2.a Functional definability and mechanistic explainability

Chalmers famously distinguishes between two explanatory projects within a science of consciousness, the hard and the easy problems:

The relationship between functions and mechanisms is taken to be a conceptual truism: “How do we explain the performance of a function? By specifying a mechanism that performs the function” (Reference Chalmers2010, 7). For instance, “[a]ll it could possibly take to explain reportability is an explanation of how the relevant function is performed” (7), where the ‘how’ in question is “a story about the organization of the physical system that allows it to react to environmental stimulation and produce behavior in the appropriate sorts of ways” (Reference Chalmers1996, 22). Thus, if a phenomenon is functionally definable, it follows that a mechanism can explain that phenomenon. In contrast, consciousness is not about the performance of functions, which entails that there is no ‘functional how’ to be explained mechanistically in the first place. Thus, the criterion of functional definability discriminates between consciousness and mechanistic unexplainabiliy on one side, and all/most other phenomena and mechanistic explainability on the other.

2.b Functions and mechanisms

Before we can test the accuracy of the criterion, we need a more precise characterization of the terms ‘mechanism’ and ‘function.’ The new mechanistic philosophy offers close to a dozen characterizations of mechanisms (Glennan Reference Glennan2017). Among these, the one that best matches Chalmers’s terminology is the proposal that a “mechanism is a structure performing a function in virtue of its component parts, component operations, and their organization. The orchestrated functioning of the mechanism is responsible for one or more phenomena” (Bechtel and Abrahamsen Reference Bechtel and Abrahamsen2005, 423). The mechanism is a physical entity consisting of an organized system of interacting parts, while the phenomenon for which it is responsible is the behaviour of the system, typically described as an input-output or stimulus-response sequence (Machamer et al. Reference Machamer, Darden and Craver2000).

Chalmers defines the term ‘function’ as “any causal role in the production of behaviour that a system might perform” (Reference Chalmers2010, 6). This understanding of the term is distinct from the systemic and evolutionary concepts of function developed in the philosophy of biology. The former requires that functional ascriptions are made in light of an explanatory account—usually, but not necessarily, of a mechanistic variety—specifying how a part of a system causally contributes to the ability of the system to behave in a certain way (Craver Reference Craver2001; Cummins Reference Cummins1975). The latter defines functions relative to a natural selection mechanism (Millikan Reference Millikan1989; Neander Reference Neander1991). Chalmers’s characterization best matches a broader notion of function commonly found in experimental biology, where the term is used to describe the results of controlled experiments designed to demonstrate causal relevance. Something is said to have a function if that something (mechanism, mechanistic component, factor, independent variable) makes a difference vis-à-vis something else (phenomenon, outcome, dependent variable) in the context of an experimental setup (animal or cell model, in vitro reconstitution system, etc.). Functional ascription is made here solely in virtue of evidence for causal relevance, without the explicit involvement of an explanatory account. For instance, when early geneticists and developmental biologists concluded that the function of the wingless gene is to determine wing development and body axis formation, they were merely describing differences between phenotypes of Drosophila with (test/mutated) and without mutations (wild-type/control) at the wingless locus. Such differences reveal that wingless is causally relevant to, and in this sense can be attributed the function of contributing to the outcomes of wing development and body axis formation (Sharma Reference Sharma1973).

It is possible, however, that Chalmers contemplates a somewhat more restrictive notion of function, of the sort assumed in functionalist theories of the mind (McLaughlin Reference McLaughlin2006). The latter define mental states in terms of causal relations to stimuli, other mental states, and behavioral responses. Here, a function is not just any cause contributing to an outcome, but more specifically a causal mediator between stimuli and responses. Many functional ascriptions in experimental science aggregate these two slightly different notions of function. For instance, since lesions to the visual cortex area V5 lead to an inability to perceive motion when presented with visual motion stimuli, while the artificial stimulation of V5 neurons affects the perception of motion (as contrasted to normal/control subjects), it is common practice to summarize causal inferences in statements such as “the principal function of V5 is to detect and signal the presence and direction of visual motion” (Zeki Reference Zeki2015).

2.c The sensitivity of the functional undefinability criterion remains unproven

Equipped with these precisions, we can proceed to the evaluation of the first axis of Chalmers’s discrimination criterion, namely the claim that unlike other biological and psychological phenomena, consciousness is not functionally definable. The first question that comes to mind concerns the sensitivity of the criterion: Is it indeed the case that consciousness is functionally undefinable?

The empirical justification of the criterion rests on the fact that in some brain lesion patients or under experimental conditions, certain perceptual and cognitive tasks can be performed nonconsciously, thus demonstrating that task performance and behavioral responses to stimuli can be dissociated from conscious experience. Known examples include unconscious processing of visual stimuli (Milner and Goodale Reference Milner and Goodale2006), pain (Melzack and Wall Reference Melzack and Wall1982) and threatening events (LeDoux Reference LeDoux1996), blindsight (Weiskrantz Reference Weiskrantz1990), covert facial recognition (Young and Burton Reference Young and Burton1999), and implicit memory (Schacter Reference Schacter1987). By extrapolating from such cases, one may speculate that nonconscious processing allows for a reasonable degree of functionality in ‘zombie mode’ outside artificially constructed experimental contexts. In turn, this raises the possibility that equivalent function performance may exist in the absence of consciousness. Chalmers takes this to entail that consciousness is not necessary for cognitive and behavioral performance, hence his claim that consciousness cannot be a problem about the performance of functions.

The possibility envisaged by Chalmers is, however, mitigated by the fact that the extrapolation on which it relies turns out to be false. Patients suffering from deficits in awareness invariably exhibit a significant degree of dysfunctionality under routine, everyday conditions:

The recurring theme emerging from clinical observations is that patients behave as if covertly processed information is not processed at all. It is only from the external perspective of the experimenter that nonconscious processing can be evidenced and only under the external prompting of the experimenter that the patient may act on the information covertly processed. This phenomenon is particularly obvious in blindsight patients, who only exhibit good performance in the experimental context of forced-choice tasks prompting them to guess which option is correct. Left on their own devices, blindsight patients fail to spontaneously initiate visually guided behaviour in response to stimuli in their impaired visual field (Cowey Reference Cowey2010; Marcel Reference Marcel1983). This observation has been used to support the rival view that consciousness has a function in normal subjects. In particular, global workspace theories posit that “consciousness is required for some specific cognitive tasks, including those that require durable information maintenance, novel combinations of operations, or the spontaneous generation of intentional behavior” (Dehaene and Naccache Reference Dehaene and Naccache2001, 1).

What does this entail for the functional definability of consciousness? So far, nothing conclusive. Global workspace theorists argue that consciousness has a function because loss of consciousness correlates with loss of task performance, while Chalmers argues that consciousness is not functionally definable because of observed and extrapolated dissociations between task performance and consciousness. In both cases, conclusions about functional definability are based on prior knowledge of association/dissociation, not of causation/absence of causation. The problem with such inferences is that association doesn’t always entail causation and causation doesn’t always entail association.

The global access hypothesis faces the difficulty of inferring causation from association. While there is ample empirical evidence demonstrating a robust association between loss of consciousness and loss of function, this is insufficient to demonstrate the causal link required for functional ascription. For instance, Zeki’s claim that the function of V5 is to detect motion is supported by experiments in which the item which is ascribed a function (i.e., V5) is known to be independently manipulated. In contrast, in the experiments cited in support of the conscious access hypothesis, it is not clear that consciousness is independently manipulated. The awareness deficits (amnesia, blindsight, prosopagnosia, aphasia, etc.) mentioned by Weiskrantz involve natural experiments in which what is in fact manipulated (the independent variable) is brain activity, not consciousness. Assuming that brain lesion patients are comparable to healthy subjects in all respects except for localized loss of brain activity, it can be inferred that the lesioning of specific brain areas causes both loss of awareness and loss of performance for certain types of tasks. However, nothing here justifies the additional inference that loss of consciousness is responsible for loss in performance (Block Reference Block1995). Strictly speaking, these natural experiments only show that brain lesions are causally relevant to both awareness and task performance. It could be that consciousness has the function global workspace theories attribute it (Baars Reference Baars2002; Dehaene and Naccache Reference Dehaene and Naccache2001), just as it could be that functional correlates are required for stimulus awareness, as postulated by the feature integration theory (Treccani Reference Treccani2018; Treisman and Gelade Reference Treisman and Gelade1980). Or again, according to an interpretation compatible with Chalmers’s antifunctionalism, consciousness and its functional correlates could be divergent effects of a common cause (LeDoux Reference LeDoux1996; LeDoux and Pine Reference LeDoux and Pine2016). Currently accepted standards of internal validity dictate that singling out the correct causal explanation of an observed correlation requires studies capable of ruling out rival interpretations; in turn, this requires an experiment in which consciousness is independently manipulated.Footnote ²

Chalmers’s functional undefinability thesis faces the converse difficulty of inferring lack of causation given lack of association. Just as the knockout of a gene may result in no phenotypic differences because a second gene takes over the function of the knocked-out gene, it is conceivable that a factor or mechanism Z compensates for the loss of performance caused by the loss of consciousness. In this scenario, both consciousness and Z are causally relevant, and therefore play a functional role vis-à-vis task performance, but since the inhibitory effect of consciousness knockout is masked by the excitatory effect of Z, no difference in task performance is observed when comparing performance in zombie and normal subjects. This indicates that in order to infer lack of causation/function, one cannot solely rely on actual or conceivable dissociations between consciousness and function; additional information about how data was generated needs to be known or assumed.Footnote ³

2.d The functional undefinability criterion lacks specificity

To accurately discriminate between hard and easy problems, it is not enough to show that consciousness is not functionally definable (i.e., demonstrate criterion sensitivity); it must also be shown that all or most other phenomena are functionally definable (demonstrate specificity). Without such evidence, the criterion may correctly identify consciousness (the true positive) along with a lot of everything else (false positives) as hard problems.Footnote ⁴ Is it true then that all/most other biological and psychological phenomena are functionally definable?

A causal role understanding of functions entails that the only things that are not functionally definable are those which are never causes—that is, epiphenomena. These are not as rare as Chalmers seems to assume. Many phenomena in biology and psychology amount to stimulus-response causal sequences replicated and studied in the laboratory (Baetu Reference Baetu, Ramsey and Ruse2019; Bechtel and Richardson Reference Bechtel and Richardson2010; Craver Reference Craver2007; Darden Reference Darden2006). In most cases, when a phenomenon is replicated in the laboratory, it is cut off from the causal structure of the world since whatever causal role the response may play under natural conditions is not allowed to unfold; hence, for all intents and purposes, the response studied is functionally undefined. Moreover, some aspects of a response may be always epiphenomenal, and therefore functionally undefinable. For instance, sunburns are experimentally characterized as the stimulus-response phenomenon of ultraviolet radiation-induced erythema (a type of inflammatory response) (Rainsford Reference Rainsford and Rainsford2015). The most distinctive feature of sunburns is the redness of the skin (erythema), which is used to measure the magnitude of the inflammatory response. Yet it is very likely that the redness itself doesn’t play any causal role (physiological, evolutionary, or other); it is simply a sterile side effect of increased blood flow. Such counterexamples undermine the assumption that all or almost all biological and psychological phenomena are functionally defined.

2.e Functional undefinability is not an obstacle to mechanistic explanation

The second axis of the functional definability criterion is the link between functional definability and mechanistic explainability. According to Chalmers, it is a conceptual truism that all it could possibly take to explain the performance of a function, and thus solve an easy problem, is to specify a mechanism. This, however, seems doubtful. For instance, in the case of a power hammer, the variable ‘mass of the hammer’ is causally relevant to the outcome ‘force applied on the target,’ as determined by controlled experiments with different loads. The variable ‘mass of the hammer’ is therefore functionally definable, yet there is no mechanism linking mass and force. Perhaps Chalmers refers to the fact that most phenomena in the life sciences are characterized as functional ‘black boxes’ linking stimuli to responses whose inner workings are subsequently elucidated by specifying mechanisms (Baetu Reference Baetu, Ramsey and Ruse2019; Bechtel and Richardson Reference Bechtel and Richardson2010; Craver and Bechtel Reference Craver and Bechtel2007; Machamer et al. Reference Machamer, Darden and Craver2000). If so, then this is not a conceptual truism, but a contingent, empirical fact about the life sciences.

Conversely, Chalmers holds that the set of functionally undefinable phenomena—which he takes to specifically include only or almost only consciousness—are not mechanistically explainable. This, too, is doubtful. Functionally undefinable phenomena as defined by Chalmers are ‘epiphenomena.’ But any effect, epiphenomenal or not, can be in principle explained by a causal mechanism. For example, even if erythema is most probably a side effect devoid of any functional relevance, it is mechanistically explained by a local increase of blood follow. Moreover, from an epistemological point of view, epiphenomenalism facilitates the elucidation of mechanisms. Ideally, what happens or doesn’t happen after a response is generated during the replication of stimulus-response phenomenon is not part of the mechanistic explanation of that phenomenon. This requirement is satisfied if nothing further happens, if the response never feeds back into the mechanism, or if the feedback occurs at a different timescale than that of the phenomenon of interest.Footnote ⁵ What worries scientists is not that epiphenomenalism may be true. Quite on the contrary, their main concern is that what is treated as an epiphenomenal response in the laboratory may not be so under natural conditions and that this may have an impact on the physiological relevance of the proposed mechanism. Downstream effects may feed back into the mechanism, altering its structure and dynamics at physiologically relevant timescales. Such feedback loops turn any attempt to quantitatively model and predict the states and outcomes of the mechanism into a nightmare of partial differential equations (Shmulevich and Aitchison Reference Shmulevich and Aitchison2009). In contrast, what doesn’t happen after an epiphenomenon is generated is a state of perpetual nonhappening that doesn’t require any further explanation or modelling. Thus, that consciousness may not play a causal role, and thus fail to have a function in the context of a system doesn’t constitute an obstacle to mechanistic explanation; if anything, it should facilitate the explanatory project.

3.a The subjective experience accompanying cognitive and behavioral functions

The criterion of functional definability is meant to capture consciousness, functional undefinability and mechanistic unexplainability under the category ‘hard problem of consciousness,’ and all/most other biological and psychological phenomena, functional definability and mechanistic explainability under the category ‘easy problems.’ It turns out, however, that the criterion lacks the perfect sensitivity and near-perfect specificity Chalmers attributes it. The claim that consciousness is functionally undefinable is unjustified, and even if sensitivity were proven, functional undefinability would still lack the required specificity to accurately discriminate consciousness from epiphenomena such as erythema. The fact that functional definability doesn’t guarantee mechanistic explainability and functional undefinability doesn’t entail mechanistic unexplainability further undermines the accuracy of the criterion as a tool for discriminating between the hard problem of consciousness and the easy problems of [almost] everything else.

Notwithstanding, Chalmers has a fallback position:

The allocution ‘in the vicinity of’ refers to a measure of statistical dependence, in this case the association between conscious experience and functional correlates:

Thus, consciousness “goes beyond problems about the performance of functions,” in the sense that it correlates with certain cognitive and behavioral functions, yet an explanation of these functions doesn’t explain consciousness.

Chalmers takes this ‘association/accompanying without explanation’ criterion to single out a highly distinctive feature of consciousness, and argues that a conceptual mistake test demonstrates the accuracy of the criterion:

3.b The sensitivity of the criterion of the accompanying phenomenon is uncertain

Proof of sensitivity rests entirely on the claim that no conceptual mistake is made when one says that consciousness remains unexplained even if functional correlates of consciousness are explained. Presumably, Chalmers takes this to follow from the tacit assumption that it is false that all it means to be conscious is to perform the functions of discrimination, integration, and reporting of information. But what justifies this assumption? Two possible responses may be envisaged. The first would be to treat this assumption as a definitional matter. If we go down this path, we run into the problem of conceptual relativism. For example, since Tononi and Koch (Reference Tononi and Koch2015) are committed to the view that consciousness is nothing else but integrated information, they would insist that one does make a conceptual mistake when claiming that an explanation of how information is discriminated and integrated leaves consciousness unexplained.

A more promising option is to seek an empirical justification all parties might accept. For example, LeDoux (Reference LeDoux1996) hypothesized that adaptive behavioral responses correlate with conscious feelings of fear because the two are divergent effects of a common cause, namely threatening stimuli. A common cause model entails that an explanation of the mechanism linking stimulus and behavioral correlate will not shed any light on the mechanism linking stimulus and conscious experience. If this model turns out to be correct and can be generalized to other functional correlates, then one can conclude that, given our current understanding of causation, no conceptual mistake is made by stating that an explanation of functional correlates will not explain consciousness. For the time being, however, there is no conclusive evidence to support this view. As discussed in section 2.c, currently available evidence is likewise compatible with rival models hypothesizing that functional correlates causally determine consciousness—as, for instance, proposed by the feature integration theories (Treccani Reference Treccani2018). According to these models, the mechanisms underpinning functional correlates overlap with those of consciousness. If we accept these models, it would be a conceptual mistake to expect that an explanation of functional correlates will contribute nothing to an explanation of consciousness or to rule out the possibility that an explanation of functional correlates may suffice to explain consciousness.

Whether we pursue the definitional route or the empirical alternative, the conceptual mistake test is inconclusive: consciousness may or may not be said to “go beyond” problems about the performance of functions depending on background assumptions about how consciousness is defined and which causal model best explains the association between consciousness and functional correlates.

3.c The criterion of the accompanying phenomenon lacks any usable degree of specificity

Before proceeding to an evaluation of the specificity of the conceptual mistake test, some preliminary clarifications are needed. First, the gene is neither a phenomenon in need of an explanation, nor an explanation of anything, but a scientific concept defining a class of objects. A concept refers or fails to refer depending on whether there are things in the world that correspond to the description postulated by the concept, and the classifications it generates are objective or subjective depending on whether the things to which it refers constitute a natural kind (Boyd Reference Boyd, Beebee and Sabbarton-Leary2010; Machery Reference Machery2009). Second, when Chalmers states that “all it means to be a gene is to be an entity that performs the relevant storage and transmission function,” he is not talking about genes as understood in genetics and molecular biology, but of a functionalist gene* concept of his own invention.Footnote ⁶ Third, Chalmers’s gene* concept refers indiscriminately to chromosomes, plasmids, maternal RNA, transcription factors and their cellular localization, DNA methylation and histone acetylation, mitochondria and mitochondrial DNA, as well as a host of environmental influences (drugs, pathogens, viruses, prions) transmitted vertically or horizontally. Since a different mechanism underpins a different pattern of heredity in each case, this collection of factors and processes doesn’t constitute a natural kind.

This last point indicates that even if s₁ and s₂ are genes* (i.e., storers and transmitters of inheritance information), explaining how s₁ is a gene* may not shed any light on how s₂ is a gene*. For instance, one doesn’t make a conceptual mistake in saying, “I can see that you have explained how chromosomal DNA is a gene*, but you have not explained how maternal RNA, is a gene*.” It should likewise be clear that the fact that an item s_i falls within the extension of a concept C ₁ doesn’t necessarily entail that s_i doesn’t also fall within the extension of some other concepts C ₂, C ₃, … C_j. Thus, one can say, “I can see that you have explained how s_i is a C ₁, but you have not explained how s_i is a C ₂, C ₃, … C_j.” This is particularly obvious in the case of functional concepts: since any given cause can have many effects, it can play multiple causal roles, and thus have multiple functions, each captured by a distinct functional concept. For instance, since DNA performs the relevant storage and transmission function, DNA is a gene*. However, DNA also plays a role in maintaining the structural integrity of chromosomes. Thus, one doesn’t make a conceptual mistake in saying, “I can see that you have explained how DNA stores and transmits hereditary information from one generation to the next, but you have not explained how it is a structure supporting the chromosome.” Finally, a functionalist concept of the form ‘All it means to be X is to perform function F’ fixes only the functional outcome without positing any constraints on the mechanistic how causing the outcome. In particular, such concepts don’t prohibit a priori mechanisms involving conscious performers of a function. Thus, the protagonist of a dystopian novel wouldn’t commit a conceptual mistake by saying, “I can see that you have explained how downsized nanoproletarians store and transmit hereditary information from one generation to the next, but you have not explained how they experience their condition.”

Whichever way we turn it, the conceptual mistake test fails to deliver any kind of specificity. No conceptual mistake is made if one claims that the explanation of how a gene* is a gene* doesn’t explain how an[other] gene* is a gene*, or how a gene* does something else than perform the function of gene*, including how a gene* is conscious. In response, some may be tempted to retort that the conceptual mistake test might fare better for other concepts or functions. I have serious reserves about this rebuttal. For one thing, the objections I raised generalize to any concept that categorizes things based on functional attribution alone. A deeper concern is that Chalmers commits a category mistake when he proposes to compare a phenomenon to a concept, be it functional or not. Since a hard problem is a mechanistically unexplainable phenomenon and an easy problem is a mechanistically explainable phenomenon, a test should assess whether the criterion of the accompanying phenomenon discriminates between mechanistically explained phenomena and the phenomenon of consciousness, and not whether it discriminates between concepts and the phenomenon of consciousness.

If we correct this inconsistency and consider the relevant term of comparison, it can be easily shown that many phenomena go beyond explanations of a given set of functions. For example, in the 1960s, immunologists observed that the phenomenon of sunburns is accompanied by several biological activities (Rainsford Reference Rainsford and Rainsford2015). Yet even though the science of that time explained the performance of biological activities in the vicinity of sunburns, such as gene expression regulation and protein synthesis, a further unanswered question remained: Why was the performance of these activities accompanied by sunburns? In fact, there is an endless list of such ‘why’ questions. Why is λ-phage replication accompanied by cosmic background radiation? Why are religious beliefs accompanied by λ-phage replication? Why are lamps switches accompanied by religious beliefs? There is nothing absurd or contradictory in asking why statistically significant associations occur. Quite on the contrary, since such associations cannot be attributed to chance alone, they are paradigmatic examples of phenomena demanding an explanation (Baetu Reference Baetu, Ramsey and Ruse2019; Bogen and Woodward Reference Bogen and Woodward1988).

It can further be shown that at least some phenomena that go beyond explanations of a given set of functions are amenable to mechanistic explanations. In the case of sunburns, the mechanisms underpinning inflammatory responses were subsequently elucidated, revealing that the biological activities of gene expression regulation and protein synthesis, along with other mechanistic details, are causally involved in the production of sunburns (Clydesdale et al. Reference Clydesdale, Dandie and Muller2001). This addresses the ‘why’ question raised earlier: sunburns accompany biological activities because the latter are causally relevant to the former. Yet even though the question has been answered and the mechanisms responsible for the phenomenon of sunburns have been explained, it is still the case that questions about sunburns go beyond problems about the performance of the functions of gene expression regulation and protein synthesis. This is simply because the latter are part of the mechanism of sunburns but are insufficient to explain the phenomenon. As for the other, seemingly more outlandish, ‘why’ questions mentioned above, they are banal cases of phenomena accompanying one another in virtue of having a common cause. Thus, the mere fact that a phenomenon goes beyond the performance of functions in their vicinity is neither a unique property of consciousness, nor a reliable indicator that researchers are confronted with a mechanistically intractable hard problem.

4.a Objective functions vs. subjective data

The last criterion of demarcation between hard and easy problems hinges on the distinction between the subjective nature of first-person data vs. the objective nature of third-person data:

The above proposes an alignment of third-person data with ‘being about objective functioning’/mechanistic explainability, and of first-person data with ‘being about subjective experience’ and ‘not being about objective functioning’/mechanistic unexplainability. We can safely conjecture that ‘objective functioning’ refers to the causal link between the thing that is attributed a function (the cause) and its function (to bring about an effect). This reading is consistent Chalmers’s definition of ‘function’ as a causal role (section 2.b), as well as with the assumption that, at least in the life sciences, functions are typically explained by mechanisms (section 2.e). As for the concept of ‘data being about something,’ it seems reasonable to assume that data are informative of (e.g., measure, predict, correlate with) the something they are about. Under this interpretation, first-person data are characterized as informative of subjective experience and uninformative of mechanistically explainable functional relationships, while third-person data are thought to convey information about mechanistically explainable functional relationships. As discussed in section 2, Chalmers takes functional definability and mechanistic explainability to be characteristic of the easy problems, while the absence of these properties is the distinctive mark of a hard problem.

4.b The hard problem of consciousness is not uniformly hard

A noticeable peculiarity of the above-proposed alignment is that while first-person data are about subjective experience and not about objective functioning, it is only stipulated that third-person data are about objective functioning; nothing is said about third-person data not being about subjective experience. Presumably, this omission is meant to accommodate the widespread assumption that whenever we talk about what we see, feel, think, and so on, we engage in spontaneous report experiments generating third-person data meant to inform—and occasionally misinform—others about the private first-person data of subjective experience. This assumption is empirically supported by self-experimentation studies in which researchers adopt simultaneously a first- and a third-person perspective (Head Reference Head1920; Piccinini Reference Piccinini2009; Price and Aydede Reference Price, Aydede and Aydede2005; Price and Barrell Reference Price and Barrell2012). Among other things, such studies provided evidence validating the use of verbal reporting as a method for assessing both consciousness and consciousness contents. In turn, verbal reports are commonly used to validate other measurement techniques, such as those relying on behavioral (Teasdale and Murray Reference Teasdale and Murray2000), neurological (Owen et al. Reference Owen, Coleman, Boly, Davis, Laureys and Pickard2006) and informational (Casarotto et al. Reference Casarotto, Comanducci, Rosanova, Sarasso, Fecchio, Napolitani, Pigorini, Casali, Trimarchi, Boly, Gosseries, Bodart, Curto, Landi, Mariotti, Devalle, Laureys, Tononi and Massimini2016) correlates of consciousness.

Researchers also attempted to disentangle the effects of conscious and nonconscious processing in verbal and verbal-like reporting. Experiments in which the reporting condition is manipulated revealed that the same stimulation condition may result in different discriminability thresholdsFootnote ⁷, depending on the instructions used to elicit reports (e.g., report, guess, rate confidence, introspect, etc.), the modality of the report (verbal, pointing, clicking, blinking) and other constraints imposed on reporting (e.g., leisurely vs. as fast as possible; yes/no forced choice vs. yes/no/I don’t know). By far, the best understood dissociation concerns ‘guess’ vs. ‘awareness’ verbal reports (Dienes Reference Dienes2008; Dienes et al. Reference Dienes, Gerry, Kwan and Goode1995), the contrasting of which emerged as the method of choice for demonstrating “nonconscious perception in both normal subjects and blindsight patients” (Marcel Reference Marcel1994, 80). A statistically significant distinction can be made between a ‘subjective’ and an ‘objective discriminability threshold’ corresponding to “detection level at which subjects claim not to be able to discriminate perceptual information at better than a chance level” [when instructed to report awareness] vs. “detection level at which perceptual information is actually discriminated at a chance level” [when subjects are instructed to guess] (Cheesman and Merikle Reference Cheesman and Merikle1984, 391). According to a popular interpretation, the subjective threshold is neither about the stimulus, nor about the subject’s perceptual acuity and its impact on task performance, which are measured by the objective threshold, but rather about the subject’s conscious awareness of the stimulus and decisions about how to act on this awareness.

If we accept adequately controlled verbal reports and other correlates as legitimate examples of third-person data about consciousness, then, according to the first- vs. third-person data demarcation criterion, some third-person data are both about consciousness and about mechanistically explainable objective functions. But if this is the case, then that at least some aspects of consciousness—namely those measured by third-person data—can be explained mechanistically whether or not first-person data are amenable to mechanistic explanation.

4.c The partial transparency of experience

In response to this caveat, Chalmers may turn tables and insist that whether or not third-person data about consciousness are available and explainable, first-person data of subjective experience ‘go beyond’ problems about objective functioning, that these data remain unexplained, and that the hard problem of consciousness is precisely about the unexplained character of first-person data. Yet this cannot possibly be true of all first-person data. Being informative about something is a symmetrical relationship: if A is informative of B, B is necessarily informative of A (Steinhart Reference Steinhart2018, ch. 6). Thus, if in addition to being informative of objective functioning, some third-person data also convey information about the first-person data of subjective experience, then some first-person data must likewise convey information about things measured by third-person data, namely objective experience, in addition to being about subjective experience. In such cases, the properties ‘being about subjective experience’ and ‘being about objective functioning’ cannot be exclusively assigned to first- and third-person data, respectively.

It is not difficult to empirically verify that this is indeed the case. At least some first-person subjective experiences, such as feeling of color and motion, are informative of intersubjectively verifiable causal phenomena such as changes in color of a litmus paper and the motion of the moon in the sky.Footnote ⁸ But if first-person data are not exclusively aligned with subjective experience/the mechanistically intractable hard problem of consciousness and third-person data are not exclusively aligned with objective functioning/mechanistically tractable easy problems, then the criterion is noncategorical: at least for some first- and third-person data, it classifies problems as simultaneously hard and easy.

Moreover, it is impossible to determine a priori whether any given first-person datum is solely about subjective experience or about subjective experience and objective functioning. Here is an example that illustrates the problem. Pain and erythema are assessed by essentially the same measurement technique. In clinical practice, pain is usually measured by instructing patients to introspect and rate their current pain levels on a numerical scale of 0 (no pain) to 10 (worst pain imaginable) (Noble et al. Reference Noble, Clark, Meldrum, ten Have, Seymour, Winslow and Paz2005). In a similar way, clinicians usually measure the intensity of skin inflammation by visual assessment and reporting a value on a four-point scale ranging from ‘no erythema’ to ‘violet erythema with edema’ (Rainsford Reference Rainsford and Rainsford2015). From the first-person perspective of the subject assessing pain or erythema, measurements are based entirely on subjective experiences of what pain or visual awareness feel like now and how these subjective experiences compare with past and imagined experiences. From this perspective, there are no reasons to suspect that erythema measurements are any more or less likely to be about objective functioning than pain measurements. It is only after measurements are reported and shared intersubjectively that the subject finds out that some of the reported experiences systematically agree with the reports issued by other agents, while others don’t. In this particular example, an intersubjective agreement is systematically reached about erythema ratings—as indicated by the reliability of the scale, as well as consistency with instrument-based measures (Fullerton et al. Reference Fullerton, Fischer, Lahti, Wilhelm, Takiwaki and Serup1996). In contrast, it is much more common that pain ratings are not amenable to intersubjective agreement or consistent with behavioral and physiological measures, suggesting that pain assessment can generate information of a strictly subjective nature (IASP Task Force on Taxonomy 2020).Footnote ⁹

If it cannot be established a priori to what extent first-person experience is amenable to intersubjective agreement, then the premise that “that first-person data are not data about objective functioning” is not a self-evident or necessary truth, but an empirical fact contingent on what happens after any given first-person experience is shared. Since this premise is taken to justify the conclusion that “explaining the objective functions does not explain subjective experience,” it can further be objected that this conclusion may or may not be justified depending on whether the justificatory premise is true or not.Footnote ¹⁰ In other words, the overall ‘easiness’ or ‘hardness’ of the problem of consciousness is contingent on the extent to which first- and third-person data are shown to convey or fail to convey information about one another.