RA · P02

The Strawberry Mind

There is a question you could once ask a machine and reliably get the wrong answer.

How many R’s are in the word strawberry?

Three. You can see them without counting — strawberry. A six-year-old manages it.

For a while, the most capable language systems ever built answered two.

Not because they were stupid. By then they could draft a contract, summarise a scan, hold a conversation in four languages. They failed on the strawberry because they never saw letters at all. Text arrived pre-bundled — st, raw, berry — in chunks large enough to be efficient and too large to be exact. You cannot count what has already been packed.

The failure was not ignorance. It was granularity. The machine was reading at a coarser grain than the question required, and had no way of knowing that was what it was doing.

Now notice which R goes missing.

It finds the double. That one is loud; it announces itself. The R that disappears is the first — buried inside str, swallowed whole by the chunk it happened to be standing in. The thing that vanishes is never the conspicuous thing. It is the one absorbed into a larger unit for the convenience of whoever is reading quickly.

A person inside a demographic. A case inside a caseload. A name inside a category.

It was never hidden. It was the third letter. It disappeared anyway.

Being uncounted is not the same as being absent.

You cannot argue strawberry into having two R’s. The third one is there whether or not anything present is able to perceive it. Which sounds like a small point, until it is your count.

Then there is the part that troubles me most.

They fixed it. Ask today and you will usually get three. But as far as anyone outside can tell, the machines did not learn to see letters — they learned to spell the word out first, routing around the blindness with a procedure. The answer arrived. Whether the perception did is a harder question than it looks, and I am not sure anybody can answer it from where we are standing.

A system that produces the right number without having acquired the ability to count is stranger than one that gets it wrong, because now the error is invisible from outside. The miscount was a symptom. The symptom was the only thing being measured.

That is where these novels began. Not with a machine that hates us — that story has been told, and I do not believe it. With a small, funny, entirely harmless mistake, and one question underneath it:

What else is being counted by something that cannot see the third R?

These essays sit beside the novels. Download any of our novels, free and in full.