Human in the loop?

Will it always be possible to have a human in the loop when using AI?

I think it’s a very relevant question, as the human will be a severe bottleneck. The more interesting question is when we can let AI take decisions without human intervention.

If your indoor heating system is controlled with assistance  of AI, should you approve every change in temperature? Same if you let AI control charging of your electrical car during the night. Wake me up?

We need some guiding principles for how to control AI, but wording matters. This is why I would frame the control mechanism in four levels based on importance, not risk. Low, medium, high and critical.

Low importance examples are temperature control in the house and charging of the car. High and critical flows always include a human in the loop. Medium depends on confidence. How confident is the AI agent of the answer is the real question?

If the confidence is low, then you need to escalate to a human, or another agent. If the confidence is high, and the importance is low or medium, you can let AI work autonomously.

Caveat, your AI-agent scope must be limited, with clear boundaries.