I worry that legislation that attempts to regulate future AI systems with explicit thresholds set now will just get outdated or optimized against.
I think a better way would be to have an org like NIST be given permission to define it’s own thresholds and rules.
Furthermore, to fund the monitoring orgs, allow them to charge fees for evaluating “mandatory permits for deployment”.
I worry that legislation that attempts to regulate future AI systems with explicit thresholds set now will just get outdated or optimized against. I think a better way would be to have an org like NIST be given permission to define it’s own thresholds and rules. Furthermore, to fund the monitoring orgs, allow them to charge fees for evaluating “mandatory permits for deployment”.