Acquired savant syndrome, and where computation lives. After some brain injuries, people gain skills they did not have before. This chapter follows that question down four levels — a person, neurons in a dish, a mapped fly brain, and matter itself — and at each level asks the same four operator questions, with a status for every answer.
The chapter uses the same guide as every other chapter in Book 3 (see The g-series, “Translation and reintegration”). For each level it names the state and the control, says which of the four operators is claimed, supported or skipped, and gives a status: established, model, demonstration, analogy or open. The operator tests are the ones used throughout:
Acquired savant syndrome is the appearance of a new skill — in drawing, music, calculation or memory — after a brain injury, a stroke or a degenerative disease, in a person who did not show it before. Darold Treffert collected the cases and gave the name; they are rare, and most of what is known comes from case reports. The best-documented group is people with frontotemporal dementia who began to paint or compose as the disease progressed, often with damage concentrated in the left anterior temporal lobe (Miller et al., Neurology 1998).
Neuropsychology has a wider name for the pattern: paradoxical functional facilitation — a lesion that improves some function rather than only removing one (Kapur, Brain 1996). The usual explanation is release from inhibition: one region, when damaged, stops suppressing another.
Allan Snyder turned that explanation into a hypothesis that could be tested in healthy people. If the left anterior temporal lobe filters raw detail into concepts, briefly suppressing it should give access to the detail. In a small study using repetitive transcranial magnetic stimulation, some participants estimated the number of dots on a screen more accurately for a short time after stimulation (Snyder et al., Perception 2006). The effects were small, the samples small, replications mixed, and the change faded within about an hour.
Status: model. The disinhibition account is a reasonable model with thin quantitative support.
DishBrain. Kagan et al. (Neuron 2022) grew human and mouse neurons on a multi-electrode array and connected them to a simulated game of Pong. Stimulation told the culture where the ball was; when the paddle missed, the culture received unpredictable stimulation, and when it hit, predictable stimulation. Over about twenty minutes of play, rallies became longer than in controls. The result is that feedback built on predictability shapes activity in cultured neurons. The paper's use of the word “sentience” was criticised in published responses, and this chapter does not use it.
CL1 and Doom. In March 2026 Cortical Labs showed neurons on its CL1 system playing a version of Doom. It is a company demonstration, not a peer-reviewed result, and sceptics pointed out how much of the play depends on the encoding and decoding around the cells. It shows the interface works; it does not yet show learning.
Organoid intelligence. Smirnova et al. (Frontiers in Science 2023) set out a research programme for computing with brain organoids, together with its ethics. It is a roadmap, not a finding.
A link to regeneration. The substrates studied in Ch Λ · Polylaminin are first tested on the same kind of preparation — cultured neurons on a surface — so the dish is also where repair and computation meet.
Status: DishBrain established as a feedback effect; CL1 Doom demonstration; organoid intelligence open.
The fly connectome. The FlyWire consortium reconstructed the whole brain of an adult female Drosophila melanogaster from electron microscopy — about 139,000 neurons and some 50 million synapses — and published it in Nature in 2024 (Dorkenwald et al.). The map was produced by automated segmentation and then proofread by people; it is a reconstruction, checked by hand, not a formal verification.
A model on the map. Shiu et al. (Nature 2024) built a simple spiking model on the whole connectome, with every neuron a leaky integrate-and-fire unit, and used it to predict which neurons carry taste and grooming signals; several predictions were then confirmed in flies. The code is open, which matters here: it means the proposal below can be run by anyone.
A counterexample. “Less inhibition, more ability” is not a general rule. In the fly's mushroom body a single inhibitory neuron, APL, keeps odour coding sparse. Blocking it makes coding less sparse and impairs the fly's ability to tell similar odours apart (Lin et al., Nature Neuroscience 2014). Any disinhibition account has to say why removing inhibition helps in one place and hurts in another.
A proposal (not a result). Run in-silico lesions of specific inhibitory populations in the whole-brain model, and look for any readout that improves — abruptly, and in a way that persists. A null result would also be informative.
Status: connectome and model established data; the lesion test open (proposal).
Computronium. Toffoli and Margolus (Physica D 1991) used the word for matter arranged to compute as efficiently as possible — “programmable matter”. The physics that bounds it is well established: erasing a bit costs at least kT ln 2 of energy (Landauer 1961), and Lloyd (Nature 2000) worked out the ultimate limits on speed and memory for a kilogram of matter in a litre.
Tubulin. Ch T read microtubules as a kind of computronium. Its numbers did not pass the checks there (TubulinChecks.lean): the critical curvature it gives is more than a hundred times the measured GDP curvature, and the φ value it cites is off. Microtubule computation in the brain (Penrose–Hameroff) is a minority view and disputed. The reading stays an analogy.
Status: physical limits established; tubulin as computronium analogy.
| Level | C · slow variable | K · driver | F · abrupt? | U · residue? | Status |
|---|---|---|---|---|---|
| Person (acquired savant) | top-down filtering (claimed) | injury, dementia, rTMS | no time courses | yes after injury; no after rTMS | model |
| Dish (DishBrain) | predictability (definable) | feedback rule | open | open | established effect |
| Dish (CL1 Doom) | — | interface | — | — | demonstration |
| Fly connectome model | inhibitory weight (definable) | lesion size | open | none without plasticity | data + proposal |
| Matter (computronium) | — | — | — | — | limits; tubulin analogy |
Read across, the table says what the Coherence Bridge is for in this chapter. It does not claim that a person, a dish and a fly share a mechanism. It gives each field the same questions, so that a neurologist, a cell biologist and a modeller can see which of them could answer which one. The person level has the residue contrast; the dish has the time courses; the fly model has the controlled lesion. None has all four.