EcosystemSUMMIT & RESEARCH
Evidence library / Responsible AI: Ethics, Governance and Practice
AI for development · SELECTED MOMENT · 1:16

Faith partly disagrees: adding safety to minimum viable evaluation can make assurance practical without requiring a costly RCT.

Likely speaker
Rachel ColdicuttFounder · Careful IndustriesLinkedIn ↗
Becky FaithSenior Research Fellow; Development Lecturer · FCDO; Institute of Development StudiesSpeaking in a personal capacityLinkedIn ↗
Unidentified speaker

Free short preview · Research has to happen during deployment

Speaker turns in this passage

Select a time to jump to that part of the passage. Speaker changes are approximate.

Rachel ColdicuttFounder · Careful IndustriesLinkedIn ↗
Unidentified speaker
Becky FaithSenior Research Fellow; Development Lecturer · FCDO; Institute of Development StudiesSpeaking in a personal capacityLinkedIn ↗
Unidentified speaker

Faith partly disagrees: adding safety to minimum viable evaluation can make assurance practical without requiring a costly RCT.

View citation

Attributed contributors: Rachel Coldicutt; Becky Faith. “Faith partly disagrees: adding safety to minimum viable evaluation can make assurance practical without requiring a costly RCT..” Responsible AI: Ethics, Governance and Practice. ECOSYSTEM Summit, Barcelona, 16 September 2026. Session time 35:15–36:32. https://cs-ecosystem.commonshare.workers.dev/evidence/E-75459b9a0a99

Interpretation and context.

Analyst synthesis. The recording establishes that a claim was made. It does not independently establish the claimed effect.

qualification

Faith partly disagrees: adding safety to minimum viable evaluation can make assurance practical without requiring a costly RCT.

Contributors
Rachel ColdicuttFounder · Careful IndustriesLinkedIn ↗
Becky FaithSenior Research Fellow; Development Lecturer · FCDO; Institute of Development StudiesSpeaking in a personal capacityLinkedIn ↗
Unidentified speaker

Proposed mechanism: Layer proportionate safety evaluation into an existing tool.

Conditions: Responsibility for safety and equity remains institutional.

Limit: Difference of emphasis with the preceding culture argument.

evaluationcapacityassurance learning

Where it connects.

Credible judgement requires methods and context appropriate to the question, with failure and missing voices visible.

The corpus contains studies, practitioner accounts, demonstrations, proposals and questions. These establish different things. Local knowledge and negative cases can reveal what a generic measure misses.

Follow the evidence →