When Technology Has to Explain Itself

New regulation coming into force in parts of the world now requires certain AI systems to disclose plainly that a person is talking to a machine, or that what they're reading was generated by one. A decade ago nobody thought that rule would be necessary.

In a 2013 Oxford study, economists Carl Benedikt Frey and Michael A. Osborne estimated that 47% of US employment was in occupations at high risk of computerisation. That was an estimate of susceptibility, not a prediction that 47% of jobs would definitely disappear.

Most people do not experience technological change as a statistic. They experience it when a familiar task becomes automated, a job description changes or a skill they once considered secure starts to feel temporary.

It is now, because the alternative, quietly, had become the default. I read something this morning that made me put my cup down.

Not the content of it. The fact that I had to check, for a moment, whether a person or a system had written it.

The machines have to say so now, or at least that's the assumption we've started operating under, checking an interface's origin before checking what it said. I don't think that rule is only about the machines.

I think it's about how much we'd stopped asking the same question of ourselves, whether the voice guiding a decision was actually our own judgement or something borrowed and running on autopilot. I don't check every source that carefully.

That's probably the actual problem the new rule is trying to correct, not the machines themselves. The disclosure was never really about them.

That distinction feels worth holding onto for longer than the length of one news cycle about a new regulation. It's a small enough habit to build.

It's worth building before the next system comes along asking for the same blind trust. I'm trying to hold myself to the same disclosure the new law asks of the systems.

Is this actually what I think, or is it a suggestion I picked up somewhere and forgot to check.