科技人文

机器下棋的历史与启示——从“深蓝”到AlphaZero

  • 薛永红 ,
  • 王洪鹏
展开
  • 1. 华北科技学院理学院, 北京 101601;
    2. 北京师范大学哲学学院, 北京 100875;
    3. 中国科学技术馆, 北京 100012
薛永红,副教授,研究方向为科学思想史与科学社会史,电子信箱:aristotle@ncist.edu.cn

收稿日期: 2019-07-05

  修回日期: 2019-08-23

  网络出版日期: 2019-10-19

基金资助

2018年度国家社会科学基金重点项目(18AZX008);中央高校基本科研业务费项目(3142018057)

Brief history and enlightenment of machine chess: From “Deep Blue” to AlphaZero

  • XUE Yonghong ,
  • WANG Hongpeng
Expand
  • 1. College of Science, North China University of Science and Technology, Beijing 101601, China;
    2. College of Philosophy, Beijing Normal University, Beijing 100875, China;
    3. China Science and Technology Museum, Beijing 100012, China

Received date: 2019-07-05

  Revised date: 2019-08-23

  Online published: 2019-10-19

摘要

以历史为线索,从设计思路和技术特征两个方面对“深蓝”和AlphaGo进行了梳理和概括。“深蓝”依赖人类在国际象棋领域的经验,借助强大的算力与算法实现了对人类的超越;20年后的AlphaGo,虽然最初的版本也是利用人类经验而获得成功的,但是它的不断进化却揭示了一个重要事实:人类经验具有局限性。放弃人类经验、完全采用机器自对弈经验的AlphaZero,不但具有最强的围棋对弈能力,而且同时具备国际象棋和日本将棋的最高棋力,3种最强技能集于一身。机器下棋的这一历史线索揭示了在棋类游戏中,囿于人类自身认知能力的局限,人类几千年积累下来的经验较之于机器在短期内所形成的“经验”已不占优势。在巨大的算力和不断完善的算法的支撑下,借助于机器自身“经验”,机器可以做得比人类更好。未来,“放弃人类经验,依靠自身经验”的机器将有可能在更为复杂的领域取得突破性进展。

本文引用格式

薛永红 , 王洪鹏 . 机器下棋的历史与启示——从“深蓝”到AlphaZero[J]. 科技导报, 2019 , 37(19) : 87 -96 . DOI: 10.3981/j.issn.1000-7857.2019.19.012

Abstract

Taking the historical evolution as the main line, this paper combs and summarizes "Deep Blue" and AlphaGo from the aspects of design philosophy and technical features. Relied on human experience of chess, "Deep Blue" achieved transcendence with humans by means of computational power and algorithms. Twenty years later, AlphaGo, although its original version was also successful by using human experience, and its evolution revealed an important fact that human experience has its limitation. AlphaZero, which gives up human experience and adopts machine self-playing experience, convincingly defeated a world champion program in the games of chess and shogi (Japanese chess) as well as Go. It is clear that in chess games, limited by human cognition ability, experience accumulated by human beings for thousands of years is no longer superior to the "experience" formed by machines in a short term. Machines can do better than humans with the help of their own "experience", supported by enormous computing power and ever-improving algorithms. In the future, machines that "give up human experience and rely on their own experience" will likely make breakthroughs in more complex areas.

参考文献

[1] Shannon C E. Programming a computer for playing chess[J]. Philosophical Magazine, 1950, 41(314):256-275.
[2] 尼克. 人工智能简史[M]. 北京:人民邮电出版社, 2017. Ni Ke. A Brief History of Artificial Intelligence[M]. Beijing:Posts & Telecom Press, 2017.
[3] Newborn M. 旷世之战——IBM深蓝夺冠之路[M]. 邵谦谦, 译. 北京:清华大学出版社, 2004. Newborn M. Deep Blue-An artificial intelligence milestone[M]. Shao Qianqian, trans. Beijing:Tsinghua University, 2007.
[4] Hsu F H. IBM's Deep Blue chess grandmaster chips[J]. IEEE Computer Society Press, 1999, 19(2):70-81.
[5] 吴岸城. 神经网络与深度学习[M]. 北京:电子工业出版社, 2016. Wu Ancheng. Neural network and deep learning[M]. Beijing:Publishing House of Electronics Industry, 2016.
[6] Silver D, H A, Maddison C J, et al. Mastering the game of Go with deep neural networks and tree search[J]. Nature, 2016, 529(7587):484-489.
[7] Silver D, Schrittwieser J, Simonyan K, et al. Mastering the game of Go without human knowledge[J]. Nature, 2017, 550(7676):354-359.
[8] 李开复, 王咏刚. 人工智能[M]. 北京:文化发展出版社, 2017. Li Kaifu, Wang Yonggang. Artificial Intelligence[M]. Beijing:Cultural Development Press, 2017.
[9] 吴军. 智能时代:大数据与智能革命重新定义未来[M].北京:中信出版集团, 2016. Wu Jun. The age of intelligence:Big data and the intelligent revolution redefine the future[M]. Beijing:China CITIC Press, 2016.
[10] Hinton G E, Osindero S, Teh Y. A fast learning algorithm for deep belief nets[J]. Neural computation, 2006, 18(7):1527-1554.
[11] 陶九阳, 吴琳, 胡晓峰. AlphaGo技术原理分析及人工智能军事应用展望[J]. 指挥与控制学报, 2016, 2(2):114-120. Tao Jiuyang, Wu Lin, Hu Xiaofeng. Principle analysis on AlphaGo and perspective in milltary[J]. Application of Artificial Intelligence, 2016, 2(2):114-120.
[12] Silver D, H A, Maddison C J, et al. Mastering the game of Go with deep neural networks and tree search[J]. Nature, 2016, 529(7587):484-489.
[13] Silver D, Schrittwieser J, Simonyan K, et al. Mastering the game of Go without human knowledge[J]. Nature, 2017, 550(7676):354-359.
[14] Silver D, Hubert T, et al. A general reinforcement learning algorithm that masters chess, shogi, and Go through self-play[J]. Science, 2018, 362(6419):1087-1118.
[15] AlphaZero:Shedding new light on the grand games of chess, shogi and Go[EB/OL].[2019-07-05]. https://deepmind.com/blog/alphazero-shedding-new-light-grand-games-chess-shogi-and-go.
[16] Sterken C, Manfroid J. Astronomical photometry:A guide[M]. Springer Science & Business Media, 1992.
[17] AlphaGo之父:关于围棋, 人类3000年来犯了一个错[EB/OL].[2019-07-05]. https://www.thepaper.cn/newsDetail_forward_1660773. Father of AlphaGo:A mistake about go humans made for 3000 years[EB/OL].[2019-07-05]. https://www.thepaper.cn/newsDetail_forward_1660773.
[18] 董春雨, 薛永红. 机器认识论何以可能[J]. 自然辩证法研究, 2019, 35(8):3-10. Dong Chunyu, Xue Yonghong. Why is machine epistemology possible?[J]. Studies in Dialectics of Nature, 2019, 35(8):3-10.
[19] 马丁·戴维斯. 逻辑的引擎[M]. 张卜天, 译. 长沙:湖南科学技术出版社, 2001. Martin Davis. Engines of Logic[M]. Zhang Butian, trans. Changsha:Hunan Science and Technology Press, 2001.
[20] Alvarado R, Humphreys P. Big data, thick mediation, and representational opacity[J]. New Literary History, 2017, 48(4):729-749.
[21] 哈萨比斯在剑桥大学的演讲"超越人类认知的极限"[EB/OL].[2019-07-05]. http://scholarsupdate.hi2net.com/news.asp?NewsID=22161. Demis Hassabis's talk at Cambridge University about "Exploring the frontiers of knowledge"[EB/OL].[2019-07-05]. http://scholarsupdate.hi2net.com/news.asp?NewsID=22161.
文章导航

/