SUMMIT & RESEARCHSelect a time to jump to that part of the passage. Speaker changes are approximate.
Rachel Coldicutt explains why AI evaluation must continue during use, asking who is excluded when a tool fails and whether it works in each new context.
Attributed contributors: Rachel Coldicutt. “Research has to happen during deployment.” Responsible AI: Ethics, Governance and Practice. ECOSYSTEM Summit, Barcelona, 16 September 2026. Session time 18:13–20:49. https://cs-ecosystem.commonshare.workers.dev/moments/rachel-research-during-deployment
Analyst synthesis. The recording establishes that a claim was made. It does not independently establish the claimed effect.
A tool working in one setting does not establish effectiveness elsewhere; failures must be assessed by who is excluded.
Proposed mechanism: Context-specific evaluation and remediation constrain indiscriminate scaling.
Limit: Illustrative percentages are not study estimates.
Evaluation should continue during use as models and contexts change.
Proposed mechanism: Repeated review catches change after deployment.
Conditions: Resourced ongoing assessment.
Limit: Suggested quarterly cadence is not a tested universal standard.
Explore the linked codes above to compare related passages.