It is easy to say that AI needs quantitative risk assessment, but it is not a machine in the same sense as a plane or a nuclear reactor. Those machines have tolerances which can be guaranteed to a confidence level. They have measurable qualities. Sure one can devise aggregate metrics, measures of reliability or trustworthiness in an ai, but that's like giving your uncle an honestly coefficient.
It is easy to say that AI needs quantitative risk assessment, but it is not a machine in the same sense as a plane or a nuclear reactor. Those machines have tolerances which can be guaranteed to a confidence level. They have measurable qualities. Sure one can devise aggregate metrics, measures of reliability or trustworthiness in an ai, but that's like giving your uncle an honestly coefficient.