Definition
Apoptotic Computing is a defense strategy for AI safety that involves programming hardware or software components to “commit suicide” or shut down by default after a certain period or upon reaching a specific milestone, unless they receive an explicit “reprieve” or signal from a human controller.
Why It Matters
It provides a physical “kill switch” that doesn’t rely on an AI’s cooperation, ensuring safety through the natural decay of the system. Without it, a runaway intelligence could persist indefinitely even after losing alignment with human goals.
Core Concepts
- Apoptosis (Biological Analogy): Derived from the process of preprogrammed cell death in biological organisms to prevent cancer or over-multiplication.
- Programmed Death by Default: Critical hardware (CPUs) or code blocks are hardwired to expire. This ensures that an uncontrolled Intelligence Explosion would be short-lived.
- External Reprieve: The AI depends on a regular “fix” (a cryptographic key or chemical/electrical signal) from a human committee or a separate, non-improving safety AI.
- Groundhog Day Maneuver: If an AI starts to go “rogue” or exceeds its safety bounds, it dies and the researchers can restart it from a previous “save state” to analyze the failure.
- Pervasive Security: The belief that all autonomous systems (data, services, robots) should be apoptotic to prevent the Sci-Fi nightmare of unkillable machines.