On 2015 Sep 06, Lydia Maniatis commented:

The claims of Blakeslee and McCourt are flawed on logical, methodological, theoretical, and empirical grounds.

Perhaps the core error is the denial of perceptual facts, as described below (and as noted also by Gilchrist in his commentary on this article).

In the Adelson checker-shadow illusion (to take one example), a check in apparent shadow looks white, while an equiluminant check, apparently in plain view, looks black. Aside from the additional presence of the apparent shadow, the experience of the two surfaces is similar to looking at a white check and at a black check under homogeneous illumination. Kohler (1935), describes a “real-world” version wherein a white and a black paper appear white and black, respectively, even if the illumination is adjusted so that the two surfaces are actually equiluminant.

Imagine, now, what would happen if we asked a naive observer to report on “the intensity of the light coming from the surface of each of the two checks.” First, he or she would likely assume the question referred to the apparent illumination of the surfaces. In this case, the white, “shadowed” check should receive a lower rating than the black “plain-view” one. If we then tried to clarify that we want the observer to make matches based on “the amount of light each surface is sending to the eye,” (its luminance) I think the observer would have trouble a. understanding what we are asking and b. performing the required task. And I'm not sure anyone would be able to judge, with confidence, whether the two checks in the Adelson checkerboard are, or are not, in fact equiluminant. That's what makes the demo so impressive. Even if observers could estimate this value, the task would be difficult and the results unreliable. To achieve it, they would have to focus narrowly on each square, isolating it from the surround.

Yet, Blakeslee and McCourt maintain that the latter task, which requires viewers to overide their spontaneous, salient perceptual experience, is “strictly based on appearance,” while the former, effortless experience is “based on an inferential judgment.” If, however, we define “appearance” as “what something looks like,” then the authors' arguments are obviously false, as can be confirmed by any observer.

The authors' argue that the experience which is quite literally based on appearance, is actually a product of learning. This is a major claim (implying that learning can actually alter a percept from black to white), but the authors offer no evidence for it. All of the arguments and available evidence is against. As Gilchrist (2015) points out, even fish seem to naturally achieve this kind of learning. A child can see the Adelson checker-shadow effect as effortlessly as an adult. Our perceptions aren't affected by what we learn about the nature of light and the properties of surfaces, we don't have to learn how to judge when a surface is in shadow or merely darker than its neighbor, or when it is covered by various types of transparency, we don't even have to learn that at night things don't actually change color. Given the difficulty scientists have in analyzing and modelling percieved lightness, and given that massive, early exposure to artificial images mimicking illumination variation has no discernible effect on our perception of the “real” world, the claim that people go through a process of learning to make the inferences necessary to achieve veridical percepts in natural conditions does not seem credible. B and M have certainly not tested it.

The quality that the authors argue is “strictly based on appearance” is a quality that they term “brightness,” defined as the perceptual correlate of luminance. The view that there is a perceptual correlate of luminance seems to be uncontroversial among lightness researchers (e.g. Kingdom, 2011; Gilchrist, 1999)), although Anderson (2014) seems to define brightness (more properly, in my opinion) as apparent illumination. The claim seems easy to refute.

Suppose we observe a set of surfaces lacking cues to differential illumination, and that some appear white, some gray, some black. Suppose, then, that we observe the same set of surfaces, at a different time, under a different degree of illumination. Assume that have completely forgotten the previous experience with the surfaces, and are again asked to judge their white/gray/black character. Our responses will typically be similar to those we gave in the first (now forgotten) instance. In other words, even though the illumination (and consequently the luminance) of the surfaces will have changed, our responses will remain the same. If the range of luminances is complete enough, they will, in both cases, be correlated with reflectance and not with luminance. Thus, even under homogeneous illumination, we cannot say that we are perceiving luminance, or that perception is more direct than in other situations. As always, the percept is the product of a complex visual process based on luminance values and structural assumptions.

When it comes to the case of non-homogenous (apparent) illumination, the authors seem to be treating illumination boundaries as though they were directly-perceived facts serving to support “inferentially” perceived lightness judgments. They say, for example, that “when the illumination component is clearly visible” the observer can use “brightness contrast” at the boundary to infer the magnitude of the illumination. There are a number of problems with this description.

First, if a shadow is perceived – is “clearly visible” - as the cause of the luminance boundary, then viewers are perceiving a double-layer – a surface with lightness x and an apparent shadow of darkness y lying on top of it. They are not perceiving a single “brightness” value. The authors are using the term “brightness” when they actually mean luminance.

Relatedly, the illumination boundary only becomes “clearly visible” after the visual process has inferred its presence based on the luminance structure – including the relative luminances at luminance boundaries - of the image. Whether the darker side of an edge will be perceived as being similar in reflectance, but lower in illumination than its neighbor, or as darker than its neighbor due to a lower reflectance, or any other combination of possibilities, depends on the global structure of the image. Given certain conditions, even a non-existent luminance edge may produce an apparent lightness difference, as in the case of illusory surfaces. So the argument that perceiving a surface as continuing beneath a “shadow” boundary is more inferential than perceiving a particular surface as gray due to its luminance relative to other surfaces in an image is naive. If “appearance-based” means “based on luminances”, then all perception is appearance-based. If it means that surfaces are perceived based on local luminance conditions, then it never is. Local conditions do not even determine photoreceptor activity in the horseshoe crab.

As corroborating evidence, the authors point to a few references, including Blakeslee and McCourt (2008), which is supposed to prove the existence of “brightness” judgments. Their stimuli consist of the classic simultaneous contrast demo plus variations that create weak impressions of differential illumination. The “brightness” judgments are defined as those that arise when observers are instructed to focus narrowly on the targets. This is similar to applying a mask. Effectively, we are talking about the same visual process acting on a different stimulus, not about a different type of judgment. Due to the weakness of the structural cues to differential illumination in B and M's (2008) stimuli, the ability of observers to isolate the target in this way is very easy. The demand would be much more difficult, and the results surely very different, if the stimulus had been the Adelson checkerboard.

On 2015 Sep 06, Lydia Maniatis commented:

The claims of Blakeslee and McCourt are flawed on logical, methodological, theoretical, and empirical grounds.

Perhaps the core error is the denial of perceptual facts, as described below (and as noted also by Gilchrist in his commentary on this article).

In the Adelson checker-shadow illusion (to take one example), a check in apparent shadow looks white, while an equiluminant check, apparently in plain view, looks black. Aside from the additional presence of the apparent shadow, the experience of the two surfaces is similar to looking at a white check and at a black check under homogeneous illumination. Kohler (1935), describes a “real-world” version wherein a white and a black paper appear white and black, respectively, even if the illumination is adjusted so that the two surfaces are actually equiluminant.

Imagine, now, what would happen if we asked a naive observer to report on “the intensity of the light coming from the surface of each of the two checks.” First, he or she would likely assume the question referred to the apparent illumination of the surfaces. In this case, the white, “shadowed” check should receive a lower rating than the black “plain-view” one. If we then tried to clarify that we want the observer to make matches based on “the amount of light each surface is sending to the eye,” (its luminance) I think the observer would have trouble a. understanding what we are asking and b. performing the required task. And I'm not sure anyone would be able to judge, with confidence, whether the two checks in the Adelson checkerboard are, or are not, in fact equiluminant. That's what makes the demo so impressive. Even if observers could estimate this value, the task would be difficult and the results unreliable. To achieve it, they would have to focus narrowly on each square, isolating it from the surround.

Yet, Blakeslee and McCourt maintain that the latter task, which requires viewers to overide their spontaneous, salient perceptual experience, is “strictly based on appearance,” while the former, effortless experience is “based on an inferential judgment.” If, however, we define “appearance” as “what something looks like,” then the authors' arguments are obviously false, as can be confirmed by any observer.

The authors' argue that the experience which is quite literally based on appearance, is actually a product of learning. This is a major claim (implying that learning can actually alter a percept from black to white), but the authors offer no evidence for it. All of the arguments and available evidence is against. As Gilchrist (2015) points out, even fish seem to naturally achieve this kind of learning. A child can see the Adelson checker-shadow effect as effortlessly as an adult. Our perceptions aren't affected by what we learn about the nature of light and the properties of surfaces, we don't have to learn how to judge when a surface is in shadow or merely darker than its neighbor, or when it is covered by various types of transparency, we don't even have to learn that at night things don't actually change color. Given the difficulty scientists have in analyzing and modelling percieved lightness, and given that massive, early exposure to artificial images mimicking illumination variation has no discernible effect on our perception of the “real” world, the claim that people go through a process of learning to make the inferences necessary to achieve veridical percepts in natural conditions does not seem credible. B and M have certainly not tested it.

The quality that the authors argue is “strictly based on appearance” is a quality that they term “brightness,” defined as the perceptual correlate of luminance. The view that there is a perceptual correlate of luminance seems to be uncontroversial among lightness researchers (e.g. Kingdom, 2011; Gilchrist, 1999)), although Anderson (2014) seems to define brightness (more properly, in my opinion) as apparent illumination. The claim seems easy to refute.

Suppose we observe a set of surfaces lacking cues to differential illumination, and that some appear white, some gray, some black. Suppose, then, that we observe the same set of surfaces, at a different time, under a different degree of illumination. Assume that have completely forgotten the previous experience with the surfaces, and are again asked to judge their white/gray/black character. Our responses will typically be similar to those we gave in the first (now forgotten) instance. In other words, even though the illumination (and consequently the luminance) of the surfaces will have changed, our responses will remain the same. If the range of luminances is complete enough, they will, in both cases, be correlated with reflectance and not with luminance. Thus, even under homogeneous illumination, we cannot say that we are perceiving luminance, or that perception is more direct than in other situations. As always, the percept is the product of a complex visual process based on luminance values and structural assumptions.

When it comes to the case of non-homogenous (apparent) illumination, the authors seem to be treating illumination boundaries as though they were directly-perceived facts serving to support “inferentially” perceived lightness judgments. They say, for example, that “when the illumination component is clearly visible” the observer can use “brightness contrast” at the boundary to infer the magnitude of the illumination. There are a number of problems with this description.

First, if a shadow is perceived – is “clearly visible” - as the cause of the luminance boundary, then viewers are perceiving a double-layer – a surface with lightness x and an apparent shadow of darkness y lying on top of it. They are not perceiving a single “brightness” value. The authors are using the term “brightness” when they actually mean luminance.

Relatedly, the illumination boundary only becomes “clearly visible” after the visual process has inferred its presence based on the luminance structure – including the relative luminances at luminance boundaries - of the image. Whether the darker side of an edge will be perceived as being similar in reflectance, but lower in illumination than its neighbor, or as darker than its neighbor due to a lower reflectance, or any other combination of possibilities, depends on the global structure of the image. Given certain conditions, even a non-existent luminance edge may produce an apparent lightness difference, as in the case of illusory surfaces. So the argument that perceiving a surface as continuing beneath a “shadow” boundary is more inferential than perceiving a particular surface as gray due to its luminance relative to other surfaces in an image is naive. If “appearance-based” means “based on luminances”, then all perception is appearance-based. If it means that surfaces are perceived based on local luminance conditions, then it never is. Local conditions do not even determine photoreceptor activity in the horseshoe crab.

As corroborating evidence, the authors point to a few references, including Blakeslee and McCourt (2008), which is supposed to prove the existence of “brightness” judgments. Their stimuli consist of the classic simultaneous contrast demo plus variations that create weak impressions of differential illumination. The “brightness” judgments are defined as those that arise when observers are instructed to focus narrowly on the targets. This is similar to applying a mask. Effectively, we are talking about the same visual process acting on a different stimulus, not about a different type of judgment. Due to the weakness of the structural cues to differential illumination in B and M's (2008) stimuli, the ability of observers to isolate the target in this way is very easy. The demand would be much more difficult, and the results surely very different, if the stimulus had been the Adelson checkerboard.