Fleming & Michel turn a welcome spotlight on the temporal profile of consciousness. They make a compelling case that knowing how long it takes for incoming information to be processed into a conscious representation helpfully constrains theories of consciousness. But their link to model-based planning hinges on the precise delay. A lag of 350–450ms implies that large swathes of visually driven behaviour are guided by non-conscious representations. A 150–200ms lag lets conscious vision guide more behaviour, including in the aquatic animals which the authors think are unlikely to have conscious visual states. Ironically, perhaps, this underscores their point: reliable timing can cleave between theories. Here, I question where that cleaver will cut.
For direct evidence, they rely mainly on postdictive effects. These must be handled carefully: a delayed stimulus can alter what the viewer is conscious of. If participants are consciously aware of two contents in quick succession, it is more difficult to elicit accurate information from their reports. A compelling case is when stimulus-1’s properties are experienced only within a representation that combines information from stimuli 1 and 2. However, the extensive 50–200ms postdiction evidence does support moderate confidence that it does often take ∼200ms for a stimulus to contribute to conscious contents.
The evidence for “long-lasting” postdictive effects, occurring in the 350–450ms range in some contexts, is less extensive. For a second argument, the authors turn to functional considerations: because of the poor transmission of light in water, there is no strong selection pressure for aquatic animals to integrate visual information in the 300–500ms range. That is only needed by terrestrial animals in order to build a stable world model as the basis for choosing model-based actions. If consciousness evolved in terrestrial animals for model-based planning, then there is an evolutionary reason to expect a time-delay to consciousness of up to 500ms for visual information.
The authors point to good evidence that consciousness facilitates model-based planning. This is already a very useful data point. It can act as strong positive evidence for consciousness in experiments – the abductive equivalent of a sufficient condition.
Nevertheless, the fact that capacity A facilitates capacity B does not entail that enabling B is why A evolved. A case in point is image-forming eyes. Model-based planning based on conscious vision, of the kind the authors discuss, requires image-forming eyes to collect the information needed to form representations of distal objects and events. But the authors do not (and cannot consistently) think that image-forming eyes evolved to enable model-based planning. So the question is: did visual consciousness evolve in order that animals could engage in model-based planning, or was it a preadaptation that subsequently enabled model-based planning after the water-to-land transition?
There is good evidence that aquatic animals already had the capacity to recognise, categorise and track distal objects. Their visual sensory distance, although generally less than in terrestrial animals, still substantially exceeds their motor distance (Coombs & Trestman, Reference Coombs and Trestmanforthc). So a plausible rival hypothesis is that visual consciousness arose already in aquatic animals and has, amongst its evolutionary functions, the function of enabling an organism to condition its behaviour on distal objects and to direct its behaviour in relation to them.
Here, the predator-prey interactions that drove the Cambrian explosion are central. Predators detect distal objects, distinguish prey objects from non-prey objects, and engage in patterns of behaviour tailored to catching an object categorised as prey object, given its nature, location and trajectory. Prey do the converse. There may be no simulation and step-by-step planning, but selecting and executing an appropriate behaviour from the animal’s repertoire requires that it can represent distal objects in its environment.
Fleming & Michel rightly point out, following McIver, that some behavioural reactions are very quick. The cost of not evading a predator is so high that animals have hair-trigger responses to some signals. Fleming & Michel identify the emergence of visual consciousness with a reality-monitoring function that also addresses “Hamlet’s problem”. But Hamlet’s problem arises in principle for any behaviour at all. There is always some cost to moving, and no animal can move in two different directions at once, so even hair-trigger reactions are based on an implicit reckoning of how much evidence to accumulate before initiating a response. It is just that in these cases, the answer is strongly on the side of accepting false positives. As sensory horizons become larger, relative to motor distance, the balance shifts in favour of longer integration. That selection pressure may already be non-trivial for the behaviour of aquatic animals in relation to distal objects.
This argument suggests that aquatic animals may integrate information on a substantial timescale in forming representations of distal objects, plausibly in the 100–200ms range. Thus, on a timescale intermediate between hair-trigger responses to proximal stimuli, and long-timescale model-based planning in relation to distal objects, there may be selection pressure for integrating visual information in the 100–200ms range in order to construct stable representations of distal objects as the basis for choosing between behavioural options, for example categorising a moving object as either predator or prey, and initiating an appropriate behaviour. That is still compatible with the detailed control of an ongoing action being guided by unconscious representations, as is found in the primate dorsal stream. Furthermore, the 100–200ms range is consistent with the majority of the evidence of postdictive effects that the authors rely on.
If visual consciousness was already present in fish, then the functional distinction between consciously-initiated behaviour on the one hand, and very fast reaction or action modulation on the other, actually predates the water-to-land transition. If so, consciousness is a preadaptation that facilitates model-based planning but did not evolve for that purpose.
The evolutionary question will thus, in turn, be informed by experimental evidence about the time-delay to consciousness. The authors are right that evidence about timing is a valuable constraint on theories of consciousness, but whether it is 150–200ms or 350–450ms makes a big difference to which theories it undermines and which it supports. Timing evidence is a meaty cleaver but it is not yet clear where it will cut.
You have
Access
Fleming & Michel turn a welcome spotlight on the temporal profile of consciousness. They make a compelling case that knowing how long it takes for incoming information to be processed into a conscious representation helpfully constrains theories of consciousness. But their link to model-based planning hinges on the precise delay. A lag of 350–450ms implies that large swathes of visually driven behaviour are guided by non-conscious representations. A 150–200ms lag lets conscious vision guide more behaviour, including in the aquatic animals which the authors think are unlikely to have conscious visual states. Ironically, perhaps, this underscores their point: reliable timing can cleave between theories. Here, I question where that cleaver will cut.
For direct evidence, they rely mainly on postdictive effects. These must be handled carefully: a delayed stimulus can alter what the viewer is conscious of. If participants are consciously aware of two contents in quick succession, it is more difficult to elicit accurate information from their reports. A compelling case is when stimulus-1’s properties are experienced only within a representation that combines information from stimuli 1 and 2. However, the extensive 50–200ms postdiction evidence does support moderate confidence that it does often take ∼200ms for a stimulus to contribute to conscious contents.
The evidence for “long-lasting” postdictive effects, occurring in the 350–450ms range in some contexts, is less extensive. For a second argument, the authors turn to functional considerations: because of the poor transmission of light in water, there is no strong selection pressure for aquatic animals to integrate visual information in the 300–500ms range. That is only needed by terrestrial animals in order to build a stable world model as the basis for choosing model-based actions. If consciousness evolved in terrestrial animals for model-based planning, then there is an evolutionary reason to expect a time-delay to consciousness of up to 500ms for visual information.
The authors point to good evidence that consciousness facilitates model-based planning. This is already a very useful data point. It can act as strong positive evidence for consciousness in experiments – the abductive equivalent of a sufficient condition.
Nevertheless, the fact that capacity A facilitates capacity B does not entail that enabling B is why A evolved. A case in point is image-forming eyes. Model-based planning based on conscious vision, of the kind the authors discuss, requires image-forming eyes to collect the information needed to form representations of distal objects and events. But the authors do not (and cannot consistently) think that image-forming eyes evolved to enable model-based planning. So the question is: did visual consciousness evolve in order that animals could engage in model-based planning, or was it a preadaptation that subsequently enabled model-based planning after the water-to-land transition?
There is good evidence that aquatic animals already had the capacity to recognise, categorise and track distal objects. Their visual sensory distance, although generally less than in terrestrial animals, still substantially exceeds their motor distance (Coombs & Trestman, Reference Coombs and Trestmanforthc). So a plausible rival hypothesis is that visual consciousness arose already in aquatic animals and has, amongst its evolutionary functions, the function of enabling an organism to condition its behaviour on distal objects and to direct its behaviour in relation to them.
Here, the predator-prey interactions that drove the Cambrian explosion are central. Predators detect distal objects, distinguish prey objects from non-prey objects, and engage in patterns of behaviour tailored to catching an object categorised as prey object, given its nature, location and trajectory. Prey do the converse. There may be no simulation and step-by-step planning, but selecting and executing an appropriate behaviour from the animal’s repertoire requires that it can represent distal objects in its environment.
Fleming & Michel rightly point out, following McIver, that some behavioural reactions are very quick. The cost of not evading a predator is so high that animals have hair-trigger responses to some signals. Fleming & Michel identify the emergence of visual consciousness with a reality-monitoring function that also addresses “Hamlet’s problem”. But Hamlet’s problem arises in principle for any behaviour at all. There is always some cost to moving, and no animal can move in two different directions at once, so even hair-trigger reactions are based on an implicit reckoning of how much evidence to accumulate before initiating a response. It is just that in these cases, the answer is strongly on the side of accepting false positives. As sensory horizons become larger, relative to motor distance, the balance shifts in favour of longer integration. That selection pressure may already be non-trivial for the behaviour of aquatic animals in relation to distal objects.
This argument suggests that aquatic animals may integrate information on a substantial timescale in forming representations of distal objects, plausibly in the 100–200ms range. Thus, on a timescale intermediate between hair-trigger responses to proximal stimuli, and long-timescale model-based planning in relation to distal objects, there may be selection pressure for integrating visual information in the 100–200ms range in order to construct stable representations of distal objects as the basis for choosing between behavioural options, for example categorising a moving object as either predator or prey, and initiating an appropriate behaviour. That is still compatible with the detailed control of an ongoing action being guided by unconscious representations, as is found in the primate dorsal stream. Furthermore, the 100–200ms range is consistent with the majority of the evidence of postdictive effects that the authors rely on.
If visual consciousness was already present in fish, then the functional distinction between consciously-initiated behaviour on the one hand, and very fast reaction or action modulation on the other, actually predates the water-to-land transition. If so, consciousness is a preadaptation that facilitates model-based planning but did not evolve for that purpose.
The evolutionary question will thus, in turn, be informed by experimental evidence about the time-delay to consciousness. The authors are right that evidence about timing is a valuable constraint on theories of consciousness, but whether it is 150–200ms or 350–450ms makes a big difference to which theories it undermines and which it supports. Timing evidence is a meaty cleaver but it is not yet clear where it will cut.
Acknowledgements
The author would like to thank Matthias Michel for helpful discussions about the time-delay to consciousness and Steve Fleming for constructive discussion of issues arising from the target article.
Financial support
This work was supported by UK Research and Innovation (UKRI) under the UK government’s Horizon Europe funding guarantee (selected as ERC Consolidator, grant number 101043666), PI: Stephen M Fleming.
Competing interests
The author declares no competing interests.