Cat or Not Cat? What a Silly Game Taught 200 People About AI

A surprise superpower of mine is leading icebreakers. So when I was asked to help open a team offsite about working in the AI era, I skipped the slides about transformers and token windows. Instead, I put up pictures of cats.

The game is called Cat or Not Cat?, and I've since run it with more than 200 colleagues. It takes about fifteen minutes, needs no technical background, and lands the single most important idea I know about AI at work.

The rules

There are only three:

1. See it. A picture appears on the screen.
2. Shout it. Is it a "Cat" or "Not Cat"?
3. Speed it. Trust your instincts. Don't overthink it.

Rounds 1–10: easy money

The first ten rounds are deliberately easy. A Maine Coon kitten on a green rug: Cat! A golden retriever mid-sprint: Not cat! An extreme close-up of a Siamese eye: Cat! A flock of sheep from above, a parrot, a mountain goat, a rabbit hiding in the garden: Not cat, not cat, not cat!

There are a couple of gentle curveballs. One is an oil painting of a Persian cat with a red bow. Most people shout "Cat!" without blinking, and a few pedants in the back hesitate. Hold that thought.

By round ten the room is loud, confident and fast. That's the point. This is what classification looks like when everyone shares the same unspoken definition. It's also what AI looks like at its best: quick, consistent and correct on the easy stuff.

Then: the tiger

Then I put up a close-up of a tiger and ask, simply: Is it a cat?

The room splits. Half shout "Cat!" Half shout "Not cat!" Someone starts explaining the family Felidae. Someone else points out you can't keep one in an apartment. For the first time, speed doesn't help at all.

So I ask the question that actually matters: What is a cat, and why are we asking?

• If we're stocking a pet store, the tiger is definitely not a cat.
• If we're running a biology lab studying felines, the tiger is 100% a cat.

Same picture. Same "data." Opposite correct answers. The only thing that changed was the purpose.

Purpose makes the complex simple. Without it, even the simplest task becomes a conflict.

Rounds 11–15: same game, now with a purpose

Then we play again, with one change: a definition at the top of the slide. Definition: Domestic Pet.

A black cat walking down a moonlit alley: Cat. A lion with a golden mane: Not cat. A ginger cat asleep on the sofa: Cat. A beagle puppy: Not cat. And the final round, a leopard on a branch: Not cat (pet).

Nobody argues anymore. The decisions are fast and unanimous, not because the pictures got easier but because we finally agreed on what we were trying to do.

The AI paradigm

This is where I connect it back to work:

• Compute is cheap. An AI model can classify a million "cats" in seconds. It will happily do the tiger round at scale.
• Purpose is rare. AI can't tell you why the classification matters, or which definition of "cat" your business should use.
• That's our value. In the AI era, our job isn't to out-classify the machine. It's to provide the purpose that makes the complex simple: defining the question, choosing the definition, and deciding what happens with the answer.

This is also why the game resonates with data people. Every metric dispute I've seen in 15 years ("What counts as headcount?", "When is a customer churned?") is a tiger round. AI doesn't make those arguments go away. It makes them more expensive to skip, because now the wrong definition gets applied a million times in seconds.

Why it works

• It's participatory. People shout, laugh and disagree out loud, which is more memorable than any slide.
• Nobody needs to know what a token is. The concept lands without any jargon.
• The fear flips into agency. People walk in worried AI will replace their judgment. They walk out realizing judgment is the scarce ingredient.

In the AI era, we aren't just here to execute tasks. We're here to provide the purpose that makes sense of the complex.

Steal this format

• Open with 8–10 easy rounds to build speed and confidence.
• Hide one or two near-misses in there (a painting, a cartoon) to plant doubt.
• Drop the tiger. Let the room argue before you explain anything.
• Replay with an explicit definition and watch the disagreement disappear.
• Close by mapping it to real work: which "tiger" definitions does your team argue about, and who owns them?

The cats are optional. The purpose isn't.