Racing Intensity
Racing intensity measures the competitive pressure that drives AI developers to prioritize speed over safety. Even when actors prefer safe outcomes, the logic of competition creates a multipolar trap where rational individual behavior produces collectively dangerous results. Investing in safety while competitors cut corners means falling behind—and falling behind feels existential to labs and nations alike.
Evidence shows intensifying dynamics: ChatGPT’s 2022 launch triggered industry-wide acceleration, compressing safety evaluation timelines by 40-60%. Release cycles shortened from 18-24 months to 3-6 months. Safety budgets declined from 12% to 6% of R&D spending. The Racing Dynamics Impact Model quantifies these effects: racing reduces safety investment by 30-60% and increases alignment failure probability by 2-5x.
| Metric | Score | Notes |
|---|---|---|
| Changeability | 50 | Moderate potential through coordination; defection remains possible |
| X-risk Impact | 65 | High direct contribution through safety corner-cutting |
| Trajectory Impact | 50 | Affects which actors lead and what norms get established |
| Uncertainty | 45 | Well-documented and empirically observable |
Related Content
Section titled “Related Content”Risks:
Responses:
Models:
Key Debates:
- Are we actually in a race, or do actors perceive one where none exists?
- Does racing actually reduce safety margins, or can safety work proceed in parallel?
- Can meaningful racing slowdowns be coordinated, or is defection inevitable?