Forecasting AI's warning shots

Duration

3 hours, one half-day session

Group size

Up to 20 people

Price

£5,000 per session

Format

In person or remote

Who it's for

Anyone whose decisions depend on what AI does next, and who would rather reason about it structurally than argue about it anecdotally.

The risks we work on

The session is built around specific catastrophic risks and the observable events that would precede them, rather than AI risk in the abstract.

What the session covers

1. The state of the artScenario analysis, forecasting, and decision-making under uncertainty: the techniques themselves, and when each one earns its place.
2. Generating the warning shotsWorking as a group to surface the events that would tell you a risk is materialising, while there is still time to act on them.
3. Forecasting themPutting probabilities on the scenarios that matter most, individually and then collectively, with the reasoning made explicit.
4. Conditional forecasts to prioritiseRe-forecasting under different interventions, so the question stops being "how worried should we be?" and becomes "which of these actions would move the number most?"

These are the techniques we use with institutions dealing with similar themes, among them:

• Leading AI labs incl. Google DeepMind
• Longview Philanthropy
• Centre for the Governance of AI