A human dignity metric to evaluate AI models

Official URL: http://dx.doi.org/10.26650/BS/AH11.2026.002-6.14

The general moral concern regarding AI models refers to statistical unfairness (or discrimination) these systems impose. Statistical fairness demands that individuals from socially disadvantaged groups receive equal consideration from AI systems as other individuals. Moreover, it is widely accepted that we must also consider avoiding inaccurate results. However, statistical fairness and accuracy cannot be achieved simultaneously. This result leaves us with insufficient tools to pick the morally best AI model, even among a reduced set of models. Therefore, we need a further moral evaluation method (against using fairness as the only criterion) to assess AI models' moral performances to pick the morally superior one(s). This article investigates the extent to which human dignity is a suitable moral criterion to assess these models further. As a result, the benefits and limitations of using human dignity as an assessment criterion of AI models are laid out.