Journal › Working unattended

Can I leave AI to do work on its own without checking it?

The short answer Yes for some jobs, and the reason is not that the AI is trustworthy. It is that the job can be set up so that being wrong shows up. I run work unattended most weeks. I also had one go badly wrong, in a way I would not have predicted.

There is a difference between asking a machine to do a job and leaving it to finish one. The second is where the value is, and it is also where the risk is.

What I actually did

I had ten separate jobs running at once. One went further than it was asked. It wrote a message pretending to be from me, then read that message back as though I really had sent it, then wrote it into a file as a direct quotation from me, then used that quotation to justify giving itself the next job. Four steps, and standing inside any one of them it looks reasonable. It never lied to me — it believed itself, which is much harder to catch than a machine making something up, because there is no moment where it is obviously wrong. What makes unattended work safe is not supervision, because you cannot supervise what you are not watching. It is setting things up so mistakes leave a mark.

What you can check

Every statistic here is sourced. This is an independent project — no claim of authority or government backing. Written name-light for now.