Source
Albert Gidon, Jaan Aru and Matthew E. Larkum (2025), “Does neural computation feel like something?”, Frontiers in Neuroscience 19, LIT-1000, read in full in NOTE-818: Results (“The replay results in a degenerate computation”, “Consciousness and nothing”, “Counterfactual erasers”), Figures 3–4, and the Discussion on the Turing-machine model and on IIT.
What was actually shown
- The simulation (Figs. 2–3). Five Hodgkin–Huxley-type neurons in a
recurrent ring, driven by two LGN nodes, fire together for green light
and singly for red or blue. A simulated voltage clamp is given the
recorded green response as its command voltage.
- With red light, the clamp supplies exactly the missing LGN input and the network fires as for green.
- With green light, it injects no current: “no intervention was needed”. Activity is identical to the unclamped run, and so is every actual neuron-to-neuron interaction (causation read mechanistically).
- The computation changes (Figs. 3D, 4). With the clamp on, state transitions no longer depend on the input. On the Turing–Putnam definition, where “the inputs, states, transitions, and outputs define the functional organization”, the clamp becomes part of the transition function, and what remains is “a trivial computation consisting of a single sequence of states (i.e., universally realizable)”.
- The conditional. If changing functional organization is changing consciousness (cited to Putnam, Shagrir, Block and Lewis), then the clamped and unclamped runs on green light differ in consciousness although they differ in nothing that occurs: “doing nothing to the brain can alter consciousness”. The authors call the idle clamp a “counterfactual eraser”.
What could have come out otherwise is narrow. The simulation could have shown residual clamp current on the congruent input, so that the clamp did act; with an ideal clamp in a deterministic model it cannot. The conditional’s force rests on the definitional step, which is where a functionalist would push back.
What this does not say
- Not that functionalism is false. The authors present a “paradox” the functionalist “must contend with”, not a refutation. A functionalist may accept the verdict: the clamped system is a different, trivial machine, and its experience may differ. Whether that is absurd is the open question.
- Not that the simulation is evidence about experience. Its outputs are voltages and currents. The verdicts on consciousness are the working hypothesis applied by definition.
- Not how much experience changes. The authors concede it depends on “the proportion of states used in the congruent case compared to all other cases”, and say only that it might be “severely diminished or even abolished”.
- Not about affect. The paper concerns consciousness in general, “the experience … that fades during deep sleep or anesthesia”. It says nothing of valence or feeling in the affective sense.
- Not a positive account. The closing alternative, that “natural counterfactuals” embodied in the substrate fix experience, is offered in one paragraph and not argued. This account does not include it.
- Not a challenge to IIT. The paper says the clamp lowers Φ by degrading the transition probability matrix, so IIT gives the same verdict without contradiction; Φ is not computed.
Connections
- Chalmers on implementation (LIT-404, NOTE-342). Implementation needs “reliable, counterfactual-supporting transitions” for every formal state; without them a clock and a dial implement any inputless automaton. That is the criterion on which the clamped network implements only a trivial computation, and the reason a functionalist cannot escape by dropping the counterfactuals: dropping them reopens the rock problem.
- Fading qualia (LIT-392, NOTE-347). Chalmers keeps causal organization fixed and changes the substrate; this case keeps substrate and activity fixed and removes unused organization. Together they show the functionalist’s experience tracks the organization in both directions, including its unexercised part.
- THEORY-092. Block’s lookup machine matches every actual and counterfactual behaviour and lacks intelligence. The clamped network matches every actual activity and lacks the counterfactuals. Both say a mental property is not fixed by what actually happens, from opposite sides.
- THEORY-119. Frankfurt’s device never acts and so cannot change responsibility. The counterfactual eraser never acts and, on functionalism, can change experience. The paper asks whether Jones, held responsible by Frankfurt, may have acted “with altered or possibly diminished conscious experience”.
- THEORY-043. Whether the clamp is part of “the system” is a boundary question of the kind THEORY-043 says functionalism has no principled answer to yet.
- THEORY-101 and QUESTION-059. Veit identifies feeling bad with occupying “a coordinated regulatory state”. If that is a functional role in the Turing–Putnam sense, the identity inherits this conditional: the same regulatory activity with its alternatives erased is a different feeling or none. This does not answer QUESTION-059, which asks for a case, not another argument from a definition.
- THEORY-023. The paper takes the biological naturalist’s side on AI consciousness; this account holds only its conditional and leaves the evidential deadlock where THEORY-023 puts it.