Improving artificial intelligence with games

• 成果类型: Editorial Material
• 署名作者: Wurman, Peter R.; Stone, Peter; Spranger, Michael
• 署名单位: University of Texas System; University of Texas Austin
• 刊物名称: SCIENCE
• ISSN/ISSBN: 0036-10840
• DOI: 10.1126/science.adh8135
• 发表日期: 2023-07-14
• 页码: 147-148
• 关键词: level; go; ai

摘要

For thousands of years, humans have played board games as a pleasurable way to pass time and match wits. Games such as Senet were so prized as to be buried with other valuables for the afterlife in Egyptian tombs similar to 3500 BCE (1). Skill in strategic board games such as chess and Go came to represent a pinnacle of human intelligence, and thus it is no surprise that such board games have been viewed as yardsticks by which to measure machine intelligence (2). In recent years, computers have overtaken human skill in a wide array of strategy games. Although it may seem that the recent dominance of artificial intelligence (AI) in classic strategy games foreshadows the end of what games can teach us about AI, researchers are just beginning to tackle the substantially more complex and varied challenges posed by modern video games.