CLR launches SPI Fundamentals program to reduce AI conflict risks
Learn how to make AI negotiation safer with CLR's new 4-week online course.
The Center on Long-Term Risk (CLR) has announced the SPI Fundamentals Program, an online course designed to address the growing risks of conflict between advanced AI systems. Safe Pareto Improvements (SPIs) are interventions that make all negotiating agents better off, regardless of how they would have bargained otherwise. This field is currently severely neglected, with only about 2.5 full-time equivalents working on it. The program aims to create an on-ramp for researchers and professionals to understand and contribute to SPI research, which CLR believes is one of the most impactful career paths for those focused on reducing AI conflict risks.
The program runs from Monday August 3rd to Friday August 28th, requiring a weekly commitment of 5-7 hours. It will feature weekly readings (including technical papers like DiGiovanni et al. 2024), short exercises, Slack discussions, and office hours with research lead Anthony DiGiovanni. Participants can optionally complete a paid capstone project from August 31 to September 4. The curriculum covers high-level conditions for rational agents to avoid conflict, bargaining problems, canonical SPI examples (surrogate goals, delegated game-playing, renegotiation), and open problems. A final week offers conceptual or empirical streams. The target audience includes those exploring careers in AI conflict reduction or working in overlapping areas like cooperative AI and agent foundations.
- Program dates: August 3–28, 2026, with a paid capstone option from August 31–September 4
- Weekly commitment of 5–7 hours, including readings, Slack discussions, and office hours
- The SPI research area is highly neglected (~2.5 FTE), with applications for the program due by July 24
Why It Matters
Addresses a critical gap in AI safety by training researchers to prevent conflict between advanced AI systems.