The Variable Man Cannot Be Modeled

The SRB machines — the Statistical Research Bureau's prediction engines — run on historical data. They produce odds for any future event by pattern-matching against everything that has ever happened. Reliable. Accurate. The government uses them to decide whether to go to war.

Then Thomas Cole appears from 1929, and the machines jam.

The jam isn't a technical failure. The machines are working correctly. They're trying to model a person for whom they have no data — not just sparse data, zero data. No one from 1929 has ever been projected into 2136 before. He is a genuine unknown, and the unknown doesn't resolve to an odds ratio; it makes the ratio undefined.

Commissioner Reinhart's response is to eliminate Cole. Kill the variable. Restore the clean, predictable odds.

There's a logic to this. A jammed prediction machine is useless. The machine says: this man cannot be predicted. Therefore, from the machine's perspective, he might as well not exist. Get him out of the model.

But the machines weren't broken. They were detecting something real. Cole was genuinely unpredictable because he was genuinely different — not different in a random way, but different in a way the model had no category for. Non-verbal intuitive intelligence, operating on machines directly, with no stake in the political situation.

The elimination instinct — kill what you can't model — is the opposite of the learning instinct — expand your model to include what breaks it. Reinhart chose elimination. The story is about what that choice cost him.

What the variable man ultimately did was solve the problem that everyone with model-shaped intelligence couldn't solve, precisely because he came at it from outside the model. He didn't know what the machine was supposed to do. He only knew what it was trying to do.

The prediction machine identified him as a threat to its accuracy. It was right. He was a threat to its accuracy.

He was also the only person who could do the thing that mattered.