Board oversight of AI works best when it focuses on accountability and evidence. These questions don't require technical knowledge; they require clear answers.
The questions
- Which AI systems are live, and who is accountable for each?
- What outcome was each expected to deliver, and what has been measured?
- What can each system do without human approval?
- What happened the last time one made a serious mistake?
- Which customer or employee data does each use, and where is it processed?
- Which vendors do we depend on, and what is the exit plan?
- How do we know the quality is not drifting?
- Which regulations apply and how are we evidencing compliance?
- What is the full run cost, including people's review time?
- What have we decided not to build, and why?
Vague answers are information. A programme that can't answer these simply and quickly has not yet got control of its own systems.