The disease that does not exist

A researcher in Sweden recently invented a fake skin condition. She called it bixonimania - a made-up name for a made-up illness, supposedly causing sore, pink-tinged eyes after prolonged screen exposure.

Then she planted it. A fake university. A fake researcher whose name, translated from Bosnian, literally means "the Lying Loser." A fake preprint, which, for those unfamiliar, is an academic paper published before peer review, essentially a working draft that anyone can submit to an open repository. Preprints let researchers share findings quickly, before the slow machinery of formal review catches up. They're almost like the tabloids of academia - useful for speed, but you've got to remember they haven't been verified.

This particular preprint stated explicitly in its methods section that it was entirely fabricated. It thanked Professor Ross Geller for his time. It listed funding from the Galactic Triad and the Lord of the Rings. It acknowledged colleagues on the Starship Enterprise for use of their lab.

She expected it to get caught. It didn't.

The blogs got scraped. The preprint got indexed. The fake condition made it into AI training data - pulled in by Common Crawl, the nonprofit that has been photographing the public internet since 2007 and whose archives form the backbone of most large language model training sets. Popular chatbots began suggesting bixonimania to users describing tired, strained eyes. Not as the first suggestion - they'd work through conjunctivitis and allergies first, but eventually, after enough symptom narrowing, they'd land there.

And then the fake preprint got cited in an actual peer-reviewed journal paper. Someone, almost certainly AI-assisted, put it in a reference list. A reviewer let it through.

What failed here? People stopped asking a question that had always been non-negotiable - is this source real? That gate existed for a reason. But the tool made it easy to not bother with the gate.

This failure has a name.

Constraint discrimination is the ability to distinguish what can be changed from what cannot. Most people treat constraints as obstacles - things to be removed wherever possible.

Constraint discrimination is the ability to look at what's in your way and ask, is this load-bearing, or is it just friction?

But there's a second question inside this one that's equally important.

Within the set of things that can change - the genuinely negotiable constraints - there is a subset that should not change yet.

Why? Because they exist for a reason you may have forgotten, or never consciously identified in the first place.

The peer reviewer skipped past why source verification existed as a constraint. They just knew it was slow, and the tool made it skippable. So they did. It was negotiable. It could move. It was also load-bearing.

This is a very specific kind of failure. Removing a constraint that seems movable, without asking why it was there. And it is happening inside commercial organisations right now, because a genuinely powerful tool has made it frictionless to stop asking why certain gates exist.

Most of us reading this are already using AI. The question isn't whether you've relaxed constraints. You have. The question is whether you know which ones, and whether you know why they were there in the first place.

Part two takes this into the three places it tends to matter most - your personal life, your individual work, and the team you lead. With examples that are, I hope, more uncomfortable than abstract.

Constraint discrimination in the wild. The framework itself is under How We Work.

Read part two here.