Regis A. James, PhD, has spent well over a decade building scalable, end-to-end AI systems in domains ranging from molecular diagnostics of genetic disease patients to electroretinogram hallmark extraction from mouse models to company-wide employee pairing for institutional knowledge transfer (mentor matching). In doing so, the contexts in which he has collaborated have presented numerous opportunities to develop and implement circumspect practices for effective due diligence towards reducing the risk and, therefore, maximizing the explainability of ultimate decisions. To demonstrate these principles, Regis will discuss how these considerations were applied in MAGNETRON AI, a recent patent-pending autonomous AI system that he built with colleagues, and highlight how these principles can be re-applied within other institutional contexts. 

Topics covered:

• Appropriately isolating sensitive data via governance controls
• Ensuring AI literacy among collaborators to maximize the predictability of AI system behavior
• Maximizing data quality to enable comprehensiveness, reproducibility, and high-throughput processing within AI systems
• Considering the consequences of autonomy granted to AI systems during the algorithmic development process
• The necessity of incorporating explainability into AI systems
• Achieving risk-mitigation consensus with colleagues