Perceptual Precision

Adrienne Prettyman
Bryn Mawr College

 

Abstract: The standard view in philosophy of mind is that the way to understand the difference between perception and misperception is in terms of accuracy. On this view, perception is accurate while misperception is inaccurate. However, there is some evidence (albeit controversial evidence) that perceptual experience actually involves widespread inaccuracy. I add to that evidence in the paper. Then I point toward a way of understanding the difference between perception and misperception, not in terms of accuracy, but in terms of precision. That is, I argue that perceptual experience is designed to enable more fine-grained discrimination among the properties that are most useful for action, even if that involves inaccuracy.

 

Target Presentation from Adrienne Prettyman

1. Introduction

It is a standard view in philosophy of mind that the distinction between successful perception and misperception (like hallucination or illusion) can be understood in terms of accuracy. A subject misperceives when the world is presented to her in perception as being a way that it is not. However, recent evidence suggests that inaccuracy is surprisingly common in everyday perceptual experience. In some of the cases that I will discuss, perception even seems to be successful despite being inaccurate. In this paper, I present and add to the evidence that inaccuracy is widespread in ordinary visual experience, and point to one possible way forward. I suggest that the way to understand the difference between perception and misperception is not in terms of accuracy, but in terms of perceptual precision: the fineness of grain of discrimination among the actual properties of an object. On the view that I propose, perception is successful when it enables us to discriminate among those properties that are most useful for action, even if that involves inaccuracy.

Most philosophers distinguish perception from misperception by appeal to veridicality. Unlike successful perception, misperception involves a mismatch between appearance and reality, or a failure to accurately represent an object or its properties. For example, in distinguishing perception from illusion, Smith (2002) writes that an illusion is:

“any perceptual situation in which a physical object is actually perceived, but in which that object perceptually appears other than it really is, for whatever reason.” (p. 23)

Likewise, Fish (2010) introduces the distinction between successful perception and illusion as follows:

 “Fully successful cases of perception — cases in which an object is seen and seen correctly or ‘as it is’ — will be termed perception or, sometimes, veridical perception… In contrast, illusion refers to cases in which an object is seen but seen incorrectly or ‘as it is not.’” (Fish p. 3)

Macpherson (2009) similarly writes:

“Traditionally, philosophers have contrasted perception with sensation. Perception was taken to be a process that involved states that represented — or that were about — something. For example, typical visual experience had at the beach might represent sand, crabs, or the blueness of the sea. These experiences might accurately represent the beach or misrepresent it, if undergoing illusion or hallucination.” (p. 503)

These are just a few of many examples. As I will show, this way of drawing the distinction is challenged by empirical cases of inaccurate perception that don’t intuitively fit into the same category as standard illusion cases. The cases that I have in mind are ones in which perception is inaccurate but precise. As I’ll understand it, a perceptual experience is accurate iff it specifies some way that the world is, and the world actually is that way; it is inaccurate otherwise. In contrast, a perceptual experience is precise iff it enables the subject having that experience to make fine-grained discriminations among the actual properties of an object. It may seem at first glance that precision requires accuracy, but this turns out not to be the case, as I will show. In order to make fine-grained discriminations among the actual properties of an object, we don’t need to perceive those properties accurately. In some of the cases that I will discuss, precision is even improved by inaccuracy.

One advantage of appealing to precision is that it can help us to understand the distinction between successful perception and illusion even if inaccuracy turns out to be widespread in ordinary perceptual experience. There is some evidence that perception is frequently inaccurate, though how to best interpret this evidence remains controversial. For example, in the study of visual attention, Carrasco and colleagues have suggested that attention modulates visual appearance in a variety of ways, such that attended objects appear “bigger, faster, earlier, more saturated, stripier” (Block 2010 p. 41) than unattended ones. As Carrasco and colleagues write, “attention augments perception… by emphasizing relevant details at the expense of a faithful representation of sensory input…” (2004, p. 1162). Carrasco and colleague’s interpretation of the evidence has been challenged by psychologists (see Schneider & Komlos 2008 and Schneider 2011; see also Anton-Erxleben, Abrams, and Carrasco 2010 & 2011 for a response) and philosophers (see Block 2010, Nanay 2010, Speaks 2011, Stazicker 2011, Wu 2011, Ganson & Bronner 2013, Block 2015, Prettyman 2016, & Watzl forthcoming). It is especially controversial that attention’s effect on appearance make perceptual representations “less faithful” or inaccurate, rather than simply more determinate. Nevertheless, it is a live possibility that attention augments appearance in the way that the original experimenters suggest.

Watzl (forthcoming) develops Carrasco and colleagues’ suggestion, arguing that the best explanation of the evidence is one on which we frequently misrepresent properties in perceptual experience. On Watzl’s view, while perception aims to present the subject with an accurate representation of the world, attention aims to make that representation usable. When the aims of perception and attention conflict, perceptual experience will be determined by a tradeoff between accuracy and usability. As a consequence, we should expect subtle but widespread misrepresentation in ordinary perceptual experience.

Arguments for widespread inaccuracy in perception also have historical precedent. As Simmons (2008) has argued, versions of this view can be found in Descartes and Malebranche. Malebranche, for instance, claims to show “that the world you live in is not at all as you believe it to be, because actually it is not as you see it or sense it” (as quoted in Simmons 2008 p.81). The reason is that perception is narcissistic rather than veridical (a term that Simmons borrows from Akins 1996). It aims to tell the subject about the world as it relates to the preservation of the body, and not as it objectively is. As Simmons summarizes, Descartes and Malebranche argue for a view on which the function of perception is to “show us what bodies are like not as they are in themselves, but as they are related to us and, in particular, to our self-preservation” (Simmons p. 84, emphasis original). As a result, perception frequently misrepresents the objective world.

A contemporary defense of the view that perception is narcissistic is developed by Akins (1996). Using thermoreception as an example, Akins shows that perceived temperature is exaggerated for temperatures that are more likely to cause damage or discomfort, such as rapid change caused by placing cool skin in warm water, or submerging one’s head (as compared to one’s hand) in a cold mountain lake. This exaggeration comes at the cost of a consistent and accurate representation of temperature. Rather than conclude that thermoreception is an inept or defective veridical system, Akins thinks this shows that thermoreception is not aimed at producing an accurate representation of the temperature of objects in the first place. Perceived temperature is often inaccurate because thermoreception is a narcissistic rather than a veridical system. It aims to present temperature as it is related to our needs and interests (such as our interest in being comfortable or being warned of damage), and not as it objectively is. Under the assumption that thermoreception is narcissistic, we should expect perceived temperature to be distorted. In particular, we should expect distortions that serve to better communicate the relevance of objective temperature for our own comfort and safety. Akins shows that this is exactly what we find across a variety of cases.

Some philosophers may be happy to accept that thermoreception is not veridical, but resist the general conclusion that our perceptual systems do not aim at accuracy. After all, philosophers have long distinguished between properties like temperature or taste, which are in part dependent on the perceiver, and objective properties like shape or motion, which aren’t. In section 2, I will add to the evidence that inaccuracy is widespread in ordinary perceptual experience, and show that even objective properties like distance, slant, and motion are distorted in perceptual experience. I will also suggest a way forward in sections 2 and 3. On the view that I will suggest, even if inaccuracy is widespread in ordinary perceptual experience, misperception is not.

 

2. Distinguishing Accuracy from Precision

It is fairly uncontroversial that accurate representation of the world is not necessary for precise action. Consider an example from Warren and Whang (1987). In order to pass through a doorway, a subject needs to perceive its size relative to her own. She could be inaccurate with respect to the size of the door and the size of her body, yet accurately perceive that the door is taller than she is. Likewise, a watchmaker looking through a magnifying glass may adeptly fix a watch even though she misrepresents the size of its parts, and a spear fisherman might catch a fish despite distortion caused by the water (Durgin et al. 2010). A dramatic real-world illustration of the dissociability of successful action from accurate perception comes from prism goggle experiments in the 1960s. Subjects fitted with light inverting lenses were able to learn to perform complex actions, like driving a car or fencing, even though they perceived the world as being upside-down (Kohler 1961; Linden et al. 1999).

Misrepresentation of an object’s properties can even be useful. Consider the case of color. If color anti-realism is true, representing objects as colored may nonetheless enable us to perform visual search more quickly, or make discriminative judgments more accurately (Millikan 1984; Mendelovici 2012). For example, representing a berry as red may help me to quickly and accurately distinguish ripe fruit from the surrounding leaves, even if the berry does not have the property of being red. This is a hypothetical case in which perceptual inaccuracy increases perceptual precision: the ability to make discriminative judgments among the actual properties of an object. Misrepresenting the berry as red enables me to identify those objects that have the property of being a berry and being ripe. My point is not to defend color anti-realism, and nothing in what follows will depend on this example. I mention the case of color because it is one area where philosophers have discussed the possibility that widespread and reliable inaccuracy might support successful action.

These examples provide precedent for the view that widespread misrepresentation does not always impede — and can even support — successful action. In the next two sub-sections I turn to empirical evidence that strengthens support for this view. A series of studies in psychology suggest that there is systematic and widespread misrepresentation in perceptual experience. These studies further suggest that misrepresentation serves a useful function: increasing perceptual precision for the sorts of discriminative judgements that are useful for the subject in her environment.

 

2.1  Misrepresentation of Space

Depth perception involves the representation of egocentric distance, or distance from an observer to a target, and surface orientation, the slant or tilt of an object. There is evidence that both are distorted in perceptual experience. This evidence suggests not only that visual representations of space are systematically inaccurate, but also that these distortions support successful action by increasing perceptual precision.

Take the case of egocentric distance. It is widely accepted that subjects underestimate when asked to judge their distance from a target (I will say more below about why this should be taken to indicate misperception rather than just misjudgment) (Gilinsky 1951; Da Silva 1985; Loomis, da Silva, Fujita, & Fukushima 1992; Foley, Ribeiro-Filho, & da Silva 2004; Li, Phillips, & Durgin 2011). This has been demonstrated using both verbal (Foley et al 2004) as well as non-verbal measures. For example, Gilinsky (1951) constructed a scale of perceived distance by asking subjects to sit overlooking a grassy field and placing stakes in the ground at intervals that appeared equal from the subject’s perspective. As the intervals became objectively farther away from the subject, Gilinsky had to make them objectively larger in order to retain an appearance of equality. She concluded that apparent distance is not a linear function of objective distance, and instead is distorted such that distances farther from the subject are underestimated. Similarly, Toye (1986) and Wagner (1985) conducted experiments in which subjects were asked to compare the lengths of two intervals on the ground plane. When one of the intervals lay “in depth,” or along the sagittal plane, they tended to judge that it was shorter than a “frontal” interval of equal length, which lay along the frontal plane. Using a similar paradigm to Toye and Wagner, Loomis and colleagues (1992) asked subjects to adjust the length of a frontal interval to match an in-depth interval placed in the distance. They also found that subjects adjusted the frontal length to be consistently smaller than the in-depth interval, a result predicted by the view that distance is underestimated in visual perception.

A second example of widespread distortion in perceptual experience comes from perception of optical slant. A battery of studies provide evidence that subjects’ perception of the slant of hills is exaggerated (I will say more below about why this exaggeration should be considered a perceptual effect) (Kammann 1967; Ross 1974; Proffitt et al. 1995). For example, Proffitt and colleagues (1995) found that subjects’ overestimated the slant of a hill when they were asked to provide verbal estimates, or to adjust a representation of a cross-section of the hill.[1]  They suggest that these distortions of perceived slant serve to make subjects more sensitive to small changes in incline, like those they encounter in everyday life. The exaggeration of perceived slant has been replicated using other non-verbal tasks like angle-bisection, in which subjects are asked to judge which of two legs of an ‘L’ shape, presented on slanted ground, appeared longer (Li & Durgin 2010; see also Ooi, Wu, & He 2006; Wu Ooi, & He 2004). Durgin and Li (2009) further found that downhill slopes appear shallower than they are when viewed from the edge, rather than steeper, and that subjects are systematically inaccurate in their perception of the declination of their own gaze.

While it is fairly uncontroversial that subjects misjudge surface orientation and distance across a wide variety of viewing conditions, psychologists disagree about how to understand this pattern of findings. It is important for my argument that the distortion of distance and slant is genuinely perceptual. One possible explanation of the data, however, is that subjects’ judgments are distorted, but their perceptual experience is not. Firestone and Scholl give a compelling statement of this worry. When looking for top-down effects on perception, it is important to distinguish between effects on “…what we see or instead only our inferences or judgments made on the basis of what we see” (Firestone & Scholl 2015, section 4.2, emphasis original). When experimenters rely on verbal judgments as evidence that a subject misperceives the world, they risk conflating these two judgments. For example, in the case of perceived distance, a subject may perceive the distance to a target accurately, but underestimate when she makes a judgement about how far away the target is. That is, subjects may perceive the world accurately even if a judgment based on that perception is is inaccurate.

It is worth looking more closely at the reasons why many psychologists favor a view on which the distortion of distance and slant is genuinely perceptual, and not a mere distortion of judgment. The first reason is based on introspection. As Firestone and Scholl point out, perceptual effects that we can directly experience on our own are more plausible than those that we cannot. Consider an example. We can experience the Muller-Lyer illusion for ourselves. One line looks longer than the other even though both lines are objectively the same length (see Figure 1). If we looked at the Muller-Lyer figure and could not experience the effect, we should be suspicious of the claim that the figure induces a genuine perceptual illusion.

 

The Muller-Lyer arrows.

FIGURE 1. The Muller-Lyer Illusion. Though the line on the top looks longer than the line on the bottom, they are objectively equal in length.

 

There is reason to think that distortion of slant and distance are like the Muller-Lyer illusion in that we can experience them for ourselves. In his introduction to his 1995 paper on optical slant, Proffitt describes a recent drive through the mountains of Virginia. The roads seemed incredibly steep to him, even though he knew that the law required that roads incline at no more than 6 degrees. Other anecdotal reports of distorted space come from hikers and skiers. Climbers report that distant uphill slopes look “impossibly steep,” while skiers report that, when looking down from the top of a slope, other skiers sometimes appear to ski uphill (Ross 1974). Similarly, Durgin and colleague’s work on optical slant was inspired by a perceptual effect that researchers observed firsthand. They noticed that hills appeared shallower than their objective slant when observed from the edge (Li & Durgin 2009), an effect which didn’t fit with the dominant view of slant perception. According to the dominant view, hills should appear steeper from the top due to a general compression of distance along the lines of sight. The observation that hills appear shallower from the edge led experimenters to re-investigate the accepted theory, and eventually to propose a new model of slant perception. These are just three examples. The abundance of introspective evidence suggests that the distortion of space is genuinely perceptual, and not merely a distorted judgment.

Another way to identify genuinely perceptual effects is to use “performance-based measures” in which “subjects’ success is tied directly to how they perceive the stimuli” (Firestone & Scholl 2015 section 4.2.3). In the study of perceived slant, researchers don’t rely on verbal report alone. Several studies have shown that verbal judgments of slant are largely consistent with nonverbal measures, like angle bisection and aspect ratio tasks (Durgin, Li, & Hajnal 2010; Li & Durgin 2010). A related worry for verbal report is that subjects may exaggerate their answer to fit the experimenter’s hypothesis (Durgin, Baird, Greenburg, Russell, Shaughnessy & Weymouth 2009). For example, if a subject has guessed that the experimenters think hills look more shallow from the top than the bottom, they may exaggerate their estimates to fit that hypothesis. To address this issue, some experimenters (Li, Phillips, & Durgin 2011; Durgin & Li 2011) debrief subjects by asking them to report their beliefs about the aims of the experiment as well as the strategies that they used to perform the experimental tasks. Experimenters who include these questions can then compare the data from subjects who guessed the experimenter’s aims with the data from those that did not. In the studies cited above, the significance of the findings did not depend on subjects guessing the experimenter’s hypothesis.

If we accept that the evidence supports a genuinely perceptual distortion of visual space, then we are faced with a new question. What explains our success at navigating the world, given that perceived spatial properties (like distance and slant) are so often inaccurate? The problem arises only if we assume that successful action requires accurate perception. An alternative is that inaccuracy, far from being a problem that the visual system must overcome, is sometimes useful for guiding action. Akins (1996) has made a similar point in her argument against the view that perception is veridical. She suggests that vision is not a faulty veridical system, which aims at accuracy but frequently falls short. Rather, vision is best understood as narcissistic rather than veridical. It aims to tell us about the world relative to our own interests and activities, rather than present the world as it objectively is. Given this possibility, a more interesting question concerns whether systematic inaccuracy in vision, like that suggested by the study of slant and distance, has a useful function for supporting action.

One way that systematic inaccuracy might support action is by enhancing perceptual discrimination through scale expansion (an idea put forward by Li & Durgin 2009 and Durgin & Li 2011). Scale expansion is a “coding strategy” in which a subset of a range of values is coded more “densely”, thereby magnifying that subset. Scale expansion theory starts from the assumption that information is coded in the brain, for example, as when the visual system compresses information from the retina into units that are used in further visual processing. It then posits that it would be useful to code the most common units in a way that exaggerates the difference between them, thereby expanding that region of the scale. An example from Li and Durgin helps to flesh out this proposal. Consider the perception of your own gaze. The range of gaze declination from 0-60 degrees is a candidate for scale expansion because human gaze tends to fall straight ahead or tilt downward within a 60 degree range (Li and Durgin 2009). That is, for most actions that we perform, it is not useful to look overhead or down at our feet — though this will of course vary for individuals given the sorts of tasks they might need to perform. Coding the typical range more densely than, say, the range from 60-90 degrees, enables greater precision for the values that matter most for typical human action.

 

Li & Durgin’s (2011) illustration of the scale expansion theory of depth perception.

FIGURE 2. Li & Durgin’s (2011) illustration of the scale expansion theory of depth perception. In the diagram γ and β represent the objective declination of gaze and optical slant, respectively. In perceptual experience, perceived gaze (γp) and perceived slant of the ground (βp) are each overestimated compared to the objective declination of gaze and objective slant. This results in a compression of perceived distance.

 

Scale expansion provides an elegant explanation of the evidence that perceived distance and perceived slant are distorted (see Figure 2).  As we have seen, evidence suggests that both perceived gaze and perceived optical slant are systematically distorted. As Figure 2 illustrates, when both the declination of gaze and optical slant are overestimated, perceived distance will be compressed — an effect that has been widely demonstrated (Gilinsky 1951; Da Silva 1985; Loomis, da Silva, Fujita, & Fukushima 1992; Foley, Ribeiro-Filho, & da Silva 2004; Li, Phillips, & Durgin 2011). A further consequence of these distortions of spatial perception is that differences in slant relative to the horizontal plane are magnified (a result also suggested by Proffitt et al. 1995). While Li and Durgin’s experiment does not test this directly, the magnification of slant could help subjects to be more sensitive to small deviations in slant relative to flat ground. The research summarized in this section suggests that we are not only inaccurate with respect to spatial properties in the world around us, but also that inaccuracy serves a useful function: it makes us more precise in our judgments of objective slant for the angles that matter most for action. In the next section, I turn to an area of research where psychologists have directly tested the claim that inaccuracy can increase perceptual precision, and offer stronger support for that claim.

 

2.2  Misrepresentation of Motion

As you walk through a room, the pattern on your retina changes, even if everything in the room is perfectly stationary. This change in the retinal image due to self-motion is called visual flow (Gibson 1966). Since Wallach (1987), it has been widely accepted that visual flow appears to slow down while walking. A series of studies aim to quantify this effect, and to explore its consequences for discriminative judgment. The effect of walking on visual flow provides further support for the view that systematic misrepresentation in vision has the function of enabling greater perceptual precision.

 

Durgin’s (2009) illustration of a subject walking in a virtual hallway.

FIGURE 3.  Durgin’s (2009) illustration of a subject walking in a virtual hallway. The spots on the walls and floor flow past the walker at a rate controlled by the experimenter. In a speed discrimination experiment, a subject is presented with a hallway with spots moving at one speed, and then, after a short break, a second set of spots moving at either the same or a different speed. In each trial, the subject is instructed to either walk or stand still. The subject’s task is to judge which spots appear to move faster. While walking, the stimulus appeared to move about 50 cm/sec slower than while standing.

 

In a 2007 study, Durgin and Gigone aimed to quantify the amount by which visual flow slows down while walking (see also Durgin, Gigone, & Scott 2005), and to explore whether distortion of perceived motion increased perceptual precision. In their first experiment, subjects were presented with a virtual moving hallway using an immersive head mounted display (see Figure 3). The virtual hallway had spotted walls which were moving past the subject at a rate controlled by the experimenter. In the first condition, subjects walked while viewing the virtual hallway, while in the second condition, they stood still while viewing. Each condition lasted just a few seconds. Subjects were then asked to judge which hallway had been moving faster. On average, the speed of the virtual hallway while walking had to be 50 centimeters per second faster than those presented while standing in order for the subjects to reach the point of subjective equivalence, or the point at which the hallways appeared to move at the same speed. That is, if you present a subject with an object moving past them at a fixed rate, that object will appear to be moving more slowly when the subject is walking than when they’re standing still.

To assess the speed of visual flow, Durgin and Gigone asked subjects to report which stimulus appeared to move faster. They were careful to ask for reports about the appearance of motion, and not judgments about the motion of objects themselves (see Durgin and Gigone 2007). This instruction is important because it helps to address a similar worry to the one raised in the previous section on slant and distance: subjects may judge that an object is actually moving faster (or slower) than it appears. Just as we can be aware of a shadow on a surface while simultaneously seeing the wall as uniformly colored, so too subjects can learn to distinguish between the appearance of motion and the judgment that an object is in fact in motion.

As in the case of slant and distance, there is reason to think that misrepresentation of motion has a useful effect. Reducing the apparent speed of objects while walking enables greater perceptual precision, such that deviations from expected visual speeds can be discriminated in a more fine-grained way while walking than while standing still. To demonstrate this, Durgin and Gigone repeated their study, this time presenting subjects with flow speeds that were very close to the visual flow induced by walking. As before, they compared subjective judgment of flow speeds across two trials, walking and standing still. They found that subjects were better able to discriminate visual flow speeds that were near to walking speed when they were walking as compared to when they were standing still. The experimenters suggest the following explanation for the increase in subjects’ perceptual precision during self-motion. By reducing the perceived speed of objects while walking, small differences in speed become proportionally larger compared to the absolute perceived speed, which in turn enables more precise discrimination among actual speeds. In this way, the systematic misrepresentation of the speed of objects while walking helps subjects to perceive the motion of objects in a more precise way. It does this by making the subject more sensitive to small deviations from the expected effect of their own self-motion. As Durgin writes:

“…in the control of action, perceptual precision (the fineness of discrimination among actual values of a variable) is more important than perceptual accuracy (direct correspondence between the perceived and actual value of variable)” (Durgin 2009, p. 43).

Since precision is more important than accuracy for navigating the world, perceptual systems misrepresent in ways that enable more precise perceptual discrimination for those properties that are commonly used in action. Durgin and Gigone’s study provides an example of both the pervasiveness of misrepresentation, and its usefulness for making discriminations among the actual properties of objects.

 

3.  Illusion as a Failure of Precision

Evidence for scale expansion in the perception of space and motion supports a view on which perception is frequently inaccurate. The well-functioning human visual system distorts objective properties like the slant of a hill, the speed of a moving object, egocentric distance, and the declination of our own gaze. These distortions are a feature of vision, not a flaw. By distorting objective properties, we are able to make more precise discriminations among those properties for the purpose of guiding action. In many cases, systematic inaccuracy serves to magnify differences among the actual properties of objects in the world, increasing the precision of perceptual discrimination. If perception is illusory whenever it’s inaccurate, as many philosophers seem to think (see, for instance, the quotes from Smith, Fish, and Macpherson in section 1), then the evidence that I’ve presented would suggest that illusion is widespread even when perceptual systems are functioning properly. An alternative is to provide a new account of illusion in terms of a failure of precision. Many classic cases of visual illusion involve a failure of precision as well as accuracy. But for some cases of inaccurate perception, appealing to precision results in a better account than appealing to accuracy.

Consider first a couple of standard cases of illusion that involve a failure of precision. If you stare at a waterfall tumbling over the rocks, and then look at the stationary rock beside you, there’s a good chance that you will experience an illusion of motion on the stationary rock — a phenomenon known as the Waterfall Illusion. In virtue of seeing the rock’s surface as moving, you are in a poor position to discriminate among its objective properties on the basis of perception. The illusion of motion does not, in this case, enable more fine-grained discrimination among the actual properties of the rock. Similarly, consider the Muller-Lyer Illusion (Figure 1). This case, too, involves a failure of precision. If asked to discriminate the length of the horizontal lines based on perception alone, a subject may be misled by her perceptual experience. On the basis of her experience, she judges that one line is longer than the other. As in the Waterfall Illusion, her perceptual experience of the figure interferes with her ability to distinguish among its objective properties.

Both the Waterfall Illusion and the Muller-Lyer Illusion involve a failure of veridicality that does not increase perceptual precision. In the Waterfall Illusion, inaccuracy doesn’t help you to make more fine-grained discriminatory judgments about the actual properties of the rock. Likewise, in the Muller-Lyer Illusion, one line appears shorter than the other even though both lengths are objectively equal. While both illusions do involve inaccuracy, this is not what makes them illusory experiences on my view. Instead, they are illusions because they do not enable precise discrimination among the actual properties of the object, and not simply because they fail to present the world accurately.

The research on perceived motion and space, however, shows that not all inaccurate perception involves a failure of precision. Some perceptual experiences that are inaccurate nonetheless give us good contact with objects and their properties. Consider the case of reduced visual flow while walking. Suppose a subject, S, views the virtual hallway in Durgin and Gigone’s 2007 experiment. In trial 1, S is standing while a particular spot in the virtual hallway — call it Spot — moves past her at a rate controlled by the experimenter, say 100 cm/sec. In trial 2, S walks forward while Spot moves past her at the same rate. Yet Spot appears to S to be moving more slowly in trial 2 than in trial 1. If S’s perception of Spot were veridical, then Spot would appear to move at the same speed in both trials. A plausible explanation of S’s mistake is that S’s perceptual experience of Spot is inaccurate in at least one of the trials. But although S inaccurately perceives Spot, it is not obvious that this case involves an illusion. For one thing, if walking (or standing still) results in illusion, then illusion will be widespread in ordinary perceptual experience. This is not necessarily a reason to reject the view, but it is at least a surprising consequence. Second, unlike in the Waterfall or Muller-Lyer Illusion, S’s inaccurate perceptual experience enables her to make more fine-grained judgments about Spot’s actual properties. S inaccurately perceives Spot’s motion, but this serves to make her more precise in judging Spot’s motion for speeds that are close to her own walking speed. Intuitively, S’s perception is a success. If we think about her perceptual experience in terms of precision, this makes sense of why exactly it is a success, despite its inaccuracy.

Similar reasoning applies in the case of perceived distance and slant. Consider Proffitt’s example of traversing the Virginia mountains (Proffitt et al. 1995). For a subject standing at the bottom of a hill, it may appear that the road slants upward at 12 degrees, even if the objective slant is only 6 degrees. If she then climbs to the top and look down from the edge, the hill will appear shallower than it really is, perhaps 3 degrees. The objective slant of course stays the same. So, at least one of the subject’s perceptions of the hill is inaccurate, and probably both are, since neither matches the objective slant. Yet despite this inaccuracy, a subject normally won’t stumble when going up or downhill. By exaggerating slant from the bottom and underestimating slant from the top, a subject is better able to detect small changes in the incline of the ground. Inaccuracy serves to exaggerate inclines that are close to ground level, which may make a subject’s judgments and actions more precise than if she were to perceive slant accurately. If we appeal to precision to distinguish successful perception from illusion, then the result aligns with common sense. This is a case of successful perception.

As these examples show, the line between perception and illusion gets drawn in a slightly different way if we appeal to precision. This different way of categorizing specific problem cases is part of what motivates the view that precision offers a promising way forward for distinguishing successful perception from misperception. A view that characterizes successful perception in terms of accuracy will struggle to explain the sorts of cases discussed in this paper, in which perception is inaccurate but precise. Perceptual distortions of space and motion would be considered illusions if we rely on accuracy to draw the distinction between perception and illusion. But they are not illusions on the view that I propose, provided that they enable more precise discrimination among the actual properties of objects.[2]

 

Works Cited

Abrams, J, Barbot, A. & Carrasco, M. (2010) “Voluntary Attention Increases Perceived Spatial Frequency.” Attention, Perception, & Psychophysics. 72(6):1510–1521.

Akins, K. (1996) “Of Sensory Systems and the ‘Aboutness’ of Mental States.” Journal of Philosophy. 93: 337—372.

Anton-Erxleben, K., Abrams, J. & Carrasco, M. (2011) “Equality judgments cannot distinguish between attention effects on appearance and criterion: A reply to Schneider.” Journal of vision 11(13): 8-8.

Block, N. (2010) “Attention and Mental Paint.” Philosophical Issues. 20:23—63.

Block, N. (2015). The Puzzle of Perceptual Precision. in T. Metzinger & J. M. Windt (Eds). Open MIND. doi: 10.15502/9783958570726

Carrasco, M., Ling, S., & Read, S. (2004) “Attention alters appearance.” Nature Neuroscience: 7:308—313.

Da Silva, J.A. (1985) “Scales for perceived egocentric distance in a large open field: Comparison of three psychophysical methods.” The American Journal of Psychology. 98:119-144.

Durgin, F.H. (2009) “When walking makes perception better.” Current Directions in Psychological Science. 18(1): 43—47.

Durgin, F.H., Baird, J.A., Greenburg, M., Russell, R., Shaughnessy, K., & Waymouth, S. (2009) “Who is being decided The experimental demands of wearing a backpack.” Psychonomic Bulletin & Review, 16:964-969.

Durgin, F.H. and Gigone, K. (2007) “Enhanced optic flow speed discrimination while walking: Contextual tuning of visual coding.” Perception. 36: 1465—1475.

Durgin, F.H., Gigone, K., & Scott, R. (2005). Perception of visual speed while moving. Journal of Experimental Psychology: Human Perception and Performance, 31, 339–353.Durgin, F. & Li, Z. (2011) “Perceptual scale expansion: an efficient angular coding strategy for locomotor space.” Attention, Perception, & Psychophysics. 73:1856-1870.

Durgin, F.H., Li, Z., & Hajnal, A. (2010) “Slant perception in near space is categorically biased: Evidence for a vertical tendency.” Attention, Perception & Psychophysics. 72(7):1875-1889.

Firestone, C. and Scholl, B.J. (2015) “Cognition does not affect perception: Evaluating the evidence for ‘top-down’ effects.” Behavioral and Brain Sciences, 20:1-77. DOI: http://dx.doi.org.proxy.brynmawr.edu/10.1017/S0140525X15000965

Fish, W. (2010) Philosophy of Perception: A Contemporary Introduction. Routledge.

Foley, J.M., Ribeiro-Filho, N.P., & Da Silva, J.A. (2004). “Visual perception of extent and the geometry of visual space.” Vision Research. 44:147-156.

Fuller, S., Ling, S., Carrasco, M.(2004). “Attention increases perceived saturation.” Journal of Vision, 4(8): 329.

Ganson & Bronner (2013) “Visual Prominence and Representationalism” Philosophical Studies 164 (2):405-418.

Gibson, J.J. (1966). The senses considered as perceptual systems. Boston: Houghton Mifflin.

Gilinsky, A.S. (1951) “Perceived size and distance in visual space.” Psychological Review, 58: 460-482.

Gobell, J. & Carrasco, M. (2004) “Attention alters the appearance of spatial frequency and gap size.” Psychological Science. 16(8): 644—51.

Kammann, R. (1967) “Overestimation of vertical distance and slope and its role in the moon illusion.” Perception & Psychophysics, 2:585-589.

Kohler, I. (1961) “Experiments with Goggles.” Scientific American. 206: 62-86.

Li, Z. & Durgin, F.H. (2009) “Downhill slopes look shallower from the edge.” Journal of Vision. 9(11):6, 1-15.

Li, Z. & Durgin, F.H. (2010) “Perceived slant of binocularly viewed large-scale surfaces: A common model from explicit and implicit measures.” Journal of Vision. 10(14):13, 1-16.

Li, Z., Phillips, P., & Durgin, F.H. (2011) “The underestimating of egocentric distance: evidence from frontal matching tasks.” Attention, Perception, and Psychophysics. 73:2205-2217.

Linden, Kallenbach, Heinecke, Singer & Goebel (1999) “The myth of upright vision: A psychophysical and functional imaging study of adaptation to inverting spectacles.” Perception. 28:469-481.

Loomis, J.M., Da Silva, J.A., Fujita, N., & Fukusima, S.S. (1992) “Visual space perception and visually guided action. Journal of Experimental Psychology: Human Perception and Performance. 18:906-921.

Macpherson, F. (2009). “Perception, Philosophical Perspectives.” In Tim Bayne, Axel Cleeremans & P. Wilken (eds.), The Oxford Companion to Consciousness. Oxford University Press.

Mendelovici, A. (2012) “Reliable Misrepresentation and tracking theories of mental representation.” Philosophical Studies 165 (2):421-443)

Millikan, R. (1984) Language, thought, and other biological categories. Cambridge: MIT Press.

Nanay, B. (2010) “Attention and Perceptual Content.” Analysis. 70: 263—270.

Ooi, T.L., Wu, B., & He, Z.J. (2006) Perceptual space in the dark affected by the intrinsic bias of the visual system. Perception, 35:605-624.

Prettyman, A. (2016) “Perceptual content is indexed to attention.” Synthese: 1-16.

Proffitt, D.R., Bhalla, M., Gossweiler, R., & Midgett, J.(1995). “Perceiving geographical slant.” Psychonomic Bulletin & Review. 2:409-428.

Ross, H.E. (1974) Behavior and perception in strange environments. London: Allen & Unwin.

Schneider, K.A. (2010) “Attention alters decision criteria but not appearance: A reanalysis of Anton-Erxleben, Abrams, and Carrasco.” Journal of Vision 11, no. 13 (2011): 7-7.

Schneider, K. A., & Komlos, M. (2008) “Attention biases decisions but does not alter appearance.” Journal of vision 8, no. 15: 3-3.

Simmons (2008) “Guarding the body: A Cartesian Phenomenology of Perception.” in Hoffman, P., & Yaffe, G. (Eds.) Contemporary perspectives on Early Modern Philosophy: Essays in Honor of Vere Chappell. Broadview Press: 81-113.

Smith, A.D. (2002) The Problem of Perception. Cambridge: Harvard University Press.

Speaks, J. (2011) “Attention and Intentionalism” The Philosophical Quarterly. 60:325—342.

Stazicker, J. (2011) “Attention, Visual Consciousness, and Indeterminacy.” Mind & Language. 26:2 156—184.

Toye, R.C. (1986) “The effect of viewing position on the perceived layout of space.” Perception & Psychophysics, 40:85-92.

Wagner, M. (1985) “The metric of visual space.” Perception & Psychophysics, 38:483-495.

Wallach, H. (1987) “Perceiving a stable environment when one moves.” Annual Review of Psychology, 38:1-27.

Warren, W.H., & Whang, S. (1987). Visual guidance of walking through apertures: Body-scaled information for affordances. Journal of Experimental Psychology: Human Perception and Performance, 13, 371–383.

Watzl (forthcoming) “Can Intentionalism Explain how Attention Affects Appearance?” in Pautz, A. & Stoljar, D. (Eds.) Themes from Block: Cambridge, MIT Press. (http://folk.uio.no/sebaswat/materials/Watzl_Intentionalism_Appearances.pdf)

Wu, W. (2011) “What is Conscious Attention?” Philosophy and Phenomenological Research. 82:93—120.

Wu, B., Ooi, T.L., & He, Z.J. (2004) “Perceiving distance accurately by a directional process of integrating ground information.” Nature, 428:713-77.

 

Notes

[1] Some of Proffitt’s research has been contested, such as his finding that backpacks and sugary drinks influence distance judgments (see Durgin, Baird, Greenburg, Russell, Shaughnessy & Weymouth 2009). In contrast, the finding that subjects exaggerate the slant of hills has been replicated and is widely accepted in the literature.

[2] Redacted.

Invited Comments from Alison Ann Springle (Pittsburgh)

The Prospects for Prettyman’s “Perceptual Precision”

 

Alison Ann Springle
University of Pittsburgh

 

Prettyman may be understood as presenting and posing a solution to the following seemingly inconsistent triad:

  1. Perception is usually successful.
  2. When perception is successful it is veridical (accurate), i.e., it “specifies some way that the world is, and the world actually is that way.” (2)
  3. Perception is frequently and systematically non-veridical (inaccurate).

Prettyman’s bold solution is to reject (II) and offer an alternative account of the nature of perceptual success.

I will refer to the view expressed in (II) as “The Accuracy Thesis.” As Prettyman notes, this thesis is accepted by the vast majority of philosophers of mind. So, it is significant that Prettyman thinks it should be rejected. According to Prettyman, a growing quantity of empirical evidence supports (III). (III) together with (I) puts pressure on the accuracy thesis (II), and motivates an alternative way of conceiving of perceptual success (and thus also failure) that can account for the ways in which perception seems to be successful even when it is inaccurate. Prettyman’s proposed alternative is what I will refer to as “The Precision Thesis,” according to which perception is successful when it is precise, where “a perceptual experience is precise iff it enables the subject having that experience to make fine-grained discriminations among the actual properties of an object.” (2)

Perception’s role in guiding action appears to loom large in the Prettyman’s thinking about perceptual success. Rightly so; that a fundamental function of perception is to contribute to successful action guidance strikes me as a truism. What’s debatable is how this function is realized (manifested), where the options that seem relevant to Prettyman’s project include:

  1. Non-intentionally Realized: Perceptual experience is a matter of a subject’s having sensory access to (parts of) the world. A subject can, on the basis of making this contact with the world in a perceptual experience, make judgments that guide action.[1]
  2. Descriptively Intentionally Realized: Perceptual experience is a matter of descriptively representing the world; a perceptual experience tells the subject how the world (or relevant parts of it) is, and contributes to successful action guidance by producing accurate descriptions. A perceptual content is indicative.[2]
  3. Non-Descriptively Intentionally Realized: Perceptual experience is a matter of a subject non-descriptively representing the world; a perceptual state is assessable with respect to non-alethic but nonetheless intentional norms (as might be the case if e.g. a perceptual content is imperatival).[3]

The claim that perception or a perceptual experience is itself accurate or not presupposes (b) and is the standard intentionalist (representationalist) position.  Relationalists typically align themselves with (a): they take it that perception or a perceptual experience is not properly assessed with respect to accuracy (or any alethic norms), for perception does not itself have representational function; it is not itself in the business of representing the world. Rather, perception provides a subject with access to the world (acquaints a subject with the world), thereby putting the subject in a position to make an empirical judgment. Judgments made on the basis of perception may be accurate or not, but not the perceptual states. (c) is an intentional alternative to (b) in that it treats perceptual experiences as having content, but denies that perceptual content is essentially assessable with respect to accuracy (or other alethic norms).

It is not clear whether, according to the precision thesis, the way in which perception succeeds by enabling fine-grained discriminations is intentional or relational, so I consider the prospects for both.

In §1 I identify two distinct notions of precision as a kind of success: an intentional notion, and an instrumental notion. If Prettyman has in mind an intentional notion of precision, I presume she intends for precision to be a version of (c). [1] However, I argue that qua an intentional kind of success, precision entails and thus cannot function as an alternative to accuracy; intentional precision just is a version of (b). What’s more, insofar as intentional precision is consistent with (III), so is the accuracy thesis, so the triad is not necessarily inconsistent. Finally, I consider whether the precision thesis might be understood as a modified version of the accuracy thesis or (b), but suggest that the modification may not be desirable.

In §2 I consider how Prettyman might use the instrumental notion of precision to defend relationalism (a). I explain how (III) together with (I) might be employed to motivate dropping intentionalism in favor of relationalism. I suggest that the instrumental notion of precision might help provide a positive account of perception’s contribution to perceptual illusions without positing perceptual representation, thereby undermining a popular motivation for intentionalism,. If this is Prettyman’s project, much hangs on her evidence for (III). I argue that Prettyman does not succeed in demonstrating (III); it is not clear that the empirical cases she considers are cases in which there is frequent and systematic inaccuracy.  Consequently, I do not think Prettyman successfully motivates dropping intentionalism, nor, for that matter, seeking an alternative to or modifying the accuracy thesis.

Despite the critical nature of my assessment of the prospects for the precision thesis, I am sympathetic with and have a stake in Prettyman’s general project of identifying the right way of conceiving of perceptual success in light of the sorts of empirical findings she considers. So, it is with an eye to making progress in this common pursuit that I raise these criticisms.

1.  The Intentionality of Precision

Consider the difference between the notion of precision at play in the following statements:

  1. “The victim gave a precise description of the suspect.”
  2. “This paintbrush is very precise.”
  3. “That shot was very precise—hit the target right one the nose!”

In (i) precision is a measure of an intentional performance that essentially involves the production of an intentional object, namely, the victim’s description of the suspect. In (ii) precision is a measure of a disposition of a non-intentional object (though it is a disposition relative to an intentional agent’s intention to bring about some end). In (iii) precision is a measure of an intentional performance, but one that does not essentially involve the production of an intentional object; rather, it is the measure of the activity of a genuinely intentional subject, with respect to an intentional object, namely, the subject’s intention to hit a target.

In each of these cases, the normativity of precision (its status as a measure of success) depends on an intentional state or being, and only in (i) is precision itself an intentional property, where its intentionality seems to derive from its being a property of something independently intentional, namely, a description.

Now, of course, to say that a description is precise does seem to go beyond calling it true or accurate. There could, for instance, be two descriptions of a suspect, both of which are true, but where one is more precise than the other. It is not, however, clear that something can be precise in the intentional sense unless it is also descriptively accurate. To wit, we may say of two descriptions or depictions of some subject matter, X and Y, that X is more precise (detailed) than Y insofar as X includes more items (discrete claims), but it would seem that X can only be more precise than Y if the additional items or discrete claims are accurate (describe truly or accurately). Consider: the following two sentences describe a man in a trench coat and a yellow hat standing on a red square.

  1. “That man is wearing a trench coat and hat, and he is standing on a square.”
  2. “That man is wearing a winter coat and a blue hat, and he is standing on a green square.”

While (v) includes more details than (iv), it would be odd to say that (v) is more precise, insofar as the additional details miss the mark descriptively—they are inaccurate. Being detailed simply in the sense of including more stuff is not, generally, an intentional kind of success. Where precision has an intentional status, it appears to sub-serve a descriptive function. It would seem, then, that intentional precision is simply a measure of the fine-grainedness of descriptions (or sets of descriptions) that succeed intentionally when accurate.

But aren’t there cases in which something can be precise with respect to some feature without being accurate (i.e. without accurately attributing properties)?  Where such cases exist, I think they involve a non-intentional notion of precision, what I’ll call an “instrumental” notion of precision. Thus, a paintbrush does not describe a line precisely; rather, it is used to make a line precisely as the painter intends. Likewise, an arrow shot does not describe the target; rather, it hits it. We might understand the perceptual precision thesis on the model of instrumental precision thus: perception succeeds when it gives a subject access to the world in a way that allows the subject to produce true or accurate judgments about behaviorally relevant parts of the world. But if we do this, the perceptual precision thesis appears to be a version of (a) rather than (c). If precision is a (functional) property not of content but of the mechanistic exercises of perceptual systems/ discriminatory capacities that enable perceptual contact with the world, then it is not an intentional kind of success, and ipso facto not an intentional alternative to accuracy.

Now, Prettyman claims that while “It may seem at first glance that precision requires accuracy…this turns out not to be the case… In order to make fine-grained discriminations among the actual properties of an object, we don’t need to perceive those properties accurately.” (3) But Prettyman does not consider the sorts of arguments I’ve just given. With those arguments in hand, I’ll now try to make clear why I am skeptical that Prettyman provides any arguments that support the claim that intentional precision does not require accuracy.

Prettyman appears to raise the following point in favor of the claim that precision and accuracy come apart: “It is fairly uncontroversial that accurate representation of the world is not necessary for precise action.” (5) However, the notion of precision here is surely of a kind with (iii) and not (i) above, so it is not properly intentional. In other words, this seems to conflate intentional with instrumental success.

Prettyman also claims that misrepresentation of an object’s properties can serve useful functions such as “increasing perceptual precision for the sorts of discriminative judgments that are useful for the subject in her environment.” (7)[4]  But this seems perfectly compatible with saying that misrepresentation with respect to some properties may enable accurate representation with respect to other, more important properties. Thus, perception may succeed when it is accurate with respect to some domain—the domain it is in particular responsible for accurately representing, e.g.  biologically relevant properties. I see no reason why accuracy cannot be restricted in this way. Indeed, as Prettyman notes, Watzl (forthcoming) proposes that perception aims at accuracy and usefulness, and sometimes there’s a tradeoff and usefulness dominates. That usefulness plays this role does not mean that perception does not aim at accuracy. Rather, it suggests that perception aims at accuracy because accuracy is generally useful in guiding action, but it may be inaccurate with respect to some properties where such inaccuracy allows it to be more accurate with respect to the more important properties. By Prettyman’s own lights, this is what happens: inaccuracy with respect to some properties enables precise—but I’ve just argued that this does entail accurate—representation of the more important properties. This is all perfectly compatible with (b).

In fact, rather than finding in Prettyman’s discussion an argument that supports the claim that intentional precision can come apart from accuracy, her proposal seems to support the accuracy thesis (or (b)). Prettyman claims that in many cases illusion qua inaccurate perception also involves a failure of precision. But insofar as precision (qua intentional property) is just accurate perception of biologically important objective properties, these observations seem to support (b). For if usually when perception is inaccurate it is also imprecise (also fails to accurately represent the biologically most relevant objective properties), this suggests that accuracy usually begets more accuracy and action-guiding success, and inaccuracy usually begets more inaccuracy and therefore fails with respect to action-guiding success.[5]

It’s also not obvious that (I) (II) and (III) necessarily form an inconsistent triad.  It goes without saying that perceivers do not perceive everything there is, and it is often thought that perceptual representation approximates full accuracy; not that it constantly perfectly achieves it.[6] But Prettyman does not seem to leave room for the possibility of perception aiming to be approximately accurate of a limited domain of properties.  Consider, for instance, Prettyman’s reasons for thinking that precision, as compared to accuracy, is better able to distinguish successful perception from illusion for the following case of intuitively successful perception:

For a subject standing at the bottom of a hill, it may appear that the road slants upward at 12 degrees, even if the objective slant is only 6 degrees. If she then climbs to the top and look down from the edge, the hill will appear shallower than it really is, perhaps 3 degrees. The objective slant of course stays the same. So, at least one of the subject’s perceptions of the hill is inaccurate, and probably both are, since neither matches the objective slant. Yet despite this inaccuracy, a subject normally won’t stumble when going up or downhill. By exaggerating slant from the bottom and underestimating slant from the top, a subject is better able to detect small changes in the incline of the ground. Inaccuracy serves to exaggerate inclines that are close to ground level, which may make a subject’s judgments and actions more precise than if she were to perceive slant accurately. (8)

But contrary to Prettyman’s verdict, we can account for how perception is successful in this case in terms of accuracy: it is successful insofar as it is accurate with respect to the most important properties. If the position expressed in (b) is true and accuracy is in the service of the practical function of perception, it is not surprising that perceptual systems give priority to a certain domain of properties, namely, those that are the most useful. The way perception makes these properties available for use in action is precisely by mostly accurately describing them (as accurately as they can muster).  Accuracy and usefulness or utility are distinct albeit related standards of perceptual success, and we may appeal to them separately. So, in the above case, perception succeeds both with respect to utility and accuracy with respect to the biologically relevant objective properties; where it distorts objective properties, it may succeed with respect to utility and (perhaps; see next section) fail with respect to accuracy. In addition, insofar as systematic inaccuracies serve precisely to enable accuracy with respect to biologically important properties, as Prettyman seems to think is the case, such inaccuracies surely do not show that perception doesn’t function to accurately represent.

In light of my objections so far, it might be thought that intentional precision should be understood as a hybrid notion that combines accuracy with utility—i.e. perception aims to be accurate where accuracy has the most utility. Precision as a hybrid notion is, however, a relatively mundane proposal; rather than replacing accuracy, it modifies by restricting the domain of accurate representation to biologically relevant properties, a restriction at least some philosophers who endorse the accuracy thesis may have largely taken for granted.  In addition, it’s not obvious that intentional precision captures the way in which perception can be successful with respect to use even when inaccurate. For in the cases in which Prettyman suggests inaccuracy has utility, perception is intuitively less intentionally precise in that fewer objective properties are accurately represented.[7]

It’s also not clear that precision is, at least as it has so far been fleshed out, terribly useful as a modification of the accuracy thesis. If talk of precision is to be informative, we need some standard for what counts as course- as opposed to fine-grained discrimination. But the concept of grain appears to be highly context-sensitive. This is potentially unattractive because to determine whether a perception is successful with respect to accuracy, you need simply be given the content and determine whether the condition it specifies obtains. In contrast, to determine whether a perception is successful with respect to precision, you need some standard against which to determine (optimal) grain, and Prettyman has not provided such a standard. Consequently, it is difficult to see why, for instance, the waterfall illusion involves a failure of precision, as it would seem such an illusion could be as fine-grained (include an equal proportion of accurately represented elements) as a veridical perception as of a waterfall.

Prettyman might retort that it is only when biologically relevant objective properties are discriminated that a perception will be precise; a veridical perception is precise insofar as the relevant biological properties are accurately represented while illusion lacks precision insofar as these properties are not accurately represented.  So, it is the representation of biologically relevant properties that determines the optimal grain, and thus the standard of grain for a perceiver.

However, just what “biological relevance” comes to is also not clear, and could deliver some results that are arguably as unsavory as finding that, where accuracy is our standard, perception systematically fails. Consider: whether foodstuffs are spoiled is, presumably, biologically relevant. But humans are not very successful at discriminating (by sight, smell, and unfortunately even taste) spoiled from unspoiled food until it’s pretty far into the spoiling stages (or until it’s been at least partially ingested). Does this mean that we are generally failing with respect to precision, and thus less successful perceivers? – The point generalizes: it seems that if precision is our standard of perceptual success, then we may often be “misperceiving” when, intuitively, we simply are not perceiving (the relevant properties).  While lacking the capacity to perceive properties it would be good to have the capacity to perceive might be considered a problem relevant to perception, it is prima facie distinct from perceptual failure in the sense of misrepresentation or illusion.

In addition, given the context sensitivity of grain and biological relevance, it’s not clear how useful precision will be for assessing perceptual success. For the same given objects and properties, it seems overwhelmingly likely that what properties are biologically relevant will be species-dependent. That is, for a given object, one species need only discriminate relatively course-grained properties, while some other species will need to be able to distinguish relatively fine-grained properties (e.g. texture, specific of hue). Not only is optimal grain is likely to be species-dependent, it may also be dependent on the environment in which the creature happens to spend its life—plausibly, a snake that spends its life in a cage needn’t have nearly the discriminatory capacities of a snake in the wild. So, where accuracy provided a common, uniform standard, it’s not yet clear precision as constrained by biological relevance can achieve this. To the extent that precision may seem to provide a uniform standard, I think this is because of its relationship to the following seemingly trivial truth: a creature should be able to perceptually distinguish those objects and properties that matter to its behavior, or those objects and properties that, through the aid of other activity or cognitive processes, allow it to distinguish objects and properties that matter to its behavior. Whether a perceiver succeeds in discriminating the properties relevant to animal survival is obviously important to how biologically successful a particular perceptual system or state will be. But there’s a difference between what properties perception represents and how it represents them. The precision thesis seems to get at something true insofar as it concerns what properties get represented. But it’s not clear that what determines what properties are represented should have any consequence for the standards we use to assess how these properties are represented.

2.  Misrepresentation or Mis-ascription?

In light of the above considerations, perhaps we ought to understand precision as a matter of the sensitivity of a sensory or perceptual system—of its ability to respond selectively to and to enable the perceiver to respond selectively to biologically relevant parts of the environment. In other words, perhaps we ought to understand the precision thesis as involving the instrumental notion of precision. As a merely instrumental proposal, I have no problems with precision, except that it is not clear what its philosophical upshot would be. Indeed, instrumental precision is consistent both with relationalism and with a standard intentionalism that treats perceptual states as assessable with respect to accuracy. A relationalist may say that perceptual or sensory systems are instrumentally precise or imprecise–can succeed or fail with respect to this standard instrumentally– and, for entirely independent reasons, deny that perceptual experiences are assessable with respect to accuracy. The intentionalist may allow that perceptual systems are instrumentally assessable with respect to precision, but for entirely independent reasons, maintain that perception is genuinely representational, where perceptual representations are assessed with respect to an alethic norm like accuracy.  It seems, then, that the instrumental conception of perceptual precision would not be of much philosophical significance.

But perhaps we may understand the claim that perception systematically misrepresents (III) as a reason to prefer relationalism to intentionalism. As Prettyman notes, systematic misrepresentation “is not necessarily a reason to reject the view [that perception succeeds when accurate], but it is at least a surprising consequence.” (19) This strikes me as the right verdict: it could be that perception functions to accurately represent but frequently and systematically fails, but it would indeed be a bit odd if perception frequently and systematically failed to do the thing it’s supposed to do, or at doing the thing it does (accurately represent) to achieve the end it’s supposed to achieve (successful action guidance). At the very least, evidence of frequent and systematic failure certainly provides a reason to think twice about the accuracy thesis (b). If something else, e.g. relationalism with instrumental precision, can account for perception’s action guiding role and the behavior of perceptual systems while avoiding frequent and systematic failure, that may be a reason to favor such an alternative.

In addition, among the phenomena that motivate intentionalism are illusions. By reconceiving of perceptual success and failure, and thus illusions, in terms of instrumental, non-intentional precision, perhaps Prettyman can undermine some of the motivation for intentionalism by providing a positive non-intentional account of perception’s contribution to misjudgment. The idea would be to treat discrimination as a kind of judgment that perception enables by relating perceivers to fine-grained biologically relevant properties. Then an illusion is a mistaken judgment to which perception contributes when it either doesn’t care about the properties that are misperceived since they are not biologically relevant, or else when perception fails with respect to a related but non-intentional norm, namely, instrumental precision, i.e. when it fails to acquaint a perceiver with the fine-grained objective properties of objects that are of biological consequence. Understanding perceptual function in terms of instrumental precision also allows us to see how perception is usually doing just fine by the standards that properly apply to it: insofar as perceivers systematically misjudge some objective properties, this is because perceptual systems are only intended to acquaint us in a fine-grained way—a way that enables accurate judgments—with certain biologically relevant objective properties. Contrary to what happens when we take accuracy as our standard (if (III) is true), perception as measured against precision does not typically fail at doing the thing it is its function to do.

However, if this is Prettyman’s strategy, (III) had better be true, or at least well supported. Do the empirical results Prettyman adduces genuinely show or strongly suggest that perception frequently fails to accurately describe the world? Not necessarily.

With respect to Durgin and Gigone’s 2007 experiment, Prettyman cannot infer from the fact that subjects may make judgments with contents about the speed of Spot that perception represents the absolute speed of objects like Spot in the environment, and thus it does not follow that such properties are systematically misrepresented in perception. The perceptual phenomenology of the experience is arguably consistent with perception representing a kind of self-motion-relative speed, so that there is indeed no illusion or misperception, as subjects accurately represent a relational property, namely, the speed of Spot relative to their own motion.  The point may generalize: what perception represents, first and foremost, may be biologically relevant properties, some of which may be subject-involving or relational rather than straightforwardly physical or objective. Indeed, Prettyman appeals to a paper by Kathleen Akins (1996) in which Akins argues that perceptual or sensory systems are not “veridical.” However, Akins does not deny that perception is veridical full-stop. Rather, she argues that perception is often not veridical if we think it purports to be about objective (subject-independent) properties; Akins explicitly allows that perception may accurately represent “narcissistic properties” (e.g. “I am cold,” “I am in danger,” etc.)

Prettyman also appeals to a number of studies that suggest that spatial perception is systematically distorted. But other philosophers who have looked closely at these or related findings do not conclude that perceptual success does not consist in accuracy. For instance, Gary Hatfield (2003, 2009) argues that there is systematic distortion with respect to “objective” spatial properties. According to Hatfield, the entire visual space contracts with distance, yielding diminished phenomenal size with increasing distance even under full cue conditions, (2009, 198) and this contracted space affects phenomenally presented shape. (203) Because Hatfield’s proposed contracted space does not exhibit free mobility, it does not meet the usual conditions on a physically possible space. Thus, Hatfield denies that perception aims to phenomenally present physical sizes and shapes “as they are.” However, he does not deny that perception aims to be accurate. By Hatfield’s lights, perception represents a visual space that is systematically transformed in ways that enable action and cognition guidance.[8]  Hatfield assumes that “Visual perception allows sighted organisms to interact effectively with their environment. To serve that function, perception must represent the environment with some degree of accuracy.” (2003, 375) He asks whether, to serve that function, perception must represent the environment “as it is.” His answer: it depends on the notion of accuracy or veridicality in play. While he thinks it’s obvious that “full and accurate representation of the physical, chemical, or biological environment is not required” and denies that “perceptual representation even remotely approach such completeness,” Hatfield nonetheless treats spatial perception as representational, and even suggests that experienced space may stand in a resemblance relation to physical space that “can sustain a notion of veridicality in which information about physical spatial structure is phenomenally presented by a resembling spatial structure.” (2003, 374)

In a related vein, Hill and Bennet (2008) argue that systematic distortions in the perception of shape and size can be explained by appeal to what they call “Thouless Properties,” i.e. “computable functions of external stimuli—which means, in effect, that they are computable functions of properties of the retinal image that are projected by external objects.”(309)  They claim that perceptual appearances consist in the representation of Thouless properties, that perception is “generally and systematically concerned with Thouless properties,” and that in representing such properties “we obtain the benefits” in terms of, inter alia, action-guidance “that would otherwise be achieved only by independently representing properties of two kinds—objective physical properties and angular properties.”  (311)

In the cases she discusses, Prettyman seems to assume that if there is a descriptive content, it is as of some objective properties. Consider, for instance, her discussion of Proffitt’s example of traversing the Virginia mountains (Proffitt et al. 1995), where she concludes that “at least one of the subject’s perceptions of the hill is inaccurate, and probably both are, since neither matches the objective slant.” (19) But as we have just seen, Prettyman cannot just assume that the content, if veridical, is objective slant. In addition to the sorts of alternatives just discussed, another alternative explanation might be that perception represents steepness in terms of affordance properties, e.g., the amount of bodily effort it would take to traverse the hill.[9] Perhaps, for instance, walking up the hill is represented as being about four times more demanding than walking down the hill.

By way of arriving at a more general conclusion, consider some results Prettyman does not discuss, but may at first blush appear to support her case: Hoffman, Singh, and Prakash (HSP; 2015) have studied the effects of selection on perception using evolutionary games and genetic algorithms. They found that “that veridical perceptions–strategies tuned to the true structure of the world–are routinely dominated by nonveridical strategies tuned to fitness.”  Indeed, “Veridical perceptions escape extinction only if fitness varies monotonically with truth.” (1480) HSP take these results to provide strong evidence for the thesis that perception does not function to be accurate or veridical.

Jonathan Cohen (2015), however, has argued that this conclusion does not follow, for “Much depends on how we understand the contents of the relevant states” and “you can’t say whether something is veridical or not without first knowing what it is saying.” (4) HSP provide cases they think show that perception is systematically non-veridical, but, Cohen notes, it is “highly unobvious” that the description of the content they provide is “the right description” (4). Cohen clarifies, with respect to the HSP cases, that the “point is not that we are forced to make content assignments on which the relevant states turn out to be veridical after all; it is that, pending some argumentative support, HSP aren’t entitled to make content assignments on which they turn out to be non-veridical.” (5)

The same, I think, can be said for the contents Prettyman is supposing: until argumentative support has been provided for ascribing objective properties as the contents of perceptual representations, one may defend (II) and deny (III). So long as we allow that the properties represented in perception may not be simply and straightforwardly objective, physical, or perceiver-independent, the cases Prettyman discusses are not obviously genuine cases of frequent and systematic misperception.

 

Conclusion

I have argued that precision is not a viable intentional alternative to accuracy, but that doesn’t mean that there are not viable alternatives, i.e., (c)-type views of perceptual content in the offing. I have also argued that Prettyman does not successfully motivate the need for an alternative. But, that does not mean that, with further argument, she couldn’t. As noted in my introduction, I am sympathetic with Prettyman’s project, so I hope the issues I’ve raised help point the way towards positive progress.

 

References

Akins, K. (1996). “Of Sensory Systems and the Aboutness of Mental States.” Journal of Philosophy Vol. 93 (7) pp. 337-372.

Brewer, B. (2011). Perception and its Objects. Oxford University Press.

Burge, T. (2010). Origins of Objectivity. Oxford, Oxford University Press.

Campbell, J. (2002). Reference and Consciousness. Oxford: Oxford University Press.

Dretske, F. (1995). Naturalizing the Mind. Cambridge, MA: MIT Press.

Cohen, J. (2015) “Perceptual representation, veridicality, and the interface theory of perception.” Psychonomic Bulletin & Review, 22 (6) pp. 1512-1518.

Hall, R. (2008). “If it itches, scratch!” Australasian Journal of Philosophy, 86(4) pp. 525–535.

Hatfield, G. (2009). Perception and Cognition: Essays in the Philosophy of Psychology. Oxford: Oxford University Press.

Hatfield, G. (2003). “Objectivity and Subjectivity Revistied: Color as a Psychobiological Property.”  In Colour Perception: Mind and the Physical World, ed. By Rainer Mausfeld and Dieter Heyer. Oxford: Oxford Univeristy Press. 187-202

Hatfield, G. (2000). “Representation and Constraints: The Inverse Problem and the Structure of Visual Space.” Acta Psychologia, pp. 355-378

Klein, C. (2015). What the Body Commands: The Imperative Theory of Pain. MIT Press.

Klein, C. (2007). “An Imperative Theory of Pain.” The Journal of Philosophy. CIV(10), pp. 517–532.

Nanay, B. (2013). Between Perception and Action. Oxford University Press.

Schellenberg, S. (2016) “Perceptual Particularity.” Philosophy and Phenomenological Research, 93 (1), pp. 25-54.

Schellenberg, S. (2011). “Perceptual Content Defended.” Noûs, Vol. 45 (4), pp. 714-750.

Travis, C. (2013). Perception: Essays After Frege. Oxford University Press.

Tye, M. (2000). Consciousness, Color, and Content. MIT Press.

 

Notes

[1] See e.g. Travis 2004, Brewer 2011, Campbell 2002.

[2] See E.g. Burge 2010, Schellenberg 2012, 2016, Drestke 1995, Tye 2000.

[3] See e.g.  Mandik 2005, Klein 2007, 2015, Hall 2008.

[4] At first blush, this claim may seem to express a view much like (c), but in that case, precision is not an intentional alternative to accuracy.

[5] Of course, Prettyman also notes that not all inaccurate perception involves a failure of precision. (18) But, given that we are mere terrestrial perceivers, it needn’t shake our conception of perceptual success if, once in a while, inaccuracy with respect to some less relevant features begets greater accuracy with respect to the more biologically relevant features, and it certainly doesn’t show that perceptual systems do not function to accurately represent (at least approximately) parts of the world that are perceptually available and in particularly those that are behaviorally relevant.

[6] See e.g. Burge 2010.

[7] This is so even if inaccuracy aids in a kind of instrumental precision e.g. via coding strategies, e.g. “scale expansion” (12), that increase sensitivity to certain kinds of properties.

[8] In the case of color, Hatfield (2009) says perceived color is a “psycho-biological” property. Though he does not say so explicitly, it is natural to interpret the properties represented in visual space (the properties that result from transforming physical space) as psychobiological properties, in which case, perception accurately represents such properties, while approximating (and thus somewhat accurately) representing objective properties.

[9] See also Nanay 2013.

Invited Comments from Christopher S. Hill (Brown)

Comments on Adrienne Prettyman’s “Perceptual Precision”

 

Christopher S. Hill
Brown University
Until recently, philosophers and psychologists have tended to downplay discrepancies between the perceived or apparent values of magnitudes and their actual values. For example, many vision scientists have maintained that the perceived or apparent sizes of objects are virtually constant across a broad range of viewing distances, departing from virtual constancy only when the distances are large, or when the subject is motivated by special interests, such as an interest in drawing objects according to the principles of linear perspective.  (Palmer 1999, pp. 313-4) This picture allows for illusions, but it is committed to marginalizing them, denying that they are pervasive or truly systematic. Presumably the philosophers and scientists who embrace the picture share a background assumption to the effect that we need to represent the environment accurately in order to achieve our goals, and also an assumption to the effect that Nature had provided for this need by endowing us with a visual system that successfully tracks objective magnitudes. These assumptions are of course plausible. How could we have survived if the visual system had been out of touch with reality?

Building on the work of Frank Durgin and his associates (see, e.g., Durgin and Li 2011) Adrienne Prettyman challenges this traditional picture in two ways. First, she argues that discrepancies between the apparent properties of objects and their actual properties are in fact pervasive. The areas in which such discrepancies are prominent include perception of temperature, egocentric distance, the slant of hills and other tilted surfaces, optic flow, and color. She also reminds us of Marissa Carrasco’s work, which indicates that attention enhances apparent contrast, apparent size, apparent velocity, and apparent saturation of color. (See, e.g., Carrasco 2004) These examples suffice to make her case, but she might also have mentioned perception of size and shape: apparent size decreases with distance, except perhaps for a very small area surrounding the subject’s body, and apparent shape varies with perspective, as is illustrated by such phenomena as the changes in the apparent shape of a rotating coin, and the way parallel lines appear to converge as they recede into this distance. Second, Prettyman challenges the assumption that visual representation must be accurate in order to serve its function. Systematic distortion can be useful, even necessary. For example, if perception magnifies the apparent slant of hills, it becomes easier to discriminate between different slants, and therefore easier to answer such questions as whether the slant of a particular hill increases with distance. In general, she maintains, areas of discrepancy between apparent and actual properties are genuinely adaptive, not places where evolution was blocked by constraints or settled for kludges.

These are very exciting ideas, and Prettyman is to be thanked and congratulated for developing them. I have reservations about her proposals, but they are mild compared to my admiration for the courage and imagination that it took to write her paper. She has at the very least pointed out serious problems with the traditional picture, which in view of her arguments now seems quite Panglossian. It is to be hoped that philosophers of perception will be persuaded by her paper to devote more attention to alternative views. (For related arguments see Hill 2009, Chapter 5, Hill 2014, Chapter 12, and Hill 2016.)

Prettyman’s main proposal concerns a new way of distinguishing between successful perception and misperception. On her view, perception is successful to the extent that it “enables us to discriminate among those properties that are most useful for action.” It does not achieve this by being fully accurate, but rather by increasing the precision of our grasp of the values of magnitudes. This is normally done by “magnifying” visual angles and other properties that are the inputs to calculations concerning such magnitudes as distances, slants, and degrees of optic flow. Just as a magnifying glass enables a jeweler to make finer cuts in a gem she is shaping, so also the visual system makes it possible for us to make finer discriminations among values of magnitudes by inflating or expanding portions of the scales on which the magnitudes are represented. This proposal has an important corollary concerning the nature of illusions: what makes a visual experience an illusion, Prettyman claims, is not the fact that it is inaccurate, but rather the fact that it fails to “enable precise discrimination among the actual properties of the object.”

In evaluating Prettyman’s proposals, we have to ask what exactly she means by the two key words “accuracy” and “precision.” Here is her answer:

As I’ll understand it, a perceptual experience is accurate iff it specifies some way that the world is, and the world actually is that way; it is inaccurate otherwise. In contrast, a perceptual experience is precise iff it enables the subject having that experience to make fine-grained discriminations among the actual properties of an object.

She continues: “It may seem at first glance that precision requires accuracy, but this turns out not to be the case…”

Now this last claim is prima facie confusing, and I’m not sure that Prettyman ever quite succeeds in making it plausible. If, as she says, it is of the nature of precision to promote perceptual discrimination, then precision is entangled with accuracy after all, for it is clear that our commonsense concept of discrimination implies accuracy. If I discriminate between A and B, I am pretty clearly representing the world as being a certain way, and it must be true that the world actually is that way, so the conditions in Prettyman’s definition of accuracy are fulfilled. I think, though, that careful reflection sustains Prettyman’s claim. When we look at the O.E.D. we find that it defines “discriminate” in terms of discerning differences between features. Webster’s concurs. Now discerning that two features are different requires accurate representation, but only of the fact of difference, not of the features themselves. I can accurately represent two heights as different even if I am representing both of the heights as larger than they are. This must be what Prettyman has in mind. Precision in her sense requires accuracy, but only accuracy in representing complex magnitudes like differences between values, ratios of values, and transitions between values.

As noted, when Prettyman draws on empirical research, she mainly relies on the work of Durgin and his associates. She invokes Carrasco’s work at one point, but I would have liked to see a more extended treatment. While Prettyman and Durgin rely heavily on the notions of scale expansion and enhanced precision in accounting for the benefits of the divergence between appearances and reality, Carrasco tends to account for the benefits in terms of enhanced resolution. (Yeshurun and Carrasco 1998, Anton-Erxleben and Carrasco 2013) The notions of precision and resolution are clearly related, but it remains to be seen whether they function in similar ways in Durgin’s and Carrasco’s research. Do we have loosely related theories or theories that deploy constructs that are fundamentally the same?

I turn now to a couple of concerns I have about Prettyman’s picture. In the first place, while it is all very well to claim that accurate representation of differences between features is compatible with inaccurate representation of the features themselves, I start to worry when Prettyman and Durgin go on to claim that representation of the features definitely is inaccurate. I worry because our best theories of representation (Dretske 1995, Millikan 1989, Neander 2012) imply that covariation between representations and represented items is a necessary condition of the former representing the latter. Clearly, covariation is precluded if perception suffers from systematic inaccuracies. (Millikan requires only that covariation occur in certain privileged contexts, but there is good reason to believe that the mantle of privilege is quite broad in perceptual cases.)

Prettyman and Durgin could avoid this difficulty by changing their story. Instead of saying that the visual system inaccurately represents objective magnitudes like physical distances and sizes, they could say that it accurately represents certain different magnitudes that are viewpoint-dependent. We know that the visual system computes functions of input quantities like visual angles. It is usually assumed that it is the job of these computational processes to produce representations of objective magnitudes, but as Prettyman points out, there is a great deal of evidence that they do not perform this job successfully. Why not then make a different assumption about their job, supposing that it is to compute quantities that relational and viewpoint-dependent rather than intrinsic and objective? Here is an illustration. We know that there are mechanisms that apply constancy transformations to visual inputs that are relevant to size, but there is a great deal of introspective and experimental evidence that visual representations of size fall short of full constancy, with the divergence increasing drastically as distance from the observer increases. We would have to recognize very substantial inaccuracies here if we supposed that it was the function of constancy transformations to produce representations of objective sizes; but we can avoid positing inaccuracies if we assume instead that it is the job of the constancy transformations to produce representations of properties of objects that are “inflated” visual angles – inflated in such a way is to achieve partial constancy, but not enough to achieve full constancy. According to the picture I have in mind, inflated visual angles are real properties of objects, just as the “inflated” size property being of a height H such that two times H is twelve feet is a real property of objects. Thus, according to the picture, visual representations of size stand for properties that objects really have, and can therefore be said to be accurate. It’s just that the properties in question are relational and viewpoint-dependent rather than intrinsic and objective. Moreover, the picture authorizes us to say that visual representations of size are useful. It is clearly more useful to represent inflated visual angles than visual angles themselves, because, being the products of constancy transformations, inflated angles tend to be more similar to each other than actual angles, and therefore lend themselves more readily to categorization. (It’s easier to see that a currently perceived object belongs to a category if it is more similar to stored exemplars of the category than if it is less similar.) Equally, it is more useful to represent size-quantities that exhibit partial constancy than quantities that exhibit full constancy. This is because differences in apparent size encode information about distances from the observer: in a line of pine trees receding into the distance, differences in apparent size are a valuable cue as to the ordinal distance relations of the trees. In sum, appropriately inflated visual angles compress two kinds of information into a single quantity, and representing that quantity therefore promotes efficient coding. If we represented size information and distance information separately, we would need two streams of experience rather than one. (I develop this picture in Hill 2014 and Hill 2016.)

As noted, this line of thought is meant to illustrate the possibility of a position that recognizes widespread discrepancies between the contents of perceptual experiences and objective properties, while also maintaining that perceptual experiences tend to be veridical. But it also illustrates a second concern that I have about the line that Prettyman is developing. She offers a more or less uniform account of why it is useful to “misrepresent” the world. On her account, “misrepresentation” is useful because it enhances precision and therefore discrimination. On my account of the “failure” of size constancy, however, “misrepresentation” has a different kind of value: it makes it possible to encode size information and distance information simultaneously, thereby promoting coding efficiency. It is possible that Prettyman’s Procrustean approach is the right way to think about perceptual “misrepresentation,” but as of now, I am inclined to favor a more pluralistic approach.

 

References

Anton-Erxleben, K. and Carrasco, M. (2013). “Attentional enhancement of spatial resolution: linking behavioural and neurophysiological evidence,” Nature Reviews Neuroscience, 14188-200. 

Carrasco, M., Ling, S., & Read, S. (2004) “Attention alters appearance.” Nature Neuroscience: 7: 308—313.

Dretske, F. (1995). Naturalizing the Mind. Cambridge, MA: MIT Press.

Durgin, F. H. and Li, Z. (2011). “Perceptual scale expansion: an efficient angular coding strategy for locomotor space,” Attention, Perception, & Psychophysics. 73:1856-1870.

Hill, C. S. (2009). Consciousness. Cambridge: Cambridge University Press.

Hill, C. S. (2014). Meaning, Mind, and Knowledge. Oxford: Oxford University Press.

Hill, C. S. (2016). “Perceptual Relativity,” in Philosophical Topics 44: 179-2000.

Millikan, R. (1989). “Biosemantics,” Journal of Philosophy, 86: 281–97.

Neander, K. (2012). “Teleological Theories of Mental Content,” The Stanford Encyclopedia of Philosophy. https://plato.stanford.edu/entries/content-teleological.

Palmer, S. (1999). Vision Science. Cambridge, MA: MIT Press.

Yeshurun, Y. and Carrasco, M. (1998). “Attention improves or impairs visual performance by enhancing spatial resolution,” Nature 396: 72-5.

6 thoughts on “Perceptual Precision”

  1. Reply to Alison Springle

    I would like to thank Alison Springle for taking the time to provide such detailed comments about my work. Her response is rich and thought-provoking. I very much appreciate her positive suggestions regarding precision and the relational view of perception, which I will think more about. I am also grateful to her for pointing me toward some additional literature at the end of her comments.

    Is Perception Relational or Intentional?

    As Springle points out, the “accuracy thesis” (which is the view that I argue against) is shared by philosophers who hold a standard representationalist position. Since I offer an argument against the view that perception aims at accuracy, my own account is in tension with representationalism (my view also raises problems for representationalism for other reasons; see Hill’s comments & my response). But I have not argued for a relational account in this paper, and I’m not sure that my view pushes in the direction of a relational view (e.g. when I perceive an inflated angle, what exactly am I related to?). Springle is right to push me on this point, since the next step is to develop a positive account of perceptual content in light of the distinction between precision and accuracy. One way forward is to develop a revised intentional account of perceptual content as viewpoint dependent (e.g. see Hill’s comments) or indexed to states of the subject (as I’ve previously argued in the case of attention, see Prettyman 2016). This is something that I’m still thinking about.

    Precision and Fineness of Grain

    In section 1 of her comments, Springle correctly interprets my account of perceptual precision as the ability to make fine-grained perceptual discriminations. She then distinguishes between two additional meanings of ‘precision’: intentional and instrumental. Springle argues that if what I mean by perceptual precision is “intentional precision,” then precision does entail accuracy after all.

    It may be true that the word ‘precision’ is sometimes used to pick out a measure of the “fine-grainedness of descriptions” (Springle’s “intentional precision”). But this is not what psychologists studying perceptual precision mean by the term ‘precision’. Instead, ‘perceptual precision’ is defined roughly as I define it in my paper, as a behavioral ability (e.g. can the subject indicate that this angle is different from that one?). Even if Springle is right that what she calls “intentional precision” entails accuracy, this is not what I mean by perceptual precision. So, I don’t think that the arguments in this section succeed in raising a challenge for my claim that precision (in my sense) does not require accuracy.

    One difference between my account of precision and Springle’s concerns fineness of grain. On my view, precision does not require fine-grained descriptions. In fact, a more coarse-grained descriptions (I prefer to use speak in terms of representations) may enable greater precision in some cases. Consider an example. Suppose I represent doorways as larger than me or smaller than me. This is quite coarse-grained (note that I don’t think that we actually represent doors in this way!). My point is that this coarse-grained representation of relative size would be all that I need to make the discriminations that matter for action, such as passing through doors without bumping my head. Furthermore, I may be better at this task than if I represented the ratio between my size and the door’s size in a more fine-grained way. Suppose there is a door that is very close to my size and just barely passable. If I represent the ratio of that door in a fine-grained way, it may be difficult for me to distinguish that door from one that is just slightly smaller and not passable. In a hypothetical case like this one, a coarse-grained representation may enable more fine-grained discrimination of actual door size (relative to me). The upshot is this: Fine-grained discrimination does not require fine-grained representation or description.

    My account of perceptual precision may be closer to what Springle calls “instrumental precision,” though I wouldn’t want to accept all of what she says here. Springle suggests that instrumental precision “is a measure of a disposition of a non-intentional object.” But I don’t define precision as a measure of a disposition. On my view, perceptual precision is (roughly) an ability to discriminate among variables on the basis of perception.

    Though it is not the view that I take in this paper, I am very sympathetic to Springle’s suggestion that even if objective properties are represented inaccurately, “biologically relevant properties” may be represented accurately. This seems to me like a possible response to the evidence, depending on what is meant by a biologically relevant property. In developing this view it would be important to establish that these properties can provide conditions for accuracy (e.g. are they really properties of objects in the world, or are they subject-relative properties? If the latter, can we specify accuracy conditions for such properties in a way that doesn’t include reference to mentality or appearance?).

    Is Inaccuracy Really Widespread?

    Springle argues that the evidence that I provide does not establish that inaccuracy is widespread. Her main issue with my argument seems to be that in each of the empirical cases, subjects could be representing relative properties accurately, even if they misrepresent objective properties. I think Springle is right on this point, though whether subjects do in fact represent relative properties accurately is a further empirical question. My argument focused on showing that perceptual precision does not require accurate perception of objective properties. Once that is established, philosophers have a choice: embrace widespread inaccuracy, or argue for the view that subjects represent some other kind of property (like relative properties).

    Springle suggests that other philosophers have examined similar literature on spatial distortions, and have not been moved to abandon the view that perception aims at accuracy. There is indeed a centuries-long discussion in philosophy that uses cases like the ones that I have discussed to motivate variants of perceptual relativity (Springle discusses Hatfield, Hill & Bennet and Akins). What this paper adds to the mix is a new approach: rather than rescue veridicality by developing an account of relative properties, perhaps we should reject veridicality as the aim of perception. To my knowledge this is an under explored position debate, but I hope that alone will not count against it!

  2. Reply to Hill

    I am very grateful to Chris Hill for these supportive yet challenging comments. They have helped me to clarify my own thinking in this paper, as well as to think more deeply about how my view differs from accounts of perceptual relativity (such as Hill’s own view). I will continue to think about the issues that Hill has raised for a long time to come. I have done my best to address his main criticisms below, though there is much more to be said.

    Does precision require accuracy?

    Hill is right that my point in distinguishing accuracy from precision is to suggest that we can make precise discriminations among objective values even if we represent those values inaccurately. Hill suggests that in order for perception to be precise in my sense, we must accurately represent complex magnitudes. I’m not sure whether that is the case, and here’s why. Consider the example of discriminating between two angles. Suppose these have an objective slant of 5 and 6 degrees. I might be good at identifying the difference between 5 and 6 degrees precisely because they appear more different to me than they actually are. So my perception of the slants is precise even if it is inaccurate with respect to complex magnitudes, such as the objective difference between two values.

    Of course, even in this scenario, I accurately represent the 6 degree slant as larger than the 5 degree slant, even if I’m wrong about how much larger. I am only inaccurate with respect to objective properties as well as more fine-grained complex magnitudes, but I veridically perceive 6 degrees as larger than 5. Nonetheless I think it is quite surprising that precision does not require accuracy with respect to either objective values or more fine-grained complex magnitudes, and this is the point that I hoped to emphasize in drawing the distinction between accuracy and precision.

    Precision & Resolution

    Hill suggested that I could say more about work by Carrasco and colleagues on attention and perceptual resolution, and I’m happy to take the opportunity to do so here. For those who are not familiar with Carrasco’s work: over a series of studies, she and her colleagues have presented evidence that attention alters perceptual appearance. To summarize just a few of the findings in this literature: Attended contrast gradients appear boosted compared to unattended ones (Carrasco et al 2004); items presented in an attended spatial region appear faster than items presented in unattended regions (Stelmach & Herdman 1991; Spence & Parisse 2009); and attended items appear more saturated in color than unattended ones (Blaser, Sperling & Lu 1999; Fuller, Ling & Carrasco 2004). Since these examples were introduced to the philosophical literature by Block in 2010, philosophers have attempted to make sense of attention’s effect on appearance in a number of ways. One of the most popular accounts is that attention makes perceptual content more determinate (Nanay 2010, 2011; Stazicker 2011).

    I think the view in this paper suggests a different reading of the Carrasco cases than the determinacy reading. I focused on the Durgin examples because there is nothing indeterminate about the content represented (I happen to think this is true in many of the Carrasco-style cases as well, though I am in the minority). Slants simply look more steep than they actually are. Dots in the virtual hallway simply appear to move more quickly than they actually move. I think that once I have motivated the view that we are frequently inaccurate with respect to objective properties, it is more plausible to interpret attention’s affect on appearance as involving misrepresentation. This is a deep issue if perception aims at accuracy (as Block 2010 has explained). But it is less troubling if perception aims at precision. Attention helps us to see more precisely, but not by improving the resolution of vision, increasing determinacy, or making us more (or less) accurate.

    For those who are familiar with the Carrasco cases involving perceived contrast, you may notice that it is not obvious that attention makes perception more precise. In Carrasco’s study, attending to a patch of lesser contrast seems at first glance to obscure the objective difference between contrast patches. An attended patch of (say) 22% objective contrast may appear the same as a patch of unattended 28% objective contrast. But attention’s effect on appearance will only obscure difference in contrived experimental scenarios where the eyes and attention are held fixed; in the real world, that doesn’t happen very often. (If you don’t believe me, take a look at Carrasco’s study and try shifting attention without moving your eyes. It’s not easy to do!) In the real world beyond the constraints of psychology experiments, your eyes and attention jump around quite rapidly. When your attention shifts to the 28% contrast patch, and it is “boosted” to appear 34%, this is very obviously higher in contrast than the 22% patch. Given time to explore, attention’s effects on perceptual appearance will serve to magnify difference among the actual variables of objects and enable greater perceptual precision.

    I did not rely on attention and Carrasco-style cases to make my point in this paper because I recognize that my interpretation of those studies is controversial. That said, I think that the evidence for the effects of attention on perceptual appearance do indeed contribute to the empirical case for my view (or at least the view that perception is not aimed at accurate representation of objective properties — more on that below!)

    Consequences for Representationalism

    I agree with Hill that widespread misrepresentation of the sort that I argue for in this paper raises a serious problem for tracking theories of representationalism. The view that I defend in this paper undermines the covariation between representations and the items that they represent (In thinking through this point, I have found Mendelovici’s 2012 paper very helpful. This issue is also briefly discussed in Hill’s recent 2016 paper). My view thus has an important consequence which I have thought about a great deal, but did not explore in the present paper. If I am right, then the most promising account of representational content is wrong. It may be that these and similar empirical lessons on perception lead us to reject representationalism as an account of perceptual content.

    Hill has a suggestion for how I could avoid this consequence for representationalism. On his preferred account, perception is inaccurate with respect to objective properties, but still largely veridical with respect to relational, view-point dependent properties.

    Overall I think that the view I present in this paper is largely concordant with Hill’s preferred account, and I find much of what he says very plausible. One main difference between our views concerns whether there is good enough reason to posit the kinds of properties that will ground veridicality, given that so much of perceptual content is view-point dependent. For example, in order for inflated angles to be veridical, there must be some “inflated angle” property of the slant that grounds that veridicality. Hill agrees:

    “According to the picture I have in mind, inflated visual angles are real properties of objects, just as the “inflated” size property being of a height H such that two times H is twelve feet is a real property of objects. Thus, according to the picture, visual representations of size stand for properties that objects really have, and can therefore be said to be accurate.”

    It seems to me that there is an important difference between an inflated visual angle and the property of being a height H such that two times H is twelve feet. In the latter case, the complex property can be fully explained by the objective height H. It is simply another way of describing H mathematically. But in the case of relative properties like an inflated angle, the perceiver plays an essential role in determining which property is instantiated. The inflated angle is not fully explained by the objective angle in the way that the complex property above is fully explained by H. In the case of relative properties like an inflated angle, the property is a function of a relation between a subject and an object. So in what sense is an inflated angle a property of an object? Certainly not in the same sense as being a height H such that two times H is twelve feet.

    My bigger question is this: What reason do we have for attributing these inflated properties to objects, other than to rescue veridicality and representationalism? I prefer to explore an alternate route: What happens when we let go of a pre-commitment to the idea that perception is veridical or that representation requires covariation, and let the empirical lessons lead the way to a theory of perception?

    Blaser, E., Sperling, G., & Lu, Z.-L. (1999). Measuring the amplification of attention. Proceedings of the National Academy of Sciences, USA, 96, 11681-11686.
    Block, N. (2010) “Attention and Mental Paint.” Philosophical Issues. 20:23—63.
    Carrasco, M., Ling, S., & Read, S. (2004) “Attention alters appearance.” Nature Neuroscience:  7:308—313. 
    Fuller, S., Ling, S., Carrasco, M.(2004). Attention increases perceived saturation [Abstract]. Journal of Vision, 4( 8): 329.
    Mendelovici, A. (2012) “Reliable Misrepresentation and tracking theories of mental representa- tion.” Philosophical Studies 165 (2):421-443)
    Nanay, B. (2010) “Attention and Perceptual Content.” Analysis. 70: 263—270.
    Spence, C. & Parise, C. (2009) “Prior-entry: A review.” Consciousness and Cognition.
    Stazicker, J. (2011) “Attention, Visual Consciousness, and Indeterminacy.” Mind & Language. 26:2 156—184.
    Stelmach, L. B. & Herdman, C.M. (1991) “Directed Attention and Perception of Temporal Order.” Journal of Experimental Psychology: Human Perception and Performance. 17: 539—550.

  3. Hello Adrienne,

    This is a very cool paper! The idea that successful perception need not be veridical sounds convincing. The empirical cases you describe, in which perceptual distortions improve precision, sound convincing as well.

    For simplicity, I limit my discussion to Carrasco’s findings. On your account, attention distorts perception, making it inaccurate (e.g., apparent size increases) in order to increase precision. What about perception outside attention? Is it accurate (but imprecise)? You don’t (unless I missed it) say so explicitly, but it seems like a plausible view, right? If attention distorts perception, then outside attention perception is (usually?) not-distorted, and hence is veridical (I ignore Durgin’s cases, of walking, of seeing slopes of hills, etc.).

    Suppose this is right. Now, on your view, perception aims at precision, not accuracy (and precision does not require accuracy). If so, then I wonder: why is it that perception outside attention is accurate? It can’t be an accident, right?

    I hope that the worry is clear. Let me try to reformulate it. If perception aims at precision, not accuracy, and if precision does not require accuracy, we should expect perception to be inaccurate yet precise, regardless of attention. But (apparently) this is not the case: on the picture coming from Carrasco’s lab, we need attention to distort experience and thereby increase precision. Without attention, perception is veridical (but imprecise).

    Does that make sense?

  4. Thank you for your question, Assaf! I think this gets at a really interesting issue.

    Carrasco does say that attention boosts perceived contrast. In my reply to Hill, I adopted Carrasco’s way of speaking. But the evidence doesn’t actually show that attention boosts perceived contrast. What it shows is that an attended contrast of 22% looks indistinguishable from an unattended contrast of 28%. This could be because attention boosts contrasts, OR because contrast outside attention is distorted/suppressed.

    So it may be that perception of contrast is more accurate outside attention. But I don’t think we have been given good empirical reason to believe this from Carrasco’s work, so I wouldn’t want to assert it. Even if perceived contrast is boosted, I would be careful about moving from perception of a specific property (like contrast) to a general claim that perception is more accurate outside the focus of attention. We’d need to look at specific properties and types of attention (spatial, object-based, etc) to make that further claim.

    A final point in response to your worry. In the example of perceived contrast, attending increases precision by inflating the perceived difference between the attended and unattended contrast patches. Given this, I may not have to say that the unattended patch is veridically but imprecisely perceived. Attention makes perceptual experience more precise because one contrast is distorted in a way that magnifies the difference between them, regardless of which contrast is represented veridically (maybe neither!)

    Thank you again for raising this worry. It’s something that I’ll think more about, especially when considering how I should describe Carrasco-style cases.

  5. I would like to respond briefly to Adrienne Prettyman’s replies to my comments on her paper. I will ask for a clarification of her views about one issue, and try to locate the difference between our views about another.

    According to Prettyman, the perception of certain magnitudes (e.g., slant, distance, optic flow) is precise but inaccurate. She also claims that enhanced precision is useful to the perceiving subject because it makes for successful discrimination among the values of a perceived magnitude. One of my concerns in my comments was to explain how this package of claims is consistent. On the face of it there seems to be a problem, because successful discrimination clearly requires accuracy of some kind. More precisely, if a subject can discriminate successfully between, say, the slant of hill H1 and the slant of hill H2, then the subject’s representations of the two slants must be accurate in some respect. What I proposed was that perception accurately registers certain relations among the slants of hills — for example, the relation being greater than. This proposal explains how perception can be inaccurate while enabling successful discrimination, because it is possible to represent the fact that the slant of H1 is greater than the slant of H2 accurately even though one is misrepresenting both the slant of H1 and the slant of H2. Prettyman accepts this interpretation in her comments. But this leaves us with an important question. As Prettyman notes, it is possible to register the fact that the slant of H1 is greater than the slant of H2 without registering the numerical difference between the two slants – the number that is obtained by subtracting the slant of H2 from the slant of H1. Should we conclude that perception only represents the relational fact that the slant of H1 is greater than the slant of H2, or should we suppose that it also represents the size of the gap between the two quantities? (Of course, we can represent the size of the gap between two quantities accurately even though our representations of the quantities are inaccurate.) If I understand her, Prettyman prefers the former view. But I want to make sure that this reading is correct, because it seems that there are many contexts in which information about numerical differences or ratios would be invaluable. Consider, for example, a situation in which H1 is much steeper than H2, but in which H1 offers better footholds. You must decide which hill to climb. Here it would be very useful know the numerical difference between the slopes, or the ratio, or at least to be able make qualitative judgements concerning large families of relations like being greater than by a tiny amount, being greater than by slightly more than a tiny amount, and so on. So I wonder: have I construed Prettyman correctly in supposing she believes that purely qualitative relational facts like the slope of H1 is greater than the slope of H2 are the only ones that are registered, or does she have some way of accommodating examples like the one I just described?
    In my comments I aired my views (i) that the visual system computes functions of input quantities like visual angles, (ii) that it is the outputs of certain of these functions that we experience directly in conscious perception, and (iii) that the outputs the functions in question are relational, view-point dependent properties of external objects. I also maintained that these viewpoint-dependent properties are objective, meaning thereby that there is no need to appeal to mental states of the perceiver or processes in the perceiver’s visual system to define them. I take it that claims (i) and (ii) are perfectly orthodox. What is special to my view is that the functions cited in (i) and (ii) never yield true constancies, with the result that the properties that are their outputs cannot be said to be constant, intrinsic properties of objects. Rather, as claim (iii) maintains, they are viewpoint-dependent properties. More specifically, they are properties that to some degree reflect changing aspects of the perceiver’s relations to the object – changes in distance, changes in angle of view, changes in relative velocity, and so on. They are more constant than the fleeting projections on the retina, but less constant than intrinsic properties of objects like physical sizes and physical shapes. Now Prettyman seems to doubt that viewpoint-dependent properties of this sort count as objective. Thus, she says that if a perceiver apprehends a viewpoint-dependent property, “the perceiver plays an essential role in determining which property is instantiated.” But viewpoint-dependent properties can be perfectly objective. Consider the visual angle V that an object subtends with respect to the point in space where the agent’s lens is located. The fact that the object subtends V is a perfectly objective fact about the object. The fact is completely independent of the subject’s mental states and visual processing. Indeed, it would have obtained even if the observer had never existed. Now let f be a function that is computed by the human visual system. Clearly, the fact that f yields a certain value when it is applied to V is another perfectly objective fact. The “subjectivity” of a computational process has no tendency to show that facts involving the function computed the process are subjective. (My pocket calculator can compute the addition function, but that has no tendency to show that facts involving the addition function are calculator-dependent.)

    All of this is obvious, so why did Prettyman challenge my claim about objectivity by saying that “the perceiver plays an essential role in determining which property is instantiated.” Here is a conjecture: she was thinking of cases in which the visual system is arguably concerned with properties that are viewer-dependent rather than viewpoint-dependent. It is plausible that this often happens. Thus, for example, it is plausible that the visual system takes the perceiver’s eye height into account in calculating the distance from the perceiver to the object of awareness. (See, e.g., Palmer 1999, p. 232.) Now the distance from a perceiver’s eye to the ground is a property of the perceiver, not of the object that is perceived, so it is plausible that computations pertinent to distance take viewer-dependent properties into account. If this was Prettyman’s point, then we are in full agreement. But I would note that facts about the eye heights of observers are perfectly objective. In general, viewer-dependent facts are just as objective as viewpoint-dependent facts, provided that they are facts about the body of the viewer.

    Reference
    Palmer, S. (1999). Vision Science. Cambridge, MA: MIT Press.

  6. Thanks for asking both of these questions, Chris. I regret that I am posting my response so late in the session, and I hope we will have the chance to continue this conversation.

    In response to your clarification question:

    I agree with you that it can be useful to represent numerical differences or ratios, and that we probably do represent these and not just relational facts (such as that slant H1 is great than H2). I can see how this was unclear in my original response to your comments. As I understand it, the evidence for scale expansion shows that subjects are inaccurate about complex magnitudes as well as objective properties. (For example, the difference between 5 and 6 degree slant appears greater than the difference between 25 and 26 degrees). So what explains the subject’s ability to make fine-grained discriminations, and how do I avoid the charge of inconsistency which you pointed out in the original comments? My suggestion is that the subject does represent H1 as greater than H2, even if she has inflated just how much greater. So we can describe one and the same perceptual content as accurately representing the relational fact about which slant is greater, while inaccurately distorting the magnitude of the difference between the slants.

    I think your example (involving a steep hill with good footholds versus a more shallow hill) gives a good illustration of this. I agree that it would be helpful for a climber to have information about numerical differences and not just relational facts like “greater than”. But the most helpful information for the climber may not be accurate information about numerical difference. I think the evidence in my paper suggests that the distortion of complex magnitudes would actually help a climber to make the decision of which hill to climb. This is because the distortions of objective slant serve to magnify differences between slants that might otherwise be difficult to discern.

    On locating the difference between our views:

    When I wrote that “the perceiver plays an essential role in determining which property is represented,” you are right that I was thinking of cases more like the perceiver’s eye height (“viewer-dependent” rather than “viewpoint-dependent” properties). But I think you are also obviously correct that eye height is a perfectly objective fact. Nothing about eye height requires us to appeal to mental states of the perceiver or processes in the visual system.

    Other cases may be more problematic, however, if they do require an appeal to mental states of the perceiver rather than facts about her body. For example, there is some anecdotal evidence that slants of hills are perceived differently by expert skiers than novices (this was mentioned in Ross 1974 which I reference briefly in my paper). More generally there’s evidence that experts perceive the world in specialized ways (e.g. see Connolly 2016 for a review). Consider the expert skier and the novice, and suppose it is right that they experience the slant of a hill differently in virtue of the expert’s visual learning. In this case, it seems to me that mental facts about the perceiver (or at least facts about her visual processing) are playing a role in determining which property she perceives. I’m curious what you would think about this and other cases involving expert perception.

    Covert attention may provide another case in which properties are not just viewpoint-dependent, but viewer-dependent in a way that challenges objectivity. I’m thinking of the Carrasco-style cases in which attention is shifted covertly, that is, without movement of the eyes. Accounting for the effect of covert attention on appearance seems to require that we make reference to something mental: where the subject is focusing. At the very least it looks like we will need to reference facts about visual processing (like the effect of visual attention) in order to account for the property perceived. Again, this seems to show that properties are viewer-dependent in a way that may be problematic for objectivity.

    References:
    Ross, H.E. (1974) Behavior and perception in strange environments. London: Allen & Unwin.
    Connolly, K. (2017) “Perceptual Learning,” Stanford Encyclopedia of Philosophy https://plato.stanford.edu/entries/perceptual-learning/

Comments are closed.