People use fast and flat simulation to reason about new games

NOMIS researcher Tom Griffiths and colleagues have found that people can rapidly assess a game they have never played and play it sensibly on their first try, using only a few quick, shallow mental simulations. The study, published in Nature, sheds light on how people reason their way through unfamiliar situations, and could inform the design of AI systems that work out not just how to solve a new problem, but whether it is worth thinking about at all.

Games have often been used to study how people and machines plan, reason and make decisions. Much of that research looks at highly skilled players or advanced AI systems. But in everyday life, people also have to flexibly navigate decision-making in situations they have never faced before.

To study how people can think reasonably about systems or problems for the first time, Griffiths and team ran large-scale behavioral studies with more than 1,000 participants and 121 two-player strategic board games, almost all unfamiliar to the people in the experiments. Participants rated games from a written description of the rules alone, played them for the first time against another newcomer, and predicted the moves other first-timers would make. The researchers showed that people are systematic and adaptively rational in how they play a game for the first time or evaluate a game before ever playing it — for example, how fair it seems or how fun it is likely to be.

The team developed a model called the Intuitive Gamer to explain this behavior. The model has two parts. The first is a simple game-playing system that looks only a very short distance ahead (flat or depth-limited) and uses general rules of thumb to choose its next move. The second is a fast reasoning system that estimates what might happen in the game by mentally running through a small number of possible play-throughs. The model matched both the moves people made and the judgments they formed before playing, suggesting that people can quickly build an understanding of a new situation by mentally testing a few simple possibilities.

The authors write, “Our work offers insights into how people rapidly evaluate, act and make suggestions when encountering novel problems, and could inform the design of more flexible and human-like artificial intelligence systems that can determine not just how to solve new tasks but also whether a task is worth thinking about at all.”

Read the Nature publication: People use fast and flat simulation to reason about new games