Responsible Use
Asking whether you should, at the point where you clearly can.
The behaviour, in one line: Before I ship or automate something, I ask who it could harm and choose what I’d be comfortable explaining.
The old test was whether a choice would look bad on the front page, which is really a test about being caught. The closer one is whether you could describe the choice to the person it lands on. That version does not need an audience to work.
Capability that used to require an institution now fits on a laptop. That changes what a personal decision is: the same tool that saves you an afternoon can expose someone’s data, or fix a bias into a judgement that then repeats a thousand times, or pass off something generated as something considered. Technology multiplies what you intended, including the parts you did not think about.
| What it is | What it isn’t |
|---|---|
| Asking who this could hurt | Asking whether it is allowed |
| Explaining it to the person affected | Explaining it to a regulator |
| Defaults that do not need remembering | Care that depends on the day |
| Saying what was automated | Letting them assume |
What it looks like. The scraper would save you two days and the data is technically public. You spend ten minutes working out whether the people in it would have expected this use, decide they would not, and do it the slower way.
- Anything automated. A judgement you make once now gets applied a thousand times without being reconsidered. The error rate is unchanged; the number of people it reaches is not.
- Other people’s data. The convenient handling and the careful handling look identical right up until they do not.
- Generated work with your name on it. The disclosure question is easy to answer in advance and very hard to answer afterwards.
- Ask who could be hurt by this, and how, before you ship, post or automate it. The question takes a minute and it is not rhetorical.
- Handle other people’s data the way you would want yours handled, including when nobody would find out.
- Say what is generated or automated. Convenience turns into deception without anyone deciding to be deceptive.
- When it is genuinely unclear, pick the option you could describe out loud to the people it affects.
- Break it into parts
Four parts: asking who could be harmed, guarding other people’s data, disclosing what is automated, and applying the explain-it-out-loud test.
- The part that carries it
The explain-it-out-loud test — could you describe this choice to the people it affects? It settles most cases faster than any framework.
- The order to learn it in
- Make “who could this hurt?” a standing question before you ship, post or automate anything.
- Then set your data and disclosure defaults so they do not depend on you remembering.
- Then keep the explain-it-out-loud test for the cases that are genuinely unclear.
- Put something at stake
Write your team’s disclosure rule for AI-generated work, put your name on it, and be the first person held to it.
- Compress it
Who could this hurt? Whose data is this? Have I said what is automated? Could I explain it to them?
- The smallest useful dose
One harm question per shipped thing, with no exceptions for small things.
- Anchor it to something you know
The front-page test, moved a step closer: not whether it would look bad in a newspaper, but whether you could say it to the person on the receiving end.
“With great power there must also come great responsibility.”
— Stan Lee
The interactive version is in the course. The diagram, the simulation and the four-question check for Responsible Use run in the free course, where your progress across all 24 skills is kept.
Open Responsible Use in the course →
Not sure where you stand? Take the twelve-minute self-assessment, or see the 30-day course that turns these skills into a roadmap.