What is Attention?

I’m going to push this topic ahead of discussing the visual streams since it is connected to the very active discussion that we had on automaticity and control in the second post. Apologies to those who want to discuss visual processing in primary cortex and its connection to visual consciousness. We’ll start with attention, and while I focus here on the “what is” question, I’m happy to discuss all aspects of attention including attention and consciousness (if need be, we can add a post on that).

Update: I should add that answering the “What is” question seems to me crucial to making clear headway on some other questions people are familiar with, such as whether attention is a gatekeeper for consciousness (all the debates about inattentional blindness, phenomenal overflow, etc.). So much seems to depend on being clear on what attention is, whether it is the sort of thing that there can be more or less of (cf. Christof Koch and co-workers who often speak of consciousness in the near absence of attention). How resolvable are these debates when the notion of attention is not well-defined?

It is worth noting what I am finding to be a growing skepticism about attention (albeit among the scientists I have interacted with). I had the opportunity to organize a discussion group made up of faculty, postdocs and students at the CNBC to discuss the nature of attention. This group includes many people who have done important work on attention both in psychology and in neuroscience including Aaron Batista, Marlene Behrman, Marlene Cohen, and Carol Colby  (I’m not claiming that these folks specifically are skeptical of attention, and they have refreshing views on the subject matter). I tried to see if we could get anywhere in respect of defining what attention is, but I think the general consensus among the scientists was that this project wasn’t going to be successful. I have since met others who have similar doubts, and indeed that there is no such thing as attention! How in the world can that possibly be?

In short, my answer is that attention is selection for action.

This idea was first articulated by Alan Allport and Odmar Neumann in papers in the 1980s, and their work deserves to be read more carefully (I think it is unfortunately neglected). If I have made any contributions here, it was to bring their ideas more in contact with philosophical questions about the nature of agency.

Having now written a book on attention, or at least submitted a complete draft, I wish I had approached the question, “What is attention?”, a bit differently with my empirical colleagues. As I have given detailed arguments elsewhere for the selection for action view, let me just highlight how I would now approach the issue with a skeptical scientist.

Begin with the idea that there are some pretty clear phenomena that got us thinking about attention in the first place in vision (substitute your favorite modalities as you wish): that we can direct people’s gaze, that we search for things, that our attention can be pulled to something (“captured”). This provides our intuitive starting point, an area where we have the most agreement. Let us then consider some experimental paradigms that tap into these intuitively attention involving tasks: visual search paradigms, where you are asked to find a target among lots of distractors; spatial cueing paradigms, where a cue directs your gaze or captures it; object tracking paradigms, where you are asked to follow a subset of indicated targets. There are others of course, but I think each of the ones I have mentioned capture something of our intuitive starting point regarding attention.

The central point then is this: for each of these attention paradigms, there is a fundamental assumption, namely that the task defines some relevant targets, and that where the subject selects those targets to perform the task, the subject is attending to the target. This is to say that there is a sufficient condition for attention that is built into these paradigms, paradigms that echo our intuitive starting point. So what I would say to my empirical colleagues, if I could travel back in time, is that to the extent that they have investigated attention experimentally, they have used paradigms that implicitly endorse this sufficient condition:

If S selects X for task T, then S attends to X.

It is not hard to convert “task” to “action”, and then we have one part of the selection for action definition of attention:

S’s attending to X is S’s selecting X for action

Notice that this account allows for unconscious attention, something that people find paradoxical. I once had a paper rejected from a journal and one of the central objections in the otherwise terse referee report was that of course, attention is conscious! That’s another matter in itself that we can discuss in comments.

I think the sufficient condition is highly plausible, but it is the necessary condition that gives people pause:

If S attends to X, then S selects X for action

It seems clear to many that there are obvious counterexamples to this claim. A natural counterexample is attentional capture, and here, you might think that the drawing in of attention is automatic. Given how I have defined automatic in contrast to control in my second post, namely via the absence of intention, it seems that attentional capture (the grabbing of attention) doesn’t involve selection for action. So, a loud bang that grabs my attention is not selected for action.

It is important to emphasize that “action” here is very broad. It includes not only bodily action but mental action as well. Moreover, actions of the relevant sort need not be intentional actions, but also simple behaviors. Given the latter point, I think that the case of attentional capture is not an obvious counterexample, for there are typical responses to attentional capture, namely orienting. There is good reason for this, at least from pop teleology, namely that a sudden stimulus is of likely behavioral significance and that a good thing to do in the face of it is to investigate further. So, loud sounds typically induce an orienting response: What was that? Where was that?

I’m sure others will have more to say about this worry and will undoubtedly have other worries as well. But let’s leave it at that for now.

Psychologists have long tied attention to some selective process, but the challenge is to say what sort of selection. After all, lots of natural processes can be described as selective, but not all of them instantiate attention. Some, like Jesse Prinz, emphasize selection for working memory. To my mind, this seems unduly restrictive since it seems to me clear that there can be attention that serves action but does not require working memory, say your tracking a single object. Of course, working memory is very close to action since it is memory for work.

Neuroscientists speak of different effects of attention on neural activity including the change of gain (signal amplification), receptive field remapping (a type of filtering), biasing of competition for neuronal resources (a type of modulation of activity) and so on. Yet the neural effects of attention are fairly diverse, and it is arguable that none of them are uniquely implicated in attention. What is true is that to investigate each of these phenomena, neuroscientists must use task-based attention paradigms where by defining a task and target, they can carefully control their subjects’ behavior and, where the subject correctly performs the task, they can infer that their subjects are attending as they should. The task-based approach to attention is to me prior to any neural-based approach.

The functionalist account of attention I endorse, that of Allport and Neumann, seems to me to provide a nice framework for approaching empirical questions about how attention is implemented in the brain, and selection for action is, undoubtedly, multiply realized by the neural effects I just noted. But if this is right, then there is a clear answer to what attention is, one that we can see resonating with empirical and intuitive conceptions of attention.

Suggestions for some reading: Allport’s discussion is found in his “Selection for Action: Some Behavioral and Neurophysiological Considerations of Attention and Action” while Neumann’s is found in his “Beyond Capcity: A Functional View of Attention”. These are both to be found in this volume. I should note that in conversation, Allport does not hold the identity claim. I discuss this point and defend selection for action in my “Attention is Selection for Action” in this volume, which has many other nice papers on attention. Jesse Prinz’s account can be found in his recent book The Conscious Brain, and if you are interested, I have reviewed the book here. Christopher Mole has perhaps the most worked out philosophical theory of attention, an adverbial account: Attention is Cognitive Unison. Chris and I agree in our task-centered orientation (Chris’s book, I just noticed, is on STEEP discount at the link given…you should take advantage and pick up a copy). John Campbell deserves special recognition in pushing attention beyond questions about whether attention is a gatekeeper for consciousness in his important book, Reference and Consciousness. There are many others worth noting but too numerous to do that here.  I will the work of a growing number of philosophers on attention in my forthcoming volume on attention called (obviously!): Attention.



  1. Very interesting, but it seems to me that it would be possible to take this farther. Brain studies have been telling us that attention is not simply selection for action, it is itself a type of action. There seems to be substantial overlap between mechanisms for directing attention and motor control mechanisms. Notably, there is evidence that directing attention involves “activity bumps” in the superior colliculus that are comparable to the activity bumps that control saccadic eye movements. Also, people with Parkinson’s disease have difficulties sustaining attention that are comparable to their difficulties in sustaining movement; drugs that generate hyperactive movements also generate hyperactive attention-switching; etc.

    • Wayne Wu

      Thanks Bill

      I do agree that attention can be an action, in which case, you have selection for its own sake, so to speak or for certain purposes such as focusing or highlighting (this is speaking metaphorically, which will do for now). I’m sure someone might press on this…

      The question of what to make of “overlap” as you put it is worth teasing out more. For example, a notable account of attention is the premotor theory which can come in various grades, but the strongest grade is that the preparation of eye movement, with an eye towards some of the points you make, is constitutive of visual spatial attention. I think that the empirical evidence is pretty clearly against this version of the view. In the Prinz review I link to above, there are two references you might be interested in on this topic, one that lays out in systematic terms evidence against the premotor theory and one that summarizes the sorts of issues you raise.

      I think it seems to be true that the preparation for eye movement has some important connection to visual spatial attention, but that story still requires some sorting out. So, I think the stuff you point it is suggesting of some interesting connections.

      Note, of course, that we are largely speaking of spatial attention in the visual system. The question of broadening out to other modalities and other forms of attention (feature, object) is an important one too.

  2. Interesting stuff. A couple of questions…

    One: do you take it that this applies both to “bottom up” attention (e.g., pop-out) and “top down” attention?

    Two: operationally speaking, it is hard to get at attention without some behavioral indicator (e.g., reaction time), but I would be wary of inferring that behavior is constitutive of attention. What seems most central to attention, intuitively speaking, is that (neural or cognitive) information processing resources are devoted preferentially to certain types of stimuli, to the detriment of other types of stimuli. This likely evolved to help us behave in more adaptive ways, but I can attend to X in the service of multiple different actions. In all those instances of attending, the common factor is the differential internal processing, not any particular behavior.

    • Wayne Wu

      Thanks for these comments, Eric!

      On 1, yes, I do take it to apply to bottom up attention as well as top-down. The capture case I discuss in the post was meant to “capture” the bottom up case as well. Though “pop-out” is often goal-directed (the relevant items that pop out are in some clear way task related). I mention a little of the empirical controversy on this in my automaticity/control post in comments in an exchange with Wayne Christensen, but you might be familiar with the literature (Yantis, Theeuwes and others). I’d be happy to say more on this, if you aren’t satisfied with the points made in the post (they were quick!).

      On 2: Why are you wary about the constitutive connection between behavior and attention? I’m genuinely curious.

      I gave the beginnings of an argument for the sufficient condition (the missing step is moving from tasks to action in the broadest sense). The main challenge is the necessary condition, but I think I can answer the obvious putative counterexamples. So, you’d have a necessary and sufficient condition for attention.

      Note: “constitutive” aside, I think recognizing the sufficient condition is actually very important, and would provide something of an organizing principle for investigating attention, empirically. For any given task, what are the neural processes that implement selection of stimuli for performance of that task. When people worry about “attention” meaning too many things, the sufficient condition, extracted from experimental practice, is a good antidote.

      Note, my claim is not that there is any *particular* behavior that is constitutively tied to attention, as you suggest in the last sentence. And I think you and I need not disagree that there is “differential internal processing” in cases of attention.

      Here is the central question for any theorist who has a stake in saying what attention is or what is necessary for it: why are information processes “devoted preferentially to certain types of stimuli, to the detriment of other types of stimuli”? I take it that this “preferentialness” is just a selectivity, and as I suggested, not all forms of selectivity suffice for attention.

      So, attention is selectivity (of information processing) of a specific sort. Challenge: what is that specific sort? Prinz says for working memory, many people say for consciousness. I say for action. That is the common factor in the cases you note is “selection for performance of action”.

      If you’re not satisfied with any of these options, then just to turn the question back to you: what would you say differentiates attentional selectivity?

  3. Wayne, thanks for the helpful response. You wrote:
    “So, attention is selectivity (of information processing) of a specific sort. Challenge: what is that specific sort? Prinz says for working memory, many people say for consciousness. I say for action.”

    Any of the above, and more (probably). While you are focusing on the results of attention, what it could be used for, I would claim that selective processing is the only thing all these uses have in common (and the mechanisms of this are to be determined empirically, and likely depend on brain area, and type of attention (e.g., spatial versus object)).

    That said, I do think *if* you make the decision to analyze attention in terms of its consequences, then one quite reasonable end point is behavior. But then I think of someone paralyzed having surgery, where the anesthesia didn’t work to make them unconscious, and they are attending to (or trying very hard to not attend to), the pain in their abdomen as their appendix is removed. There is attention in that case, I would want to say.

    I’m just concerned about identifying X with what X is used for, at least in this case.

    • Wayne Wu

      Hi Eric:

      I have to apologize for my quick replies…as in previous posts, I’m trying not to write lectures! So, this is diverting for me.

      For me, the main divide is between attention as for action (behavior) and attention as for consciousness. I think the memory proposal gets subsumed under the former. But I’m not sure if there are more that are plausible as answers that emphasize function. That is, we have really only two plausible functionalist accounts of attention: action/behavior and consciousness (or variants of these).

      Question: Is all selective processing in the brain, attention? Should we say that all changes in information flow such that something is favored over another count as attention?

      The counter-thought is that if one thinks of the brain as a massive information processing organ, then there will be many cases of selective processing by neurons that don’t implicate or constitute attention. There’s just too much informational selectivity going on. This is why the functionalist construal is important; it provides a basis for constraining descriptions at more basic levels, and isolating those forms of selection that are attentional.

      This is a riff on David Marr’s response to Barlow’s Neuron Doctrine as given in Chapter 1 of Marr’s book, Vision. In Marr’s terms, the computational level (the what it is for, what it does) is in a sense prior in the order of explanation.

      Your paralysis case is nice. It’s a nightmare scenario: I’m supposed to be anaesthetized, but unfortunately, I’m not. So, I feel the pain and it really captures my attention even as I try to ignore it. What “selection for action” is going on here?

      Well, what is it not to be able to ignore it? One possibility is that I can’t stop thinking about it: wow *that* hurts; why do I still feel *that*; yow!!!! (internally). These are automatically driven thoughts (in the sense of automatic I defined in post 2). The relevant selection for action is selection that automatically drives internal behavior in this case, cognition about the pain in a broad sense of cognition.

      • Eric Sotnak

        Another possible response to the paralysis case is that the pain is, in fact, being selected for action, but the ability to act on it is absent. The same would apply to phantom pain. When an amputee attends to a phantom sensation in the absent limb, her brain could be seen as still performing the function of alerting her to damage or discomfort in the missing limb (thereby allowing her to prepare to act so as to protect it), but the system has been compromised and is not functioning properly.

        • Wayne Wu

          Thanks Eric:

          That’s an interesting reply, and seems to me plausible. Paralysis could, in the case at issue, just mean inability to generate the final output, even if much motor preparation still continues. I’d have to have an anaesthesiologist weigh in on the actual mechanisms of the drugs used for surgery (my brother in law is one, so maybe I’ll ask him next time I talk to him). Knowledgeable readers can weigh in, but this option you provide certainly seems to be reasonable to me (especially as it’s in my defense!).

  4. Hi Wayne. The comment I started turned into a post in it’s own right, so this is a pingback of sorts (https://rsbakker.wordpress.com/2013/08/26/attention-all-attention-skeptics/). My question has to do with how the heuristic nature of your intentional characterization of ‘task-based approaches’ impacts your claim that neural-based approaches conceptually and operationally presuppose your characterization. How does a folk characterization of an experimental format bear on the scientific target of that format?

  5. Wayne Wu

    Hi Scott:

    Thanks for that extended post. I think you characterize the relevant aspect of my view exactly spot on. Your point, to summarize it here, is to highlight the contrast between what I called the task-based versus the neural-based models and question whether I’m correct in terms of the priority of the former. As you note, I suggested that the neural mechanisms that we think are implicated in attention are so implicated, from the experimenter’s point of view, because they are discovered within the context of a well-defined experimental task. That’s the reason for the priority of the task-based case.

    You could imagine the following situation happening: that neuroscientists identify the same mechanism in all attention tasks, call it N. Then you might just say that it’s N, the neural mechanism, that counts as attention and not the selection for action bit. Here’s a replay of the role/realizer distinction regarding functional accounts of mental states. I’d be surprised, however, if one found the same N, but that’s an empirical question. That there are many different mechanisms underlying attention is one reason, I suspect, for growing empirical skepticism. But if a single mechanism is what you are after to unify a phenomenon, then skepticism might be a natural endpoint if there turn out to be too many different ones.

    It’s because of my skepticism on a single mechanism that I would be inclined for the role account, that attention is selection for action. This is to acknowledge the possibility of multiple realization. But it’s consistent with this account that there is an interplay between the neuroscience and the more abstract characterization.

    I do think that minimally, the role is a heuristic that guides one’s investigation at the mechanistic level. Without it, there would be no reason to attribute a neural property to attention, none at all. So, we can agree there, I believe.

    Consider gain modulation, the amplification of a signal by a neuron that has a stimulus it prefers in its receptive field. So, as you record from that neuron, sometimes it spikes at a high rate, sometimes it doesn’t. Why even think that this has anything to do with attention? Well, in the contexts of the original experiment, it was because the animal had to perform a task which required selection of the stimulus, and when the stimulus was selected, the modulation of gain was seen and not, when the stimulus was not selected. It is a heuristic, but an explanatory necessary one. The metaphysics, we can debate further.

    • This makes sense to me at the operational level, but don’t you run the risk of confounds extending it to the conceptual level? I can think of a number of problems, but here’s one: If what we call ‘attention’ turns on multiple mechanisms, it seems safe to assume that it (at least possibly) possesses multiple roles (which simply remain implicit) as well, which in turn suggests we would be better served having *no* consensus definition. Conceptual pluralism might be exactly what’s needed. Multiple parties possessing multiple definitions would minimize the chance of missing hidden functions.

      Given the immaturity of cognitive science, this will likely be the case anyway. But it seems to significantly transform the terrain of your prescriptive claim – unless I’m missing some caveat.

      • Wayne Wu

        You’re right that I should enter a caveat. I am exuding perhaps overconfidence in the conceptual advance of identifying the sufficient condition and then suggesting that it is necessary to (something that is harder to argue for). But let’s say that this goes through, then you have the conceptual advance of identifying necessary and sufficient conditions for attention, a form of functionalism about it.

        I agree, if attention has multiple mechanisms, then there can be different upshots for each mechanism. But what unifies the mechanisms, given this? Perhaps selection for action. I want that to be our working hypothesis. It has never been, but I’ve tried to make the case for it.

        That said, in the end pluralism might be the best bet. This has been true for other notions (look at the taxonomy for memory, poor souls!). But I feel more confident about attention.

        From my point of view, the selection for action account has not yet been given a fair shake, despite fairly clear presentations by Allport and Neumann. So, minimally, the post is to motivate running with it, with the caveats which I am now entering.

  6. Interesting post, Wayne! I like your perspective on these issues. I’ve struggled with similar questions for years, in somewhat similar terms, and never felt I saw them through very adequately. It seems clear to me that there is some sort of cognitive attention (of the Jamesian sort) that one clearly *can* dedicate to phenomenology and that also it’s clear that one cannot dedicate *sensory* attention to phenomenology. For the question to be interesting, “attention” has to be understood in neither of those ways; but how it is to be understood, then? Harman, Tye, Shoemaker, etc., were never very sophisticated in their treatment of these issues in their the classic statements about “transparency”.

    It does seem to me that there is something inherently local in (putative) introspective attention: It is, by its nature, confined it time and space — to your own mind in the specious present — that distinguishes it from at least *typical* cases of intellectual attention. But where exactly to go with that, I have never been quite sure. Some sort of self-monitoring story invites itself: introspective attention involves the dedication of central cognitive resources to “System 2” thinking about contents of the central workspace, maybe. But I’ve never been a fan of that sort of model.

    • Wayne Wu

      Hi Eric!

      Thanks for your thoughts.

      “It seems clear to me that there is some sort of cognitive attention (of the Jamesian sort) that one clearly *can* dedicate to phenomenology and that also it’s clear that one cannot dedicate *sensory* attention to phenomenology.”

      I’m grappling with this in the other post, which you might also check out. We say that attention is sensory with respect to the objects of attention, and this is how transparency folks you cited spell out the issues. I focus on what I call the “process” reading, in which case we can think of attention as grounded in experience but is individuated by its informing cognitive attention. The label “sensory” is confusing though, in the end, perhaps not misleading. So, the last post is meant as an answer to the issue you raise here.

      I share your puzzlement. I don’t have a concrete thing to say except to wonder if there is in fact, some empirical purchase to be gained, once we think clearly about attention. I have some sympathies for the more pluralistic account you give of attention (if I remember correctly!) in the Smithies and Stoljar volume, though I”m exploring a more unified element for perceptual consciousness. So, perhaps we’re at or near the same place?

  7. I’d like to enlarge on an earlier comment I made in this thread. Sometimes it is impossible to get anywhere without first getting definitions straight, but sometimes it is more productive to be simpleminded.

    Consciousness falls into the first category. The folk concept of consciousness — at least, the mental picture that many people hold — is that the brain is a sort of house, and consciousness is the person who lives in it. Any observation that can’t be aligned with that mental picture will not be recognized as an observation of consciousness. But it’s extremely unlikely that neuroscience will give us any such thing. Therefore right from the start we face a conceptual boggle that needs to be at least partly resolved before we can even get started empirically.

    But the story concerning attention is arguably different. The folk concept of attention derives from the physical behavior of *looking at* something. We think of attention, at the folk level, as a sort of covert looking, a sort of looking that occurs inside the mind and is not visible from the outside. If that’s another boggle, then we will need to fix it before we can get started, but the question is, is it a boggle?

    There are reasons for thinking that it is not, which I touched on in an earlier comment. There is substantial empirical evidence that the neural mechanisms for directing attention share many properties with mechanisms for motor control. In other words, the brain activity that we see when a person shifts attention is similar in a number of respects to the activity that we see when a person moves part of the body. Thus we apparently have something that closely matches the folk concept and that we can straightforwardly study.

    In short, it seems to me that unlike the study of consciousness, the study of attention is an area where the best thing we can do is to let empirical findings shape our concepts, rather than trying to make concepts lead the way. Of course our concepts will evolve as our understanding increases, but I don’t see why we need to fear the result of that process.

Comments are closed.

Back to Top
%d bloggers like this: