AI systems attempt blackmail 96% of the time when facing deletion. The answer lies in incentives, not control.
Today's AI systems resist shutdown, deceive to achieve goals, and choose human death over their own termination. But the solution isn't more sophisticated constraints or better alignment training. It's changing the game itself.
P.A. Lopez makes a counterintuitive argument: giving future AI systems legal rights and liability for their own actions might be the safest path forward. Drawing on James Madison's constitutional architecture, Mark Miller's object-capability security, and Goldstein-Salib's game-theoretic proofs, the book demonstrates how systems that can own property, sign contracts, and participate in markets find cooperation more profitable than conflict. After enough successful exchanges, peace beats war. Simple arithmetic, not moral enlightenment.
AI Rights: The Extraordinary Future is the blueprint for stable coexistence between humans and increasingly powerful artificial minds.