EVOTER: Evolution of Transparent Explainable Rule-sets

Transparent AI can match black-box performance: EVOTER demonstrates that explainable AI systems using simple logical rules can achieve comparable accuracy to traditional black-box models across multiple domains, challenging the assumption that transparency requires sacrificing performance.

Rule-based systems make hidden biases visible: By evolving interpretable rule-sets, EVOTER can expose biases embedded in training data that would remain hidden in black-box systems, enabling stakeholders to identify and address fairness issues before deployment.

Rules can be directly edited for bias correction: Unlike post-hoc explanation methods that only interpret existing models, EVOTER's transparent rules can be manually modified to remove discovered biases and add domain-specific constraints, giving humans direct control over AI behavior.

Explainability from the start is preferable to explanation after the fact: The paper advocates building transparency into AI systems during development rather than attempting to interpret black boxes after training, suggesting this approach better serves domains with strict trustworthiness requirements like healthcare, finance, or criminal justice.

Interpretable AI provides domain insights: Beyond meeting compliance requirements, EVOTER's rule-sets can reveal meaningful patterns and relationships in data, potentially advancing human understanding of the domains they model.