Building on the well-known distinction between the slow ventral visual system, whose contents are conscious, and the fast dorsal sensorimotor system, whose outputs are not (Milner & Goodale, Reference Milner and Goodale1995), Fleming and Michel (hereafter F&M) argue that the former evolved for model-based decision-making, which is only likely to be adaptive for land-living creatures with extended sensory horizons. They also suggest that model-based decision-making involves capacities for offline simulation, which in turn requires a higher-order capacity for reality-monitoring. In this comment, we challenge both sets of claims. We first present evidence from studies of optimal foraging and social behavior in fish showing that most, if not all, species of fish are capable of model-based decision-making of a simple sort. We then argue that since even simple kinds of model-based decision-making are unlikely to be a product of the fast sensorimotor system, F&M face a dilemma once the weaker and stronger senses of “model based” are distinguished: either their argument for the significance of the water-to-land transition fails, or that transition is irrelevant to the evolution of conscious vision.
To support their hypothesis that conscious vision evolved with the water-to-land transition, F&M argue that the close visual horizons typical of aquatic environments require only fast and reflexive actions, which makes model-based planning associated with consciousness redundant for aquatic animals. But they mistakenly assume that predator–prey interactions are the only sources of evolutionary pressure. Aquatic creatures also need to forage successfully and to select appropriate mates. Neither needs to involve extended sensory horizons, and both can be model-based, going well beyond mere sensorimotor responding. Indeed, the evidence shows that many species of fish engage in model-based decision-making that is unlikely to be a product of a fast dorsal sensorimotor system.
There is, for example, a significant literature demonstrating optimal foraging in fish (Werner & Mittelbach, Reference Werner and Mittelbach1981; Grossman et al., Reference Grossman, Rincon, Farr and Ratajczak2002), which is generally assumed to require capacities to build models of the environment and then compute and compare the expected values of the various alternatives. But the main focus of this comment will be on mate choice in fish and adaptive use of number representations in fish.
In species belonging to a wide variety of taxa – including fish – female mate-choice can be influenced (generally positively) by the mate choices of other females (Wade & Pruett-Jones, Reference Wade and Pruett-Jones1990). A female guppy that observes one of two males associating with another female is more likely to select and swim towards the former when the other female is no longer present (Dugatkin, Reference Dugatkin1992). Indeed, a female that has previously shown a preference for a particular male is much more likely to reverse her choice after observing the previously unpreferred male interacting with another female (Dugatkin & Godin, Reference Dugatkin and Godin1992). In naturalistic conditions, such observations would not require an extended sensory horizon, since guppies typically live in shoals. Yet it is difficult to see how such behavior could be the product of a fast (model-free) sensorimotor system. On the contrary, it requires the observing female to build and recall a simple model of the situation (albeit not capacities for imaginative simulation).
Capacities to discriminate the approximate number of items in a set (often called “numerosity”) have also been found in a wide variety of taxa (Dehaene, Reference Dehaene2011), and there are multiple demonstrations of such abilities in fish (Dadda et al., Reference Dadda, Piffer, Agrillo and Bisazza2009; Piffer, Miletto Petrazzini, & Agrillo, Reference Piffer, Miletto Petrazzini and Agrillo2013; Mehlis et al., Reference Mehlis, Thünken, Bakker and Frommen2015; Agrillo & Bisazza, Reference Agrillo and Bisazza2017). While numerosity computations are likely to be used in the context of foraging decisions too, most experimental investigations have utilized the preference that shoaling fish have to belong to larger groups. (This is a preference also shown in many other taxa, and is likely an adaptive protection against predators.) A fish presented with two groups of conspecifics can discriminate which group has the larger number of members and will opt to join the larger group. Moreover, in at least one of the extant experiments, presentation of the members of each set was sequential, not simultaneous, requiring the fish to build a representation of the numbers involved over a significant period of time (Dadda et al., Reference Dadda, Piffer, Agrillo and Bisazza2009). Again, it seems very unlikely that the resulting behavioral choice could be the work of a fast and memory-free sensorimotor system.
These findings highlight an important conflation in F&M’s argument. They use “model-based” planning and decision-making to refer to a group of capacities that they suggest co-evolved in tandem with the water-to-land transition: building a model of the world based on perceptual input, offline executively-directed planning and simulation that relies on visual imagery, and a higher-order reality-monitor that distinguishes perception from simulation, tagging certain representations as reliably representing the world. But there is no evidence that a capacity to build and rely on world-models must co-evolve with the other abilities. Indeed, model-based decision-making can be fully online, drawing on learned information to respond to perceptually presented options without relying on the sorts of conscious reflection that involve simulation and mental imagery. We assume that this is true of most fish, as well as many land-based creatures.
F&M may well be right that consciousness requires capacities for offline simulation and higher-order reality-monitoring. And they may well be right that fish lack such capacities. But this has taken us a long way from the distinction between the slow ventral visual system (which supports model-based decision-making) and the fast dorsal sensorimotor system with which they began. And there is no reason to think that it was the extended sensory horizons characteristic of land-based living that provided the pressure for the evolution of such sophisticated capacities.
The distinction between weaker and stronger readings of the term “model-based” presents F&M with a dilemma. If conscious vision is supposed to have evolved to support model-based decision-making in the weak sense, then their argument for the water-to-land transition fails, as fish also have such capacities. If, instead, consciousness requires much more sophisticated kinds of cognition – such as executively-controlled forms of simulation, mental imagery, and higher-order reality-monitoring – then the transition is irrelevant to the evolution of conscious vision.
You have
Access
Building on the well-known distinction between the slow ventral visual system, whose contents are conscious, and the fast dorsal sensorimotor system, whose outputs are not (Milner & Goodale, Reference Milner and Goodale1995), Fleming and Michel (hereafter F&M) argue that the former evolved for model-based decision-making, which is only likely to be adaptive for land-living creatures with extended sensory horizons. They also suggest that model-based decision-making involves capacities for offline simulation, which in turn requires a higher-order capacity for reality-monitoring. In this comment, we challenge both sets of claims. We first present evidence from studies of optimal foraging and social behavior in fish showing that most, if not all, species of fish are capable of model-based decision-making of a simple sort. We then argue that since even simple kinds of model-based decision-making are unlikely to be a product of the fast sensorimotor system, F&M face a dilemma once the weaker and stronger senses of “model based” are distinguished: either their argument for the significance of the water-to-land transition fails, or that transition is irrelevant to the evolution of conscious vision.
To support their hypothesis that conscious vision evolved with the water-to-land transition, F&M argue that the close visual horizons typical of aquatic environments require only fast and reflexive actions, which makes model-based planning associated with consciousness redundant for aquatic animals. But they mistakenly assume that predator–prey interactions are the only sources of evolutionary pressure. Aquatic creatures also need to forage successfully and to select appropriate mates. Neither needs to involve extended sensory horizons, and both can be model-based, going well beyond mere sensorimotor responding. Indeed, the evidence shows that many species of fish engage in model-based decision-making that is unlikely to be a product of a fast dorsal sensorimotor system.
There is, for example, a significant literature demonstrating optimal foraging in fish (Werner & Mittelbach, Reference Werner and Mittelbach1981; Grossman et al., Reference Grossman, Rincon, Farr and Ratajczak2002), which is generally assumed to require capacities to build models of the environment and then compute and compare the expected values of the various alternatives. But the main focus of this comment will be on mate choice in fish and adaptive use of number representations in fish.
In species belonging to a wide variety of taxa – including fish – female mate-choice can be influenced (generally positively) by the mate choices of other females (Wade & Pruett-Jones, Reference Wade and Pruett-Jones1990). A female guppy that observes one of two males associating with another female is more likely to select and swim towards the former when the other female is no longer present (Dugatkin, Reference Dugatkin1992). Indeed, a female that has previously shown a preference for a particular male is much more likely to reverse her choice after observing the previously unpreferred male interacting with another female (Dugatkin & Godin, Reference Dugatkin and Godin1992). In naturalistic conditions, such observations would not require an extended sensory horizon, since guppies typically live in shoals. Yet it is difficult to see how such behavior could be the product of a fast (model-free) sensorimotor system. On the contrary, it requires the observing female to build and recall a simple model of the situation (albeit not capacities for imaginative simulation).
Capacities to discriminate the approximate number of items in a set (often called “numerosity”) have also been found in a wide variety of taxa (Dehaene, Reference Dehaene2011), and there are multiple demonstrations of such abilities in fish (Dadda et al., Reference Dadda, Piffer, Agrillo and Bisazza2009; Piffer, Miletto Petrazzini, & Agrillo, Reference Piffer, Miletto Petrazzini and Agrillo2013; Mehlis et al., Reference Mehlis, Thünken, Bakker and Frommen2015; Agrillo & Bisazza, Reference Agrillo and Bisazza2017). While numerosity computations are likely to be used in the context of foraging decisions too, most experimental investigations have utilized the preference that shoaling fish have to belong to larger groups. (This is a preference also shown in many other taxa, and is likely an adaptive protection against predators.) A fish presented with two groups of conspecifics can discriminate which group has the larger number of members and will opt to join the larger group. Moreover, in at least one of the extant experiments, presentation of the members of each set was sequential, not simultaneous, requiring the fish to build a representation of the numbers involved over a significant period of time (Dadda et al., Reference Dadda, Piffer, Agrillo and Bisazza2009). Again, it seems very unlikely that the resulting behavioral choice could be the work of a fast and memory-free sensorimotor system.
These findings highlight an important conflation in F&M’s argument. They use “model-based” planning and decision-making to refer to a group of capacities that they suggest co-evolved in tandem with the water-to-land transition: building a model of the world based on perceptual input, offline executively-directed planning and simulation that relies on visual imagery, and a higher-order reality-monitor that distinguishes perception from simulation, tagging certain representations as reliably representing the world. But there is no evidence that a capacity to build and rely on world-models must co-evolve with the other abilities. Indeed, model-based decision-making can be fully online, drawing on learned information to respond to perceptually presented options without relying on the sorts of conscious reflection that involve simulation and mental imagery. We assume that this is true of most fish, as well as many land-based creatures.
F&M may well be right that consciousness requires capacities for offline simulation and higher-order reality-monitoring. And they may well be right that fish lack such capacities. But this has taken us a long way from the distinction between the slow ventral visual system (which supports model-based decision-making) and the fast dorsal sensorimotor system with which they began. And there is no reason to think that it was the extended sensory horizons characteristic of land-based living that provided the pressure for the evolution of such sophisticated capacities.
The distinction between weaker and stronger readings of the term “model-based” presents F&M with a dilemma. If conscious vision is supposed to have evolved to support model-based decision-making in the weak sense, then their argument for the water-to-land transition fails, as fish also have such capacities. If, instead, consciousness requires much more sophisticated kinds of cognition – such as executively-controlled forms of simulation, mental imagery, and higher-order reality-monitoring – then the transition is irrelevant to the evolution of conscious vision.
Financial support
This research received no specific grant from any funding agency, commercial, or not-for-profit sectors.
Competing interests
None.