Episode 8
Authentic Intelligence: Context, Explainability, Bias, and Human-in-the-Loop
with Keshavan Seshadri
About This Episode
Keshavan Seshadri, a Senior ML Engineer at Prudential Financial, makes the case that intelligence without transparency is not intelligence at all. The conversation covers explainability requirements for AI systems, how bias creeps into models and what it takes to detect it, and why human-in-the-loop design is not a limitation yet a feature that strengthens AI deployments.
Key Insights
Explainability in AI is not a nice-to-have; it is a prerequisite for any deployment where decisions affect people's lives or livelihoods. When healthcare AI systems make recommendations that clinicians cannot understand, clinicians cannot defend them to patients, cannot override them when appropriate, and cannot improve them when they fail.
Bias in machine learning models is a systemic issue that requires ongoing monitoring, not a one-time audit. Bias emerges from training data, can amplify during deployment, and changes as the population using the system evolves. Organizations that treat bias detection as a launch criteria rather than continuous practice will continue to experience failures.
Human-in-the-loop approaches strengthen AI systems by introducing accountability at the points where models are most likely to fail. Rather than viewing human involvement as a necessary evil, effective organizations recognize it as a design feature that improves outcomes and maintains trustworthiness.
Topics Explored
The episode covers AI explainability techniques, algorithmic bias detection methods, human-in-the-loop AI system design, model transparency and accountability practices, responsible AI deployment frameworks, and the engineering practices that separate trustworthy AI from black-box automation. Discussion includes specific techniques for making models interpretable and strategies for ongoing bias monitoring.
About the Guest
Keshavan Seshadri is a Senior ML Engineer at Prudential Financial, where he works on building machine learning systems that are transparent, fair, and operationally reliable. His technical depth informs a practical perspective on responsible AI.
Questions This Episode Answers
Why does AI explainability matter in healthcare?
Clinicians and patients need to understand why an AI system made a particular recommendation before they can trust or act on it. Explainability is not just a technical feature; it is a requirement for clinical adoption, regulatory compliance, and patient consent. Black-box models that cannot explain their reasoning face justified resistance from healthcare professionals.
How do you detect and mitigate bias in AI models?
Bias detection requires systematic evaluation across demographic groups, clinical populations, and data sources throughout the model lifecycle. Mitigation is not a one-time fix but an ongoing process of monitoring, testing, and adjusting as populations and data distributions change. Responsible AI development treats bias detection as a core engineering practice.
What is the case for keeping humans in the loop with AI?
Human oversight ensures that AI recommendations are evaluated against clinical context, patient preferences, and situational factors that models cannot fully capture. The human-in-the-loop approach also provides a critical safety net for catching errors before they reach patients. Removing human oversight to gain efficiency creates risks that far outweigh the benefits.
"Authentic intelligence requires transparency. If a model cannot explain why it made a recommendation, clinicians have every right to reject it."
Related Episodes
Listen & Subscribe
Catch every episode of The Signal Room on your preferred platform.
Guest
Keshavan Seshadri
Senior ML Engineer at Prudential Financial
Never Miss an Episode
Get the AI Health Pulse newsletter with episode highlights and healthcare AI insights.
Subscribe to NewsletterSponsor
Interested in reaching healthcare AI leaders? Get in touch.