AI Job Matching: Framework Aims to Spot and Limit Bias

Researchers at an unspecified institution propose a two-stage framework to govern bias in AI systems that match workers to jobs. The system first assesses how skills are extracted from candidates, particularly when using chatbot interfaces, identifying potential biases and categorizing them as either requiring immediate correction or simply logging for future consideration.

Second, the framework uses a recommendation system where candidates, companies, and regulators each create job rankings. These are then combined using social choice methods to produce a single, transparent recommendation.

The framework aligns with guidelines like the Fraunhofer AI Assessment Catalog, utilizing distributional auditing and counterfactual testing to create a bias inventory. Predefined fairness thresholds would trigger adjustments to recommendations, while minor deviations are flagged as bias reports. The researchers hope this approach will increase fairness and accountability in AI-driven hiring processes.