Is Your AI Helping Society, or Hurting It?

I watched a clip of Jon Stewart talking with Cory Doctorow this week, and Doctorow handed me a phrase I have been needing for two years. Reverse centaur. It is the cleanest test I have found for telling real AI from the kind that is quietly making everything worse. Here is the clip that set me off.

Doctorow is the writer who gave us "enshittification," his word for how platforms decay. First they are good to you to get you hooked. Then they treat you worse to please their business customers. Then they squeeze the business customers too, until nothing is left but a husk that still technically works. His argument on the show is that AI is the next thing getting run through that same machine, and that a lot of what gets sold as intelligence is really the old corporate wish to pay fewer people, dressed up in a demo.

I do not agree with every turn he takes. But on the core of it he is right, and I think leaders should hear it from someone who sells AI for a living, not only from the critics.

Centaur or reverse centaur

The word comes from chess. After the machines got better than people, the strongest chess in the world was played for a while by what they called centaurs: a human and an engine as a team, the human steering, the machine calculating. The person was in charge and the tool made them stronger.

A reverse centaur flips it. Now the machine is in charge and the human is bolted on to do the parts the machine cannot, and to take the blame when it fails. The person is not being augmented. They are being conscripted to keep pace with something that sets the pace, cover its mistakes, and absorb the consequences it cannot feel.

That is the whole distinction, and once you have it you cannot stop seeing it.

The test I actually run

When a company walks me through an AI project, I ask one question underneath all the others. Is this making our people centaurs or reverse centaurs?

A centaur deployment gives a person a faster horse. The analyst still decides, the model just did the four hours of gathering first. The support rep still owns the customer, the AI drafted the reply and pulled the account history so the human could actually think. The manager reads a clean summary and makes the call. In every case a person is more capable and still holds the wheel.

A reverse centaur deployment points the other way. The AI makes the decision and a human is kept on to rubber-stamp it, or to be the face when it is wrong. The system sets a quota no person chose. The worker cannot see why it decided what it decided and cannot overrule it, but their name is on the outcome. They have all of the accountability and none of the authority. That is not a job. That is a liability sponge with a pulse.

Why the reverse centaur is also a bad bet

Here is the part I add as the operator in the room. Reverse centaurs do not just feel bad. They fail as a business decision, and they fail on a predictable arc.

You buy the AI on the promise that it replaces the expensive humans. The demo was clean because someone hand-fed it three easy cases. In the real world it is confidently wrong a meaningful slice of the time, so you keep a human on to catch it. But you have stripped that human of the authority to fix anything, so they catch less and less, because nobody stays sharp babysitting a machine they were told to trust. Quality drifts down. The people who could have told you why leave, because the job got miserable. And now you depend on a vendor who knows you already fired the alternative. That is the enshittification arc, running inside your own company.

The centaur version compounds the other way. Your people get faster and stay accountable. They build real judgment about where the tool is good and where it lies. That judgment becomes something a competitor cannot buy off the shelf.

What to do this week

1. Run the test on your biggest AI project. Centaur or reverse centaur. Be honest. If a human is being kept on mainly to absorb blame, you have a reverse centaur.
2. Find the accountability gap. For any AI making decisions, ask who is on the hook when it is wrong and whether that person can actually override it. If they are on the hook and cannot override it, fix that before you scale it.
3. Reframe the goal from replacement to leverage. The question is not how many people this lets you cut. It is how much better it makes the people you keep. That reframe changes which projects you fund.
4. Watch the sales pitch. If a vendor's whole case is the salaries you will save, they are selling you a reverse centaur. The good tools sell you what your people can suddenly do.

Doctorow's warning and my day job do not actually conflict. I want companies to use AI. I just want them to use it to make people stronger, because that is the version that also happens to work over time. The technology does not decide which one you build. You do.

If you want the full conversation, it is worth the hour: AI and the Enshittification Era with Cory Doctorow on The Weekly Show with Jon Stewart (Apple Podcasts). His take is sharper than mine, and you should hear it straight from him.