Forecasting
Overview
Section titled “Overview”This section compiles forecasts about AI development trajectories, capability timelines, and safety outcomes. Tracking predictions over time helps calibrate expectations and identify who has good track records.
Forecast Categories
Section titled “Forecast Categories”When might transformative AI arrive?
- Expert surveys and estimates
- Compute-based projections
- Historical accuracy of past forecasts
Detailed timeline predictions:
- 2025-2030 scenarios
- 2030-2040 scenarios
- Post-2040 considerations
Key Forecast Sources
Section titled “Key Forecast Sources”| Source | Focus | Method | Track Record |
|---|---|---|---|
| Metaculus | AI milestones | Prediction market | Good calibration |
| Epoch AI | Compute trends | Trend extrapolation | Strong on hardware |
| Expert surveys | Timeline estimates | Elicitation | Variable |
| Superforecasters | Specific questions | Tournament forecasting | Best overall |
Current Consensus Ranges
Section titled “Current Consensus Ranges”| Question | Low | Median | High |
|---|---|---|---|
| P(AGI by 2030) | 10% | 25% | 50% |
| P(AGI by 2040) | 40% | 65% | 85% |
| P(Catastrophe | AGI) | 5% | 15% |
Estimates represent rough synthesis of public expert views; wide disagreement exists.
Forecasting Best Practices
Section titled “Forecasting Best Practices”- Track calibration - Did predictions come true at stated probabilities?
- Decompose questions - Break complex questions into more tractable components
- Update regularly - Revise forecasts as new information arrives
- Acknowledge uncertainty - Use ranges, not point estimates