Exposing Assumptions in AI Benchmarks through Cognitive Modelling
This work addresses the need for more rigorous and transparent evaluation in AI, particularly for researchers developing benchmarks, by promoting critical examination of assessment foundations.
The paper tackles the problem of vague and poorly validated cultural AI benchmarks by proposing to expose implicit assumptions using explicit cognitive models, specifically Structural Equation Models, to theoretically ground benchmark construction and guide dataset development for improved construct measurement.
Cultural AI benchmarks often rely on implicit assumptions about measured constructs, leading to vague formulations with poor validity and unclear interrelations. We propose exposing these assumptions using explicit cognitive models formulated as Structural Equation Models. Using cross-lingual alignment transfer as an example, we show how this approach can answer key research questions and identify missing datasets. This framework grounds benchmark construction theoretically and guides dataset development to improve construct measurement. By embracing transparency, we move towards more rigorous, cumulative AI evaluation science, challenging researchers to critically examine their assessment foundations.