AWS for Software Companies Podcast Ep139: Human-in-the-Loop by Design: Building AI Systems Responsibly
Sep 1, 2025
Join Oliver Sharp, Chief AI Officer at Highspot; Umberto Milletti, R&D Officer at Demandbase; and Kayvan Alikhani, Global Head at Archer, as they dive into the future of AI governance. They discuss the importance of keeping humans in the loop and tackling data privacy challenges. Expect insights on managing compliance across different regions, reducing bias in AI through detailed assessments, and prioritizing expert review for accountability. With a focus on building transparent AI systems, this panel sheds light on responsible AI development.
AI Snips
Chapters
Transcript
Episode notes
Isolate Customer Data For Training
- Keep customer first-party data isolated in tenant-specific models to avoid unintended training leaks.
- Use opt-in second-party sharing and public third-party sources for broader training while preserving confidentiality.
Geographic Complexity Demands Precision
- AI compliance requirements vary widely by geography, making a one-size-fits-all policy ineffective.
- Vendors must be crisp on regional rules and advise non-technical customers about local constraints.
Use Granular Behavioral Questions
- Replace generic performance prompts with granular, behavior-focused questions using the SBI framework to reduce bias.
- Ask for timestamped examples so feedback is objective, actionable, and verifiable by managers.