technology & power

摘要

"Human-in-the-loop" (HITL) is a phrase that shows up frequently in discus-sions regarding AI. One topic that draws attention is the potential displacement of jobs due to increased automation by intelligent machines. Another is the possibility of serious mistakes resulting from using AI systems for critical decision making, such as parole sentencing or the amount paid by the state Med-icaid program to individuals with developmental or intellectual disabilities.1 Placing humans in the loop is often suggested as a remedy for such cases. What exactly does HITL mean, and why is it considered to be a solution for these problems? A HITL AI system requires confirmation by a human before executing an action. This makes such a system much less autonomous than a "hu-man-on-the-loop" or a "human-off-the-loop" AI system. A human-on-the-loop AI system acts automatically unless a human supervisor overrides the decision. A human-off-the-loop AI system is completely autonomous. It does not require any confirmation before executing an action, and that action cannot be aborted once the AI system activates it.