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 2,801–2,900 of 3,633
Time-Variant Variational Transfer for Value FunctionsPreprint
Canonaco, G., Soprani, A., Roveri, M., & Restelli, M. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2005.12864
“The reinforcement learning (RL) framework [30] is becoming increasingly more effective in dealing with complex problems [37, 29, 25] at the cost of requiring a huge amount of experience in order to achieve these impressive results.”
BackgroundSwitchblade - a Neural Network for Hard 2D Tasks.Preprint
Ozoliņš, E., Freivalds, K., & Šostaks, A. (2020) · arXiv (Cornell University)
Evolving ab initio trading strategies in heterogeneous environments
Dewhurst, D. R., Li, Y., Bogdan, A., & Geng, J. (2020) · Genetic and Evolutionary Computation Conference (GECCO) · doi:10.1145/3377930.3389806 · Cited by 1
“Recent years have seen substantial progress toward this goal [49, 56, 57].”
BackgroundNAPPO: Modular and scalable reinforcement learning in pytorch.Preprint
Bou, A., & Fabritiis, G. D. (2020) · arXiv (Cornell University)
Playing Catan with Cross-dimensional Neural NetworkPreprint
Gendre, Q., & Kaneko, T. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2008.07079
“ation of a game, as 1, 0, -1 for win, draw, loss, respectively. Given that agents are not enhanced by game-specic knowledge, the agent as well as its opponent must start as random players. AlphaZero [9] begins by gathering game records of random players, then gradually updates the agent by their experiences and period…”
BackgroundMobile Networks for Computer GoPreprint
Cazenave, T. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2008.10080 · Cited by 2
“Later Alpha Zero successfully applied the same principles to the games of Chess and Shogi [5].”
BackgroundMethodA Structured Approach to Risk Assessment of Machine Learning Applications
Fackrell, J., Glomsrud, J. A., & Eisinger, S. (2020) · Smart innovation, systems and technologies · doi:10.1007/978-3-030-57745-2_33
Constrained Bayesian Neural Networks
Lorch, L. (2019) · Digital Access to Scholarship at Harvard (DASH) (Harvard University)
Bestärkendes Lernen als Teilgebiet des Maschinellen Lernens
Lorenz, U. (2020) · Springer eBooks · doi:10.1007/978-3-662-61651-2_1
TotalBotWar: A New Pseudo Real-time Multi-action Game Challenge and Competition for AIPreprint
Estaben, A., Díaz, C., Montoliu, R., et al. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2009.08696
“For instance, deep reinforcement learning has enabled computers to learn how to play games such as Chess [1], Go [1], Atari games [2], and many other games [3].”
BackgroundLatent Weights Generating for Few Shot Learning Using Information Theory
Li, Z., & Ji, Y. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-58799-4_72
Matrix Shuffle- Exchange Networks for Hard 2D Tasks
Ozoliņš, E., Freivalds, K., & Šostaks, A. (2021) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn52387.2021.9533919 · Cited by 1
“AlphaGo Zero is based on a similar ResNet architecture and has been successfully applied to grid-like games [28], [25], making it a viable option for matrix tasks.”
BackgroundMethodChess as a Testing Grounds for the Oracle Approach to AI SafetyPreprint
Miller, J. D., Yampolskiy, R. V., Häggström, O., & Armstrong, S. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2010.02911 · Cited by 4
“Through self-play and reinforcement learning, it created its own evaluation criteria from scratch; see Silver et al (2018) and Sadler and Regan (2019).”
BackgroundMethodAmortized Variational Deep Q NetworkPreprint
Zhang, H., Wang, Y., Sun, J., & Xu, Z. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2011.01706
“For example, the agents trained with RL on StarCraft and Go have surpassed the top level of human [1, 2].”
Formal Verification of Tree Ensembles in Safety-Critical Applications
Törnblom, J. (2020) · Linköping studies in science and technology. Thesis · doi:10.3384/lic.diva-170862
Praeger, M. F., Xie, Y., Grant‐Jacob, J. A., et al. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2011.04424
Curiosity Based Reinforcement Learning on Robot Manufacturing CellPreprint
Hameed, M. S. A., Khan, M. M., & Schwung, A. (2020) · arXiv (Cornell University) · doi:10.13140/rg.2.2.11814.75847
LBGP: Learning Based Goal Planning for Autonomous Following in FrontPreprint
Nikdel, P., Vaughan, R., & Chen, M. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2011.03125
Efficient ESD Generator Modeling Using Reinforcement Learning
Kumar, A., Yang, E.-C., Shih, M.-C., Li, Y., Chuang, W.-T., & Chang, N. H. (2020)
Advice is Useful for Game AI: Experiments with Alpha-Beta Search Players in Shogi
Takeuchi, S. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-65883-0_1
Digital Me Ontology and Ethics
Kocarev, L., & Koteska, J. (2022) · Futures of Education Culture and Nature - Learning to Become · doi:10.7146/fecun.v1i.130229 · Cited by 15
Research Programs Based on Machine Intelligence Games
Tamburrini, G., & Altiero, F. (2020) · Philosophy of engineering and technology · doi:10.1007/978-3-030-54522-2_11 · Cited by 1
Towards sample-efficient episodic control with DAC-MLPreprint
Freire, I. T., Amil, A. F., Vouloutsi, V., & Verschure, P. F. M. J. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2012.13779
“Reaching human-level performance in historically considered uniquely-human intellectual domains (such as chess [1], Go [2], and real-time strategy video games [3]) has led to renewed claims about its resemblance to human cognition.”
BackgroundAchilles Heels for AGI/ASI via Decision Theoretic AdversariesPreprint
Casper, S. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2010.05418 · Cited by 4
“An example is AlphaMu [58] (and its predecessor AlphaZero [60]) which greatly surpasses human performance in the games of Go, Chess, and Shogi, demonstrating that human minds do not represent the apex of intelligence.”
BackgroundSeeking Nonhuman Advice: Ancient and Modern
Hutchinson, Z. (2020) · doi:10.1109/ai4g50087.2020.9311038
Zero-Knowledge Agent Trained for the Game of Risk
Bethdavid, S. (2020) · Uppsala University Publications (Uppsala University)
Solving QSAT problems with neural MCTSPreprint
Xu, R., & Lieberherr, K. J. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2101.06619
Simple Rules for a Complex World with Arti?cial IntelligencePreprint
Fernández‐Villaverde, J. (2020) · RePEc: Research Papers in Economics
Decisions and Performance Under Bounded Rationality: A Computational Benchmarking Approach
Zēgners, D., Sunde, U., & Strittmatter, A. (2020) · Open access LMU (Ludwid Maxmilian's Universitat Munchen) · doi:10.5282/ubm/epub.74746 · Cited by 8
“Only very recently more general machine learning techniques in the form of neural networks have been embodied in chess engines such as Google’s non-public AlphaZero (Silver et al., 2018).”
BackgroundFidelity-based Deep Adiabatic Scheduling
Ovits, E., & Wolf, L. (2021) · International Conference on Learning Representations
PyTorchRL: Modular and Distributed Reinforcement Learning in PyTorchPreprint
Bou, A., & Fabritiis, G. D. (2020) · arXiv (Cornell University)
ScrofaZero: Mastering Trick-taking Poker Game Gongzhu by Deep Reinforcement LearningPreprint
Shi, N., Li, R., & Youran, S. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2102.07495 · Cited by 1
“, 2017) and Go (Silver et al., 2018) by using deep reinforcement learning.”
BackgroundMethodInfluentialCombining Off and On-Policy Training in Model-Based Reinforcement LearningPreprint
Borges, A., & Oliveira, A. L. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2102.12194 · Cited by 6
“AlphaZero took three days to achieve superhuman performance in the game of Go, using a total of 5064 TPUs (Tensor Processing Units).”
BackgroundMethodInfluentialMoores, B. M. (2021) · Radiation Protection Dosimetry · doi:10.1093/rpd/ncab005
Robust and Efficient Planning using Adaptive Entropy Tree Search.Preprint
Kozakowski, P., Pacek, M., & Miłoś, P. (2021) · arXiv (Cornell University)
Okawa, Y., Sasaki, T., & Iwane, H. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.03656
“, 2015) or strategy board games (Silver et al., 2018); however, we cannot ignore this in many kinds of engineering problems.”
BackgroundEvaluation of Loss Function for Stable Policy Learning in Dobutsu Shogi
Nakayashiki, T., & Kaneko, T. (2020) · Conference on Technologies and Applications of Artificial Intelligence (TAAI) · doi:10.1109/taai51410.2020.00044
“AlphaZero [1] showed super-human performance in Go, chess, and shogi, by using reinforcement learning techniques.”
BackgroundZero-shot Policy Learning with Spatial Temporal RewardDecomposition on Contingency-aware ObservationPreprint
Xu, H., Chen, B., Gao, Y., et al. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1910.08143
“(3) More generally, one can define auxiliary tasks to learn a meaningful intermediate representations (Huang et al., 2019a; Silver et al., 2018; Dosovitskiy & Koltun, 2016; Riedmiller et al., 2018; Agarwal et al., 2019).”
BackgroundRapid Play Digital Games for Emergency Management Instruction
Seater, R. M. (2021) · Advances in educational technologies and instructional design book series · doi:10.4018/978-1-7998-4087-9.ch004
User-Oriented Smart General AI System under Causal InferencePreprint
Peng, H. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.14561
“AlphaGo Zero [1] outperforms the best human players in chess, shogi and Go games.”
BackgroundOn AO*, Proof Number Search and Minimax SearchPreprint
Gao, C. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.16692
The Philosophical Significance of Deep Learning
Suzuki, T. (2021) · Kagaku tetsugaku · doi:10.4216/jpssj.53.2_151
Mapping the Internet: Modelling Entity Interactions in Complex Heterogeneous NetworksPreprint
Mandlík, Š., & Pevný, T. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2104.09650 · Cited by 3
Two-stage training algorithm for AI robot soccerPreprint
Kim, T., Vecchietti, L. F., Choi, K., et al. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2104.05931
“Recently, deep reinforcement learning (DRL) has been widely applied to deterministic games [1], video games [2, 3, 4], and complex robotic tasks [5, 6, 7, 8].”
BackgroundAlpha-T: Learning to Traverse over Graphs with An AlphaZero-inspired Self-Play FrameworkPreprint
Wang, Q. (2021) · Research Square · doi:10.21203/rs.3.rs-415344/v1
“DeepMind designed a series of algorithms for the more complex games of Go, chess, and so on, such as AlphaGo, AlphaGo Zero, and AlphaZero, and applied them to defeat professional human masters, officially ushering in the era of artificial intelligence [28].”
BackgroundInfluentialTowards Utilitarian Combinatorial Assignment with Deep Neural Networks and Heuristic Algorithms
Präntare, F., Tiger, M., Bergström, D., Appelgren, H., & Heintz, F. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-73959-1_10
Computational Prometheus: How reinforcement was stolen, and how to steal it backPreprint
Jensen, G., & Freestone, D. M. (2018) · doi:10.31234/osf.io/4mtns
Beloften en teleurstellingen van artificiële intelligentie voor wetenschappelijke ontdekkingen
Heeffer, A. (2021) · Algemeen Nederlands Tijdschrift voor Wijsbegeerte · doi:10.5117/antw2021.1.003.heef
Brain–minds: What’s the best metaphor?
Stein, D. J. (2021) · Elsevier eBooks · doi:10.1016/b978-0-323-90239-7.00005-5
Nicholas, G. (2020)
NNCFR: Minimize Counterfactual Regret with Neural Networks.Preprint
Li, H., Wang, X., Guo, Z., Zhang, J., & Qi, S. (2021) · arXiv (Cornell University)
Unbiased Self-PlayPreprint
Ohsawa, S. (2021) · arXiv (Cornell University)
Why HAL 9000 is not the future of intelligence analysis
Pili, G. (2021) · The Journal of Intelligence Conflict and Warfare · doi:10.21810/jicw.v4i1.2566
“…are useful in the presence of a large amount of data to be processed and it is irrelevant whether those data are machines’ creations themselves (such as the case of AlphaZero programs – see Silver et al., 2018; Kasparov, 2018; Modern War Institute, 2018; Pili, 2019a), or are given to them as input.”
BackgroundTowards Learning to Play Piano with Dexterous Hands and TouchPreprint
Xu, H., Luo, Y., Wang, S., et al. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.02040
“Reinforcement Learning (RL) has been proved to outperform humans in games such as Atari [1], Go [2] and achieve impressive results on continuous control tasks [3].”
BackgroundTwo-pulse switching scheme and reinforcement learning for energy efficient SOT-MRAM simulations
de Orio, R. L., Ender, J., Fiorentini, S., et al. (2021) · Solid-State Electronics · doi:10.1016/j.sse.2021.108075
“First RL breakthroughs were achieved using games like chess or Go [26], and these types of algorithms have also successfully been applied in physics, e.”
MethodPlanning Spatial Networks.Preprint
Darvariu, V.-A., Hailes, S. M. V., & Musolesi, M. (2021) · arXiv (Cornell University)
内野, 明. (2020) · Medical Entomology and Zoology · doi:10.20627/jsimconf.80.0_59
Towards Utilitarian Combinatorial Assignment with Deep Neural Networks and Heuristic AlgorithmsPreprint
Präntare, F., Tiger, M., Bergström, D., et al. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2107.00317
Limited depth bandit-based strategy for Monte Carlo planning in continuous action spacesPreprint
Quinteiro, R., Melo, F. S., & Santos, P. A. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.15594 · Cited by 1
Expert Q-learning: Deep Q-learning With State Values From Expert ExamplesPreprint
Li, M., Yazidi, A., Goodwin, M., & Engelstad, P. E. (2021) · arXiv (Cornell University)
Width-based Lookaheads with Learnt Base Policies and Heuristics Over the Atari-2600 BenchmarkPreprint
O’Toole, S., Lipovetzky, N., Ramírez, M., & Pearce, A. R. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.12151
“AlphaZero is a planning and learning agent that achieved state-of-the-art performance on the games of Go, Chess and Shogi [15].”
BackgroundResonance for analog recurrent neural networkPreprint
Qu, Y., Zhou, M. X., Khoram, E., et al. (2021) · Research Square · doi:10.21203/rs.3.rs-480598/v1
Dube, S. (2021) · Springer eBooks · doi:10.1007/978-3-030-68624-6_7
Robuste Lernverfahren: Einblicke und Ausblicke
Markert, K., & Böttinger, K. (2021) · Datenschutz und Datensicherheit - DuD · doi:10.1007/s11623-021-1471-9
A Comparison of Contextual and Non-Contextual Preference Ranking for Set Addition ProblemsPreprint
Bertram, T., Fürnkranz, J., & Müller, M. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2107.04438
“This is the approach used in Tesauro’s TDGammon (Tesauro, 1995), in DeepMind’s AlphaZero (Silver et al., 2018), and in many similar approaches.”
MethodArtificial Intelligence: Opportunity or Risk?
Adams, J. C., & Engel, J. (2021) · Springer eBooks · doi:10.1007/978-3-030-59075-8_9
Deep-learning applications to the multi-objective optimisation of IACT array layouts.
Fraga, B., de Almeida, U. B., & de Bom, C. (2021) · International Cosmic Ray Conference · doi:10.22323/1.395.0908
Position: Who Gets to Harness (X)AI? For Billion-Dollar Organizations Only.
Dodge, J. E. (2021) · IUI Workshops
An Evolutionary Approach to Combinatorial Gameplaying Using Extended Classifier Systems
Oberoi, K., Tandon, S., Das, A., & Aggarwal, S. (2021) · Lecture notes in electrical engineering · doi:10.1007/978-981-16-3067-5_54
Inverse Problem of Nonlinear Schrödinger Equation as Learning of Convolutional Neural NetworkPreprint
Wang, Y., & Li, Z. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2107.08593 · Cited by 4
“In the past ten years, it came back as deep learning and achieved great success in applications such as computer vision [16, 20, 25], machine translation [10] and game playing [48].”
BackgroundAutomated Societal Option Maximization
Füllsack, M. (2021) · Constructivist Foundations
Modelling Recognition in Human Puzzle Solving
Prystawski, B., Gelpí, R., Lucas, C., & Buchsbaum, D. (2021) · Proceedings of the Annual Meeting of the Cognitive Science Society · Cited by 1
Schulze, S., Leuschner, J., & King, E. J. (2021) · Signals · doi:10.3390/signals2040039 · Cited by 3
“While reinforcement learning has progressed towards actor-critic methods [cf. 2] and, famously, the AlphaGo Zero [31], AlphaZero [32], and MuZero [33] algorithms based on Monte Carlo tree search (MCTS), we stay relatively close to the original approach, but we extend the formulation by adding deterministic values, com…”
MethodEffects of sampling and horizon in predictive reinforcement learningPreprint
Osinenko, P., & Dobriborsci, D. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2108.04802
Subgoal Search For Complex Reasoning TasksPreprint
Czechowski, K., Odrzygóźdź, T., Zbysiński, M., Zawalski, M., Olejnik, K., Wu, Y., Kuciński, Ł., & Miłoś, P. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2108.11204 · Cited by 29
“Among notable successful examples of this approach are Alpha Zero [42], or solving Rubik’s cube using a learned heuristic function to guide the A ∗ algorithm (see [1]).”
AI (Artificial Intelligent) - A Perspective from VisualizerPreprint
Yapp, A. (2018) · SSRN Electronic Journal · doi:10.2139/ssrn.3165329
A Novel Methodology for Measuring the Abstraction Capabilities of Image Recognition Algorithms
Hudáky, M. G., Lehotay–Kéry, P., & Kiss, A. (2021) · Journal of Imaging · doi:10.3390/jimaging7080152
“Silver et al. tested it on chess, shogi, and go, but others thought the algorithm many more games since then [5,6].”
BackgroundLearning to Shape Rewards using a Game of Switching Controls.Preprint
Mguni, D., Wang, J., Jafferjee, T., Perez-Nieves, N., Song, W., Yang, Y., Tong, F., Chen, H., Zhu, J., Du, Y., et al. (2021) · arXiv (Cornell University)
Optimizing Quantum Variational Circuits with Deep Reinforcement LearningPreprint
Lockwood, O. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2109.03188 · Cited by 16
“Similar to other RL encoding schemes [12], each sheet/plane in this array represents information about a certain feature.”
BackgroundResultDeep reinforcement learning applied to power control in D2D communications
Pires, L. B. (2021) · RIUnB Institutional Repository (University of Brasília) · doi:10.26512/2021.06.d.42018
Collaborating with an AI Swarm in an Adversarial Game
Matsunami, N., & Karakama, S. (2021) · Transactions of the Japanese Society for Artificial Intelligence · doi:10.1527/tjsai.36-5_g-l45
Scheduling in Parallel Finite Buffer Systems: Optimal Decisions under Delayed Feedback.Preprint
Tahir, A., Alt, B., Rizk, A., & Koeppl, H. (2021) · arXiv (Cornell University)
Post-Selections in AI and How to Avoid ThemPreprint
Weng, J. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.13233
“In contrast to Post-Selections likely used by [21], [14], [19], [79], [23], [66], [24], [25], [27], [76], [29], [31] including probably AlphaGo [26], AlphaGo Zero [28], AlphaZero [85], AlphaFold [30] and MuZero [86] and many others, we define and reported developmental errors that includes all errors occurred through…”
Trust, but Verify: Alleviating Pessimistic Errors in Model-Based Exploration
Czechowski, K., Odrzygóźdź, T., Izworski, M., Zbysiński, M., Kuciński, Ł., & Miłoś, P. (2021) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn52387.2021.9534062
Planning and Learning using Adaptive Entropy Tree Search
Kozakowski, P., Pacek, M., & Miłoś, P. (2022) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn55064.2022.9892556 · Cited by 6
“AlphaZero [4] augments vanilla MCTS with neural network heuristics trained in a self-improvement loop.”
BackgroundMethodInfluentialCut the CARP: Fishing for zero-shot story evaluationPreprint
Matiana, S., Smith, Jr., Teehan, R., Castricato, L., Biderman, S., Gao, L., & Frazier, S. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.03111 · Cited by 9
“In other domains such as image generation (Karras et al., 2019; Patashnik et al., 2021; Galanos et al., 2021), strategic game-playing (Silver et al., 2018; 2016), and planning (Karkus et al., 2017; Fan et al., 2019), powerful models for evaluating objects of interest have lead to more powerful…”
BackgroundPPL-MCTS: Constrained Textual Generation Through Discriminator-Guided MCTS DecodingPreprint
Chaffin, A., Claveau, V., & Ewa Kijak, C. (2021) · arXiv (Cornell University)
Martens, J., Ballard, A., Desjardins, G., Świrszcz, G., Dalibard, V., Sohl‐Dickstein, J., & Schoenholz, S. S. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.01765 · Cited by 67
Learning-Based Path Planning for Long-Range Autonomous Valet Parking.Preprint
Khalid, M., Wang, L., Wang, K., Pan, C., Aslam, N., & Cao, Y. (2021) · arXiv (Cornell University)
No-Press Diplomacy from ScratchPreprint
Bakhtin, A., Wu, D., Lerer, A. K., & Brown, N. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.02924 · Cited by 49
“We also use larger values of (cid:15) on the first two turns to quickly introduce diversity from the opening state, similar to the use of larger early exploration settings in other work, for example AlphaZero [29, 28].”
Self-Supervision is All You Need for Solving Rubik's Cube
Takano, K. (2021) · Transactions on machine learning research · doi:10.48550/arxiv.2106.03157 · Cited by 7
“Therefore, the proposed method cannot apply to multiplayer games like chess, in which no goal state is available in advance, whereas reinforcement learning is an appropriate and effective approach (Silver et al., 2018).”
MethodArslan, D., & Eryiğit, G. (2021) · International Conference on Innovations in Intelligent Systems and Applications (INISTA) · doi:10.1109/inista52262.2021.9548636 · Cited by 1
Automating Network Operation Centers with Superhuman PerformancePreprint
Altamimi, S., Altamimi, B., Shirmohammadi, S., & Cote, D. (2021) · doi:10.36227/techrxiv.15176175.v1 · Cited by 1
“This great promise of AI has been demonstrated for games already [4],”
BackgroundQuantification of Joint Redundancy considering Dynamic Feasibility using Deep Reinforcement Learning
Chai, J., & Hayashibe, M. (2021) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra48506.2021.9561048
“…(DL) and deep reinforcement learning (DRL) have achieved extraordinary breakthroughs in recent years, ranging from playing Atari video games [8] and board games such as Go and chess [9] to identifying robotic dynamics models [10] and various other applications in the robotics field [11][12].”
BackgroundLearning to Shape Rewards using a Game of Two PartnersPreprint
Mguni, D., Jafferjee, T., Wang, J., et al. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.09159
Modular Reinforcement Learning Framework for Learners and Educators
Versaw, R., Schultz, S., Lu, K., et al. (2021) · doi:10.1145/3472538.3472583
“AlphaGo and its successors [20] have achieved worldwide fame.”
BackgroundMultitask Adaptation by Retrospective Exploration with Learned World ModelsPreprint
Zholus, A., & Panov, A. I. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2110.13241
“Several options were proposed for using the environment model in the agent learning process: to generate additional “virtual" agent steps in the environment [17], to simulate full episodes to more accurately approximate the Q-value [18, 14], and to directly learn the policy only on “imaginary" trajectories [19, 15].”
MethodHumanity in the Era of Autonomous Human–machine Teams
Chen, S. (2021) · Springer eBooks · doi:10.1007/978-3-030-77283-3_15
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.