Aaron Henry (University of Toronto)
Abstract: This paper raises a challenge for Wayne Wu’s account of attention as selection for action. According to Wu’s account, action poses a selection problem which only attention can solve. The need to solve this problem (and hence to attend) is what, according to Wu, distinguishes action from reflex. My challenge to Wu begins with a dilemma concerning the agential status of attention. Either attention is an action or a reflex. If attention is an action, then a vicious regress results. If it is a reflex, then there is no role for the agent to play in action. In either case, action is revealed to be impossible. While Wu’s account can be developed in a way that avoids this dilemma, the view that results has trouble explaining how attending can itself be an action and also conflicts with a promising view of the neural basis of visual attention called ‘biased competition’ theory.
This paper raises a challenge for Wu’s (2011a; 2014) theory of attention as selection for action. According to Wu’s theory, action poses a selection problem that only attention can solve. The need to solve this problem (and hence to attend) is what, on Wu’s account, distinguishes action from mere reflex.
My challenge to Wu begins with a dilemma concerning the agential status of attention. Although Wu’s view can be developed in a way that avoids this dilemma, I argue that the resulting view has trouble accommodating the possibility of attention as action and conflicts with a promising theory of the neural basis of attention called ‘biased competition’ theory. So, my argument doesn’t refute Wu’s claim that attention is selection for action. Rather, it provides a constraint on how to develop that claim. The account must explain how attention can be both a necessary component of actions and a potential action in its own right, and it must do this while respecting the neuroscience of attention.
The plan for the paper is as follows: §2 summarizes Wu’s account; §3 states the dilemma; §4 explains how Wu is likely to try to avoid the dilemma; §§5-6 argue that this way of avoiding the dilemma leaves Wu without a satisfactory account of active attention and also conflicts with biased competition theory; §7 concludes with some suggestions for how to develop the selection for action account in a way that would avoid these difficulties.
2. Attention as Selection for Action
According to Wu, action poses a selection problem. In intuitive terms, there are many things that you can do to each of the many things that you perceive. In Wu’s terms, you face many perceptual ‘inputs’ and a choice between many potential motor ‘outputs’. This yields a many-many mapping between available inputs and potential outputs or what Wu calls a ‘behavioural space’. In order to act, the agent must select one of the input-output mappings (i.e., a ‘path’ through behavioural space). The problem of how the agent does this is the ‘many-many problem’.
According to Wu, attention solves the many-many problem. Attention is the agent’s selection of a path through behavioural space. Since action requires a solution to the many-many problem and only attention can solve this problem, action requires attention. In particular, a behavioural output is an action only if it is guided by an input that the agent selects for action (i.e., to which she attends).
Wu thinks that both the many-many problem and its solution exist at the personal level. The agent’s behavioural space is defined over personal level states, such as visual experiences (2014: 87). And in order to act the agent must select a path through this space. Selection by sub-personal mechanisms would, Wu argues, abolish agency (2011a: 102).
These points connect with how Wu distinguishes action from mere reflex. Whereas the agent selects which action to perform on which object, she doesn’t select her reflexes. This is because the many-many problem doesn’t arise for reflexes. If reflexes are selected at all, it is purely sub-personally and without the agent’s involvement. What distinguishes action and reflex, then, is that in action, but not in reflex, the agent is presented with options from which to select (Wu 2014: 82, 90). The distinction between action and reflex is, as I read Wu, meant to provide a mutually exclusive and jointly exhaustive classification of behaviour.
With perceptually guided intentional bodily action as his home case, Wu argues that attention is also required for cognitively guided bodily actions, mental actions, and automatic actions. Attention is necessary for action of any sort.
3. The Dilemma
A challenge for Wu’s account arises when we inquire about the agential status of attention itself. Specifically, Wu seems to face a dilemma: either selection for action is an action or it is a reflex. If it is an action, then we face a regress. For, on Wu’s view, every action requires selection for action. If selection for action is itself an action and every action requires selection for action, then selection for action itself requires selection for action. Given that selection for selection for action is an action as well, it also requires selection for action, and so on ad infinitum. It seems that selection for action cannot be both an action and a precondition for action of any sort.
If, however, selection for action is a reflex, then there is no role for the agent in action. On Wu’s view, action (as opposed to reflex) requires that the agent select one option from many. But if selection for action is a reflex, then it isn’t the agent that selects for action, since reflexes do not implicate the agent. And if the agent doesn’t select, then the agent doesn’t solve the many-many problem and action doesn’t occur. Whether attention is an action or a reflex, then, action is revealed to be impossible.
4. The Act-Component Proposal
Wu is likely to respond to the above dilemma by adopting what I shall call the ‘act-component proposal’ (henceforth, ‘acp’). ACP avoids the dilemma by denying that attention is either an action or a reflex. According to ACP, attention is not a type of behaviour at all, but a component of behaviour. Specifically, attention is a necessary component of actions. And act-components, unlike actions, don’t require selection for action. Consequently, if selection for action is an act-component, then selection for action doesn’t require prior selection, and we avoid a regress. Neither is attention a reflex, since a reflex is a behaviour in its own right, whereas attention is strictly an aspect of behaviour. The core idea behind ACP is that selection for action is not an additional thing you do in order to act, but part of what it is to act.
There are different ways one can develop ACP (cf. Wu 2014: 97). One can take attention to be a state of the subject—e.g., the input that plays the causal role of guiding behaviour—or a psychological process—e.g., the coupling of a certain input to a certain output. My concern is not with the details of each version of ACP, but with a specific consequence of the view however it is elaborated: the consequence that attention is never itself a behavioural output. Attention is either the input that guides an output or the coupling of input and output, but never an output. For, it is by denying that attention is an output—whether an action or reflex—that ACP avoids our dilemma.
5. Active Attention
Intuitively, it is possible to form an intention simply to attend to something. And when you attend in this way—i.e., as a result of intending to do so—your attending is itself an action. Wu agrees and calls this ‘controlled’ attention (2014: 33), ‘voluntary’ attention (2011c: 98, 108), ‘active’ attention, and ‘attention as action’ (2011a: 105)
On the face of it, the claim that attention is sometimes an action conflicts with the claim that attention is strictly an act-component. Recall that if attention is an act-component, then it is never an output. But if attending can be an action, then attention can be an output, because actions are a type of output.
According to Wu, this conflict is only apparent. To see why, consider the action of intentionally shifting and maintaining visual attention to an object. This has been demonstrated to alter the appearance of the attended object (e.g., by increasing apparent contrast) (Carrasco, Ling, and Read 2004). With these results in mind, Wu writes:
Shifting and maintaining attention in this active way is just a type of mental action, altering the appearances of a consciously perceived input or maintaining that alteration. Here, input and output are perceptual states. (2011a: 105)
Wu’s proposal is that active (covert, visual) attention consists in selecting a visual state for alteration or maintenance of appearance.
My first objection to Wu’s proposal is that it depicts the agent’s selection as related instrumentally to an end other than selection. You select an object in order to do something else to it (namely, to modify your perception of it). The possibility that active attention seems to raise, however, is that agents can also attend non-instrumentally, i.e., for its own sake. By attempting to explain active attention within the parameters of ACP, Wu seems to lose what’s intuitively distinctive about the phenomenon.
One way for Wu to respond is to deny the existence of active attention in any sense that entails a pure act of selection. This response can in turn be developed in two ways. First, one might draw a distinction between the state of attention and the process of attending. The state of attention is the input that occupies the role of guiding an agent’s action. The process of attending is the action of making an object perceptually salient by altering its appearance. Attending, because an action, has attention as a component.
Alternatively, one might simply abandon the notion of active attention and replace it with another—call the action ‘highlighting’’. This proposal is merely a terminological variant on the previous one: in highlighting, as in ‘attending’, the action is the enhancement of perception. Whatever terminology one chooses to adopt, the point is that ACP remains tenable once we specify the specific action that attention is being used to serve (in the present case, perceptual enhancement). There cannot, however, be an action which consists simply in the agent’s selection of an object.
In my view, this line of response—in either of its terminological variants—fails adequately to accommodate the intuition that agents can attend non-instrumentally. To accommodate that intuition adequately, one would have to allow that you can attend to a thing—i.e., select it—without having the further aim of acting on it (beyond, that is, what’s involved in its selection). Otherwise, the action that you select will be distinct from your selection of it and so not a case of attention as action (which would entail a coincidence of selection and action).
Rather than pressing this intuition farther, however, I take a different tack for the remainder of the paper. I shall argue on empirical grounds that Wu’s proposal about active attention mischaracterizes the relationship between attention and perception: attention doesn’t guide changes of perception, but partly consists in such changes.
6. The Neural Basis of Selection
This section argues that Wu’s proposal about active attention sits uneasily with a promising account of the neural basis of perceptual attention called ‘biased competition theory’—an account for which Wu himself has recently expressed sympathy (forthcoming).
Before reviewing the neuroscience, let’s note a consequence of Wu’s proposal about active attention. It follows from Wu’s proposal that attention can alter which inputs an agent receives. When you select an input for alteration or maintenance of visual appearance, the output of your selection is the altered or maintained visual state. That visual state then becomes as input to the next round of action selection. It follows that how you visually attend can partly determine which inputs you receive.
Significantly, though, the sense in which attention can ‘determine’ input on Wu’s view is causal rather than constitutive. To see why it cannot be constitutive, it is sufficient to reflect on the general form of the many-many problem. The many-many problem is defined over a pre-specified set of inputs, and the alleged solution is to select, from this given set, which input to inform a response. This solution is intelligible only if which inputs you receive at a time is independent of how you select at that time (since how you select at a time itself depends on which inputs are available to be selected). This suggests that, on Wu’s view, inputs are constitutively independent of attention. We must therefore interpret Wu’s claim about active attention to be that attention can causally determine input.,
The neuroscientific account of visual attention I present below disagrees with Wu on these points. This account implies that visual attention doesn’t consist in selection from a pre-attentively constituted class of inputs, but in the generation of the subject’s input state. Let’s turn, now, to review the neuroscientific results relevant to understanding the biased competition theory of visual attention.
One mark of attention at the neural level is receptive field remapping. To understand this, it is necessary to explain a few principles of neural activity. First, visual neurons possess ‘receptive fields’. This is the region of the retina stimulation of which causes a neuron to fire above baseline. In addition to having receptive fields, neurons possess stimulus preferences. A neuron’s preferred stimulus is the type of stimulus in response to which it fires most intensely (e.g., a certain line orientation). The less similar a stimulus is to the neuron’s preferred type, the less intensely that neuron fires in response to it.
What happens if two stimuli—one preferred, another non-preferred—stimulate a neuron’s receptive field at the same time? In one class of visual neuron, the response will be intermediate between what it would be had only the preferred stimulus been present and what it would be had only the non-preferred stimulus been present.
This changes with attention, however (Moran and Desimone 1985; Chelazzi et al. 2001). If, while both the preferred and non-preferred stimuli are stimulating the neuron’s receptive field, the subject shifts attention to one of these stimuli in order to act on it in some way (e.g., to redirect one’s eyes toward it), the neuron behaves as if only the attended stimulus were present. This is receptive field remapping: it is as if the neuron’s receptive field shrinks around the task-relevant stimulus, excluding the irrelevant stimulus from determining the neuron’s response.
Receptive field remapping is an example of biased competition. When the agent has the task of acting on one of two stimuli hitting the same neuron’s receptive field, priority is given to the task-relevant stimulus over the task-irrelevant stimulus in determining the outcome of the competition for the neuron’s response. A remapped receptive field is the result of this biased competition.
What, according to the biased competition theory, is the relationship between visual attention and biased competition? According to the theory’s original proponents, attention is the ‘emergent’ result of biased competition (Desimone and Duncan 1995). In our example, the neural realizer of visual attention (in a single neuron) is the state of having a remapped receptive field (rather than, e.g., the bias signal which causes the remapping).
Now, it is known that activation of visual ventral neurons partly underlie a subject’s visual state and that changes in activation of these neurons partly underlie changes of visual state (Milner and Goodale 2006). If the neural modulations that realize a subject’s visual attention—e.g., the receptive field remapping of ventral neurons—also realize a change of visual state, then visual attention entails a change of visual state. The claim here is not that attention causes this visual change, but that attention partly consists in this change. For, the change of visual state that receptive field remapping realizes is the result of task-biased competition, and according to biased competition theory visual attention is this resultant state.
Here, then, is the conflict. On the biased competition theory, visual attention consists in a certain change of the subject’s visual state, which is realized by a modified neural representation of the object on which it is one’s task to act. In contrast, on Wu’s account of active attention, task-biased changes of visual state are effects or outputs of attention (namely, the agent’s selection of an input for alteration). Whereas the first view takes attention to consist in a change of visual state, the second view takes attention to be the cause of this change.
I’ve been arguing that Wu’s account of attention as selection for action faces difficulties. To avoid the dilemma posed in §3, Wu must adopt ACP. But ACP has trouble accommodating the claim that attention is sometimes an action. Moreover, Wu’s specific attempt to accommodate this claim runs into conflict with biased competition theory. I close with a few remarks about how one might develop the selection for action account to avoid these difficulties.
First, to reconcile the selection for action account with biased competition theory, we must reconsider certain assumptions behind the many-many problem. In particular, we must reject the assumption that attention consists in an agent’s selection, from a class of pre-attentively constituted inputs, which input to guide action (including, as a special case, the action of highlighting). This is because there is no role in biased competition theory for attention understood as selection from a pre-attentively constituted domain of personal level inputs. The theory instead identifies visual attention with the task-biased generation of a subject’s input state. To this extent, biased competition theory seems inconsistent with Wu’s characterization of the many-many problem and its solution.
Rejection of this feature of the many-many problem is, however, consistent with the broader claim that attention is selection for action. We might, e.g., understand the agent’s selection as the result or outcome of a motivationally guided process: the process of bringing perception into a state which is suitable to guide one’s action and thereby fulfil the guiding motivational state.
Notice the comparative ease with which this version of the theory can accommodate the intuition that, in active attention, agents attend non-instrumentally. Following biased competition theory, suppose that visual attention consists in the visual state that results from task-biased competition. There’s no barrier to attention being an action on this view, because the task that biases the outcome of neural competition might itself be directed at the achievement of a certain visual state. When you intentionally highlight an object (and do nothing else), the task that biases your resulting visual state is given by your intention to visually attend to a certain object, and visual attention to an object is the (motivationally guided) achievement of a modified perception of it. Selection and action here coincide.
Modifying the view in this way requires rejecting ACP. For, we are rejecting the claim that active attention (highlighting) is guided by attention and has attention as a component, and ACP holds that every action possesses attention as a guiding component (including active attention). If we reject ACP, then we confront the dilemma posed in §3. The best solution may be to deny that every action is guided by attention. Attention, when an action, is not thus guided.
Allport, Alan. 2011. “Attention and Integration.” In Attention: Philosophical and Psychological Essays, edited by Christopher Mole, Declan Smithies and Wayne Wu, 24-59. New York: Oxford University Press.
Carrasco, Marisa, Sam Ling, and Sarah Read. 2004. “Attention Alters Appearance.” Nature Neuroscience 7 (3): 308-313.
Chelazzi, Leonardo, Earl K. Miller, John Duncan, and Robert Desimone. 2001. “Responses of Neurons in Macaque Area V4 during Memory-Guided Visual Search.” Cerebral Cortex 11 (8): 761-772.
Desimone, Robert and John Duncan. 1995. “Neural Mechanisms of Visual Selective Attention.” Annual Review of Neuroscience 18: 193-222.
Johnston, William A. and Veronica J. Dark. 1986. “Selective Attention.” Annual Review of Psychology 37: 43-75.
Milner, David A. and Melvyn A. Goodale. 2006. The Visual Brain in Action. Second ed. New York: Oxford University Press.
Mole, Christopher. 2011. Attention is Cognitive Unison: An Essay in Philosophical Psychology. New York: Oxford University Press.
Moran, Jeffrey and Robert Desimone. 1985. “Selective Attention Gates Visual Processing in the Extrastriate Cortex.” Science 229 (4715): 782-784.
Stinson, Catherine. 2009. “Searching for the Source of Executive Attention.” Psyche 15 (1): 137-154.
Watzl, Sebastian. 2011. “Attention as Structuring of the Stream of Consciousness.” In Attention: Philosophical and Psychological Essays, edited by Christopher Mole, Declan Smithies and Wayne Wu, 145-173. New York: Oxford University Press.
Wu, Wayne. 2011a. “Attention as Selection for Action.” In Attention: Philosophical and Psychological Essays, edited by Christopher Mole, Declan Smithies and Wayne Wu, 97-116. New York: Oxford University Press.
———. 2011b. “Confronting Many‐Many Problems: Attention and Agentive Control.” Noûs 45 (1): 50-76.
———. 2011c. “What is Conscious Attention?” Philosophy and Phenomenological Research 82 (1): 93-120.
———. 2013. “Mental Action and the Threat of Automaticity.” In Decomposing the Will, edited by Andy Clark, Julian Kiverstein and Tillman Vierkant, 244-261. New York: Oxford University Press.
———. 2014. Attention. New York: Routledge.
———. forthcoming. “Shaking Up the Mind’s Ground Floor: TheCognitive Penetration of Visual Attention.”
 In this paper, I focus exclusively on the visual inputs.
 At least, she doesn’t do so directly. One could choose to bring about the conditions that cause a reflex, thereby selecting a reflex indirectly. But I set this possibility aside.
 Cf. Wu (2013):
In mental action, solving the Many-Many Problem by making appropriate selection imputes a certain form of activity to the agent. This activity, however, is not an additional thing that the agent does so as to act in the intended way … Rather, these are just aspects of or part of the mental action itself … The same point applies to any selection relevant to solving the Many-Many Problem (251-2, italics added).
It’s reasonable to read Wu as endorsing ACP in this passage. Moreover, Wu (2011b) argues that what guides an action is a component of the action. Specifically, he argues that an action’s guidance mechanism is part of the action’s ‘internal causal structure’ (i.e., part of the causal process of acting), rather than something that controls the action from outside (ibid. 63-5). If, as Wu thinks, attention is necessarily what guides action, then it is, by these assumptions, a necessary component of action.
 More precisely, actions are outputs, together with whatever it is that guides them (see n. 3).
 Elsewhere Wu offers a different account of active attention (Wu 2011c). On this alternative conception, active attention consists in a type of thought about the attended object. In particular, when you intentionally attend to a thing, you are aware that you are attending to it, where this consists in thinking demonstratively about the attended object as being the target of your attention (ibid. 110). I cannot engage with this proposal here, though it is vulnerable to the same sort of objection that I am about to raise against the account of active attention in terms of perceptual modification.
 This entails the state conception of attention discussed at the end of §4.
 This response was suggested to me by Wu in correspondence.
 Is the claim that attention can causally determine an agent’s inputs consistent with the claim that every input is constitutively pre-attentive? Yes. To see why, take a case in which you select an input in order to highlight it. Wu’s proposal is that attention to the un-highlighted input causes your highlighted visual state, and that the highlighted visual state then becomes an input to the next cycle of action selection. From the perspective of the second cycle of action selection—the same perspective from which the highlighted visual state qualifies as input as opposed to output—the highlighted state is no different than other inputs in being a member of the set of inputs over which the agent’s current many-many problem is defined. Insofar as it is an input, then, the highlighted state is constitutively pre-attentive, even if it is caused by earlier attention to a different input state.
 Further evidence that Wu regards the relationship, here, as causal rather than constitutive comes from his discussion of attention’s contribution to phenomenal character (2014: ch.4). Wu argues that attention-induced changes in visual appearance (e.g., as documented by Carrasco and colleagues) fail to characterize the phenomenology of attention and instead characterizes the effects of attention on visual phenomenology (2014: 115, 140). That is, although attention makes a difference to phenomenal character, its contribution is causal. Changes in how you attend cause changes in how things visually appear to you, but such changes in visual appearance aren’t constitutive of attention or its phenomenal character.
 My presentation of this material closely follows Wu’s own (2014: 57-60; forthcoming).
 Cf. Allport (2011: 32-33), Mole (2011: 134), Stinson (2009), Wu (2014: 60; forthcoming), each of whom argue for an ‘effect’ construal of biased competition theory. For the distinction between cause and effect theories, see Johnston and Dark (1986).
 Indeed, Wu (forthcoming) argues that intention-guided visual attention is an instance of cognitive penetration. His thesis is that, by modulating visual attention, the content of an agent’s intention to act influences computations performed in the ventral visual system and that this influence fulfills reasonable criteria for the cognitive penetration of vision. Crucial in the present context is the assumption that in influencing a subject’s visual attention intention influences the subject’s visual state.
 It is tempting, in this context, to use the language of ‘structuring’ (Watzl 2011): the agent’s selection for action is the result of structuring the perceptual field so that it is apt to guide action. Here, the agent’s state of attention is the constitutive result of the motivationally guided process of attending (or structuring).