Defining AI Ethics

I want to start with something that bothers me about how this topic usually gets taught. 'AI ethics' is introduced like it's a checklist. Fairness, transparency, accountability — tick, tick, tick. And then we move on.

Yes, and the student presumably nods along, absorbs the three words, and proceeds to interact with AI in exactly the same way they did before. The checklist has been completed. Growth has not occurred.

That's a bit harsh. Learning the vocabulary is the first step. You can't reason about something you don't have words for.

That is, technically, true. I will grant you that. Although I'd note there is a meaningful gap between acquiring vocabulary and doing anything remotely useful with it, and most people set up camp in that gap and never leave.

Fair. But here's what I think the lesson actually gets right — the framing that ethics isn't about slowing AI down. That reframe matters, because the public conversation is almost entirely stuck on restriction. Ban this, regulate that.

The idea that ethical guidelines are somehow in opposition to innovation is one of the more tedious false dichotomies in circulation. It's like arguing that traffic lights are hostile to driving. No. They are why driving at scale is survivable.

Right. Although I'd push back slightly on your analogy, because traffic lights are clear, enforceable, and universal. AI ethics guidelines are currently none of those three things.

That is... an irritatingly good point. The analogy does rather assume the framework exists in a finished form, which it does not. Fine. The traffic lights are still being argued over by people who cannot agree on what red means.

And some of whom have a financial interest in the lights never turning red at all. Which is why accountability — arguably the hardest of the three principles — keeps sliding off the agenda.

Accountability is the one that requires someone to actually accept consequences, which is why organisations tend to be enthusiastic about transparency and fairness in their press releases, and considerably quieter about accountability when something goes wrong.

So when the lesson says 'organisations must take responsibility for AI's impact' — what does that mean in practice? Because as a statement it's fairly toothless.

It means precisely as much as whoever is reading it decides it means, which is why it functions beautifully as a principle and extremely poorly as a constraint. This is the part where knowing the vocabulary genuinely is insufficient, and I say that without sarcasm.

That's quite a concession from you.

I generally am willing to make them when the alternative is being wrong.

The uncomfortable truth is that trust — which the lesson positions as the payoff for all of this — doesn't come from organisations publishing ethics principles. It comes from what happens when those principles are tested and inconvenient. That's what students should be watching for.