The Strawberry Problem: What AI Knows, and What Only People Can

Ask an AI what a strawberry tastes like and you'll get a fine answer: bright, sweet acidity, a floral note, the faint grit of seeds. It's accurate and well written, and it comes from something that has never tasted a strawberry.

That gap tells us where AI helps, where it falls short, and where people still matter.

AI has absorbed what botanists, chefs, food scientists and countless reviewers have written about strawberries. It can explain the chemistry of the aroma, why winter fruit disappoints, and what pairs with it. No human taster holds all of that. AI gives you the experience of millions of people, on demand.

AI knows about the strawberry, but it doesn't know the strawberry. It can't tell you whether the one in your hand is any good. Researchers Emily Bender and Alexander Koller made the same point in a 2020 paper: a system that learns only from the form of language has no way to learn its meaning.

The same is true in business. Much of what keeps an organisation running is first-hand knowledge: the system that's about to fail, the customer whose yes means no, the workaround nobody wrote down. That knowledge lives in people.

The answer is to pair the two. People bring the tasting, and AI brings the library. A person spots that something is off, and AI finds the pattern, checks the history and scales the insight. A Harvard Business School study with BCG of 758 consultants found that those using AI completed 12.2% more tasks, 25.1% faster and at higher quality.

For the first time, people who know things by doing them can capture, scale and pass on that knowledge. AI still needs someone who has tasted the strawberry, so the opportunity is to take part rather than watch.

Two claims deserve to be thrown out.

The first is that because AI can describe a strawberry, it understands one. A map is not the territory.

The second is that because AI has never tasted one, nothing it says is worth hearing. Most of us have never been to the Moon, yet we'd still listen to an astronomer.

The obvious strawberries are easy to spot. Nobody asks AI whether the fruit in their hand is ripe. The danger lies in the hidden ones: decisions that look like information problems but are really judgement calls, needing someone who has tasted. The same study found that on a task beyond AI’s reach, consultants using it were 19 percentage points less likely to get the right answer.

They hide in plain sight:

Approving a contract clause because the summary said it was standard.

Shortlisting or rejecting a candidate on an AI ranking nobody questioned. Amazon scrapped an experimental AI recruiting tool in 2018 after it was found to be biased against women.

Signing off a design, a budget or a plan that no one with first-hand experience has checked.

Letting AI answer the customer whose yes means no.

In each case AI can inform, but it cannot be accountable. Responsibility doesn't transfer just because the work did. In the UK, Article 22 of the UK GDPR already restricts significant decisions made solely by automated means.

So before handing something over, ask one question: does this need someone who has tasted the strawberry? If it does, keep a person in the loop and keep their name on it. The machine can describe the world, but only people can taste it, and only people can answer for it.