Notes on Georgi Gardiner's 'Attunement: on the cognitive virtues of attention'
Unlike many, I think we’re in a golden age of analytic philosophy. At least in the parts of the discipline with which I’m most familiar, every year brings a whole range of new interesting work. There are no doubt many factors that contribute to this, and I’m not the sociologist needed to discern them, but one thing I’ve noticed is that analytic philosophers are talking about many more features of our world that they were before. Each year in epistemology, there is a paper that calls us to think about another aspect of our epistemic lives. In part, at least, I think this is due to a shift from purely ideal epistemology to the study of the non-ideal features of our epistemic lives, that is, those we have precisely because we’re limited creatures with limited time. Where before we perhaps looked at the beliefs we have and asked which ones are supported by our evidence, which count as knowledge, and which justify action and assertion, now Jane Friedman draws our attention to the process of inquiry that leads us to have the evidence we do, Julia Staffel considers the reasoning processes that lie between evidence and belief, Rima Basu points to the possible ethical evaluation of the beliefs we have, Carolina Flores and Jack Idris Sagar talk of our epistemic styles, and, in the paper I’m thinking about at the moment, Georgi Gardiner makes the case that our habits of directing our attention in various ways have significant epistemic consequences and can be evaluated epistemically.
Gardiner’s aim in the paper is not to give a theory of how we should direct our attention, but rather: (i) to highlight how important different ways of directing it can be in our epistemic lives; (ii) to argue that it is our habits and patterns of directing attention that are best evaluated epistemically, rather than any particular instance, and so to suggest that virtue theory is best placed to undertake this evaluation; (iii) to point out that it is not only individuals but also the groups they form that have habits of directing attention that can be evaluated epistemically, and to argue that such evaluations are ‘non-summative’ so that each individual can be doing well while the group is performing poorly, and some individuals can be doing poorly while the group is performing well; (iv) to highlight aspects of our life in this information-rich era that can only be understood by thinking about the ways in which we direct our attention and in which our attention is directed by others who wish to manipulate our thinking.
In fact, Gardiner’s paper connects well with Friedman’s work mentioned above. In Friedman’s central example, she imagines herself standing in front of the Chrysler Building in Manhattan counting its windows because she wants to find out how many it has—we might imagine she’s a glazier who hopes to win the contract to replace them, or she’s a trivia fanatic who’s keen to know whether this building or the Empire State Building contains more glass. Friedman imagines that, in some sense, she has a whole bunch of evidence that would support various beliefs, but she doesn’t form those beliefs because doing so would require cognitive effort that is better spent on focusing her attention on the windows in the building. For instance, she might have overheard a conversation between a couple next to her about their financial situation, but she doesn’t form the belief that they’re in financial need, even though her evidence supports that; out of the corner of her eye, she might have seen a gentleman in a yellow trilby pass by, but she doesn’t form the belief that there’s a gentleman in a yellow trilby nearby, though again her evidence supports it; and so on. Since her evidence supports these beliefs and yet she doesn’t form them, she violates a putative epistemic norm that says you should believe a proposition if, and only if, you have sufficient evidence for it. Yet we don’t consider that she’s failed in any epistemic way; she’s just directed her epistemic machinery in a direction that better serves her epistemic and practical ends. Friedman uses this apparently unproblematic violation of the putative norm to set up a clash between the epistemic and the zetetic.
In a sense, Friedman’s is already a case in which we epistemically evaluate how someone directs their attention. It also suggests, slightly in tension with Gardiner’s thesis, that quite often we do evaluate particular instances of directing attention, rather than simply patterns. Elsewhere, I’ve suggested that we should approach Friedman’s puzzle using the tools of credal epistemology and, in particular, the ‘value of information’ approach that begin with David Blackwell and I. J. Good, and which has recently enjoyed a resurgence of interest, particularly in cases in which your evidence isn’t luminous to you and where the value of the credences you form on the basis of your evidence is epistemic rather than practical (see here, here, here, and here). And I’d like to suggest that the same approach might be useful when we turn to the epistemology of attention.
This approach is teleological or perhaps consequentialist, rather than virtue-theoretic. I absolutely agree with Gardiner that sometimes the troubling aspects of how we direct attention only become apparent when we look at the patterns of our behaviour, but I don’t think this should send us down a virtue-theoretic route, since we can also evaluate individual instances of how we direct attention, as we do in the case of Friedman’s window counter, and we can evaluate patterns of behaviour that are smaller and more local than the character traits that virtue theory considers, such as when someone, during a week-long period of uncharacteristically heightened anxiety, focuses only on the possibilities that lead to mortal danger.
My summary of the Blackwell-Good framework here must be a little telegraphic if it’s not to swamp the post, but I set it out in greater detail in the piece linked in the previous paragraph. The idea is that your credences, or subjective probabilities, play two roles in your life: they represent the world, and they guide your action. How well they play each of these roles can be measured: we call the two measures the epistemic and pragmatic utilities of your credences. Having set up these measure, we can then evaluate your credences using them, but we can also use them to evaluate any other epistemic behaviour by looking to the credences it leads you to form and asking for their epistemic and pragmatic utility. For instance, we can evaluate your evidence-gathering activities by asking for the epistemic and pragmatic utilities of the credences you end up with after you’ve updated on the evidence they lead you to gather; we can evaluate your reasoning processes by looking at the epistemic and pragmatic utilities of the credences that issues from them; we can evaluate your epistemic style, perhaps, by consider how much epistemic and pragmatic utility it produces; and, I suggest, we might evaluate instances and habits of directing attention by asking for the epistemic and pragmatic utility of the credences you end up with.
What is the epistemic or pragmatic utility of a credence? Here, I’ll follow accuracy-first epistemology and take the epistemic utility of a set of credences to be their accuracy: so, for a credence in a true proposition, the epistemic utility is higher for higher credences and lower for lower ones; and for a credence in a false proposition, it is higher for lower credences and lower for higher ones; and perhaps the epistemic utility of a set of credences is the sum of the epistemic utilities of the individual credences in it. Over the past 25 years, formal epistemologists have shown how a number of formal epistemic norms for credences can be established by appealing to such measures of epistemic utility.
Following Blackwell and Good, we take the pragmatic utility of a set of credences to be the utility of the outcomes of the choice that the credences would lead you to make. For instance, if you must choose whether or not to build flood defences for your city, the pragmatic utility of your credences will be determined by which option they will lead you to choose and the utility of the outcome of your choice. So if they lead you not to build them and the city does flood, the pragmatic utility of the credences will be the pragmatic utility of a flooded city. So a notable thing about the pragmatic utility of credences is that they are determined partly by the world you inhabit, partly by the decisions you’ll use them to make, and partly by how much utility you give to the different possible outcomes of those decisions.
With all this in hand, we can turn now to attention. An easy case to start: Friedman’s window counter. There are two ways in which she might direct her attention: to the windows of the building or to the conversation of the couple beside her. Given limited time and cognitive capacity, she cannot do both. The epistemic utility of each of these options is just the epistemic utility of the credences she’d have after she had updated on the evidence she would get from it. For most measures of epistemic utility, you can give greater weight to those propositions that are more significant to you; the ones for which it matters more to you whether you are accurate or not about them. So, our window counter can give greater weight to propositions concerning the number of windows and less weight to those concerning the couple’s financial situation. And, as long as she’s a reasonably reliable window counter and so will get pretty accurate credences by focussing on that, she’ll do better in expectation by the lights of epistemic utility by directing her attention to the windows.
The case is even clearer for pragmatic utility, for we might imagine that she knows she’ll face some decision in the future for which she will be much better equipped with accurate information about the number of windows than with accurate information about the couple’s financial situation. If she is a builder hoping to win the glazing contract, she’ll face the decision how much to charge for that job, and it will need to be low enough to win the contract and high enough to cover her costs, whereas we might assume she’s unlikely ever to encounter the couple again, and that outcome of most decisions she’ll face will be independent of their financial situation. So, in expectation, she’ll do better by the lights of pragmatic utility by directing her attention to the windows.
So, how we direct our attention can determine what evidence we get and what evidence we get determines what credences we have, and those can be evaluated for their epistemic and pragmatic utility. And, as Gardiner notes, our attention is manipulable, and so we can be coaxed into having particular credences our manipulator wants us to have, even when they present us with all of the available evidence. For instance, knowing we have only a limited capacity to absorb new evidence, they might begin by presenting the evidence that favours the conclusion they wish us to draw, only presenting the counter-evidence much later on, after our view has been fixed, we’re tired and distracted and unable to take on new information. Or they might simply present the evidence favouring their conclusion in the most eye-catching format, while the counter-evidence is displayed mundanely, knowing again that with limited time, we’re more likely only to attend to the former.
But attention can also affect which credences you use in your decision-making, or that you report in discussions. Gardiner has two interesting cases: in one, the mother of a vegan daughter focuses a lot of attention on the possibility that her daughter isn’t getting enough iron from her diet; in the other, an investor focuses a lot of attention on a meteorological possibility that would affect the wisdom of a particular investment. Depending on the details of the case, these ways of directing attention could have good or bad consequences. In Wayne’s case, the possibility on which he focuses has a very low probability, but would be extremely bad if it were to come about. Since people are often inclined to simply ignore such possibilities in their decision-making, Wayne’s excessive attention to it might compensate for this error, leading him and his investment group to give that possibility exactly the weight it should receive. On the other hand, it might be that this possibility was already accorded the right amount of weight and Wayne’s fixation leads it to receive more weight than it should, which leads the investment group to withhold their money from a pretty decent prospect.
The sort of teleological approach I’m suggesting here can also help us to understand the interesting phenomena of group attention that Gardiner discusses. She points out that groups can contain individuals who direct their attention in ways that are bad for them but end up helping the group, and there are groups in which all individuals direct their attention in good ways, but the group itself performs poorly. This recalls what Conor Mayo-Wilson, Kevin Zollman, and David Danks call the independence thesis in the philosophy of science, which says that epistemically rational individuals might form epistemically irrational groups and epistemically irrational individuals might form epistemically rational groups. For instance, each individual in a group might focus on just one feature of a situation, leading them to collect evidence in a very biased fashion; but if each focuses on a different feature of the situation, when they share their evidence in the group, this leads to a very balanced and rich set of evidence, since there will be no overlap between the pieces of evidence gathered. Again, in this situation, we can take a teleological approach: each individual will end up with credences that have lower epistemic and pragmatic utility in expectation than they would have had if they had sampled the evidence more evenly; but the group will have end up with credences that have higher epistemic and pragmatic utility in expectation than they would have had if the individuals had all sampled the evidence evenly and ended up with a total body of evidence that was even but more shallower, since they all overlapped.
In general, I think these teleological approaches are useful in non-ideal epistemology. There, we look at different components of our epistemic practices—how we inquire, how we reason, how we direct attention, and so on. Each of these practices leads ultimately to the credences we have. So we can evaluate them by the expected epistemic or pragmatic utility of the credences they’ll lead to. Non-ideal epistemology recognises that we are limited, and so the range of epistemic practices we can engage in is limited—Friedman’s window counter can attend to the evidence about the windows or to the evidence about the couple’s financial situation, but she can’t do both. And the teleological approach is well suited to that as well: when you assess a particular practice we engage in, you simply compare it only with others that are actually available to limited creatures like us. So you don’t criticise the window counter for not attending to both sources of evidence, since this is simply beyond her capacities. While some of the main versions of this teleological approach to epistemology has often focused on general norms for ideal epistemic agents, it is ideally suited to answering the sorts of questions that interest non-ideal epistemologists as well.