Posted 1/25/2024, 5:00:00 PM
Proposal for Built-In AI Governance Chips to Prevent Misuse
- Proposal to encode rules governing AI training and use directly into silicon chips to prevent misuse
- Could constrain capabilities based on computing power limits enforced via hardware
- Precedent in using verification tech like seismometers to monitor nuclear weapons proliferation
- Proof-of-concept shows how chip security features could restrict unauthorized AI model use
- Technical and political challenges but government interest in governable AI chips