Nice one! I do think AI agent law can be fairly common sense. Inherent rights and personhood inadvisable, premature. Liability on some balance of principals (user, model developer, scaffold and tooling devs) depending on reasonable context.
To facilitate this in common (or any other) law, I'd imagine AI systems need to
- be identifiable (includes model, scaffold, instance)
- record the scope of their authority as the basis for any given activity
- (probably) record activity traces, or otherwise associate activity with instance identity
Counterparties to agents should handshake to check identity and authority.
This is all technically nearly trivial, but presumably faffy to get consensus and initial details on.
Running agents beyond some (to determine) threshold of capability and authority *without* following that basic protocol should be discouraged or prohibited - perhaps a trickier problem.
Dean, interested to know if you'd suggest different generic basic foundations, or how you'd add nuance to this. How has your thinking on agents and law updated since this article?
Nice one! I do think AI agent law can be fairly common sense. Inherent rights and personhood inadvisable, premature. Liability on some balance of principals (user, model developer, scaffold and tooling devs) depending on reasonable context.
To facilitate this in common (or any other) law, I'd imagine AI systems need to
- be identifiable (includes model, scaffold, instance)
- record the scope of their authority as the basis for any given activity
- (probably) record activity traces, or otherwise associate activity with instance identity
Counterparties to agents should handshake to check identity and authority.
This is all technically nearly trivial, but presumably faffy to get consensus and initial details on.
Running agents beyond some (to determine) threshold of capability and authority *without* following that basic protocol should be discouraged or prohibited - perhaps a trickier problem.
Dean, interested to know if you'd suggest different generic basic foundations, or how you'd add nuance to this. How has your thinking on agents and law updated since this article?