A unit for how much a machine actually understood. Not how fluent it sounds. Not how fast it runs. How much hidden reality its explanations can predict, after the explanations pay for themselves.
Two short films from the AIEN channel: the story of the Turing, and how machine understanding gets measured.
One Turing (T) is one bit of predictive power a machine's explanation earns on data it has never seen, minus what the explanation itself cost to say. Long, memorizable lists score near zero. Compact rules that keep working on new data score high.
Some data is shown, some is sealed away. The machine must predict the sealed part.
"Add 5 each time" is 4 words and nails every hidden number. A rule that just recites the shown numbers is 15 words and predicts nothing new. Length counts against you.
Turings = hidden events predicted − cost of the explanation.
Score is always measured against a declared baseline, on sealed data, with every artifact re-checkable by anyone.
The machine saw 3, 8, 13, 18, 23. Three numbers are sealed. Pick the explanation you would submit. The scorer is honest: it only counts sealed predictions, and it charges rent for every word.
Drag the sliders. Feel the tradeoff every explanation faces: predict more, or say less. The meter scores your rule live, exactly the way the real scorer thinks.
Each hidden event you nail is worth +1 T. Each word of explanation costs 0.1 T of rent. The baseline, blind guessing, always scores zero.
Three duels. Two explanations enter, one earns more Turings. Judge them. Watch out: round three has a twist.
On September 29, 2026, a candidate explanation was scored against sealed data under a frozen measurement profile. It cleared its pre-registered bar by more than ten times. Every artifact is public and re-checkable.
The Turing asks how much structure was discovered. A second question, raised by Gödel in 1956, asks how hard that structure was to discover, compared to checking it. Finding a proof can be enormously expensive while verifying it is cheap. That gap is measurable too.
The deep question: does explanatory compression systematically collapse search? If an abstraction worth a large T makes future search fall from near-exhaustive enumeration to a small search tree, compression is algorithmic understanding. This is proposal, not result: no measurements exist yet, and no finite experiment can ever settle P vs NP. The honest first step is instrumenting discovery cost and verification cost on work already running.
A single answer to a single instance.
A reusable method that finds answers efficiently.
Compact structure explaining why the method works across cases.
Expensive search paired with cheap verification is not something anyone invented. It is how the living world builds complexity without a designer. The Turing is the first unit written down for what the loop produces.
Variation is generated wastefully, at the cost of uncounted failed organisms across deep time. Verification is free: did it survive and reproduce, or not.
A vast random library of antibodies, most of them useless. One simple binding check selects the winners for mass production.
The hypothesis is the expensive part. The experiment is the cheap, decisive part. The Turing protocol is this loop, formalized.
Everything is public: the formal definitions, the engineering history, and the replication protocols anyone can run to check the measurements.
From the imitation game to the Turing. Six definitions, the measurement protocol, falsification criteria, and the first canonical measurement. 26 pages.
Read the PDFThe engineering history the unit grew out of: the architecture, the failures that stayed in the record, and the new addendum on understanding versus search. 25 pages.
Read the PDFThe exact procedures for calibrating the instrument and executing a measurement run, so anyone can reproduce or challenge a claimed result.
Open on GitHubThe Turing is free for public use. The price of admission: report your baseline, your measurement profile, and your evidence.
Citation details