Fleming and Michel provide an important analysis of consciousness that takes into account the time required for conscious processing in relevant ecological conditions. On the basis of this analysis, they conclude that 400ms of processing time is required before an animal (all their studies are based on mammals, mainly humans) can be deemed visually conscious. They argue that the long processing time, combined with the limited sensory horizons available in water, means that most aquatic animals cannot consciously experience vision. Here, we argue that this is a problematic conclusion given what we know about the habitat and evolution of basal fish and the transition to terrestrial life. We end by clarifying the difference between Fleming and Michel’s HOT view of consciousness and Ginsburg and Jablonka’s (Reference Ginsburg and Jablonka2019) view, which they mischaracterize.

First, fish visual-based learning can be time-consuming: fish can learn following several to many seconds temporal gap between the CS and US (trace conditioning), showing that they have working memory well exceeding 400ms (e.g., Nilsson et al., Reference Nilsson, Kristiansen, Fosseidengen, Fernö and van den Bos2008; Nordgreen et al., 2020). There is evidence that sharks navigate using cognitive maps (Brown & Schlussel, Reference Brown and Schluessel2023), and social fish are able to identify individual group members, and the cleaner wrasse is capable of mirror-self-recognition (for a review, see Kohda & Shumpei, Reference Kohda and Shumpei2025).

Second, fish, which evolved in shallow water during the late Cambrian era, were several centimeters long, vegetarian animals (predatory fish evolved later) and were preyed upon by Anomalocaris and possibly other arthropods. Accepting the (untested) assumption that 400ms is the standard time for conscious visual processing across the whole animal kingdom, and building on the analysis of Nilsson, Warrant, and Johnsen (Reference Nilsson, Warrant and Johnsen2014) model, we have calculated the detection range of the prey (fish). We assume that the environment is a shallow (2 m) lagoon with coastal-type scattering conditions (murky Jerlov type II water). The prey is a 10 cm fish (best case scenario for the predator) with a 5 mm pupil (worst case scenario for the prey), while the predator is 5 cm long (and therefore difficult to detect). The predator moves at 20 cm·s⁻¹ (≈ 4 body lengths, while the fish can produce burst speeds of 50 cm·s⁻¹ (≈ 5 body lengths). We also assume the predator discerns the prey from larger distances than vice versa. Apparent contrast decays exponentially with distance according to the beam attenuation coefficient and scales inversely with pupil size (assuming shot-noise limitation), and detection occurs when the apparent contrast drops to the visual threshold determined by photon catch (photon budget × pupil area × integration time). Using equation 31 from Nilsson, Warrant, and Johnsen (Reference Nilsson, Warrant and Johnsen2014), Supplementary Material, which underestimates available light at shallow depths, we calculated the distance R from which the prey can discern its predator:

Even in murky coastal water, the prey can discern the predator at ∼7 m away, more than 70 body lengths. In clear water, the range exceeds 45 m. Even if the predator is much larger and faster (e.g., a 50 cm Anomalocaris moving at 100 cm·s⁻¹; Usami, Reference Usami2006), the detection distance is essentially the same and the time to contact remains long enough (several second) for the prey to plan evasive maneuvers (see Mugan & MacIver, Reference Mugan and MacIver2020, Supplementary Material Figure 6, for similar estimates in slightly different conditions). There were, therefore, no sensory-horizon constraints on the evolution of conscious vision in shallow water, and if it evolved in these conditions, many descendant fish species, which migrated from the shore to the open sea, already had the underlying architecture that enabled conscious vision (excluding deep-water fish).

Third, the fish that evolved into tetrapods were also shallow-water-dwelling fish that varied in length from centimeters to more than a meter, living mostly in lagoons and tide pools by the shore (Clack, Reference Clack2012). Given their habitat and evolutionary history, they were likely to have good conscious vision, which was employed when they moved to land. The transition to land did not seem to lead to a cognitive leap, as expected by the authors’ hypothesis: the brains of the early tetrapods did not become bigger. In fact, the relative brain size of amphibians, which are the first land animals, is generally smaller than that of fish and has a simplified architecture. The brains of reptiles are, on average, the same size as those of fish, with the relative brain size of rays and sharks exceeding that of both reptiles and birds (Heldstab et al., Reference Heldstab, Isler, Graber, Schuppli and van Schaik2022). The ten-fold increase in brain size mentioned in the target article occurred after mammals had evolved, many millions of years after the first fish left the water.

Finally, we would like to clarify the difference between our UAL (unlimited associative learning) model, which the authors refer to in footnote 8, and their own view of consciousness. UAL is not a model-free type of learning as Fleming and Michel claim. Ginsburg and Jablonka devoted most of chapter 8 in their 2019 book to explain that UAL is a model-dependent capacity. Hence, we agree with Fleming and Michel that consciousness requires mental representation and model-based cognition. Mappings of the world, the body, prospective action, and the relations between them are necessary for constructing the UAL cognitive, consciousness-entailing architecture (Ginsburg & Jablonka, Reference Ginsburg and Jablonka2019, chapter 8). The difference between our view and Fleming and Michel’s view is that we do not believe that consciousness requires a second-order model, which models these first-order models for precision evaluation. It is sufficient to have several consistent samplings of the world and the body for a model to be deemed good enough for decision-making (this is the “What I tell you three times is true” principle; see Valiant, Reference Valiant2013). We agree, however, that for Popperian organisms (mostly birds and mammals), which can envisage different possible scenarios and choose among them, an additional level of representation is required (Ginsburg & Jablonka, Reference Ginsburg and Jablonka2019, chapter 9).