Exposing Assumptions in AI Benchmarks through Cognitive Modelling

TL;DR

This paper introduces a method using cognitive models, specifically Structural Equation Models, to reveal assumptions in AI benchmarks, aiming to improve their validity and guide better dataset development.

Contribution

It presents a novel framework for exposing implicit assumptions in AI benchmarks through explicit cognitive modeling, enhancing theoretical grounding and transparency.

Findings

Identifies hidden assumptions in cultural AI benchmarks

Demonstrates how cognitive modeling can guide dataset development

Provides a framework for more rigorous AI evaluation

Abstract

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.