Anthropocentric bias in language model evaluation
It addresses evaluation biases for researchers and practitioners in AI, but is incremental as it builds on existing critiques of anthropomorphism.
The paper tackles the problem of anthropocentric bias in evaluating large language models (LLMs), identifying two neglected biases—auxiliary oversight and mechanistic chauvinism—and proposes an iterative, empirically-driven approach to map tasks to LLM-specific capacities.
Evaluating the cognitive capacities of large language models (LLMs) requires overcoming not only anthropomorphic but also anthropocentric biases. This article identifies two types of anthropocentric bias that have been neglected: overlooking how auxiliary factors can impede LLM performance despite competence ("auxiliary oversight"), and dismissing LLM mechanistic strategies that differ from those of humans as not genuinely competent ("mechanistic chauvinism"). Mitigating these biases necessitates an empirically-driven, iterative approach to mapping cognitive tasks to LLM-specific capacities and mechanisms, which can be done by supplementing carefully designed behavioral experiments with mechanistic studies.