2.1 Unconscious vision?

One (putative) unconscious mental phenomenon that currently receives a great deal of attention in the philosophical and psychological literature is unconscious vision. The most prominent case of putatively unconscious vision is the so-called blindsight in patients with lesions in specific areas of the visual cortex (V1). These patients with unilateral cortical lesions report not seeing anything in the contralateral visual field. Still, in forced choice visual detection tasks, they can detect stimuli in this part of their visual field with great accuracy (Weiskrantz Reference Weiskrantz1995). Whether blindsight in patients really provides evidence for unconscious vision remains controversial. One reason is that it cannot be excluded that the brain damage leads to a distortion of the capacity to report visual stimuli (Persuh Reference Persuh2018). This objection can be avoided by focusing on nonpathological cases. (Putative) unconscious vision could be observed in healthy participants by using different paradigms, such as binocular rivalry (Yang, Zald, and Blake Reference Yang, Zald and Blake2007).

Binocular rivalry is an experimental paradigm in which each eye is presented with a different picture, resulting in the two pictures alternately fading in and out of consciousness. Binocular rivalry is used in different variants to investigate putative unconscious vision (see Phillips and Block Reference Phillips, Block and Nanay2016 for an overview). When presenting a so-called Mondrian (patterns of colored or black-and-white rectangles) to one eye, the conscious perception of the image presented to the other eye can be suppressed for minutes (i.e., participants will report that they do not see the other image). However, if the other eye is presented with a picture that depicts, for example, nudes, it will attract or repel the participant’s spatial attention depending upon their gender and sexual preferences (Jiang et al. Reference Jiang, Costello, Fang, Huang and He2006).Footnote ¹ This was established by having participants categorize Gabor patches that were presented after the Mondrian/nude trial in either the location of the nude or the location of the Mondrian. The participants were asked to indicate the orientation of the Gabor patch. One result was that male participants were more accurate in categorizing the Gabor patches if they appeared at the location of the picture of the nude if the nude was female. They were less accurate when the Gabor patches were at the location of a previously presented nude male. Block (Phillips and Block Reference Phillips, Block and Nanay2016) argues that these findings provide evidence for unconscious vision, as the participants reported not seeing anything but the Mondrian; hence, whatever was happening was not conscious to the participants. In addition, the fact that pictures of nudes attracted or repelled the participants’ attention depending on their gender and sexual orientation can only be explained by the assumption that they are perceived. Hence, there is unconscious vision.

However, not all researchers are convinced that the results of these studies provide evidence for unconscious vision (Newell and Shanks Reference Newell and Shanks2014; Phillips Reference Phillips2018). One criticism, which Phillips (Reference Phillips2018) calls the problem of the criterion, targets the most-common method used to establish that participants are not conscious of the stimuli: subjective reports of the absence of conscious vision. It is argued that such reports are compatible with degraded conscious vision (Peters et al. Reference Peters, Kentridge, Phillips and Block2017). A second objection, which Phillips (Reference Phillips2018) calls the problem of attribution, holds that it is unclear whether the empirical methods justify the postulation of unconscious vision (i.e., of a mental phenomenon). According to Phillips (following Burge Reference Burge2010) vision is an individual-level phenomenon. The problem of attribution arises for empirical studies that “fail to show that the states they implicate are attributable to the individual” (Phillips Reference Phillips2018, 481). Evidence for whether a process is attributable to the individual would consist of showing that the process is “available to central coordinating agency” (Phillips Reference Phillips2018, 494). Studies using unconscious visual primes are confronted with this problem because the primes generate representations that are not available to central coordinating agency, and, thus, it is unclear why these representations should be attributable to the individual. Rather, after the presentation of the prime the perceptual system “is more ‘fluent’” in processing further matching stimuli (Phillips Reference Phillips2018, 494). Phillips sees the same problem for the nude study, stating “[W]hy think that stimulus-driven, reflex-like attentional responses count as manifestations of central agency, and so witness individual-level perception?” (Phillips Reference Phillips2018, 495).

The problem of the criterion and the problem of attribution primarily concern methodological issues. Still, both presuppose conceptual assumptions. Phillips’s formulation of the problem of attribution clearly presupposes an answer to the conceptual unconscious mind worry: For a state or process to be mental it must be attributable to the individual, which is the case if it is available to central coordinating agency. However, it is unclear whether a view of the mental in terms of attributability to the individual is adequate. I will discuss this in more detail in section 3.

2.2 Unconscious attitudes?

A further example of a putatively unconscious mental phenomenon are implicit attitudes, or “implicit biases.” Generally, attitudes are taken to be mental states that correspond to negative, positive, or neutral evaluations of different items. A large part of implicit attitude research focusses on social attitudes (i.e., people’s evaluations of different social groups, such as African Americans or women). One crucial finding is that participants who report no negative evaluations of specific social groups, in implicit tests (see following text) show behaviors that suggest a negative attitude toward these groups. One standard conclusion is that implicit attitudes, thus, are unconscious (Saul Reference Saul2013; Banaji, Bhaskar, and Brownstein Reference Banaji, Bhaskar and Brownstein2015).

Different methods have been suggested to detect implicit attitudes, including masked affective priming (Degner et al. Reference Degner, Wentura, Gniewosz and Noack2007), introspecting affective reactions (Hahn et al. Reference Hahn, Judd, Hirsh and Blair2014), the implicit associations test (IAT) (Greenwald, Mcghee, and Schwartz Reference Greenwald, Mcghee and Schwartz1998), and explicit judgment tests (e.g., where participants are asked to evaluate different made-up CVs by male and female applicants and decide who should be hired) (Uhlmann and Cohen Reference Uhlmann and Cohen2005). Many psychologists and philosophers interpret the results of such studies as evidence for an unconscious mind. For example, Mandelbaum (Reference Mandelbaum2016) argues that “[e]xamining the workings of implicit bias can illuminate a host of foundational issues in cognitive science, such as the entities that populate the unconscious mind, and how rationally responsive unconscious thought can be” (Mandelbaum Reference Mandelbaum2016, 629).

The methodological mind worry arises because it is unclear how these different methods could in principle measure the same construct. For example, introspecting affective reactions is adequate for detecting implicit attitudes only if implicit attitudes are affective states. However, the IAT is usually interpreted as detecting associations between concepts (Greenwald, Mcghee, and Schwartz Reference Greenwald, Mcghee and Schwartz1998). Other authors argue that implicit attitudes cannot be associations but must be propositionally structured beliefs (Mandelbaum Reference Mandelbaum2016).

Thus far, no agreement has been reached about what kinds of entities implicit attitudes are supposed to be (Feest Reference Feest2020). However, the unconscious mind worry concerns an even more basic question: Given the empirical evidence, are we justified in assuming that implicit attitudes are states of the mind at all? There is evidence, for example, that IAT scores are influenced by physiological factors, such as sleep, age, and hunger. This could indicate that, similar to Phillips’s criticism presented in the previous section, the IAT simply measures the “fluency” of the brain in processing information rather than any mental phenomenon. A further explanation that would speak against categorizing implicit attitudes as mental states is that they are “truly implicit” (Johnson Reference Johnson2019). Contents are truly implicit if they are not represented in the system but are a consequence of the specific architecture and transformation rules between contents. This discussion shows that unless the conceptual unconscious mind worry is solved, the discussion of whether unconscious bias research provides evidence for an unconscious mind is likely to be unproductive.

4.1 The inference to the unconscious mind

Why do scientists label some unconscious phenomena “mental” and others as not? To answer this question, it is helpful to take a closer look at the justification underlying the postulation of unconscious mental phenomena in general. The arguments for the existence of unconscious vision and unconscious attitudes, as well as all other claims to the existence of an unconscious mind based on empirical evidence, take the form of an inference to the best explanation:

Unconscious vision

1. 1. Participants correctly react to visual stimuli above chance.

2. 2. Participants do not consciously see the stimuli.

3. 3. The best explanation for how participants can adequately react to stimuli despite not consciously seeing them is that they see them unconsciously.

4. 4. Thus, the explanation in terms of unconscious vision is true, and unconscious vision is real.

Unconscious attitudes

1. 1. Participants show discriminating behaviors toward women.

2. 2. Participants are not conscious of having any negative attitudes toward women.

3. 3. The best explanation for why participants discriminate against women despite not being conscious of any negative attitudes is that the attitudes are unconscious.

4. 4. Thus, the explanation in terms of unconscious attitudes is true, and unconscious attitudes are real.

Generalizing these two examples, I will call the following argument the Inference to the Unconscious Mind (IUM):

1. 1. Behavioral Observation: Participants show a certain behavior B.

2. 2. Unconsciousness: Participants are not conscious of a mental cause (M) of B.

3. 3. Best Explanation: The best explanation for the conjunction of 1 and 2 is that B is due to M and that M is unconscious.

4. 4. Inference to the Reality of Unconscious Mind: Thus, the explanation in terms of unconscious mental causes is true, and the unconscious mental cause is real.

As indicated in section 2, a great deal of discussion in the philosophical and psychological literature targets premise 2. However, for the purposes of this article, I will set these worries aside and assume the truth of premise 2. For the new strategy for characterizing the mental, premise 3 is crucial. This premise states that assuming that there is a mental cause that is unconscious provides an explanation of the behavior at issue (I will leave the discussion of the claim that these explanations are the best explanations for another paper). Two observations can be made. First, unconscious mental phenomena are postulated primarily as explanatory posits. Second, the explanatory power of the unconscious mental phenomenon seems to stem from its being mental in the same way as a corresponding conscious cause would have been. Thus, one implicit assumption seems to be that unconscious mental phenomena are just like the corresponding conscious phenomena minus consciousness, and that it is this similarity that does the explanatory trick. My strategy makes use of this insight. If we find out how unconscious phenomena are similar to conscious phenomena in an explanatorily relevant way, we have determined what “mental” is supposed to grasp in premise 3. Thereby, we have identified the reasons for labelling some processes “mental” and others as not: only the former are relevantly similar to corresponding conscious phenomena.

In which way are the explanatory roles of conscious and unconscious mental phenomena alike, and how does the explanatory role of nonmental unconscious phenomena differ? In a nutshell, I will argue that categorizing an unconscious phenomenon as “mental” is accurate if (a) its underlying mechanism produces and explains a behavior that is similar to a behavior produced and explained by a corresponding conscious phenomenon, and (b) the mechanism underlying the unconscious phenomenon shares explanatorily relevant components with the mechanism underlying the corresponding conscious phenomenon. To defend this view, I must show (1) that explanations invoking unconscious mental phenomena and those invoking corresponding conscious phenomena are mechanistic explanations, (2) that they explain similar behaviors, and (3) that the underlying mechanisms share explanatorily relevant components and how explanatory relevance is determined. I will elaborate on each of these points in the next subsections.

4.2 Mechanistic explanation and the unconscious mind

What type of explanation are unconscious mental explanations supposed to deliver in scientific contexts? The general justification for the postulation of unconscious mental phenomena, as described in the IUM, is that people are not conscious of mental causes and that the best explanation for why they still show the respective behavior is that these mental causes are unconscious. Researchers want to understand how the observed behavior comes about given that it cannot be due to a conscious perceptual state. They are interested in the causal mechanism that is responsible for the behavior. Hence, mental explanations, in the relevant context, are supposed to provide mechanistic explanations of behaviors.

Mechanistic explanations have become a core topic in contemporary philosophy of science. One common assumption of the so-called new mechanists is that a mechanistic explanation of a phenomenon P consists of a description of a mechanism that is responsible for P. “Responsible” in this context is usually understood in terms of either causation or constitution (Craver Reference Craver2007). In causal/etiological mechanistic explanations, the mechanism that explains the phenomenon causes the phenomenon and, thus, temporally precedes it. In constitutive mechanistic explanations, the mechanism and the phenomenon occur in the same space time region and are mutually dependent.

Mechanisms are taken to be collections of entities and their activities in a certain spatial, temporal, causal, and hierarchical organization (Bechtel and Abrahamsen Reference Bechtel and Abrahamsen2005; Craver Reference Craver2007; Illari and Williamson Reference Illari and Williamson2012; Glennan Reference Glennan2017). Note that mechanistic components (i.e., entities and activities) can be described in different terminologies and with different grades of detail. They may, for example, be described in neurobiological terms (e.g., “neuron,” “axon,” “diffusing”), functional terms (e.g., “inhibitor,” “store”) or filler terms (e.g., “activating,” “inhibiting”). Explanations in terms of functional or filler terms are often described as “mechanism sketches” that still count as mechanistic explanations but need to be mapped onto a neurobiological/physical structure to be valid mechanistic explanations (Piccinini and Craver Reference Piccinini and Craver2011).

Mechanisms are individuated relative to the phenomena for which they are responsible. That is, the components of a mechanism for P are those entities and activities (in a certain organization) that are all relevant for the production, and thus, the explanation of P, and all entities and activities (and organizational aspects) that are relevant for the production and explanation of P, are components of the mechanism for P. Which entities, activities, and organizational features are taken to be relevant for the production and explanation of P depends on whether they are causally relevant (in etiological mechanistic explanations) or constitutively relevant (in constitutive mechanistic explanations) for the phenomenon. According to a prominent proposal (Craver Reference Craver2007), causal as well as constitutive relevance are determined by means of intervention-experiments. There is an ongoing discussion of how to best understand this proposal (see, e.g., Kästner and Andersen [Reference Kästner and Andersen2018] for an overview). Luckily, the details of this discussion are not relevant here. The important aspect here is that philosophical discussions of mechanism discovery and individuation are closely tied to actual scientific practice—which is captured by the idea that the former rely on intervention experiments. Thus, prima facie, the mechanistic approach to scientific explanation does not only capture the general idea of unconscious mind research (explaining how behaviors are brought about) but can also be applied to the search for mechanisms in unconscious mind research.

Are mechanistic explanations invoking unconscious mental phenomena etiological or constitutive mechanistic explanations? The answer to this question depends on what exactly those explanations explain. If the explanandum is, for example, “How does unconscious vision work?,” the explanation would refer to a mechanism that constitutes unconscious vision. However, while this “how” question may be a question that researchers want to answer, the explananda relevant in the present context are more fine-grained and relative to specific experimental setups. Furthermore, what is at issue here is the type of explanation that motivates the postulation of an unconscious mental phenomenon in the first place (see IUM in section 4.1). In the context of the IUM, unconscious vision is the explanans, not the explanandum. Mechanistic explanations invoking unconscious mental phenomena take the form of “P happened because of unconscious vision/attitudes/” and express that there is a mechanism that is responsible for P that does not lead to reportability in the relevant sense but still is a mental mechanism (an instance of vision, believing, or the like). Thus, in the mechanistic context, “unconscious vision” or “unconscious attitude,” for example, can be understood as filler terms to be filled in with details about the mechanism for P where this mechanism occurs in the absence of consciousness.

In the following section, I will argue that the explananda of explanations invoking unconscious mental phenomena are contrastive questions concerning behaviors observed in experiments. For example, why could the patient accurately react to the stimuli rather than not? A mechanistic explanation of this contrastive explanandum picks out specific components of the mechanism underlying the patient’s behavior that are causally relevant for the contrast referred to in the question. These are causal explanations rather than constitutive ones because whatever happened in the brain that led to the patient’s accurate reaction happened before the reaction.

4.3 The explananda of the unconscious mind

What exactly are the explananda of explanations invoking unconscious mental phenomena? On the assumption that my reasoning in the previous section is correct, their explananda should be of the type that can principally be mechanistically explained. In contemporary philosophy of science, it is commonly assumed that what is to be explained by mechanistic explanations are phenomena that are picked out by “how does x work” questions (Craver Reference Craver2007). These, in turn, are answered by answering a set of contrastive “why” questions (Craver and Kaplan Reference Craver and Kaplan2020). For example, to explain how the action potential is transmitted along the axon, one has to answer a bunch of contrastive “why” questions, such as “Why does the potential peak at 40 mV rather than at a higher value?,” “Why does the axon membrane hyperpolarize rather than go back to resting potential?,” and “Why is the potential transmitted only in the direction of the axon terminal rather than back toward the soma?”

What contrastive questions do unconscious mental phenomena answer? Generally, (conscious) mental phenomena are supposed to explain how a certain behavior is brought about by explaining why a certain behavior occurred rather than some other behavior (or no behavior at all) (“Why does B occur rather than not?”). For example, conscious perceptual states explain why a participant can react to visual stimuli with only a few mistakes rather than with many mistakes. Explicit negative attitudes toward Black people explain why 21.5 percent of a sample of undergraduates were neutral or agreed with the statement “Avoiding interactions with Blacks is important to my self-concept” (Axt Reference Axt2018, 8). Postulating unconscious mental phenomena serves the same explanatory purpose. Take pathological blindsight, the ascription of an unconscious perceptual state explains why the behavioral reactions toward visual stimuli in blindsight participants are quite accurate rather than not accurate at all. Furthermore, unconscious perceptual states explain why, in the binocular rivalry experiment described in section 2.1, participants’ reactions to Gabor patches that are presented at the location of the sexually preferred nude are more accurate compared to Gabor patches that were presented at the location of the Mondrian—rather than reaction times being equal in both cases. In addition, unconscious attitudes explain why self-reported nonracist people show discriminating behavior against members of certain social groups rather than treating members of different social groups equally in line with what they report. In a nutshell, conscious as well as unconscious mental phenomena are postulated as parts of mechanisms that are supposed to explain behavioral contrasts of the form “Why did B occur rather than not-B?”

However, the similarities between scientific explanations invoking conscious phenomena and those invoking unconscious mental phenomena do not end here. The behavior that is supposed to be explained by conscious phenomena and the behavior that is supposed to be explained by unconscious mental phenomena are of the same type—on a not too fine-grained reading. This is the insight that motivates the functional approach to individuating mental capacities described in section 3. The behavioral contrast in both cases is between a type of behavior that shows some kind of accuracy, stability, or correctness, compared to absence of that particular behavior. For example, conscious and unconscious vision are supposed to explain more or less accurate reactions toward visual stimuli in contrast to random reactions. The same holds for attitudes/preferences. Conscious and unconscious attitudes explain why participants show more or less stable preferences toward something over something else rather than treating both equally (see section 3). Someone who has the conscious attitude that women should not be police chiefs will very consistently reject female applicants for the job of a police chief, and someone who is implicitly biased against women will be likely to reject female applicants but may do so less consistently as their decision may be influenced by, for example, contextual factors.

To summarize, I have argued that explanations invoking unconscious mental phenomena are similar to explanations invoking conscious phenomena in several ways. First, they both are supposed to deliver mechanistic explanations of how a behavior of type B is brought about by answering a question of the form “Why did B occur rather than not?” Second, the behavior B that is to be explained by the unconscious mental phenomenon is of the same type (not too fine-grainedly individuated) as the behavior that is to be explained by the corresponding conscious phenomenon. In the following section, I will explain how conscious and unconscious mental phenomena explain similar behavioral contrasts in a similar way—and thereby provide a novel approach to answering the conceptual unconscious mind worry in line with the general strategy outlined in section 4.1.

4.4 Explanatory relevance and the unconscious mind

In this section, I will argue that scientific explanations invoking unconscious mental phenomena explain their explananda in a relevantly similar way as conscious phenomena explain their explananda—and that this is the crucial insight for addressing the unconscious mind worry. More specifically, I will argue that the categorization of an unconscious phenomenon as “mental” is adequate if the mechanism that is responsible for the behavior B* of type B when consciousness is absent and the mechanism that is responsible for the behavior B** of type B when consciousness is present share components that are relevant for explaining the contrast “Why B rather than not-B?” To get to this, I will first explain how to think of explanatory relevance in the mechanistic framework.

According to a recent mechanistic proposal (Kohár and Krickel Reference Kohár, Krickel, Calzavarini and Viola2021), to answer contrastive “why” questions of the form “Why P rather than P’?,” we have to look at the different mechanisms that would produce phenomenon P’ and compare them to the actual mechanism that is responsible for phenomenon P. More specifically, we must compare the actual mechanism for P to the maximally similar mechanism for P’ and list all the differences between the two mechanisms that are shared by all possible mechanisms for P’.Footnote ² We thereby list all those components of the mechanism for P whose absence would be necessary for P’ to arise, and those components that are lacking in the mechanism for P whose presence would be necessary for P’ to arise. All items in this list can be called the difference-makers for P vs. P’.

To illustrate this idea, take the example of the action potential again. If one wants to explain why the action potential is transmitted in the direction of the axon terminal rather than back to the cell soma, one has to compare what is happening in the actual case to what would have to be the case for the potential to travel backward and list all those elements that are present in the actual case that prevent the contrast phenomenon to be the case, and those elements that are lacking in the actual case that would be necessary for the contrast phenomenon to occur. Thereby, one reaches an answer such as “because, in the actual mechanism, the voltage-gated Na channels are inactive, whereas in the most similar possible mechanism for the action potential’s traveling backward, the voltage-gated Na channels are still active.” This answer lists all the difference makers for the “action potential being transmitted in the direction of the axon terminal” versus the “action potential being transmitted backward.”

We can apply this framework to behavioral explanations. If we want to explain why some behavior B was observed rather than not, we must compare the mechanism for B with the maximally similar mechanism that does not bring about B. For example, if we want to explain why participants react to stimuli without any mistakes rather than with many mistakes, we have to compare the mechanism that is active when participants react to stimuli without mistakes with the maximally similar mechanism that is active when participants make many mistakes. The differences between these mechanisms will be the difference makers for B (reaction without mistakes) versus not-B (reactions with many mistakes).

The crucial idea is that an unconscious phenomenon is to be categorized as “mental” if the difference makers for B versus not-B when B is produced by an unconscious phenomenon are also on the list of differences makers for B versus not-B when B is produced by a conscious phenomenon. Together with the procedure for finding difference makers for B versus not-B, this gives us the following protocol for deciding whether an unconscious phenomenon is mental, or not (B* of type B is the behavior produced by an unconscious phenomenon; B** of type B is the behavior produced by a conscious phenomenon):

1. (1) Finding Difference Makers for B* of Type B versus not-B: Compare the mechanism underlying B*—call it “M_unconscious”—with the maximally similar mechanism that does not produce any behavior of type B, and list all the differences (list of difference makers 1). These difference makers explain why the unconscious phenomenon leads to a behavior of type B rather than not.

2. (2) Finding Difference Makers for B** of type B versus not-B: Compare the mechanism underlying B**—call it “M_conscious”—with the maximally similar mechanism that does not produce B, and list all the differences (list of difference makers 2). These difference makers explain why the conscious phenomenon leads to a behavior of type B rather than not.

3. (3) Same difference makers? Compare the list of difference makers 1 with the list of difference makers 2. If these lists overlap, the mechanisms share explanatorily relevant components, and, thus, the conscious and the unconscious phenomenon explain behavior B in a relevantly similar way. Thus, the phenomenon that M_unconscious is responsible for is to be categorized as a mental phenomenon.

To clarify the proposal, let us look at an example. Conscious vision as well as unconscious vision are supposed to explain why participants react to stimuli with an accuracy greater than chance. Unconscious vision is a mental phenomenon (and thus deserves the label “vision”) if the mechanism underlying unconscious vision (partly) consists of those components (or a subset thereof) that are the difference makers for “reaction with accuracy greater than chance” versus “chancy reaction” in the case of conscious vision. To test whether this is the case, we have to do different types of experiments and determine the mechanisms that are active in these experimental setups.

Assume that we investigate putative unconscious vision with help of a binocular rivalry task as described in section 2.1. To determine whether the phenomenon that is responsible for the observed behavior really is vision (i.e., a mental phenomenon), we have to do the following (where the relevant behavior B is a shift in covert spatial attention):

1. (i) Finding Difference Makers for B* of Type B versus not-B: Compare M_unconscious underlying the covert attention shifts triggered by a binocular rivalry task using Mondrians and pictures of nudes that are not consciously perceived (B* of type B) to the maximally similar mechanism active in the same binocular rivalry task that does not lead to B. à list of difference makers 1

2. (ii) Finding Difference Makers for B** of type B versus not-B: Compare M_conscious underlying covert attention shifts when the stimuli are consciously perceived (B** of type B) with a maximally similar mechanism active in a similar context with consciously perceived stimuli that does not lead to B. à list of difference makers 2

3. (iii) Same difference makers? Compare the list of difference makers 1 and the list of difference makers 2. If the components of M_unconscious that make the difference between B* versus not-B are also components of M_conscious that make the difference between B** versus not-B, the mechanisms are relevantly similar (i.e., they explain B versus not-B in a relevantly similar way), and the unconscious phenomenon associated with M_unconscious is to be categorized as “mental.”

The condition described by (iii) seems to be satisfied in the case of unconscious vision as investigated in binocular rivalry tasks. Empirical studies suggest that the difference makers for B (attention shifts) versus not-B (no attention shifts) may be early visual cortex activation (Klink and Roelfsema Reference Klink and Roelfsema2016). These difference makers are involved in M_unconscious as well as M_conscious. Furthermore, in a recent study, Luo et al. (Reference Luo, VanRullen and Alamia2021) found that consciously perceived and suppressed stimuli both lead to alpha activation (only that the activation induced by consciously perceived stimuli is stronger). These alpha rhythms that can last for as long as one second are directly related to visual stimulation and, thus, would not occur if there was no vision at all. Furthermore, the alpha activation induced by consciously visible and suppressed stimuli in both cases propagates as a travelling wave from occipital to frontal regions, and they do so with the same strength (Luo, VanRullen, and Alamia Reference Luo, VanRullen and Alamia2021, 6). The authors speculate that the main difference is that the alpha activation induced by consciously visible stimuli propagates further in the visual hierarchy, whereas activation generated by the suppressed stimuli disappear at an earlier stage of visual processing (ibid.). However, these higher levels of the visual hierarchy seem to be correlated with conscious higher-level processes rather than to conscious perception per se, as suggested by several studies on vision (such as Kouider et al. [Reference Kouider, Dehaene, Jobert and Le Bihan2007]; see following text). Thus, the mechanisms underlying conscious and unconscious vision as measured in binocular rivalry tasks seem to share components that are explanatorily relevant for the attention shift. Thus, unconscious vision should be categorized as a mental phenomenon.

The same would have to be shown for (putatively) unconscious vision observed in other experimental setups, such as priming studies. Indeed, there is evidence that the mechanisms underlying subliminal priming and the mechanisms underlying supraliminal priming overlap in the relevant way as well (Kouider et al. Reference Kouider, Dehaene, Jobert and Le Bihan2007). Note that studies investigating (putatively) unconscious vision usually aim at finding the neural correlates of consciousness. Therefore, they aim at describing the differences between the mechanism underling the unconscious process and the mechanism underlying the conscious one. Thus, the similarities between the mechanisms are usually not explicitly mentioned. However, the comparison of these two mechanisms is usually framed in a way that suggests that the two mechanisms are taken to be more or less the same minus whatever consciousness adds. For example, in a study comparing brain activity evoked by subliminally and supraliminally presented stimuli, Kouider et al. (Reference Kouider, Dehaene, Jobert and Le Bihan2007) conclude that their findings “fit with the suggestion that subliminal activation remains confined to a few specialized processors, whereas stimuli that pass the visibility threshold become available to a much broader range of processors including more anterior cortical sites” (Kouider et al. Reference Kouider, Dehaene, Jobert and Le Bihan2007, 2026).

They also observe that the main difference between the mechanisms triggered by subliminal and supraliminal primes is the activation of a restricted set of occipito-temporal areas, without extension into higher associative regions, and suggest that the many additional regions are linked to higher-level cognitive processes, such as recognition, report, evaluation, or memory storage, rather than the conscious vision as such (Kouider et al. Reference Kouider, Dehaene, Jobert and Le Bihan2007, 2026). Similarly, Railo et al. (Reference Railo, Salminen-Vaparanta, Henriksson, Revonsuo and Koivisto2012) provide evidence that the “unconscious perception of color … is mediated by early cortical activation in the same time window as conscious vision” (Railo et al. Reference Railo, Salminen-Vaparanta, Henriksson, Revonsuo and Koivisto2012, 828). Thus, priming studies suggest that the mechanisms underlying conscious and unconscious vision are relevantly similar with respect to explaining attention shifts and target detection.

What about implicit attitudes? On the assumption that implicit attitudes are unconscious (see section 2.2), are we justified in treating the empirical evidence as evidence for an unconscious mind? To answer this question, we first must determine the relevant mechanisms for the comparison. Assume we use the CV study as presented in section 2.2 to investigate attitudes. The behavior type B that we want to investigate is (more or less) stable discrimination behavior—in the present study, the likelihood to choose a male candidate over a female candidate despite equal qualifications.

1. a) Finding Difference Makers for B* of Type B versus not-B: Compare M_unconscious underlying the discriminatory behavior in the CV study (B* of type B) with the maximally similar mechanism that does not produce any behavior of type B. à list of difference makers 1

2. b) Finding Difference Makers for B** of type B versus not-B: Compare M_conscious underlying discriminatory behavior induced by explicit prejudice (B** of type B) in a similar CV study with the maximally similar mechanism that does not produce B. à list of difference makers 2

3. c) Same difference makers? Compare the list of difference makers 1 with the list of difference makers 2. If the lists overlap, the mechanisms share explanatorily relevant components, and, thus, the conscious and the unconscious phenomenon explain behavior B in a relevantly similar way. The phenomenon that M_unconscious is responsible for is to be categorized as a mental phenomenon.

Empirical findings show that the relevant component in M_unconscious that explains why the mechanism leads to discriminatory behavior rather than not is amygdalar activation (Stanley, Phelps, and Banaji Reference Stanley, Phelps and Banaji2008). The amygdala is also active in M_conscious and this seems to explain discriminatory behavior as well. The difference between B* and B** may be due to the fact that in M_unconscious the anterior cingulate cortex (ACC) and dorsolateral prefrontal cortex (dPFC) are active but not in M_conscious. The ACC and dPFC both seem to be involved attenuating amygdalar activation (Stanley, Phelps, and Banaji Reference Stanley, Phelps and Banaji2008, 167). Thus, the condition formulated in c) seems to be satisfied, and implicit attitudes turn out to provide evidence for an unconscious mind (assuming that they are indeed unconscious).

Support for my account also comes from considerations regarding the mechanisms underlying consciousness and unconscious processes in general. For example, Lau and Rosenthal (Reference Lau and Rosenthal2011) present empirical evidence for higher-order thought theories. They argue that “recent computational modeling work … suggests that unconscious and conscious processes probably form a serial hierarchy rather than two parallel channels” (Lau and Rosenthal Reference Lau and Rosenthal2011, 11). All types of models of consciousness that assume this kind of nonparallel approach to conscious versus unconscious processes in principle provide evidence for my account (if the processing steps that are shared by conscious and unconscious processes indeed are the difference makers for B vs. not-B).

Is there a negative example; that is, one where a putative unconscious mental phenomenon turns out to be nonmental? This will be the case for all phenomena explained by dual process models that assume that a given mental activity is realized by two wholly distinct types of mechanisms, depending on whether they are conscious or not. For example, the different memory systems of procedural (unconscious) and declarative (conscious) memory seem to provide a negative example. Both memory systems rely on wholly distinct neural networks (Bayley and Squire Reference Bayley and Squire2007; Buffalo and Squire Reference Buffalo and Squire2015). While declarative memories are encoded in the hippocampus, entorhinal cortex, and perirhinal cortex, and are consolidated and stored in the temporal cortex and elsewhere, procedural memories are encoded and stored by the cerebellum, putamen, caudate nucleus, and the motor cortex. Hence, procedural memory is not a mental phenomenon as it is not conscious, nor is the mechanism relevantly similar to the mechanism underlying conscious declarative memory. One consequence of this might be that we should not think of skilful behavior (such as playing the piano or riding the bike) that relies on procedural memory as a capacity of the mind. In contrast, learning a skill may be, as this usually involves consciousness. However, I will leave a defense of these claims for another paper.

4.5 Why mechanisms?

Why should similarity between mechanisms be the criterion for unconscious mentality? As argued, one reason is that conscious and unconscious mental phenomena are postulated as explanatory posits that play the same role in the mechanistic explanation of certain behaviors. There is a further argument. As argued by Taylor (Reference Taylor2020), mental phenomena such as perception are taken to be natural kinds (see also Shepherd and Mylopoulos [Reference Shepherd and Mylopoulos2021]). Unconscious phenomena are mental if and only if they belong to a mental natural kind. As Taylor stresses, the most prominent approach to natural kinds is Boyd’s homeostatic property cluster view (Boyd Reference Boyd1989), which outlines that natural kinds are clusters of properties that are held together by an underlying mechanism. Thus, sharing the same mechanism is a necessary condition for unconscious processes to be of the same natural kind as their conscious counterparts. As different instantiations of the same mechanism bring about the same cluster of properties, my account is in line with Taylor’s views.

One potential objection against the claim that my account is in line with the homeostatic property cluster view is that, according to my proposal, the mechanisms underlying conscious phenomena and those underlying unconscious mental phenomena are not the same—they only share the difference makers for B versus not-B. This, however, is not sufficient to be of the same natural kind. I have two thoughts on this objection. First, my account could be used to improve the homeostatic property cluster view. The view is confronted with the problem that it is unclear how similar mechanisms need to be to count as the same mechanism and how similarity is to be understood (see, e.g., Craver Reference Craver2009). Similarity in the homeostatic property cluster view could be spelled out in terms of explanatory relevance along the lines of my account. Second, the connection between natural kinds and “similar but not the same” mechanisms (or mechanistic models) is highlighted in Rob Rupert’s (Reference Rupert2019) “tweak and extend” account of natural kinds. According to this account, the instances of a natural kind bear a family resemblance to each other that is “determined by the causal-explanatory roles of the components” of these instances (Rupert Reference Rupert2019, 37). If two phenomena cannot be modeled by the same model even after tweaking or extending the model, then it turns out that the two phenomena do not belong to the same natural kind. Developing these ideas further will require more space and, therefore, will be left for future research.