Defeating looks

Lilacs are my favourite flowers. In May, when Stockholm is lush and green, they bloom in abundance in parks and gardens everywhere. But they rarely make an appearance in the flower shops. When I spot some, I never resist. Once, I emerged from a florist’s with a bunch of white lilacs in my arms. The friend who was waiting for me outside the shop exclaimed: “But you said you wanted purple! Why did you get these?” Only then realizing my mistake, I explained: “Well, these looked purple in there.”

Examples like this can easily be multiplied; they are witnesses of a folk-psychological practice of citing looks as reasons for both beliefs and actions. This is a practice most naturally understood as that of citing our perceptual experiences as reasons (cf. a. o. McDowell 1994, p. 165f; Glüer 2009). The practice is natural, deeply entrenched, and forms a central part of our intuitive or pre-theoretical conception of our cognitive relation to the world surrounding us. Accommodating the intuitive reason providing role of perceptual experience is a desideratum on any successful account of perceptual experience.¹

In the lilac example, I cite the way the flowers looked in the shop as a reason for an action, for buying them. That, of course, is due to folk-psychology’s aversion to pedantry. What the flowers’ looking purple in the shop was a reason for, was first and foremost the belief that they are purple. Which, together with my desire for purple lilacs, provided me with a (practical) reason for buying them. Even though my friend is a philosopher, too, that’s not the kind of explanation I would normally want to entertain her with. Here, however, we are interested in precisely the epistemic (or theoretical) part of the story of the lilacs: Experience’s reason providing role for first order empirical belief.

Now, my friend could have told me that that particular flower shop always is lit in strange, purplish light. And surely, had I known that, I would have been more careful in judging the color of my lilacs. My reason for believing that they are purple would have been undermined by my belief that the light is purple. Since I had no good reason to believe that the light was strange, however, I can not only provide a perfectly good explanation for my forming a false belief—rather, by citing my experience I can show that, in a certain sense, I was quite justified in forming it. This illustrates some features of the reason providing role we are interested in: the reasons in question are defeasible, and even if they are misleading, they can be good, or justifying, in a certain “subjective” sense. This is the sense in which I will use “justification” in this paper. The idea will be that experience is a “prima facie justifier”, i.e. that the reasons it provides are good unless defeated.

Of course, we cannot assume that there is just one way of accommodating the reason providing role of experience. A natural thought is that precisely how an account would go about this will depend in part on what an experience is according to that account. In this paper, I am going to assume intentionalism, the view that perceptual experience is a conscious mental state with representational content.² On a suitably uncontentious construal of propositions, as well as of attitudes towards them, intentionalism is the view that experience is a propositional attitude (cf. Byrne 2005, p. 245). Intentionalism would prima facie seem very well suited to accommodating reason providing: Having propositional content, the thought is, would allow experiences to provide their subjects with reasons in pretty much the same way other propositional attitudes do. To make intentionalism hospitable to reason providing is, however, more difficult than one might think.

Construing experiences as propositional attitudes does take the wind out of the sails of a classical Davidsonian argument to the effect that experiences don’t provide reasons or justification. The argument concludes this from the premises that having propositional content is a necessary condition on being a justifier, and that experiences don’t have propositional content.³ The claim that having propositional content is necessary for being a justifier is controversial. What shouldn’t be controversial, however, is that there are justifiers that have propositional content. Moreover, it shouldn’t be controversial that there are justifiers that not only have propositional content, but for which having such content isn’t just a feature accidental to their justificatory powers. For such justifiers it is at least natural to expect that what they provide justification for is (at least in part) a function of their contents. Let’s call such justifiers “propositional justifiers”. Having propositional content, experience would seem eligible for being a propositional justifier.

But being a propositional justifier requires more than just having propositional content. It is quite plausible to think that what disqualifies for instance desires from providing epistemic justification is that they are not the right kind of attitude: To provide (epistemic) reasons for (first order, empirical) belief, an attitude needs to be of a certain kind; more precisely, it needs to be “belief-like” in a certain respect. There is widespread agreement that experience, if it is a propositional attitude, is an attitude of this belief-like kind. Such attitudes have what Searle calls “mind-to-world direction of fit” (cf. Searle 1983, p. 7ff). They represent the world as actually being the represented way, as actually fulfilling their condition of correctness or truth. Such states have been called “committal” (Burge 2003, p. 452) or “stative” (Martin 2002, p. 386f), but the maybe most suggestive metaphor characterizing these attitudes is that of assertion. Pryor thus calls them states “assertively representing a proposition” (Pryor 2005, p. 187f).

The easiest way of making sure that experience is the right kind of propositional attitude would, of course, be to construe it as belief. Since belief uncontroversially is the paradigmatic propositional justifier, if experiences were beliefs, there would not be any additional questions as to what it is about the experiential attitude that qualifies experience for being one, too. This is not usually considered an option, however; usually, it is taken for granted that there are well-known knock-down arguments against belief theories of experience.⁴ But if it indeed were the case that we know that experiences are not beliefs, it would seem at least a matter of some urgency to pinpoint what precisely it is that makes an attitude into a propositional justifier—without by the same token making it into a belief. This is not a question that has gotten all that much attention so far, but there is reason to think that it is at least surprisingly hard to actually come up with informative necessary and sufficient conditions here that are met by all the right kinds of attitudes (including experience) (cf. Briesen 2015). Let’s call this the “attitude problem”. The attitude problem is an unsolved worry for intentionalists thinking of experience as a propositional justifier. The attitude problem is premised on the assumption that experiences are not beliefs, however. There thus is a way around it: Construing experiences as beliefs.

Even if we had a solution to the attitude problem, our troubles would be by no means over. If experience is a propositional attitude of the right kind, the basic thought now is, it can provide its subject with reasons in pretty much the same way other such attitudes do. But the paradigm case is that of belief. When it comes to belief, it is natural to think that a belief that p provides its subject with a reason for believing q in virtue of an inferential or evidential relation between p and q. Belief thus is a paradigm case of what I shall call an “inferential justifier”. When it comes to empirical reasoning, the inferences in question don’t have to be deductively valid—rather, the underlying relations can be thought of as relations of probabilification. This would seem to provide a plausible understanding of the defeasible nature of empirical reasons. And again, all of this looks just right for experiential reasons; reasoning from experience would seem to be empirical reasoning par excellence, and clearly the reasons provided would seem defeasible.

In previous work, I have suggested, and argued for, a doxastic account of experience. A main part of the argument is that construing experience as a kind of belief makes it comparatively easy to accommodate its reason providing role (Glüer 2009). I have also argued that if experience is to be a propositional justifier providing defeasible reasons, its content is best construed as different from P-belief content (Glüer 2016b). Meeting this “difference requirement” at the same time allows us to escape the classical knockdowns for belief theories of experience, as these invariably, if implicitly, are premised on the same content assumption (cf. Glüer 2009). My suggestion is to use “phenomenal” or “looks-contents” instead of standard contents. What results is a form of intentionalism, “phenomenal intentionalism”, that is non-standard both in construing experience as belief and in construing its content as phenomenal. In this paper, I shall further develop the phenomenal belief account. In particular, I shall work on the epistemology of perception that can be built upon it.

I shall proceed as follows: In the next section, I shall give a brief outline of the phenomenal belief account and illustrate both how phenomenal contents allow it to escape the classical anti-doxastic arguments and how stuttering provides independent reason for the difference requirement. In Sect. 3, I shall spell out the relevant notion of an inferential reason and look at some objections to the claim that experience is an inferential justifier. After that, I shall in Sect. 4 look into the question of when experiential reasons are good reasons and argue for what I shall call “Pollockianism” about experiential reasons: They are good unless defeated. I shall also argue that this version of Pollockianism allows for straightforwardly using standard Bayesianism for modelling defeasible reasoning from experience. In Sect. 5, some comments will be provided on Bayesian worries that have been raised for other Pollockianisms (such as dogmatism), and in the final section, I shall investigate the worry that “phenomenal Pollockianism” involves an inferential gap in need of bridging.

As worked out so far, the account I have suggested is an account of visual experience. It construes visual experience as a (peculiar) kind of belief with “phenomenal” or “looks-contents”.⁷ Phenomenal intentionalism agrees with standard intentionalism that the objects of visual experience are ordinary material objects, and that experience ascribes properties to these objects. Phenomenal contents differ from standard contents with respect to the kind of property experience is construed as ascribing to its objects.

Think of an experience of purple lilacs again. According to standard intentionalism, this experience has a content of the form o is purple. The basic idea behind phenomenal or looks-contents then simply is to instead construe the content as of the form o looks purple. Thus, phenomenal contents can be generated from standard contents by looks-modifying their predicates. The more general idea, however, is to generate looks-contents from standard contents by looks-operating on them: Where standard intentionalism assigns the content p to an experience neutrally described as an experience as of p, phenomenal intentionalism assigns Lp.⁸ In what follows, I shall call all such contents “looks-contents” or “Lp contents”.

So, the claim is that visual experience is a kind of belief, and that these beliefs have looks-contents. Two questions immediately arise: What kind of belief are experiences? And what kind of properties are the properties that figure in looks-contents? One possible answer to the first question is that (visual) experiences are precisely those beliefs that have looks-contents. Alternatively, some other characteristic could be used to specify which of the beliefs with looks-contents are experiences, for instance their sensory phenomenology. Thus, visual experiences might be those beliefs that represent their looks-contents in a visual way. Whether it is plausible to think that phenomenal contents can be shared between experience-beliefs and other beliefs depends at least in part on our answer to the second question, i.e. on what the relevant phenomenal properties are.

We can approach that question by asking more generally what appearance properties are—if we think they are properties of material objects. Most people think that appearances, so understood, are relational properties. Shoemaker, who I think was the first to propose that experience represents appearance properties, offers two main suggestions (cf. Shoemaker 1994, 2000). Occurrent appearance properties are properties such as that of presently causing an experience of a particular phenomenal kind \(k_{p}\) in a subject S. Dispositional appearance properties are properties such as being disposed to cause \(k_{p}\) experiences in subjects of a particular kind (under particular circumstances). Shoemakerian appearance properties thus are relations involving a “phenomenal” element: relations to (phenomenal kinds of) experience or sensations. Alternatively, appearance properties could be understood as “non-phenomenal” properties—as relational properties objects have in virtue of their intrinsic properties and features of the perceptual situation or perspective that are not dependent on the subject, for instance.⁹ Whether construed as phenomenal or not, those who think that experience does represent appearance properties usually think that experiences represent both these properties and “intrinsic” sensible properties such as color or shape.¹⁰ As far as I can see, only Antony (2011) and myself suggest that the only properties “present” in our experience are (phenomenal) appearance properties. For reasons going beyond the scope of this paper, I think the appearance properties represented in visual experience are best construed as phenomenally relational, roughly but not quite along Shoemakerian lines.¹¹ If that is right, more precisely, if what experience ascribes are occurrent phenomenal properties, then it might be possible to argue that experiences are the only beliefs doing so. But I think that nothing I shall argue in this paper hangs on whether looks are phenomenally relational or non-phenomenally relational.¹² I shall therefore not pursue this question any further here.

What I shall do instead in the remainder of this section is run us through two arguments against construing experiences as beliefs. The first one is one of those classical knockdown arguments I mentioned in the introduction: the argument from known illusion. The other is the “stuttering inference argument”. The second will take us directly to the problem of defeasibility.¹³

The argument from known illusion takes as its starting point certain very stable illusions such as the Müller-Lyer illusion. Looking at a Müller-Lyer diagram, the lines look unequal in length. And they do that regardless of whether you know that they are in fact of the same length or not. That is, the illusion does not go away just because you have a background belief to the effect that the lines are of the same length. They look as if they were of unequal length no matter what you believe and how used you are to the illusion. This kind of “belief-independence” (Evans 1982, p. 123) is also known as the “modularity” (in Fodor’s sense) of perceptual experience. And it provides us with a powerful argument against classical belief theories of experience, theories that identify experience either with belief, or with the formation of belief, and possibly even those that identify experience with dispositions to belief (for a classical belief theory, cf. for instance Armstrong 1968).

In its simplest form, the argument proceeds from the premise that an experience as of the Müller-Lyer lines’ unequal length is a belief. The second premise is that the subject has background knowledge to the effect that the lines are of the same length. From this, we derive the conclusion that the subject has outrightly contradictory beliefs. This conclusion is bad, but not clearly absurd. However, in cases of known illusion the subject can be fully aware of both experience and background belief. Moreover, there need not be anything irrational or otherwise malfunctioning about the subject of a known illusion. This does strike me as absurd. Beliefs just don’t behave like that. The argument from known illusion concludes that experiences are not beliefs.¹⁴

But the argument from known illusion has a hidden premise: the same content assumption. That the subject of a known illusion comes out as having contradictory beliefs follows from a belief theory of experience only on the assumption that for instance the experience you have when the Müller-Lyer lines look unequal in length to you has the content that the lines are of unequal length. If we instead construe the experience as ascribing looking unequal in length to the lines, there is no contradiction.

As I understand it, phenomenal conservatism is a novel form of foundationalism with fairly far-reaching ambitions of epistemological unification: seemings are supposed to provide non-inferential justification for beliefs of quite different types (cf. Moretti 2015b, p. 298; Piazza and Moretti 2015, p. 1273).¹⁶ Now, this form of foundationalism would seem particularly attractive precisely because the foundational justifiers it works with are not beliefs. This is in strong contrast to classical foundationalism where the foundational justifiers are beliefs: introspective beliefs about perceptual experiences. Suggesting that seemings could just as well be construed as beliefs (with seems-contents) would thus be a bit of a party-spoiler.

As far as I can see, the arguments for the claim that seemings are not beliefs are of the very same kind as the classical anti-doxastic arguments against belief theories of experience: they proceed from the observation that things can seem F to you even if you know perfectly well that they are not. Phenomena like known sensory illusion, but also the appearance of correctness that for instance the naive comprehension schema of set theory is said to possess even for those who know that it is not correct (cf. Huemer 2007, p. 31f), are cited as witnesses here. I am not totally convinced of the relevant similarity between these two kinds of cases, but even if we grant it for the time being, what we get is precisely a sort of generalized known illusion argument. And the generalized argument relies on the (generalized version of the) same content assumption just as much as the perceptual version does. Once the suggestion that seemings could just as well be understood as beliefs (with seems-contents) is on the table, we at least need additional argument to rule it out.¹⁷ \(^{,}\) ¹⁸

Pointing out that arguments like the argument from known illusion are premised on the same content assumption only shows that they don’t prevent us from construing experiences as beliefs—if we are prepared to give up that assumption. It’s another thing to argue that experiences are best construed as beliefs. That’s a fairly ambitous goal, especially as (EC), and thus the same content assumption, indeed are rather natural for an intentionalist. In this paper, I shall work towards this goal by developing the argument that the phenomenal belief account not only provides a good account of reason providing but also of the justificatory role of experience. As I said already in the introduction, thinking about the reason providing or justificatory role of experience from an intentionalist perspective pushes towards phenomenal intentionalism not only because a doxastic account provides a simple answer to the attitude problem. It also puts independent pressure on the same content assumption. This brings us to the argument I have called the “stuttering inference argument”.

It is generally held that defeasible reasoning is reasoning from premises that leave open the possibility of the conclusion’s falsity. Such inferences can be retracted because of new information. The classical example of such non-monotonic reasoning is Tweety, the bird (cf. Reiter 1987, p. 149). Learning that Tweety is a bird provides you with a defeasible reason for believing that Tweety can fly. If you subsequently learn that Tweety is a penguin, your reason for believing that Tweety can fly has been defeated. But you still believe that Tweety is a bird. Above, I said that inferential relations can be relations of implication or of probabilification. In order to be defeasible, they need to be of the latter kind. More precisely, they need to be such that the conditional probability of the conclusion on the premise is not equal to 1. But if anything is an example of a conditional probability of 1, it is the conditional probability of p on p. So, if we construe experiences as inferential justifiers for the most basic first-order empirical beliefs based on them, and we give them a standard semantics, we seem bound to make a hash of defeasibility. I shall get back to the defeasibility problem in Sect. 4.

But why should we take stuttering to put pressure on the same content assumption—rather than on the idea that experience is an inferential justifier? At this point, all I am saying is that stuttering can equally well be taken to be a reductio of either premise. Thus, it puts pressure on both. This pressure adds to the pressure on the same content assumption exerted by the attitude problem. To my mind, that is sufficient to motivate an investigation into giving up the same content assumption, be it ever so intuitive. After all, that experience provides inferential reasons is ever so intuitive, too. To boost our confidence that experiential reasons are inferential, I shall in the next section spell out the relevant notion of an inferential reason in a bit more detail, and look at some objections to thinking that experience actually provides any such reasons.

Such reasons rationalize further beliefs. If p is a reason for believing that q, then p must be such that it—from the subject’s own perspective—speaks in favour of believing that q. This rationality, too, is subjective in the sense that the explanation provided by citing the reason is such that it shows that something spoke for believing q from S’ own perspective.

At the same time, however, S’ perspective needs to be recognizable as a perspective by others: Believing that p needs to be such that it would provide any subject with a reason to believe q. This objective, or at least intersubjective, aspect of reason providing can only be secured by an underlying, objective relation of inferential or evidential support between p and q. In order for the belief that p to provide its subject with a reason for believing that q, there needs to be a valid inference (of some sort) from p to q.²⁰

So far, we are concerned only with prima facie reasons. The need for an underlying, objective relation of inferential or evidential support between p and q can also be seen by reflecting on the fact that these reasons can be good reasons: they can justify beliefs based on them. But not all reasons subjects in fact base their beliefs on justify those beliefs. The distinction between good and bad reasons, between those that do, and those that do not, justify beliefs based on them, requires there to be an (objective) relation of inferential or evidential support between the believed propositions themselves, a relation that obtains independently of any particular attitudes subjects take to them.

Two more (negative) characteristics are crucial to this conception of reasons: Having reasons for one’s first-order empirical beliefs does not require forming them by means of conscious inference. Having a reason for believing that q does not require anything regarding how that belief is formed, not even that it actually be formed at all. And if S does form the belief that q on the basis of their belief that p, this inference does not have to be performed by means of any kind of conscious deliberation, or act of reasoning, to be justified.

Connected with this is the observation that having or providing reasons does not require the possession of second order states. What we are concerned with here are relations between first order propositional attitudes and their contents. A creature has reasons in the required sense as soon as its beliefs and actions can be explained by means of its further beliefs and desires. The capacity to think about these beliefs and desires is not required, and even less the capacity to think about these beliefs and desires as providing the reasons in question. What is required, however, is a certain minimal, subjective rationality. There are no reasons explanations, be it for beliefs or actions, unless a creature’s beliefs actually to some minimal but significant degree instantiate the basic inferential or evidential relations objectively obtaining between the propositions that are the contents of their beliefs.²¹

If we think of perceptual experiences as beliefs, we can accommodate their reason-providing role by simply extending our account of how beliefs, in general, provide reasons for further beliefs to them. This requires some care, however, as we at the same time want to preserve certain peculiarities of the inferential or evidential relations between experiences and other beliefs. What’s most important here is the peculiar combination of the defeasibility of experiential reasons with the modularity or belief-independence of experience: While for instance the reason for believing that the lines are of unequal length that a Müller-Lyer experience provides its subject with usually is defeated by background belief, the experience itself is impervious to the background belief. Its “message” remains unchanged.

Before returning to the issue of defeasibility, however, I would like to take a brief look at a challenge to the very idea that experiences are reason providers in the first place. Gauker, for instance, does not find the idea intuitive or natural at all, and therefore asks if any good arguments can be provided in support of it (cf. Gauker forthcoming).

Let me start by clarifying the kind of argument I take myself to have made. It’s what we could call a “here’s a way to get the job done argument”. Such arguments start from a “job description”. My argument starts by reminding us of the everyday practice of citing the way things look as reasons for our beliefs about the world around us. I shall occasionally call this practice the “looks-practice” in what follows. This practice is pervasive and fundamental to our folk-psychology, and I take accounting for it to be a desideratum on any satisfactory account of experience. My argument then has simply been that the phenomenal belief account does precisely that. This leaves open the possibility that there are other ways of getting the job done.

What I have not been sufficiently explicit about, however, is the following: The desideratum is to account for the practice of citing looks as reasons. My account does that by construing experience itself as the reason provider. I take this to be the most natural way of understanding the practice, and have therefore formulated the desideratum as that of accounting for experience’s reason providing role. But one might try to get a wedge in here. That is, one could argue that the reasons that get cited when we cite looks are in fact not the experiences themselves, but for instance the very first beliefs we form on the basis of them.²² But even though this is true, this possibility just by itself does not detract from my basic argument: the phenomenal belief account gets the job done. Moreover, it gets it done very well, among other things by implementing the most natural understanding of the practice it accounts for. Rival accounts will have to be assessed on their merits. At the end of the day, the phenomenal belief account might turn out not to be the best overall account, not even with respect to the desideratum at hand, but for the time being, it’s alive and kicking.

By contrast, Gauker thinks that proponents of the view that experience provides reasons—including McDowell (1994), Ginsborg (2006), and Pollock (1974) and Pollock and Oved (2005)—proceed on the (implicit) assumption that there is a sense of ‘looks’ exclusively tied to reporting experience content (cf. Gauker forthcoming, p. 12). Given the nature of the argument I have tried to make, I do not think that I need to make any such assumption. My argument, it seems to me, is compatible with allowing that the looks-reports we make in the looks-practice could also be construed as reporting something other than experiences. All I claim is that we can make very good sense of the practice if we indeed construe these reports as citing experiences as reasons.

Gauker then goes on to argue that even though an interpretation of ‘looks’ exclusively tied to experience-reports can indeed be provided, on that interpretation, the proposition expressed by ‘o looks F’ does not evidentially support o is F. Proceeding by elimination, he argues, among other things, that there is no relation of probabilification between o looks F and o is F. But even if that were right for the interpretation of ‘looks’ Gauker offers, it seems pretty clear that it does not hold if ‘looks’ is interpreted in terms of phenomenal properties.²³ That, I have suggested, is the sense of ‘looks’ that figures in experience contents. And on that interpretation, it is pretty clearly the case that o looks F probabilifies o is F. That is, the conditional probability of o’s being F on o’s looking F is (equal or) higher than the prior probability of o’s being F (cf. Glüer 2014, 2016b).

Looking F is a phenomenal property, a property the possession of which by an object o is a pretty good indicator of o’s also possessing another property, that of being F. Moreover, it’s a pretty good, but not an infallible indicator—thus, the conditional probability of o’s being F on its looking F is high, but not equal to 1. And that brings us back to the claim that if we construe experiences as beliefs with phenomenal contents, they provide their subjects with defeasible reasons for first-order empirical belief. We now want to know when these reasons are good reasons—which is the topic for the next section.

According to phenomenal intentionalism, an experience-belief that Lp provides its subject with a defeasible reason to believe that p. The question now is: When does the subject of such an experience have a good or all-things-considered reason for believing p? And again, what I am interested in is first and foremost justification for believing p, whether or not such a belief is actually formed—that is, what most people call “propositional justification”.

Applying the idea to the phenomenal belief account of perceptual experience, we get the view that reasons provided by Lp experiences are good reasons for, or justify, believing that p unless defeated. Merely having a reason provided by experience is not sufficient for justification, but the “more” that is required is not more belief, or more reason for belief, but rather the absence of (good) reasons strong enough to defeat them. And just as we are concerned here with reasons subjects have, we are concerned with defeaters subjects not only do not have, but do not even have good reason to have.

As I have already indicated, I think that for perceptual reasons, evidential support should be construed as probabilification. Moreover, it should be such that the conditional probability of p given Lp is (equal or) higher than the prior probability of p—but lower than 1.²⁶ That is, the experience content must be such that its truth is compatible with the falsity of the proposition the probability of which it raises. Once thinking in terms of looks, this seems simple and natural: Of course, the proposition that the lilacs look purple can be true without the lilacs’ being purple. And analogously for the Müller-Lyer lines. And the reason their looking purple gave me for believing them to be purple is easily seen to be defeated by having good reason to believe that the light is purple. But that only undermines the connection—not the way the lilacs look.

What we see here is how well-suited the phenomenal belief account is to using a Bayesian model for the epistemology of perception.²⁷ Given the assumption that experience is a kind of propositional attitude, it is only natural to expect that its justifying powers are (at least in part) a function of its content. And it is equally natural to expect that experiential content can be treated as “incoming information” on the basis of which to update your beliefs or credences.

But standard versions of Bayesianism are in fact not particularly hospitable to this natural idea—at least as long as we stick to standard versions of intentionalism. For one thing, standard Bayesian models strictly speaking do not even apply to reasoning from experience as standardly construed—they only apply to beliefs. Thus, it is initially quite unclear what the kind of defeat we are probably most interested in here—“perceptual undermining” (Weisberg 2015, p. 121)—would even amount to on such a model: If the model does not cover the connection, how could it model its being undermined? This is of course especially glaring if we update on p—for instance, the proposition that the lilacs are purple—by means of classical Bayesian conditionalization: p will get probability 1 from the start, and be stuck with it. But, as already Christensen (1992) noted, the situation does not really improve if we switch to Jeffrey conditionalization instead. Jeffrey conditionalization does allow us to assign a probability lower than 1 to incoming p and then update on it, but it is silent on the question how to determine that initial probability. Thus, incoming perceptual information normally should get a much higher initial probability than it should get when you already know that the light is iffy, but there are no rules for determining these initial values. In a way, that is not surprising, as the model is one for updating on incoming information, but to the extent that perceptual undermining intuitively already takes place before the model even kicks in, it is a serious shortcoming. The model seems bound to mischaracterize the realm of rational cognition by missing its real starting point.²⁸ Moreover, the situation is not easily remedied, either. Considering perceptual undermining as a diachronic process, Weisberg shows that Jeffrey conditionalization “doesn’t just fail to regulate perceptual undermining, it bungles it” (Weisberg 2015, p. 122; cf. also Weisberg 2009). Think of the lilacs again. Their actual color is independent of the lighting, and before I look at them information about the latter has no relevance to my beliefs about the former. But Jeffrey conditionalization is “rigid” and thus will preserve this independence even after I have come to believe that the lilacs are purple based on their looking purple. Thus, not even later learning about the lighting in the flower shop will have any relevance for my belief about the lilacs’ color. Jeffrey conditionalization in effect prevents what intuitively looks like perceptual underminers par excellence from even diachronically having any undermining effects.

And as long as we stick to standard intentionalism, trying to extend the model to cover the step from experience to P-belief does not improve the situation. As long as we are serious about it’s being the experience itself that provides the proposition to be updated on (and not, for instance, introspection), what we end up with is updating p on p. Whether p comes in with probability 1 or something lower then doesn’t actually matter much—either way, the connection wouldn’t be such that a defeater could undermine it. This problem is different from that generated by independence preservation: Even a defeater relevant to p cannot undermine the connection between p and p. Just adding a stuttering step thus makes a hash of perceptual undermining, too.

Our problem is that we need a relation between incoming evidence and “hypothesis” that a defeater can “get at” at all. Think of the way we talk about perceptual reasons again: It is central to the way we cite looks as reasons that looking F leaves open the possibility of not being F. Our looks-practice thus is structurally very similar to other practices involving non-monotonic reasoning. When it comes to the structure of defeat, the story of the lilacs is very much like that of Tweety: Learning that the lilacs look purple did provide good reason for believing them to be so—until I also learned that the light in the store was purplish. Just as learning that Tweety is a bird provided good reason for believing Tweety able to fly—until we learned that Tweety is a penguin.²⁹ We distort this structure if we construe the ‘looks’ used in our looks-practice as propositional attitude operators reporting experiences with standard contents.³⁰ To capture this structure, the relation between incoming proposition and hypothesis must be such that the truth of the incoming proposition leaves open the possibility that the hypothesis is false.³¹ Moreover, the relation must be such that perceptual underminers are relevant to it. The relation between o looks F and o is F seems to be of precisely the right kind. And there is an additional bonus of construing experience content as phenomenal: so construed, the actual deliverances of the senses are virtually guaranteed to be true. The phenomenal belief account thus allows perceptual updating to be modelled by classical conditionalization. And in any case, whether we prefer classical or Jeffrey conditionalization, construed as beliefs, experiences are automatically covered by Bayesian models.³²

To sum these considerations up: Defeasibility provides motivation for what I have called the “difference requirement”. To successfully model the defeasibility of the reasons experience provides for further belief, experience content and P-belief content cannot be the same. The same content assumption must be given up.³³

The difference requirement pushes us towards giving up the same content assumption, and thus standard intentionalism, at least as much as the attitude problem. These two forces are at least to some degree independent, as it is only natural for an intentionalist to think that experience content is not an idle wheel in the epistemology of perception. Even if we could solve the attitude problem without construing experience as belief, that is, it would be natural to expect experience to be a justifier whose justifying power is a function of precisely its content. The content of such a justifier impacts the epistemological machinery in a certain way. How? Well, it seems natural for the intentionalist to think that experience content should be treated as incoming information on which to update belief. Which in turn suggests that it should be possible to apply our most successful formal account of such reasoning to incoming perceptual information. As long as we stick to standard intentionalism, however, none of this seems to be forthcoming. Instead, we would have to look for a rather different model of how experience justifies.³⁴

Somewhat ironically, phenomenal intentionalism thus comes out as a rather conservative position once we consider the epistemology of perception. We give up one, admittedly intuitive, assumption, and a lot of the epistemic things we have always wanted from experience happen almost automatically.³⁵ Most importantly in the present context, experience gets to be an inferential prima facie justifier, straightforwardly modelled by standard Bayesianism.³⁶ In the rest of this paper, I shall investigate the suggested epistemology of perception a little further, and locate it in relation to some landmarks, and landmark problems, in epistemological space.

A question that can been raised with respect to Pollockianism is the question of whether prima facie justification—be it experiential or from some other source, such as seeming in general—amounts to “immediate justification”, that is, in Pryor’s words, justification “that does not presuppose or rest on your justification for anything else” (Pryor 2000, p. 519). Defenders of dogmatism claim that experiential justification is immediate, and phenomenal conservatives think that justification by seemings in general is so. And what Pryor calls “credulism” (Pryor 2013) is the position of someone who thinks that at least some prima facie justification is immediate.

But there are no such objections to the applicability of Bayesianism to phenomenal Pollockianism. However, this might seem to deliver it right into those objections that White, for instance, tries to raise for dogmatism by way of applying Bayesianism to it. The first of these problems is that the combination of Pollockianism with Bayesianism has the consequence that an experience as of p provides evidence not only for p but at the very same time for “skeptical alternatives” (of a certain kind) to p. For instance, if p is the proposition that the lilacs are purple, the relevant experience does provide evidence for believing that, but at the very same time, it also provides evidence for believing that the lilacs are white and lighted by purple light. This is so because the conditional probability of both these propositions given the evidence is higher than their prior probabilities.³⁸ In fact, which hypothesis the subject will get most evidence for depends on their priors.³⁹ This, I think, should be seen as a feature of Bayesianism, not a bug. It captures an observation, not a problem. It is simply true that if an experience provides evidence for believing p, then it also provides evidence for what we can call p’s “phenomenal alternatives”, for instance, ways the world immediately around the subject could be that would look precisely the way they look in having that experience.⁴⁰

There might be a problem, however, if what we want is not just phenomenal Pollockianism, but phenomenal dogmatism. Phenomenal dogmatism would claim that the justification provided by an experience-belief that Lp for believing p is immediate. The worry here would be exactly parallel to a second worry White tries to raise for standard dogmatism. As White shows, when updating on evidence e the probability of hypothesis p will stay lower than the prior probability of the negation of any phenomenal alternative to p. The worry then is that this shows that in order to get justification for believing p from incoming e the subject would already need to have stronger justification for believing the negation of any phenomenal alternative to p.⁴¹ But if this indeed is a worry, it is a worry for immediacy. Having no stake in immediacy, I am happy to remain a fence-sitter here.

Before concluding, I would like to consider one more set of worries concerning phenomenal intentionalism’s “going Pollockian” for its epistemology of perception. Let’s call these worries “bridge worries”. Bridge worries arise where a proposition p (or set of propositions \(p_{1},\ldots ,p_{n}\)) does provide a sufficiently high degree of probabilification for another proposition q, but this, by itself, does not seem sufficient for belief in p to provide its subject with reasons for believing that q. That something is a reliable indicator of something else might be necessary for empirical reasons to be good, or “material inferences” to be valid, but in general it’s not by itself sufficient. For instance: That Paul is red-haired, just by itself, does not seem to be any reason for believing that he is left-handed. This does not seem to change on the mere assumption that (unbeknownst to the relevant subject) there in fact is a strong correlation between red hair and left-handedness. It does seem to change, however, when the subject knows, or justifiedly believes, there to be such a correlation. To put it metaphorically, in cases like this, there is a gap between propositions p and q that is in need of bridging in order for belief in p to provide justification for believing q. Moreover, consideration of cases like this suggests that what is required to bridge the gap is not only an evidential relation between p and q but the subject’s awareness of such a relation. It might thus seem that in order to secure the kind of justification we are interested in here, subjects will have to have additional (justified) beliefs in what we can call “bridge principles” of some sort wherever an inference is material.

Now, an inference from Lp to p certainly is material. Does this mean that, just by itself, an Lp proposition is as little of a reason for p as Paul’s being red-haired, just by itself, is for his being left-handed?⁴² We tried to prevent experience’s reasons from stuttering by construing experience content as different from that of basic perceptual belief. But can this really be done—without thereby introducing a gap that will need bridging? The worry is that to prevent experience from stuttering we would have to “silence” it. That would be a bit of a dilemma (cf. Glüer 2014, p. 86ff).

Even though the worry needs to be taken seriously, I don’t think there ultimately is a dilemma here. For starters, going Pollockian does not amount to regarding probabilification, by itself, to be sufficient for justification. It also requires the absence of defeaters. Of course, the absence of defeaters would not make a belief that Paul has red hair provide a good reason for believing that he is left-handed, either. But neither was it supposed to. We endorsed Pollockianism for experiential reason providing only. Reasoning from Lp to p is relevantly different from “inferring” left-handedness from red hair, it seems to me.

Thinking of experiential reasons as Pollockian prima facie reasons is perfectly in line with our intuitive rationality judgments. Consider Laura. Laura never draws any conclusions about people’s handedness from their hair color, and Laura does not (have any (good) reasons to) believe anything that would defeat such inferences, either. Laura does not in any way strike us as odd or irrational. Quite the contrary. But the situation is rather different when it comes to experience and P-belief. Take John. In bright daylight, he looks at a book right in front of him. The book is red, and nothing obstructs John’s line of sight. There is nothing wrong with John’s eyes, and the surrounding conditions aren’t such that experience will be misleading. And John neither believes any of these things, nor does he have reason to. Nevertheless, John does not believe that the book is red. He readily acknowledges that the book does, indeed, look red to him. Asked about defeaters, he denies believing any of them. Nor does he have any good reason to. Yet, he assures us ardently that he does not believe the book to be red. This is immensely odd, and appears quite irrational.

There is a stark contrast between John and Laura: Intuitively, Laura is perfectly justified in not drawing conclusions about people’s being left handed from their being red haired, while John’s refusal to draw conclusions about the book’s redness appears utterly unjustified—in fact, it appears so unjustified that we might start wondering whether John knows what ‘red’ even means.⁴³ Some very basic inferential connections, the idea is, just need to be instantiated in a subject’s belief system for such a system to be at all recognizable as a belief system, a system of states with empirical content.⁴⁴ No bridge beliefs are required at this most fundamental level. The presence of a relation of evidential support (of sufficiently high degree) between the relevant propositions is sufficient.

Ultimately, however, the presence of such a relation depends on the way the world actually is. If the world “cooperates”, inferences from Lp to p will be reliable: In the absence of defeaters, we might say, such reasoning is ultimately warranted by its reliability. This much externalism, it seems to me, is unavoidable in the theory of empirical justification. Whether we can trust what the senses “tell us” in the end remains hostage to the world. And again, that is just how it is, it seems to me—it’s the human predicament.⁴⁵

This does not have anything in particular to do with the sometimes perceived need for the subject to be aware of their reasons in the strong sense of having (justified) beliefs about what is a reason for what. The latter is one of the horns of Bergmann’s “dilemma for the internalist” (Bergmann 2006): Requiring awareness of reasons in the strong sense, Bergmann argues, generates the need for an infinity of awareness attitudes stacked on top on one another. Note that this kind of regress is vicious in the sense that it might overtax our cognitive capacities. The kind of regress I am generating is of the (more clearly vicious) kind where even an infinity of further principles will not get you where you wanted to get. The further requirement of thinking of the premise in (S) as a reason for the conclusion is not needed in order to generate the regress. The other horn of Bergmann’s dilemma is the so-called “subject’s perspective objection”, an objection first raised by BonJour (BonJour 1985, p. 41ff) against externalist notions of justification: Bergmann claims that an internalist not requiring subjects to have strong awareness of their reasons will end up with having reasons for beliefs the truth of which is merely accidental from the subject’s own perspective. This is the kind of objection motivating the demand for the subject’s having some background belief as to the relevant connection for instance in cases like the red hair case. Now, phenomenal Pollockianism isn’t purely internalist in any case—as I have developed it, phenomenal Pollockianism is built upon on a relation of probabilification the presence of which depends on how the world actually is. So, Bergmann’s dilemma isn’t ours. The subject’s perspective objection might of course nevertheless apply. But we implicitly already saw why it doesn’t: If it was possible that the truth of p systematically appears to be totally accidental to someone having experience-beliefs that Lp, nothing would hinder them from being just like John in the example above—without appearing subjectively irrational at all. But that is not the case. Rather, for someone like John it is not clear that they even have a perspective on the relevant part of the world.