Project status — July 2026: Fathom is a completed research project. This page answers common questions. For the findings, see the Results and Conclusion.
What was Fathom trying to do?
To test the central claim of the essay Consciousness and Complexity by Roger Smolski: that consciousness is a matter of processing complexity, meaning how far a system draws its own past together into its present, and that it comes in degrees with no sharp break in nature. The essay admits it lacks the tool to measure this. Fathom was an attempt to build the tool and put the claim to the test.
Did Fathom prove the theory right?
No. The instrument could not reach the case that would have mattered most, a system that genuinely learns, so the theory was neither confirmed nor refuted. It stands where it stood. What Fathom established is narrower: where the current means of measuring the theory’s central quantity stop working.
Did Fathom prove the theory wrong?
No. The essay’s boldest prediction is that there is no sharp break in nature between conscious and non-conscious systems. Fathom found no such break. An earlier stage threw up what looked like a sudden threshold, but that turned out to be a feature of the measuring rather than of the world, and it was set aside. The prediction was not overturned.
Did the project prove that language models are not conscious?
No, and this is worth being clear about. Fathom’s instrument reads neurons one at a time, and a system that learns almost certainly does its real work in the patterns spread across many neurons at once, which a one-at-a-time measure cannot see. The instrument went quiet on the trained model, but that only tells us the typical single neuron carries little on its own. It says nothing about the network as a whole. A blunt tool finding nothing is not evidence that there is nothing to find.
So did the project fail?
That depends on what counts as success. The question it set out to answer, whether processing complexity switches on sharply or rises smoothly in a learning system, got no answer, because the instrument could not reach the system. In that narrow sense the decisive test was inconclusive. But a clean, honestly reported result at the edge of what a method can do is not a failure. Fathom found a real boundary: where this kind of measurement stops working, and why. That boundary may be the most useful thing it produced.
What is the difference between “it is not there” and “we cannot measure it”?
This distinction is the heart of the project. A measurement can only certify a quantity once it rises above the background noise of the surrogates. Below that level the quantity may be perfectly real and simply too small, or too well hidden, to prove. Fathom’s instrument requires a quantity to clear its significance threshold before it will report it. So “the measure is silent” always means “we cannot certify it here”, which is not the same as “there is nothing here”. Reading the first as the second is the commonest way to misunderstand the results.
Why should I trust a result that found nothing?
Because the method was fixed before the result was known. Every parameter, threshold, and interpretation was written down and sealed in advance, so the conclusion could not be adjusted afterwards to say something more flattering. A negative result produced under that discipline is more trustworthy than a positive result produced without it.
How does this connect to the essay?
The essay, Consciousness and Complexity by Roger Smolski, sets out the theory and states plainly that the tool to measure it is missing. It even names the difficulty: “The measurement problem is real and unsolved. But an unsolved measurement problem is not an untestable theory.” Fathom is one response to that. What it found is that the measurement problem is harder and more specific than it first looks: the tool works on simple systems and loses its grip on the learned, richly connected ones where the theory’s most interesting claims are staked.
Does this connect to the Canary project on this site?
Only in spirit. Canary and Fathom are separate projects on different subjects, one on AI sector news sentiment, the other on processing complexity in physical and artificial systems. What they share is a method. Both were pre-registered, both said in advance what would count as failure, and both reported their result plainly rather than dressing it up.
What happens next?
The obvious next step reads the neurons together rather than one at a time, to test whether the signal this project missed lives in the way the parts act as a group. It reuses the data already gathered, so it needs no new model runs. Whether it recovers a signal, stays silent, or fails its own controls is genuinely unknown, and its meaning will be fixed in advance, before the first number is computed. The project documented here is complete. Any follow-up will be pre-registered separately.
Can I see the underlying data and pre-registration?
The pre-registration, its full history, the measured results, and the analysis code are kept as the project record. If you would like access for scrutiny or replication, the About page has contact details. A copy of the complete essay, Consciousness and Complexity, is available on the same basis.