2019 Evolutionary Algorithms Review

Explainability is a critical gap: The paper identifies "the ability to explain and repeat the search process" as a key responsibility area for evolutionary algorithms, suggesting that many current systems cannot adequately show users how they arrived at their solutions—a significant concern for accountability in AI deployment.

Bias management requires deliberate design: The review highlights "the ability to manage algorithm bias due to data or user design" as a distinct responsibility dimension, indicating that bias in evolutionary algorithms can stem from both the data used and the choices made by developers, requiring intentional mitigation strategies.

Environmental controls matter for safety: The paper emphasizes the importance of "the ability to manage the control of the environment with limiters," suggesting that responsible evolutionary algorithms need built-in safeguards to prevent unintended or harmful outcomes when deployed in real-world systems.

Causality understanding is underdeveloped: The taxonomy identifies "the ability to understand input and output causality within a solution" as a responsibility area, implying that current evolutionary algorithms often cannot clearly demonstrate why specific inputs led to specific outputs—making it difficult to validate their decisions.

Corrective measures are essential: The review stresses "the ability to add corrective measures" as a key responsibility factor, indicating that responsible AI systems need mechanisms to identify and fix problems after deployment, rather than relying solely on pre-deployment testing.