-35891-3_18 · Cited by 21
ScanBot: Autonomous Reconstruction via Deep Reinforcement Learning
Cao, H., Xi, X., Wu, G., Hu, R., & Liu, L. (2023) · ACM Transactions on Graphics · doi:10.1145/3592113 · Cited by 15
“Bolstered by deep neural networks, it has unleashed its potential for solving a wide range of real-world tasks such as games [Hessel et al. 2018; Silver et al. 2018] and robot control [Akkaya et al. 2019; Gu et al. 2017; Peng et al. 2018b].”
BackgroundWermelinger, J., Parduzi, Q., Sariyar, M., Raabe, A., Schneider, U. C., & Seidel, K. (2023) · BMC Medical Informatics and Decision Making · doi:10.1186/s12911-023-02276-3 · Cited by 15
“When faced with good and plentiful data, ML can lead to astonishing results, as exemplified by AlphaGo [41], AlphaZero [42] and many more.”
ResultBelfadil, A., Modesto, D., Meseguer, J., Joseph‐Duran, B., Saporta, D., & Hernández, J. A. M. (2023) · Journal of Water Resources Planning and Management · doi:10.1061/jwrmd5.wreng-6108 · Cited by 18
Sublime, J. (2024) · Journal of Artificial Intelligence Research · doi:10.1613/jair.1.15315 · Cited by 33
“…et al., 2022), which was also based on a pruning algorithm before being hybridized with neural networks, before this type of model culminated for Chess with engines such as Leela Chess Zero (a Chess adapted version of AlphaGO Zero that we will discuss in the next paragraph) (Silver et al., 2018).”
MethodLyu, X., Rani, S., & Feng, Y. (2024) · IEEE Transactions on Consumer Electronics · doi:10.1109/tce.2024.3424780 · Cited by 13
“The remarkable successes of Google DeepMind with Atari games [24] and AlphaGo [25], following the achievements of IBM’s Deep Blue [26] and Tesauro’s TD Gammon [27], underscore DRL’s transformative impact in AI.”
Carbon trading supply chain management based on constrained deep reinforcement learning
Wang, Q., & Yang, Y. (2024) · Autonomous Agents and Multi-Agent Systems · doi:10.1007/s10458-024-09669-2 · Cited by 19
“DRL has demonstrated levels of performance surpassing human abilities in tasks such as Go [32, 33] and StarCraft [34].”
Service risk evaluation of telecommunication core network: A perspective of routing resilience
Xue, Z., & Liang, Z. (2024) · Reliability Engineering & System Safety · doi:10.1016/j.ress.2024.110629 · Cited by 16
Fast-developing machine learning support complex system research in environmental chemistry
Duan, Q., & Lee, J. (2019) · New Journal of Chemistry · doi:10.1039/c9nj05717j · Cited by 13
Alternative Loss Functions in AlphaZero-like Self-play
Wang, H., Emmerich, M., Preuß, M., & Plaat, A. (2019) · IEEE Symposium Series on Computational Intelligence (SSCI) · doi:10.1109/ssci44817.2019.9002814 · Cited by 11
Distributed Methods for Reinforcement Learning Survey
Czech, J. (2021) · Studies in computational intelligence · doi:10.1007/978-3-030-41188-6_13 · Cited by 17
Learning to Guide a Saturation-Based Theorem Prover
Abdelaziz, I., Crouse, M., Makni, B., Austel, V., Cornelio, C., Ikbal, S., Kapanipathi, P., Makondo, N., Srinivas, K., Witbrock, M., et al. (2022) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2022.3140382 · Cited by 26
“TRAIL begins learning through random exploration of the search space as done in AlphaZero [40] to establish performance when the system is started from a tabula rasa state”
MethodDynamic Automaton-Guided Reward Shaping for Monte Carlo Tree Search
Velasquez, A., Bissey, B., Barak, L., Beckus, A., Alkhouri, I., Melcer, D., & Atia, G. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i13.17427 · Cited by 15
“…the proposed AGRS function and apply it within a modern MCTS implementation using deep CNNs similar to those proposed for the game of Go (Silver et al. 2016) (Silver et al. 2017), various other board games (Silver et al. 2018) (Anthony, Tian, and Barber 2017), and Atari (Schrittwieser et al. 2020).”
MethodKB-Tree: Learnable and Continuous Monte-Carlo Tree Search for Autonomous Driving Planning
Lei, L., Luo, R., Zheng, R., Wang, J., Zhang, J., Qiu, C., Ma, L., Jin, L., Zhang, P., & Chen, J. (2021) · IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros51168.2021.9636442 · Cited by 14
“Further, AlphaZero [18] generalizes this approach into a single algorithm to master the games of chess, shogi, and Go.”
BackgroundLc, R., Benayoun, M., Lindborg, P., et al. (2021) · International Conference on Digital and Interactive Arts (ARTECH) · doi:10.1145/3483529.3483844 · Cited by 13
“In the realm of Go, Alp-haZero used a neural network approach repeatedly playing itself given the rules and goal states [29].”
MethodWell Construction Action Planning and Automation through Finite-Horizon Sequential Decision-Making
Saini, G. S., Erge, O., Ashok, P., & van Oort, E. (2022) · Energies · doi:10.3390/en15165776 · Cited by 17
Automated design of quantum-optical experiments for device-independent quantum key distribution
Valcarce, X., Sekatski, P., Gouzien, É., Melnikov, A. A., & Sangouard, N. (2023) · Physical Review A · doi:10.1103/physreva.107.062607 · Cited by 19
Bolus Insulin calculation without meal information. A reinforcement learning approach
Ahmad, S., Beneyto, A., Contreras, I., & Vehí, J. (2022) · Artificial Intelligence in Medicine · doi:10.1016/j.artmed.2022.102436 · Cited by 18
Chen, Z., Nakamura, Y., & Ishiguro, H. (2023) · IEEE Robotics and Automation Letters · doi:10.1109/lra.2023.3267385 · Cited by 14
“Recently, some world-shattering cases have been reported, such as Al-phaZero [7] and GT Sophy [8], in which RL-driven agents defeated human experts.”
BackgroundPeano: learning formal mathematical reasoning
Poesia, G., & Goodman, N. D. (2023) · Philosophical Transactions of the Royal Society A Mathematical Physical and Engineering Sciences · doi:10.1098/rsta.2022.0044 · Cited by 35
“Concurrently, machine learning methods have advanced substantially over the last decades, both in supervised learning [16], where a model is trained to fit a given dataset, and in reinforcement learning (RL), where the learner’s goal is choose actions that maximize reward signals in a dynamic environment [1,20,29].”
BackgroundCompetitive Physical Human-Robot Game Play
Yang, B., Xie, X., Habibi, G., & Smith, J. R. (2021) · ACM/IEEE International Conference on Human-Robot Interaction (HRI) · doi:10.1145/3434074.3447168 · Cited by 15
“Many algorithms that solve multi-agent Markov games are designed under a multi-agent reinforcement learning scheme, where agents develop emergent and complex behavior through interacting with each other and co-evolving together [8, 9, 11, 26, 28, 30].”
BackgroundHow Does AI Play Football? An Analysis of RL and Real-world Football Strategies
Scott, A., Fujii, K., & Onishi, M. (2022) · International Conference on Agents and Artificial Intelligence (ICAART) · doi:10.5220/0010844300003116 · Cited by 26
Planning for the Known Unknown: Machine Learning for Human Healthcare Systems
Chen, J. H., & Verghese, A. (2020) · The American Journal of Bioethics · doi:10.1080/15265161.2020.1822674 · Cited by 12
alpha-Rank: Multi-Agent Evaluation by EvolutionPreprint
Omidshafiei, S., Papadimitriou, C. H., Piliouras, G., et al. (2019) · arXiv (Cornell University) · Cited by 13
Self-citationPolicy Gradient Search: Online Planning and Expert Iteration without Search TreesPreprint
Anthony, T., Nishihara, R., Moritz, P., Salimans, T., & Schulman, J. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1904.03646 · Cited by 30
“AlphaZero [4, 6] also uses MCTS as a multi-step policy improvement operator, training residual neural networks to predict the policy and value of self-play games of the MCTS.”
BackgroundMethodContinual Learning in Neural NetworksPreprint
Aljundi, R. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1910.02718 · Cited by 56
“Current machine learning models represented by neural networks are able to learn and even outperform human level performance in individual tasks, as in Atari games [141] and object recognition [133].”
BackgroundSCARL: Side-Channel Analysis with Reinforcement Learning on the Ascon Authenticated CipherPreprint
Ramezanpour, K., Ampadu, P., & Diehl, W. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.03995 · Cited by 29
“RL has been using in algorithms that learn computer games, such as Atari and Go, which exceed the performance of a human [24, 34, 35].”
MethodNonstationary Reinforcement Learning with Linear Function ApproximationPreprint
Zhou, H., Chen, J., Varshney, L. R., & Jagmohan, A. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2010.04244 · Cited by 30
“2) For other special but not good chain ǧ ∈ [5], ǧ 6= g, ∀h ∈ [H], we have μǧh,k(sǧ) = 0.”
BackgroundInfluentialNavigating the landscape of multiplayer games
Omidshafiei, S., Tuyls, K., Czarnecki, W. M., Santos, F. C., Rowland, M., Connor, J. T., Hennes, D., Müller, P. F., Pérolat, J., Vylder, B. D., et al. (2020) · RePEc: Research Papers in Economics · Cited by 12
Discovering Diverse Multi-Agent Strategic Behavior via Reward RandomizationPreprint
Tang, Z., Yu, C., Chen, B., Xu, H., Wang, X., Fang, F., Du, S. S., Wang, Y., & Wu, Y. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.04564 · Cited by 54
“Most recent successes in games are based on decentralized multi-agent learning (Brown, 1951; Singh et al., 2000; Lowe et al., 2017; Silver et al., 2018), where agents compete against each other and optimize their own rewards to gradually improve their strategies.”
BackgroundMachine Learning the Gravity Equation for International TradePreprint
Verstyuk, S., & Douglas, M. R. (2022) · SSRN Electronic Journal · doi:10.2139/ssrn.4053795 · Cited by 23
“…2014), natural language processing (Vaswani et al., 2017) or chemistry (Jumper et al., 2021), as well as reinforcement learning to game playing (Silver et al., 2018), theorem proving (Kaliszyk et al., 2018; Zombori et al., 2020) and knowledge discovery (Gauthier, 2020; Davies et al., 2021;…”
SC2EGSet: StarCraft II Esport Replay and Game-state Dataset
Białecki, A., Jakubowska, N., Dobrowolski, P., Białecki, P., Krupiński, L., Szczap, A., Białecki, R., & Gajewski, J. (2023) · Scientific Data · doi:10.1038/s41597-023-02510-7 · Cited by 22
Learning and Acting in Peripersonal Space: Moving, Reaching, and Grasping
Juett, J., & Kuipers, B. (2019) · Frontiers in Neurorobotics · doi:10.3389/fnbot.2019.00004 · Cited by 13
“There have been recent impressive results from unguided end-to-end learning of multiple games [28, 27].”
BackgroundZhu, X., Li, K., Qu, H., & Hu, X. (2024) · Optics and Lasers in Engineering · doi:10.1016/j.optlaseng.2024.108777 · Cited by 16
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.