Augmented Serendipity ← Back to the essay

Cognitive one-way functions

Easy one way.
Hard the other.

A one-way function is cheap to run forward and expensive to run backward. The same asymmetry shows up in cryptography, in memory, and in how AI gets trained. Try both directions and feel the gap.

1. Mix, then un-mix

Combining two colors is instant. Recovering the recipe from the result is not.

forward · instant
mix

Now cover the sliders and ask a friend to reproduce this exact color from scratch. That is the reverse direction. reverse · hard

2. Solve vs. check

This number is the product of two prime factors. Checking a guess is one multiplication. Finding the factors means searching.

×
0Your guesses
0Checks the computer would do to verify

Forward

Multiply p × q. One step. This is what a password check, a hash, or a proof-verifier does.

Backward

Recover p and q from the product alone. For large numbers this is the wall that public-key cryptography leans on.

The lesson for AI: do not ask first how hard a task is to solve. Ask how hard it is to check. If you can verify an answer cheaply, you can often train a system to produce one, even when producing it is hard. Verifiability, not difficulty, is what makes a problem trainable.