In his article, Seth distinguishes between simulation, implementation, and realization. Simulation is defined as a way of approximating some process or the workings of a mechanism “by abstracting it as a (Turing) computation and then implementing this computation on a digital computer” (p. 13). Implementation is understood as the way in which some computational function is executed in a physical medium. Importantly, Seth sees these concepts as continuous rather than completely distinct, since “by allowing for the inclusion of finer-grained, causal, and potentially non-computational properties” (ibid.), simulations can come closer to implementations. However, simulations can never become equivalent to realizations of their target phenomena (i.e., instantiate them) because they “generally lack the causal powers and intrinsic properties of the things being simulated” (ibid., emphasis added). In our eyes, this is the crucial step in Seth’s rejection of any arguments for computational functionalism (CF) based on the possibility of simulating consciousness: simulation, no matter how fine-grained, is simply not the kind of thing that could ever realize conscious states.

We want to put pressure on Seth’s claim that simulations of consciousness must necessarily lack the causal powers or the intrinsic properties of the real thing. Further, as we will point out, Seth’s definition of simulation inevitably leads to a dilemma that undermines his position. Crucially, and contra Seth’s claims (footnote 22), our argument does not beg the question against the deniers of CF, because at this point we simply do not know whether CF is true or not. Even though we cannot demonstrate that simulating consciousness is indeed possible – after all, this is an empirical matter – we do think that Seth fails to conclusively show that it is impossible.

Starting with causal powers, we are happy to concede that a simulated rainstorm does not have the same causal powers as a real one, as it does not produce or necessitate the same effects. However, while a simulated storm cannot sink a real ship out in the seas, it can affect simulated ships (as in SimCity). What about consciousness? Assuming that consciousness is not an epiphenomenon, it is relatively uncontroversial to posit that its causal effects involve overt changes in a subject’s behavior as well as covert changes in their cognitive states. We assume that the idea of simulating behavior is uncontroversial, and Seth does not seem to rule out the possibility of simulating cognitive states, since admitting this possibility is not the same as assuming computational functionalism about consciousness. Most cognitive scientists would agree that it is, in principle, possible to simulate various cognitive functions. Indeed, many consider existing programs and machine learning systems to already do this for some specific, if narrowly defined, functions (e.g., memory, analogical reasoning, etc.). In fact, recent evidence shows that simulations of cognition, such as large language models, can have a greater effect on people’s rationality and cognitive states than other humans (Costello et al., Reference Costello, Pennycook and Rand2024) – is this not a clear example of a simulation exerting causal powers in reality? All this leads us to wonder, why couldn’t it be possible to create simulated agents whose conscious states would have the right kind of causal powers with respect to their (simulated) cognitive states and behaviors?

Seth will likely respond by pointing out that any physicalist should assume that consciousness somehow depends on or supervenes on brain states. We concur but want to make two observations. First, Seth cannot (and does not; see section 3.3) outright assume that the causal powers of conscious states require biophysical determinants without begging the question against CF. Second, while we do agree that biological cognition might depend on a complex tangle of biophysical processes, we are skeptical about whether realizing consciousness would require simulations of neural tissues all the way down to the molecular level. After all, the same requirement doesn’t seem to hold for simulating cognitive functions. This being said, our skepticism on this matter is only warranted if consciousness is not fundamentally different from cognition. Many researchers will point out that, since at least some cognitive functions can operate unconsciously, there must be some kind of extra ingredient that makes consciousness special. This extra “secret sauce” is usually identified with the intrinsic properties of consciousness.

The intrinsic properties of consciousness, i.e., the properties that determine or constitute what experiences are like to the subject, are traditionally named qualia or phenomenal properties. They are said to be intrinsic because they seem to escape attempts at explanations in terms of non-intrinsic, i.e., relational or dispositional, properties, which in turn leads straight into the hard problem of consciousness (Chalmers, Reference Chalmers1995). Philosophers and scientists responding to this problem tend to choose one of two general strategies. The first consists in arguing against the apparent irreducibility of such properties, either by identifying them with non-intrinsic ones (Dretske, Reference Dretske1995; Lycan, Reference Lycan1996; Rosenthal, Reference Rosenthal2005) or by denying their existence altogether (Dennett, Reference Dennett1991; Frankish, Reference Frankish2016). The second strategy consists of embracing the idea that phenomenal properties are intrinsic and non-reducible, for instance by adopting some version of panpsychism (Seager, Reference Seager2006). Both strategies come with commitments that confront Seth with a dilemma. On the one hand, if the intrinsic properties of consciousness are reducible or identical to other, more familiar kinds of properties or processes, as Seth seems to suggest in section 5.5, then he should recognize that there is nothing fundamentally problematic about realizing phenomenal properties by simulating their underlying processes. On the other hand, Seth could side with Searle in claiming that the intrinsic properties of consciousness are irreducible to other kinds of properties and hence constitute a genuinely novel ontological kind. However, this option comes at a very high cost. Not only does it, once again, come close to begging the question against CF, but it would also open Seth to many well-known objections to Searle’s position (see e.g., Kim, Reference Kim1995; Warfield, Reference Warfield1999). Most problematically, however, asserting the intrinsic and non-reducible nature of qualia would severely undermine the possibility of a truly naturalistic, or indeed any scientific, explanation of consciousness, which we trust is not a direction in which Seth wants the field to go.