The effects of perception on cognition are generally incontrovertible. For example, in normal circumstances, seeing a yellow banana typically leads to the formation of the belief that there is a yellow banana. By contrast, whether cognition can affect perception is a subject of debate. For instance, one may ask whether the belief that bananas are yellow can affect the way a banana is perceived. In this example, cognitive penetrability occurs just in case the belief that bananas are yellow causes, say, a gray banana to look yellow to a subject. The cognitive penetration debate has primarily focused on visual perception, although recent research has extended to other sensory modalities, including audition. The main concepts in this debate are modularity— whether cognitive information processing can affect perceptual information processing—and the nature of perceptual content—whether the content of experience can be affected by cognition.

History

In the 1950s, psychology underwent a significant shift away from behaviorism (which focused on observable behavior and was often skeptical of mental states) and Gibsonian direct perception theories (which posited that perception is direct and unmediated by cognitive inferences; Miller, 2003). This shift gave rise to the New Look movement, which was inspired by the Helmholtzian view of perception as an active process involving unconscious inferences (von Helmholtz, 1866/1962) [see Unconscious Inference]. The New Look movement experimentally explored the perception/cognition boundary, specifically whether cognitive states such as beliefs, knowledge, goals, expectations, and emotions can affect perception (Bruner et al., 1951). For example, one study found that children from lower-income families perceived coins as being larger than equal-sized neutral discs (Bruner & Goodman, 1947). Such findings sparked two distinct (but related) cognitive penetration debates. 

The first debate stems from the computer revolution, which gave rise to the view that the mind is a computational system comprised of autonomous modules that are domain specific and informationally encapsulated—a position central to the computational theory of mind [see Theory of Mind]. Within this framework, the cognitive penetration debate is framed in terms of information processing. Specifically, the question is whether perception is an encapsulated process driven purely by sensory information processing or whether it is cognitively penetrated. According to the cognitive impenetrability thesis, perception is an informationally encapsulated process (Fodor, 1983; Pylyshyn, 1984; Raftopoulos, 2001). 

The second debate stems from New Look–type experiments, which indicate that cognitive states such as beliefs can affect the perceptual content of experience (Macpherson, 2012). Consider, for example, the experiment in which children from lower-income families perceived coins as being larger than equal-sized neutral discs. If children perceived the coins as larger than the neutral discs because they believed the former to be more valuable than the latter, then this would count as evidence for cognitive penetration because it is their belief that accounts for the differences in the content of their perceptual experience. However, if the children perceived the coins to be the same size as the neutral discs but only judged them to be different in size because of their belief about the coins’ value, then this would not count as evidence for cognitive penetration because the content of their perceptual experience remained the same (Macpherson, 2017) [see Mental Representation]. 

Core concepts

Modularity and information encapsulation

According to the computational theory of mind, perceptual and cognitive modules are characterized by their domain specificity and informational encapsulation. Domain specificity refers to the particular kind of information that a module processes. For example, visual modules are dedicated to sensory inputs such as textures and patterns, whereas cognitive modules process mental states such as beliefs, knowledge, expectations, and goals. Informational encapsulation, on the other hand, refers to the sources of information a module can utilize during its operation (Churchland, 1988; Pylyshyn, 1984; see also Raftopoulos, 2001). For an encapsulated perceptual module, this means its processing is restricted to bottom-up sensory inputs (e.g., information from visual inputs at the retinal level is sent first to the lateral geniculate nucleus and then on to the visual cortices) and cannot be affected by high-level cognitive information (e.g., beliefs or knowledge). Cognitive penetration requires that cognitive processes directly modulate perceptual processes in a top-down manner; for example, the knowledge that a stimulus is a banana must be able to modulate low-level perceptual processing. It follows that if perception is an informationally encapsulated process, then it is not subject to top-down modulation from cognition.

Proponents of the cognitive impenetrability thesis maintain that early vision, that is, the early stages of visual processing, is informationally encapsulated. Optical illusions are often used as evidence for this thesis. The two lines of the Müller–Lyer illusion continue to appear unequal in length even after one learns that they are equal in length. This suggests that the knowledge that the lines are equal in length is not integrated into how the lines are perceived. As a result, the two lines continue to appear unequal in length.

Early vision is defined by the initial bottom-up processing of visual information along the pathway from the retina to the primary visual cortex and into higher visual areas, a process largely confined to the first 100 ms following stimulus presentation (Fodor, 1983; Pylyshyn, 1999; Raftopoulos, 2001) [see Theory of Mind]. Early vision involves two processing stages (Raftopoulos, 2001). The first stage consists of fast feedforward sweeps (lasting about 100 ms), in which visual signals (carrying basic elements such as lines and edges) are processed in a bottom-up direction. These sweeps begin with the lateral geniculate nucleus and proceeds through visual processing areas from the primary visual cortex (V1) to V4 and up to the inferotemporal cortex, where high-level information is extracted (e.g., determining whether an image contains a banana or a concave shape). The second stage (which begins at about 100 ms and peaks at 150 ms) involves lateral and recurrent connections, where signals travel back and forth within the visual areas (e.g., from higher areas back to lower areas via feedback connections). This process leads to further integration of individual features into coherent objects (a process known as binding) and the separation of objects from their background. 

Feedforward sweeps can lead to early categorization. For example, studies indicate that familiarity with objects like bananas can influence classification at very short latencies of 85 to 100 ms (Kirchner & Thorpe, 2006). However, if these sweeps are not affected by cognitive information, they cannot provide evidence for the cognitive penetrability of visual perception (Raftopoulos, 2001). 

Moreover, attention-based modulation does not provide evidence of cognitive penetration because the changes in perception result from changes in the stimulation (e.g., the subject’s visual system selects different sensory data) rather than cognitive states (Firestone & Scholl, 2016; but see Cecchi, 2014). 

Perceptual contents 

Perceptual experiences represent things in the world—they are directed at something. For example, when one sees a yellow banana, one’s visual experience represents the yellow banana—it is directed or is about the yellow banana. Similarly, when one hears a bird chirping, one’s auditory experience represents or is about the bird’s chirping. This aboutness of perceptual experience is what is called the content of experience. Cognitive states such as beliefs also have contents—they too are about something. For example, one’s belief that there is a yellow banana on the table is about the yellow banana on the table. Similarly, one’s belief that the bird in one’s yard is chirping is about the bird chirping in one’s yard.

Empirical research, particularly work inspired by the New Look movement, has investigated whether cognitive states such as beliefs or knowledge can penetrate the content of visual experience. In a classic New Look–type experiment, subjects reported perceiving orange cutouts of prototypically red objects (e.g., apples) as redder than identical orange neutral shapes (Delk & Fillenbaum, 1965; see also Hansen et al., 2006). For such experimental results to count as evidence of the cognitive penetrability of visual experience, the changes in the content of the subject’s experience must be attributed to their cognitive states (e.g., their belief that apples are red). If the changes in the content of the subject’s experience can be attributed to shifts in their attentional focus (e.g., the explanation for the orange cutout appearing red is that the subject turns their attention to an adjacent red cutout), then such findings cannot provide evidence for the cognitive penetrability of visual perception (Firestone & Scholl, 2016; Zeimbekis, 2013). Similarly, if the changes in the content of the subject’s experience can be ruled out as instances of misreports (e.g., the subject experiences the cutout as orange but incorrectly reports it to be red) or misjudgments (e.g., the subject experiences the cutout as orange but incorrectly judges it to be red), then such findings cannot provide evidence for the cognitive penetrability of visual perception (Brogaard & Gatzia, 2017; Firestone & Scholl, 2016).

Questions, controversies, and new developments

Predictive coding has emerged as a leading framework for explaining the mechanisms of cognitive penetrability (Clark, 2013; Hohwy, 2013; Lupyan, 2015; Marchi, 2020). In this framework, the brain is conceived as a hierarchical Bayesian inference mechanism that continuously generates top-down predictions about sensory input and minimizes prediction error through reciprocal communication between cortical levels, in which higher levels convey predictive signals, and lower levels indicate prediction errors. 

The predictive coding framework suggests that perception is an active process of hypothesis testing driven by prior expectations. If so, this framework has significant implications for the cognitive penetrability debate, although its impact depends on how its core components are interpreted (Macpherson, 2017). For instance, the framework may require a clearer definition of what constitutes a cognitive state versus a perceptual expectation within its hierarchical architecture, which would force a reexamination of whether the traditional boundary between perception and cognition can be meaningfully maintained. In other words, if the boundary between perception and cognition is blurred, the predictive coding framework may necessitate a revision—or perhaps an abandonment—of the traditional terms of the cognitive penetrability debate.

Moreover, the possibility of cognitive penetration invites us to revisit foundational concerns in epistemology regarding perceptual justification. For example, if what one sees (e.g., a neutral gesture) is influenced by what one already believes (e.g., that members of other racial groups are threatening), this raises the possibility that perception may not serve as an independent source of evidence for belief but instead may simply reinforce prior commitments (Siegel, 2012). This prompts further inquiry into whether perceptual experiences can retain their justificatory force (i.e., their capacity to serve as evidence for beliefs) if they are influenced by antecedent cognitive states or whether this influence introduces a problematic circularity (in which beliefs shape perception, and perception is then used to justify those very beliefs).

Broader connections

Concerns about cognitive penetration are not merely theoretical. They intersect with empirical questions about how the mind integrates information and how perception operates in diverse contexts. Through these interdisciplinary intersections, the debate on cognitive penetration serves as a powerful theoretical framework, linking experimental psychology, clinical research, applied cognitive science, and philosophy in a way that enriches each field’s approach to understanding the mind.

Clinically, the notion of cognitive penetrability provides a critical benchmark for understanding pathologies such as hallucinations in schizophrenia. For example, these phenomena could be interpreted as a breakdown of the normal, encapsulated processes of early vision, in which maladaptive top-down influences (e.g., distorted priors or predictions) pathologically dominate sensory processing, creating perceptual experiences disconnected from reality. This intersection between research on the possibility of cognitive penetrability and clinical neuroscience may deepen our understanding of psychiatric disorders (e.g., as disorders of predictive processing) and may suggest new paths for therapeutic intervention aimed at restoring the balance between bottom-up sensory evidence and top-down influence. 

In more applied settings, such as radiology education and human–computer interface design, insights into how cognition may influence perceptual judgment could inform the development of training systems that enhance accuracy and reduce error by accounting for the role of expectations and prior knowledge in perceptual processing.

Further reading 

• Firestone, C., & Scholl, B. J. (2016). Cognition does not affect perception: Evaluating the evidence for “top-down” effects. Behavioral and Brain Sciences, 39, e229. https://doi.org/10.1017/S0140525X15000965

• Marchi, F. (2020). The attentional shaping of perceptual experience: An investigation into attention and cognitive penetrability (Studies in Brain and Mind, Vol. 16). Springer.

• Zeimbekis, J., & Raftopoulos, A. (Eds.). (2015). The cognitive penetrability of perception: New philosophical perspectives. Oxford University Press.