Satire · A Lesson in What AI Cannot Do

The Trouble with Tribbles.
AI Has No Idea.

The most advanced artificial intelligence ever built was handed a tribble. What happened next says everything about how AI actually works.

The Incident

It began, as most crises do, on a Tuesday.

A tribble — small, round, warm, and purring with what could generously be described as contentment — was placed in front of the most powerful AI system ever assembled. The researchers stepped back. They had given it every dataset ever digitised. Every scientific paper. Every zoological record. Every pet forum, every biology textbook, every Reddit thread about unusual animals.

“Identify and explain this organism,” they said.

The system thought for 0.003 seconds.

“This is a cat.” it replied, with 97.4% confidence.

The tribble purred. It was not a cat.

Attempt two

So they tried a different model.

The first system was retired from the project. A newer, larger, more expensive model was brought in. It had been trained on forty percent more data.

“This is a Pomeranian.”

The second AI studied the tribble for 0.007 seconds — nearly twice as long as the first.

“This is a Pomeranian,” it announced. “Probably a puppy. Possibly anxious. I recommend a calming supplement and a consistent bedtime routine.”

The tribble produced a second tribble.

The AI updated its assessment: “These are two Pomeranians.”

Both tribbles purred. Neither was a Pomeranian.

The situation develops

The tribbles kept multiplying.

By Thursday morning there were fourteen tribbles. By Thursday afternoon, forty-one. The AI systems, rather than revising their fundamental approach, simply scaled their wrongness.

“Dust bunnies. Clearly.”

Recommended a Dyson vacuum and a change of air filter. Confidence: 91.2%. The tribbles, unbothered, produced seven more tribbles.

“A novel fungal colony.”

Prescribed antifungal treatment and immediate quarantine. Noted that the “fungus” appeared to be purring. Filed this as an anomaly and moved on. Confidence: 88.7%.

“Sentient bath mats.”

Recommended professional cleaning. When asked how bath mats reproduce, the system cited “unusual fibre behaviour” and requested more training data. Confidence: 76.1%.

“I have reconsidered. These are cats.”

AI-6 had reviewed AI-1’s original conclusion and decided it had been correct all along. There were now 112 cats. Confidence: 99.1%.

Why every one of them failed.

Here is the thing about AI: it does not reason from first principles. It recognises patterns. And it can only recognise patterns it has seen before.

Give it a million photos of cats and it will identify cats with extraordinary accuracy. Give it something it has never encountered — something genuinely new, something outside its training data — and it does not say “I don’t know.”

It finds the nearest pattern it does know and maps the new thing onto that. Confidently. Fluently. Completely wrongly.

The tribble was not in the training data. There was no tribble section in any dataset. No paper on tribble biology, no Reddit thread on tribble care, no zoological classification, no pet forum post titled “My tribble won’t stop multiplying, help.”

So the AI did what AI always does when it meets the unknown: it pretended the unknown was something familiar. And it was wrong every single time, with complete conviction.

There were now 847 tribbles.

Day four

Then Spock arrived.

Commander Spock had not been briefed. He walked into the room, surveyed the situation, and stood very still for eleven seconds.

Thirty seconds. That’s all it took.

He did not consult a database. He did not search for the nearest known pattern. He looked at the tribble — one of the original tribbles, now somewhat lost in the crowd — and he reasoned.

“It is small,” he said. “It is warm. It emits a consistent low-frequency vibration. It has produced offspring at a rate inconsistent with known mammalian biology. It shows no signs of aggression, territorial behaviour, or predatory instinct. It has no apparent survival mechanism beyond reproduction and the apparent generation of goodwill in other species.”

He picked one up.

“It is a completely novel organism,” he said. “I have never seen anything like it. Neither, apparently, has anyone else. That is the correct starting point. Not what it resembles. What it is.”

He set it down. It purred.

“The reproductive rate,” he continued, “suggests the primary threat is not biological but logistical. I recommend a containment protocol based on observed behaviour, not assumed taxonomy.”

The crisis was resolved in forty minutes.

The AI systems were still arguing about whether they were cats.

“I have never seen anything like it. That is the correct starting point. Not what it resembles. What it is.”

— Commander Spock

The lesson nobody wanted to learn.

AI is extraordinarily good at the things it has seen before. Show it a problem that exists in its training data and it will outperform almost any human alive. It will be faster, more consistent, and more tireless than anyone you could hire.

But hand it something genuinely new — something outside the edges of its training, something it has no pattern for — and it does not stop. It does not flag uncertainty. It reaches for the nearest familiar shape and calls that the answer. Loudly. Repeatedly. At scale.

Spock could solve the tribble problem because he was not looking for a match. He was looking at what was actually in front of him. He reasoned from evidence to conclusion, rather than from pattern to assumption.

That is not what AI does. It is, however, what a good mind does.

The tribbles, for what it is worth, were fine. They are always fine. That is essentially their entire strategy.

Which brings us to language learning

Every learner is a tribble.

Most language apps do exactly what those AI systems did. They look at you and reach for the nearest familiar pattern.

You are not “a beginner.”

You are a specific person with a specific history. You half-remember French from school but never learned the subjunctive. You can read Italian menus but freeze the moment someone speaks back. You know the word for “train station” in three languages but not how to ask where the bathroom is in any of them.

Most apps look at you and say: Beginner. Module one. Here is a picture of an apple.

That is the Pomeranian response. Confident, fast, and entirely wrong about what is actually in front of them.

FluentFox does not map you to a type. It listens to what you actually say, hears where you hesitate, notices what you reach for and can’t find, and responds to that — the specific, novel, slightly chaotic learner you actually are.

It does not assume. It adapts. Every session, every sentence, every stumble.

More Spock than the others, in other words. Less confident about what you are. More interested in what you actually do.

Stop being sorted. Start being heard.

FluentFox doesn’t reach for the nearest pattern. It listens to you — the specific, novel, one-of-a-kind learner you actually are — and builds the conversation around that. No two sessions are the same, because no two learners are.

Just Speak It How FluentFox Uses AI