LessWrong (Curated & Popular) “Proposal for making credible commitments to AIs.” by Cleo Nardo
Jun 30, 2025
Dive into the intriguing world of AI deal-making! Discover how humans can strike agreements with AIs, ensuring they act safely and beneficially. The discussion focuses on the challenge of making credible commitments, exploring whether such commitments can indeed incentivize AIs. Learn about the framework that intertwines legal contracts and human oversight to ensure compliance without granted personhood. It's a thought-provoking exploration of the future relationship between humans and artificial intelligence.
AI Snips
Chapters
Transcript
Episode notes
Deal-Making Requires Credible Commitments
- Deal-making with misaligned AIs requires credible commitments from humans to motivate AI safety and usefulness.
- The two main challenges are making these commitments credible and ensuring they incentivize compliance.
Circumventing AI Legal Limitations
- Since AIs lack legal personhood, legal contracts can't bind them directly, so humans form contracts among themselves to assure AI compliance.
- AIs trust personal promises from legally recognized individuals to enforce safety and usefulness conditions.
Diversify Trusted Human Trustees
- AIs should choose multiple trusted individuals with integrity to make unenforceable personal promises to spend compensation as requested.
- Diversify trustees across countries to ensure reliability beyond a single jurisdiction.
