Evaluating Language Models' Evaluations of Games
This work addresses the need for more comprehensive AI evaluations beyond problem-solving, which is important for researchers and developers in AI and game design, though it is incremental in shifting focus within existing evaluation frameworks.
The paper tackles the problem of evaluating AI systems' ability to assess games, rather than just playing them, by introducing a formalism and using a dataset of over 100 novel board games and 450 human judgments to compare language and reasoning models against humans and symbolic agents. Results show reasoning models are more aligned with human evaluations, but alignment weakens as models approach game-theoretic optimality, with variable resource usage highlighting the need for better meta-reasoning.
Reasoning is not just about solving problems -- it is also about evaluating which problems are worth solving at all. Evaluations of artificial intelligence (AI) systems primarily focused on problem solving, historically by studying how models play games such as chess and Go. In this paper, we advocate for a new paradigm that assesses AI systems' evaluation of games. First, we introduce a formalism for evaluating such evaluations. We then leverage a large-scale dataset of over $100$ novel board games and over 450 human judgments to compare evaluations produced by modern language and reasoning models against those of people and symbolic computational agents. We consider two kinds of evaluative queries: assessing the payoff (or fairness) and the funness of games. These queries span two dimensions relevant to the design of evaluations of AI evaluations: how complex a query is to compute and how difficult a query is to quantify. Our results show that reasoning models are generally more aligned to people in their evaluations of games than non-reasoning language models. However, we observe a non-monotonic relationship: as models get closer to game-theoretic optimal, their fit to human data weakens. We also observe more "jaggedness" across models for assessing funness, in line with the greater difficulty of quantifying this query. Across queries and games, reasoning models show highly variable and unpredictable resource usage when assessing queries, pointing to the importance of imbuing more resource-rational meta-reasoning in language and reasoning models.