Science Seminar Series: What Does Playing Games Tell Us About AI? with Mark Nelson, American University

Games have played a (perhaps surprisingly) large role in artificial intelligence (AI) research, driving development of new algorithms and research areas. For decades, the canonical game was Chess. Human-level chess play eluded several generations of researchers, from AI's beginnings as a field in the 1950s until a computer beat grandmaster Garry Kasparov in 1997. After that, focus shifted to other games - Go, robot soccer, Atari, Starcraft - with a number of successes in the past 10 years. Some successes spilled beyond games, with Google DeepMind adapting the human-level AlphaGo player into AlphaFold, a system for predicting protein structure. What have we learned from this about artificial intelligence, and has playing games been a good way of learning it? This talk will dig into the history of why games intrigued AI researchers from the start, and what the successes and failures so far tell us (and don't) about computer decision-making and intelligence.