A new opinion piece in Trends in Cognitive Sciences by Ned Block, the NYU philosopher known for his China Brain thought experiment: Can only meat machines be conscious? The paper argues that computational functionalism, the dominant framework for thinking about AI consciousness, moves too fast. It assumes that implementing the right computations is sufficient for consciousness, without considering that the biological mechanisms implementing those computations might matter too.
Role vs. Realizer
Block builds his case on a distinction between roles and realizers. A role is a causal pattern defined by its inputs, outputs, and internal relations. A realizer is the physical mechanism that implements that pattern.
Consider fragility. The role is a disposition: hit it with a hammer, it breaks. Glass is fragile because of its amorphous molecular structure. Cast iron is fragile because of micro-cracks. Same role, different realizers.
Block applies this to consciousness. Computational functionalism says consciousness is the role: get the causal pattern right and consciousness follows, regardless of what implements it. What Block calls the meat hypothesis says the realizer matters too. Perhaps something about the biological mechanisms that implement conscious roles, the electrochemical processing in nervous systems, is necessary for consciousness to arise.
The Evolutionary Argument
The paper’s most concrete contribution is an argument from evolutionary biology. The two candidates for the earliest animals are comb jellies (Ctenophora) and sponges (Porifera). Sponges lack a nervous system entirely. A 2023 Nature paper established comb jellies as the likely first animals. They have a nervous system, but a strange one: purely electrical, with no chemical synapses except at sensory interfaces.
All animals considered plausible candidates for consciousness, from insects to mammals, evolved electrochemical nervous systems instead. Purely electrical processing did not lead to the evolutionary lineages associated with complex behavior and (arguably) experience. Block treats this as suggestive: electrochemical processing might enable something that purely electrical processing does not. The obvious alternative is that electrochemical systems simply enable more flexible computation (better learning, more effective inhibition), which in turn enables complex behavior, without consciousness being the relevant variable. Block is aware of this. He reviews evidence on learning in Cnidaria, dedicates a section to inhibition, and explicitly considers the possibility that meat enables functions which themselves are the basis of consciousness. But he argues the question remains open.
AI and Animals as Competitors
Block frames AI systems and simple animals as competing for consciousness attribution. The choice of extrapolation base determines who wins.
Extrapolating from computational roles (the functional organization, the information processing), cognitively sophisticated AI systems look like strong candidates for consciousness, and simple animals with crude computation look weak.
Extrapolating from biological realizers (the electrochemical substrate), animals that share human biology look like candidates, even cognitively simple ones, and AI systems that lack it do not.
Functionalism picks the first path. The meat hypothesis picks the second. Block argues that most current theories of consciousness, including Global Workspace Theory (which ties consciousness to information broadcast across the brain), Higher-Order theories, and Integrated Information Theory, are what he calls “meat-neutral” without having earned that neutrality. They assume functionalism rather than arguing for it.
What the Paper Does Not Say
It does not identify the relevant biological property. Block argues that something about electrochemical processing might be necessary for consciousness. But he cannot say what that something is. Block gestures toward what he calls “fluctuations in the ionic composition of the neurotransmitter soup between neurons” as a possible mechanism, but immediately concedes that the specific ions (sodium, potassium, calcium) are not essential. Other ions could work.
This concession is significant. If the identity of the ions does not matter, it becomes hard to see why ionicity itself would be the essential property. And if it is the mechanism rather than the material, the question becomes: why couldn’t a non-biological system implement the same mechanism? Block is aware of this line of reasoning. He distinguishes his view from strict biological materialism, calling it “subcomputationalism” (about mechanisms, not material composition), and explicitly asks in the paper what the prospects are for building artificial systems with electrochemical properties. But he offers no concrete answer to why a non-electrochemical implementation of the same mechanism would fail. Without one, the meat hypothesis slides back toward functionalism: what matters is what the substrate does, not what it is made of.
The parsimony question. Functionalism posits one requirement for consciousness: the right functional organization. The meat hypothesis says consciousness may depend on the realizers, the roles, or both. Block is aware of the parsimony objection. He argues that the apparent simplicity of collapsing what consciousness does and what it is into one thing is “illusory” if the role can be accomplished by something other than biology. The comb jelly evidence is exactly the kind of data that could justify looking beyond pure functionalism. But Block cannot yet specify what the extra condition is or what explanatory work it does. The evolutionary data motivates the search. It does not complete it.
Block is a major figure in philosophy of mind, publishing in a top journal in cognitive science. The paper is labeled an “Opinion” piece, and that label is accurate. The core contribution is a challenge: functionalists should not assume they are right by default. Fair enough. But the meat hypothesis, as presented here, is not a theory. It is a placeholder for a theory. The paper tells us to take biological substrates seriously. It does not yet tell us why.
A more recent paper from Google DeepMind pushes this line further. Where Block says the biological realizer might matter, Alexander Lerchner argues that computation itself is the wrong ontological category for consciousness, regardless of substrate. The arguments are complementary: both challenge functionalism, but from different angles.