Determining the functions of consciousness is a long-standing aim for the philosophy and science of consciousness (Shallice, Reference Shallice1972, Block, Reference Block1995) and raises a number of contentious issues (Ludwig, Reference Ludwig2022, Shepherd & Bayne, Reference Shepherd and Bayneforthcoming). Fleming and Michel make a compelling case that the evolutionary emergence of conscious (as opposed to unconscious) vision is associated with larger sensory horizons, which favor the emergence of a reality monitoring system that supports model-based planning. In this response, I build upon their work by considering how their account might be extended to two (surprisingly related) areas: the function of conscious proprioception and the function of conscious mental imagery.

First, Fleming and Michel suggest that their arguments might generalize in a way that “encompasses world models in general, regardless of their sensory origins” (section 1, p. 13). One fruitful area to look, in this connection, is conscious proprioception. Although work in this area is not as plentiful, nor often as precise, as work on vision, here we find evidence that supports features of Fleming and Michel’s picture. First, we find evidence of a potentially flexible, yet slow, window of integration. Research using electroencephalogram to track the sensation of bodily movement finds the emergence of a negativity signal associated with somatosensory awareness from 125 ms to 180 ms after stimulus presentation (Dembski, Koch, Pitts, Reference Dembski, Koch and Pitts2021). And work on the cutaneous rabbit illusion – in which touches to the arm at different times creates an illusion of touches moving up the arm in saltatory fashion – suggests reliable postdiction for the sense of touch at up to 300 ms poststimuli (Geldard, Reference Geldard1982, Kilgard & Merzenich, Reference Kilgard and Merzenich1995), potential postdiction at 400 ms (Eimer, Forster, Vibell, Reference Eimer, Forster and Vibell2005), and no evidence of postdiction at longer temporal windows.

Various features of proprioceptive awareness suggest a connection with reality monitoring and model-based planning. First, the cutaneous rabbit illusion is associated with the sensory body map, thought to be represented in S1, although the effect has been shown to be flexible, “hopping” out of the body to usable tools (Miyazaki, Hirashima, Nozaki, Reference Miyazaki, Hirashima and Nozaki2010). Second, proprioception in general is highly sensitive to the difference between actively and passively caused movements (Shepherd, Reference Shepherd2017, Grünbaum & Christensen, Reference Grünbaum and Christensen2024). Third, the sensation of movement is closely connected to efferent and afferent signals that construct an estimate of the “current state” of the body or body part. All three of these features might be explained by a reality monitoring system that tracks spatiotemporal properties of physical movement and their causes. Indeed, as Grünbaum and Christensen note, “The estimation of the current state of the limb and by extension the conscious sensations of movement most likely play important roles in motor learning and motor planning, even if they play no significant role in fast error correction” (2024, 7).

Our ability to keep track of differences between active and passive movement is not only relevant to how we understand the uptake of conscious proprioception in model-based planning. It is relevant to our understanding of the function of conscious mental imagery. Consider a prediction of Fleming and Michel’s account, that “conscious mental imagery does not wear its unreality on its sleeve” (p. 10). How might we square this with the appearance, which Fleming and Michel acknowledge, that mental imagery plays clear functional roles for planning (Shepherd, Reference Shepherd2025)? The answer may lie in the distinction between voluntary and involuntary mental imagery (Pearson & Westbrook Reference Pearson and Westbrook2015). A conscious signal that generation and manipulation of mental imagery is self-generated would assist agents in discriminating mental imagery from the consciously experienced world. This signal might begin with the monitoring of motor imagery, which involves rehearsals of moving the body (Jeannerod, Reference Jeannerod1994). But, arguably, it could be generalized to all manipulation of mental imagery, if the reality monitoring system is sensitive to the role of an agent’s intentions in generating mental imagery as well as bodily movements. This might afford more informative metacognition for action planning, allowing metacognition to inform agents not only of the difference between the world and imagery but also of the nature of potential sensory and cognitive consequences of various courses of action (Shepherd, Reference Shepherdforthcoming).

This possibility is relevant to accounts of the functions of mental imagery, as well as reality monitoring. Here, it is important to keep evolutionary explanations of the emergence of a function (Wright, Reference Wright1973) – Fleming and Michel’s primary focus – from causal role explanations that isolate the causal roles played by parts of a system, usually against the background of a mechanistic model of the system (Cummins, Reference Cummins1975). Evolutionary and causal role functions can dissociate (given, e.g., exaptation, see Shepherd & Bayne Reference Shepherd and Bayneforthcoming), and this may be true of mental imagery. Potentially, conscious vision arose in evolutionary time as a part of a reality monitoring system that supports model-based planning. Mental imagery could have later taken on important planning functions, as the reality monitoring system developed ways to track the difference between hallucinations and self-generated sensory imagination.