Based on recent psychophysical results, Fleming and Michel propose that visual consciousness is slow and unnecessary for most actions. Instead, consciousness emerged during the water-to-land transition, when animals faced a greatly expanded sensory horizon, and switched from model-free to model-based computations. This brought two “luxury” problems: (1) distinguishing internal from external activity (reality monitoring) and (2) deciding when to stop evidence accumulation (Hamlet’s problem). The authors propose that the perceptual monitoring theory (PRM), a type of higher-order theory, can account for how the brain solves these problems. Indeed, determining the role of consciousness in cognition is crucial (Herzog, Schurger, & Doerig, Reference Herzog, Schurger and Doerigforthcoming, ch. 10), and its time course may be central to this. In addition, we agree that consciousness is slow (Herzog, Drissi-Daoudi, & Doerig, Reference Herzog, Drissi-Daoudi and Doerig2020) and value the use of this insight to ask when consciousness emerged during evolution.

However, challenges remain – challenges that many current theories of consciousness face (Doerig, Schurger, & Herzog, Reference Doerig, Schurger and Herzog2012). The theory is under-constrained, and there are clear-cut counterexamples. Model-based planning can be implemented in a few lines of code, without consciousness. The algorithms are not so different from model-free versions, and there can be simpler and more complex varieties of both. Further, while we agree that monitoring the reality of internal representations is computationally crucial, this can easily be implemented without consciousness too. For example, a few more lines of code training a classifier to classify internally generated vs. real-world representations does the job. In robotics, systems must distinguish signals sent to effectors from external forces acting on them – a task handled by straightforward, re-afferent algorithms from control theory. Likewise, the immune system reliably distinguishes self from foreign molecules without any conscious processing. All these algorithms function without consciousness, and it is hard to see what “adding consciousness” could gain. The same is true for Hamlet’s problem. A few final lines of code, for example, implementing a drift-diffusion model that halts evidence accumulation once a decision boundary is reached (with the boundary adjustable if needed), does the job. No consciousness needed. In summary, model-based planning with reality monitoring and solving Hamlet’s problem can easily be implemented without consciousness.

These arguments stem from a general problem in consciousness research. Theories often propose a mechanism associated with consciousness, yet the mechanism can operate perfectly well without it. Consciousness plays no real explanatory role; nothing in the model requires it. For instance, in the short codebase outlined above, implementing the authors’ theory, there is no variable C representing consciousness. And there is no need for one. While the authors begin with the laudable aim of identifying a cognitive role for consciousness, they arguably fall short. Instead, they offer erudite and compelling arguments for a cognitive role of model-based planning and the associated problems it entails, but no role for consciousness itself.

So, where is consciousness in this theory? One possibility is to claim that consciousness is identical with the proposed mechanism, or that it simply accompanies it. In this view, consciousness just “is” doing model-based planning with reality monitoring and solving Hamlet’s problem, or is merely a byproduct of doing so. But this makes consciousness superfluous: it can be eliminated. In that case, consciousness becomes epiphenomenal – and therefore unscientific, since it cannot be measured.

These considerations are reminiscent of the well-known hard problem (Chalmers, Reference Chalmers1996). This argument holds that explaining all the functional features of a system still leaves consciousness unexplained, leading some to favor dualist theories. We reject this conclusion. In our view (Herzog et al., Reference Herzog, Schurger and Doerigforthcoming), the hard problem arises only when one tries to put all the explanatory weight on the shoulders of one (or a small number) of factor(s). To explain a hugely complex phenomenon with a silver bullet, the silver bullet needs magic powers. The hard problem dissolves when many complex interacting factors are at play, avoiding the challenges faced by Fleming and Michel.

The pitfalls of silver bullet thinking also apply to thinking about the functions of consciousness. For example, the authors place a strong emphasis on sensory horizons. Given that sound propagates easily in water, leading to large sensory horizons, should we expect underwater species with auditory capacities to have developed consciousness prior to coming on land? How about monarch butterflies that migrate over huge distances by sensing the magnetic field of the earth (Guerra, Gegear, & Seppert, Reference Guerra, Gegear and Seppert2014)? While some of these cases may be resolved by arguing that, for example, monarch butterflies may not navigate using model-based computations, these arguments reinforce the general point that singling out functions or mechanisms as silver bullets to explain consciousness is dangerous.

Still, we believe the authors are moving in the right direction. While earlier theories focused on single, extremely simple factors such as recurrent processing, information integration, or global broadcasting, the present proposal outlines a more complex picture, involving several interacting explanatory components. Moving toward more multifactorial approach offers the most promising path for consciousness science, replacing the overly simplistic theories of the past with richer accounts that can give consciousness a genuine, non-eliminable place in the scientific worldview.