In their rich, thought-provoking article, Fleming & Michel (F&M) propose that perceptual consciousness functions to aid model-based planning (MBP). They don’t mean “the effortful thought characteristic of conscious deliberative planning,” but instead posit implicit MBP, which occurs unconsciously, rapidly, and in creatures that lack deliberation. The belief in implicit MBP is widespread in cognitive neuroscience. However, it is unsupported.

Genuine MBP is deliberation, and the rapid, unconscious activations F&M cite can be explained by other mechanisms. This fact raises a dilemma: either perceptual consciousness is for deliberative MBP, which is unlikely given that deliberation is rare in the animal kingdom, or the evolution of perceptual consciousness was independent of MBP.

The ability of an animal to go left rather than right around a distant obstacle in no way requires simulation of the consequences of going left, then of going right, and finally picking left. There are indeed intelligent unconscious processes that underlie this behavior, but there is no need to invoke planning or simulation to explain them (Haith & Bestmann, Reference Haith, Bestmann, Poeppel, Mangun and Gazzaniga2020; Haith, Huberdeau, & Krakauer, Reference Haith, Huberdeau and Krakauer2015; Trapani, Stein, & McNamee, Reference Trapani, Stein and McNamee2024). Simulation involves evaluating counterfactuals against each other, and we contend that there is no convincing evidence for this without deliberation. Similarly for learning, where “explicit task structural knowledge is a primary determinant of model-based reinforcement learning and is most readily acquired from instruction rather than experience” (Castro-Rodrigues et al., Reference Castro-Rodrigues, Akam, Snorasson, Camacho, Paixão, Maia and Oliveira-Maia2022, 1126).

The battleground for the question whether alternative trajectories are implicitly simulated, and then one chosen, is the hippocampus. Non-local hippocampal reactivations, either at rest (replay) or during movement in the late phase of theta oscillations, allegedly show MBP/simulation. However, the literature does not support this interpretation. As McNamee highlights in a recent review, unconscious hippocampal sequence activations are frequently “physically unrealizable” and “are generally discontiguous and often systematically disordered, such that they could never correspond to the experience of the animal either in the past or the future” (McNamee, Reference McNamee2024, 6). That is, they are not simulations.

Cognitive scientists think of the hippocampus as a locus of maps and therefore interpret its functional activation as use of maps in planning, even when unconscious. In reality, there are many other functions hippocampal activation might play, including data augmentation and non-model-based updating (Trapani et al., Reference Trapani, Stein and McNamee2024). Jumping from neural activation to a psychological interpretation is an enormous leap and, here, fails to accommodate small-scale dynamics.

Take an article the authors think supports MBP in rats (Kay et al., Reference Kay, Chung, Sosa, Schor, Karlsson, Larkin and Frank2020), which finds rapid theta-cycling between positions in the right or left arm of a T-maze. For this cycling to constitute planning, minimally, we should see change over time with increasing bias toward the chosen direction. This is not reported. Indeed, some of the same authors recently acknowledged that the contribution “to behaviour may not be immediate, and in fact might have an indefinite horizon in the lifetime of subject[s]” (Comrie, Frank, & Kay, Reference Comrie, Frank and Kay2022, 11). In other words, we don’t know exactly how rapid cycling contributes to behavior, and the data do not convincingly resemble the real-time, anticipatory function of MBP.

The problem, in our view, is that pervasive and recalcitrant construct in neuroscience: the internal model. The internal model idea (Conant & Ashby, Reference Conant and Ross Ashby1970, Bin et al., Reference Bin, Huang, Isidori, Marconi, Mischiati and Sontag2022) grew largely out of sensorimotor control. Sensorimotor control is thought to proceed unconsciously and thus internal models must be usable unconsciously. As has recently been stated in an article aptly called “The illusion of internal models in biological movement”: neuroscientists “import experimental and analytical methods designed for programmable artifacts into explanations of sensorimotor behavior” (Mangalam, Reference Mangalam2025, 2), and these “frameworks elevate a computational metaphor to a mechanistic dogma: that the brain is a prediction machine solving abstract control problems through algorithmic simulation” (Mangalam, Reference Mangalam2025, 4). We agree, and one of us has been involved in empirical work directly refuting the idea that learning and planning of sensorimotor behavior occur via a forward model (Hadjisosif, Krakauer, & Haith, Reference Hadjiosif, Krakauer and Haith2021).

Implicit MBP is unsupported, so perceptual consciousness can’t be for implicit MBP. F&M face a dilemma: Either perceptual consciousness exists to aid explicit, deliberative MBP, or perceptual consciousness has some other function.

We assume, with F&M, that perceptual consciousness is widespread among land-dwelling vertebrates. Is deliberation equally widespread? To deliberate about a course of action, you must represent multiple possible actions. An animal that cannot consider multiple incompatible world-states, therefore, cannot deliberate.

Mody and Carey (Reference Mody and Carey2016) devised a “four-cup task,” in which participants see one reward go behind an occluder into one of two cups (1 and 2), and another reward behind another occluder into one of another pair of cups (3 and 4). 1 is revealed to be empty. An adult knows to select 2, which has a 100% probability of housing a reward, while 3/4 have 50% probability each.

Early work found that 2.5-year-olds fail, showing a ∼33% likelihood of going to each of the remaining cups (Mody & Carey, Reference Mody and Carey2016). But recent work shows that 2.5-year-olds select correctly more than chance, but only 50% of the time, with the other 50% of responses being one of the other pair (Leahy et al., Reference Leahy, Huemer, Steele, Alderete and Carey2022; Carey, Reference Carey2023). Chimpanzees also show this 50-50 pattern (Engelmann et al., Reference Engelmann, Haux, Völter, Schleihauf, Call, Rakoczy and Herrmann2023). What explains it? Leahy and Carey (Reference Leahy and Carey2020) argue that chimps and younger children can represent one of two possibilities – that rewards are in 2 and 3 and that rewards are in 2 and 4 – but cannot simultaneously consider both. So instead, they randomly select one possibility. Each possibility says that there is a reward in two specific cups, and participants pick one of those cups arbitrarily.

Chimpanzees and children under 3 appear to be unable to consider multiple alternatives and plan an action. Alternative performance-error explanations have been proposed (Alderete & Xu, Reference Alderete and Xu2023), including by one of us (Quilty-Dunn, Porot, & Mandelbaum, Reference Quilty-Dunn, Porot and Mandelbaum2023), but none provide equally plausible explanation for the cross-species 50–50 pattern. If chimpanzees and young humans lack deliberation, deliberative MBP is not as widespread as perceptual consciousness. We conclude that perceptual consciousness probably did not evolve for MBP.