In the social good side of the tech world (hashtag nptech), I hear people and institutions talk about the necessity of having “a human in the loop” when working with AI. This usually means that a responsible person reviews output from an LLM before it gets out in the wild. This is a low standard, but it’s not wrong. I hope to see more discussion of what having a human in the loop means for agentic AI, which is generally sold to us as a way to reduce human labor (for better or worse).
But let’s be serious. Humans need to be more than “in the loop.” Humans need to be held accountable for the results of their actions. Software does not build itself. Agents do not tell themselves when to run. When we talk about these tools as if they “went rogue” and “hacked into” things, we are removing the humans that built the tool, configured the sandbox, and gave the instructions.
AI is not people (nor are corporations but that’s another rant). As written in a 1979 IBM training manual, “A computer can never be held accountable, therefore a computer must never make a management decision.”
Software can’t be bound by the law, but its creators can and they must.