While not strictly Popperian falsificationism (Popper, 1962, 2005), the “Accelerating Research on Consciousness” (ARC) project deploys a methodology guided by the principle of falsification combined with adversarial collaboration (see e.g., Kahneman, 2003). In short, proponents of competing theories agree beforehand on an experimental set-up on which they predict different outcomes. The experiment is then conducted with the expectation that only one theory should be able to be corroborated, while the other(s) is (partly) falsified.

ARC is still in its early stages with results from only one sub-project (COGITATE) being published so far (Melloni et al., 2023). Nevertheless ARC has, due to the buy-in from prominent researchers advancing competing theories, along with its impressive scope and ambition, attracted a lot of attention. However, significant question marks remain pertaining to whether it will be applicable across the field. One reason is that ARC projects treat only a few theories at a time with each project taking years and being very cost intensive. Faced with more than two dozen theories, this becomes unfeasible in practice. Additionally, it is unclear how to weigh multiple ARC results against each other. To boot, it is standard procedure in science to revise one's theory in light of new empirical evidence, so multiyear multimillion dollar ARC projects may not successfully eliminate theories. This was shown in the results of COGITATE that compared predictions from Integrated Information Theory (IIT) and Global Neuronal Workspace Theory (GNWT). No strong conclusion was afforded by the data and both methodology and interpretation of data was questioned by proponents of the participating theories, especially in areas that went against their preferred theory (Ferrante et al., 2023; Melloni et al., 2023). While further issues with ARC have also been highlighted (Kirkeby-Hinrup, 2024b), there is some cautious optimism about the future of adversarial collaborations. Specifically, that tempered with the proper philosophical framework, adversarial collaborations (in theory) are promising catalysts for individual theoretical refinement and development (Corcoran et al., 2023; Negro, 2024). We acknowledge that adversarial collaboration is still in a nascent stage, and its full potential remains to be seen.

(Doerig et al. 2020) develop an approach (CRIT) centered around a set of criteria for theories. The idea is to evaluate theories against a set of criteria, with the theory that satisfies the most criteria being preferable over the others. An added benefit is that by situating theories in relation to the same set of criteria, a better overview is made possible, and the strengths and weaknesses of each theory are illuminated.

There are questions as to CRITs ability to deliver an evaluation of the set of theories that is sufficiently fine-grained to make any recommendation in our conundrum (Kirkeby-Hinrup, 2024b). Furthermore, the focus on specific criteria can render CRIT blind to the amount of, and quality of, empirical support that different theories might have. To boot, it is unclear how to select criteria in an unbiased way. Finally, given that the set of criteria are relatively small, ties are likely, and it is unclear how to resolve this in a manner that does not subvert the whole approach (Kirkeby-Hinrup, 2024b).

The third approach to comparing theories is based on the method of Inference to the Best Explanation combined with Bayesian updating and accommodated to track the relations between theories and data. In “Quantification to the Best Explanation” (QBE) (Kirkeby-Hinrup 2024b); cf., (Kirkeby-Hinrup and Fazekas 2021) suggests a four-step process to assessing and comparing theories. The four steps are assimilation, compilation, validation, and comparison. That is: collect evidence for each theory, compile the sets of evidence, validate the claimed empirical support, and compare theories based on the validated sets. The idea behind this methodology is that by quantifying the validated sets comparisons of the respective empirical support for each theory becomes feasible. Furthermore, QBE can offer focused evaluations examining specific domains (e.g., credence, replicability or scope) of the evidence proposed in favor of a specific theory along with similar filters on the kinds of evidence factored in. A strength of the method is its generalizability and sensitivity, with fine-grained comparisons across multiple theories posing no problem.

Being underdeveloped, QBE nevertheless faces significant challenges. Primarily, the exact mathematics/statistics behind the theory comparison needs further explication. Furthermore, it is unclear how to resolve any disagreements that may emerge about these mathematics/statistics. While there may be some promise in moving disagreements from a domain plagued by conceptual disagreements (philosophy/consciousness studies. See e.g., Section 2 above, and (Kirkeby-Hinrup 2024a) to a domain with a comparatively more determinate and established conceptual framework (mathematics/statistics), it is unclear if this will be sufficient to achieve any kind of consensus with respect to which theory is preferable. This again, partly owes to the socio-scientific factor mentioned in Section 2 above (changing people's mind).

(Chis-Ciure et al. 2024) Measure Centrality Index (MCI) aims to enable fruitful inter-theory classification that systematically can help evaluate and compare theories. The idea is to map the concepts and measures of theories to determine what (kinds of) data would impact a theory and how much a given piece of data would impact it (Chis-Ciure et al., 2024). By using the MCI classification interface it is possible (ideally) to identify specific empirical measures and rank their importance for different theories. The idea is that knowing the (kind of) data that can impact a theory will be relevant to assessing empirical evidence for or against theories. Additionally, knowing how much a specific piece of data would impact a theory allows for better experimental designs in terms of achieving maximal impact on the debates between theories.

Importantly, MCI is only feasible if comparison is possible in the first place, which, arguably given the above, is not always the case. Specifically, the MCI may face questions regarding conceptual bleed, and how to determine non-biased ways to align disparate conceptual frameworks to allow experiments to be devised. Finally, it is worth noting that there is one sense in which the MCI is not as much a way of assessing and comparing theories, as it is a tool to guide such endeavors. So at least one obvious strength of the MCI is as an auxiliary framework to recommend input (in terms of which measurements may be most impactful on debates) to an assessment and comparison process. For instance, the MCI and the QBE approach discussed above may complement each other.

One of the most pervasive and long running debates in the field concerns the distinction between Access (A-) consciousness and Phenomenal (P-) consciousness (Block, 1995, 2007). Call this the Access-Phenomenal Distinction (APD). According to APD, A-consciousness is involved in cognition and behavior, whereas P-consciousness is uniquely experiential. The role of A-consciousness with respect to cognition and behavior presents a problem for the measurement of P-consciousness. This problem is exacerbated by the fact that our best access to the (P-) conscious states of a subject is via subjective reports (i.e., asking the subjects what they experience). This is because conceptualization and verbal reporting are hallmark features of A-consciousness. Against this background, the hot topic considered here relates to how one should measure P-consciousness without the measurement being confounded by A-conscious processes related to reporting.

The central aspect here was the so-called no-report paradigms (Block, 2019; Duman et al., 2022; Overgaard and Fazekas, 2016; Schlossmacher et al., 2020; Tsuchiya et al., 2015; Whyte et al., 2022). The idea was to measure neural activity in a binocular rivalry (Frässle et al., 2014; Pitts et al., 2010; Sandberg et al., 2014) paradigm that did not require subjects to report visual changes. When viewed independently and when contrasting the no-report measurements with other (report-based) measurements of binocular rivalry the hope was to identify by subtraction the neural signatures related to the reporting and by removing them arrive at the correlates of P-consciousness.

Critical junctures in the debate surrounding this hot topic include (Tsuchiya et al. 2015) pointing out a possible confound related to pre-conscious and post-conscious activity, and (Block 2019, 2020) highlighting our inability to control for task-unrelated thoughts (A-conscious activity). Despite garnering significant attention, this hot topic (no-report paradigms) thus far appears to have failed to move the needle significantly in the debates involving proponents of APD and their critics. Nevertheless, novel no-report paradigms are still being developed, and in so far as one cares about the study of consciousness, one should root for these developments to eventually produce promising results.

Block's overflow claim posits that P-consciousness harbors more information than a subject can access (Block, 2007, 2011b). So, subjects can have P-conscious experiences they are not aware of having. Or put more simply: because the phenomenal domain outstrips the cognitive domain, we experience more than we can access.

As a hot topic, has fueled many theoretical considerations about how the evidence could even apply (Cohen and Dennett, 2011; Michel, 2019; Phillips, 2018). On the empirical side, overflow has been fueled by the Sperling paradigm (Block, 1995, 2007, 2011b; Sperling, 1960). In the Sperling paradigm, a matrix of letters is presented briefly, and subjects subsequently are asked to report which letters they saw. Normally, subjects say that it felt like they saw all the letters, but they are able to report only 3–5 specific letters. However, if, after the visual stimulus of the matrix has disappeared, subjects are cued to a specific row in the matrix, they retain the ability to report between 3–5 (despite not knowing which row they would be required to report). This indicates that information about the identity of all the letters is available prior to the report, supporting the subjects' claim that they see all the letters. In deploying APD and overflow to explain these results it is posited that subjects are P-conscious of all the letters, but because A-consciousness has limited capacity, not everything in P-consciousness can be accessed. In other words, the 3–5 item caps are not indicative of the amount of P-conscious content.

The hot topic gained new steam with the publication of the color diversity paradigm (Bronfman et al., 2014; Usher et al., 2018), which improved on Sperling by extracting task-irrelevant color-judgments supposedly indicating P-consciousness. In this variant, the letters in the matrix are of different colors, and the subjects are subsequently tasked with judging whether the diversity of colors was “high” or “low.” Despite attention being allocated to the Sperling task, performance was good on the color diversity judgments even for the un-cued rows.

Framing the support overflow derived from the color diversity data in contrast to workspace theories, Block (2014, p. 446) concludes that the color diversity data “reveals that there must have been conscious awareness of specific colors beyond the limits of the global workspace because a trace of that conscious awareness in the form of a diversity judgment can enter the global workspace for free.” Similarly, Jacobson says (2015, pp. 1032–1033):

As an excellent example of how hot topics may yield novel empirical developments, (Amir et al. 2023) devised a paradigm to show P-consciousness without A-consciousness. Subjects were queried about the presence of any experience while a persistent pink noise was playing. Only when the noise stopped were the subjects able to report that it had been present, yet supposedly they must have been experiencing the whole time. The reactions in the field to this paradigm have yet to fully manifest themselves, so it is unclear whether it will constitute progress in the end.

Problematically, the Amir et al. paradigm suffers from the same objection as the Sperling and color diversity paradigms. The general reply to this data by critics of APD has been to highlight that there is an equally good interpretation of the data that does not presuppose APD. According to this interpretation task performance in all three cases is driven by unconscious processes, and the sentiment is that it is superfluous to postulate P-consciousness as part of an explanation. In any case, while overflow perennially resurges as a hot topic, there is so far no indication that progress is made to an extent where anyone is changing their mind.

“Levels” was introduced to account for changes in global states of consciousness in the literature on disorders of consciousness (DOCd). As opposed to conscious contents or “local” states, “global states” refer to the overall state of the system. Global states include, for example, sleeping, dreaming, or being awake, and clinical conditions such as coma, sedation, the minimally conscious state, and Unresponsive Wakefulness Syndrome (“UWS”; formerly, “vegetative state”) (Barra et al., 2020, fig 2.1). Increasing research on DOCs challenged the long-standing view that consciousness is a binary matter (either you have it or not), suggesting some states are “more” conscious than others. The prospect of overturning an age-old view on consciousness certainly made it a hot topic. The interpretation has since been challenged, however (Bayne et al., 2016; Overgaard and Overgaard, 2010).

Moreover, the very concept of levels has been called into question as not capturing the complexity of the neural states of DOC patients (Bayne et al., 2016). Evidence suggests that levels (e.g., coma, sleep, wakefulness) cannot be tied to a single neural mechanism, and each patient shows unique characteristics (Overgaard and Overgaard, 2010).

At this point, the discussion surrounding levels has become consolidated in the field and now may be more akin to a sub-field than a hot topic. In any case, it seems that levels are largely orthogonal to debates between theories, and there has been no significant shift in opinion with respect to which theory is more plausible in light of discussions of levels. Nevertheless, since levels of consciousness are of particular societal concern (e.g., in terms of recovery predictions for various stages of coma), evaluating and comparing theories in this regard should carry at least some force. Moreover, since most theories agree that dreams are genuine experiences, NREM sleep and dreaming do provide interesting contrast cases, against which to evaluate putative neural correlates of consciousness.

One contemporary hot topic is AI consciousness. Considerations about (the possibility of) AI consciousness have quickly become a prominent public and scientific concern. The field of consciousness studies, naturally, is engaged with these concerns.

Roughly, there are two overall approaches to assessing AI consciousness. The first approach assumes that some X is necessary for consciousness and then proceeds to check if AI has X. Competing theories of consciousness propose competing candidates for X. Consequently, if we presume any single theory (ideally) this establishes an X and allows us to examine AI systems and draw conclusions about AI consciousness (Butlin et al., 2023). Unfortunately, assuming another theory—thereby establishing a different X—may yield a conflicting conclusion about AI consciousness. To solve this issue, an argument is needed to motivate picking one theory over another, otherwise any conclusions about AI consciousness will be arbitrary (and, for example, unsuitable to guide future research and ethical or political decisions). This means that to solve the issue with conflicting inferences from different Xs it is now necessary to determine which theory of consciousness is “right” so trying to establish this leads us back to the conundrum.

The second approach to assessing AI consciousness is to subject them to so-called consciousness-tests or “C-tests” (Bayne et al., 2024). A C-test checks if AI has a certain property or ability associated with consciousness. Typically, a barrage of C-tests is applied to ameliorate a central shortcoming of the approach (Bayne et al., 2024; Butlin et al., 2023). However, with respect to the core idea of the test-approach it is inconsequential if a single C-test is applied, or if a barrage of tests is deployed.

One obvious example of a property or ability associated with consciousness is expressions of consciousness, so-called C-expressions (Kirkeby-Hinrup and Stenseke, 2025). Overall, one would expect C-expressions to be highly correlated with the presence of consciousness, since systems without consciousness would have little reason to produce them, while systems that are conscious would have some reason to produce them. This means that if something tells us it is conscious (a paradigm C-expression), then this lends support to the belief that it might be. Yet, a lack of C-expressions does not entail a lack of consciousness. Similarly, the presence of C-expressions does not entail the presence of consciousness. Consequently, to bolster our credence in whichever conclusion we reach, it is useful to add further C-tests. Supposedly, the more C-tests that agree, the higher our credence should be that AI is or is not conscious (Bayne et al., 2024).

This test-approach has similarities with at least two long standing issues in philosophy. Firstly; the issue of induction (Roughly: that sequences of observations never yield certainty. See e.g., Hume, 2000; Minnameier, 2010). Depending on one's leanings with respect to metaphysics and philosophy of science, this may be an acceptable bullet to bite if our credence in the answer (i.e., number of C-tests passed or failed) was very high. In any case, “knowledge” may require something less than certainty (Olsson, 2007, 2016). The second similar issue is the problem of other minds (Kirkeby-Hinrup and Stenseke, 2025), the crux of which is that we cannot know if persons (other than ourselves) really are conscious, as opposed to just behaving as if they are. This is essentially the same situation we find ourselves in with the barrage of C-tests. Unfortunately, unlike the problem of other minds, which we can largely ignore by adopting a polite convention that “everyone thinks” (Turing, 1950), we are actually interested in reaching conclusions with respect to AI consciousness. Consequently, to solve this issue we need a C-test that actually measures consciousness (a Real-C-test), rather than a barrage of ones that are merely correlated with consciousness (behavior). However, to obtain a Real-C-test requires us to understand how consciousness is generated. The only place to look for this understanding is in humans, since after all, humans are the only beings we are certain are conscious (inferring from our subjective case, c.f. the problem of other minds). This means that a test approach to AI consciousness requires us to understand human consciousness, i.e., have/establish a theory of consciousness. This, again, leads back to the conundrum.

In sum, the two central approaches considered here each faced issues which in turn required a solution to the issues discussed in Section 2 and 3 concerning comparison of theories. In any case, given that progress in AI consciousness seems to rely heavily on progress on human consciousness, there is little indication that this hot topic will move forward debates between theories of consciousness significantly.²

In Section 3, we considered four prominent approaches to comparing theories of consciousness. We were able to identify shortcomings for each of these. The conclusion was for every approach either principled and/or practical obstacles made it unlikely that it would get us out of our conundrum. In this section, we considered cases where interest spiked with respect to certain data, a concept, or a debate, what we called hot topics. The motivation for this was that hot topics are characterized by increased levels of engagement, novelty, and progress. Furthermore, because they tend to engage proponents of more than one theory, they offer rare opportunities for truly dynamic exchange and interaction which (all else being equal) is presumably a good thing. Regrettably, in the context of making progress in debates between theories, the example cases considered above hardly moved the needle.³ This is unfortunate given that the hot topics were supposedly ideal for making progress. Consequently, with respect to guidance concerning which theory is most plausible, our most prominent approaches to comparison of theories and our best cases of empirical debates to date have yet to yield any progress.

In reaction to this, perhaps too much focus and expectation is put hot topics. Perhaps it is the large amount of steady work outside of the limelight where the progress will be made by researchers quietly grinding away. Because this work is conducted largely outside of the limelight, and much of it receives undue little attention, we will call this domain cold cases. In the next section, we will consider this domain and a critical upshot of it.

In (Smith 2019) argued for the distinction between phenomenal and access consciousness based on hydranencephaly. Hydranencephalic patients are born without either cerebral hemisphere, leaving them unable to move (due to lack of motor cortex) and with short life expectancy. According to Smith, hydranencephalics are phenomenally conscious but lack cognitive access. He bases this on their differentiated affective responses in relation to their primary caregiver vs. strangers. Because the patients are missing (parts of) the cortical areas (such as the prefrontal cortex) thought to underlie A-consciousness, Smith concludes they cannot have A-consciousness in the normal sense. Smith suggests that hydranencephalics may have another kind of affective-grounded access (consciousness) which explains their ability to discriminate between the primary caregiver and strangers.

Here, we will not weigh in on the merits of Smith's argument. This is because our interest is in what makes this a cold case. Namely, on the one hand: that Smith applies an empirical phenomenon to the APD distinction (which is a proxy for a theory), and on the other hand, that this application has received little to no attention (e.g., in terms of citation metrics).

Now, if we think empirical plausibility is important, then cases like Smith's should be counted when we assess and compare theories. However, they generally are not (unsurprisingly since this is part of the definition of “cold cases”). What should one make of this? Starting with a practical perspective one issue is that no-one is in fact counting these cases. In other words, there is some actual work to be done. Since this is “merely” a practical issue, we will leave it to the side here, and focus instead on another issue, which is that if we want to count Smith's case when we assess and compare theories, then Smith's case itself must be (independently) assessed.

Our notion of independent assessment has counterparts in the notion of replication in empirical sciences and fact checking in public discourse. We suggest such work has a crucial role to play, with respect to including cold cases when we assess and compare theories. The plausibility of a proposed connection between a piece of empirical evidence and a theory of consciousness (e.g., whether or not we should count the cold case in question when assessing and comparing the theory) may be impacted by several factors. For instance, there may be alternative (better) explanations of the outcome, alternative hypotheses that need to be tested, more than one cause of the phenomenon, issues with the experimental paradigm, unnoticed less salient outcomes, the conceptual mapping has gone awry, reasoning errors, etc. (Fink, 2015, 2016; Kirkeby-Hinrup, 2021; Kirkeby-Hinrup and Fazekas, 2021; Kozuch, 2014; Malach, 2011; Odegaard et al., 2017; Overgaard and Kirkeby-Hinrup, 2021).

In comparison, the hot topics do not suffer from such issues to a similar extent. Presumably, this is a direct effect of the additional attention hot topics enjoy. Occasionally, the actual debate (that turns something into a hot topic), will be about one or more of these factors. For just two examples of discussions of such factors are the many competing interpretations of change blindness (Block, 2011b; Lau and Rosenthal, 2011; Sergent and Dehaene, 2004; Tononi and Koch, 2008), and discussions of the COGITATE results (Ferrante et al., 2023; Melloni et al., 2023).

As with counting the cold cases, we again face a practical issue when it comes to independent assessment. However, this time around, it is a little more serious. The issue here is who performs the independent assessment (not just getting people to do it, which was the issue with counting cold cases). Call this the vetting problem. Briefly, the vetting problem concerns who is suitable to carry out the independent assessment of cold cases. It turns on the fact that most researchers knowledgeable enough to evaluate proposed empirical support (or objections) may not be suitable for this task, as they have vested—albeit often different—interests in the debates. In other words, there is a risk that when independently assessing cold cases, sympathizers of a theory may be biased, and objectors to a theory may beg the question. Fortunately, we think this problem washes out if one does not consider the assessment process as a finite one-time exchange with a limited number of participants. Instead, we propose the independent assessment be thought of as a dialogue in the community. In one sense, one might say that independent assessment really is an exercise in turning cold cases into hot topics. Formulated differently, one might say that if we want to count cold cases when assessing and comparing theories, they need to stop being cold cases, and the overall project (if we want empirical evidence to matter, when assessing and comparing theories) is to eliminate cold cases.

Concretely, independent assessment is not hard, as it merely involves scrutinizing the empirical data, its interpretation, and the argument in which it figures. The purpose is to identify if the proposed empirical argument (or objection) suffers from any of the shortcomings above. If no serious issues are found, then we can safely count the cold case when assessing and comparing theories. In case some shortcoming is identified, this is not necessarily negative, as it points to avenues of further exploration to clarify if the shortcoming can be remedied, and, if so, taking steps to remedy it. Ultimately, if a shortcoming cannot be remedied, then the cold case can be discarded and must be disregarded when assessing and comparing theories. Next, we consider two examples of independent assessment.

(D'Aloisio-Montilla 2017a,b) argues that aphantasics' performance on change detection paradigms supports overflow. Aphantasics lack the ability to generate voluntary mental imagery. According to D'Aloisio-Montilla, no-overflow accounts must appeal to an internal image to explain the results of retro-cue paradigms (Landman et al., 2003). The performance of aphantasics is on par with normal subjects, but we cannot appeal to internal imagery in their case. This is generalized to suggest it is unlikely that internal imagery is driving the performance in normal subjects. If we cannot appeal to internal imagery, D'Aloisio-Montilla argues, this supports the competing interpretation, viz. overflow. When independently assessing this argument, (Kirkeby-Hinrup and Fazekas 2021) find that the two empirical paradigms D'Aloisio-Montilla combine to make his argument differ in significant ways that subvert his conclusion.

The second example concerns the possibility of higher-order misrepresentation, which even if a perennial hot topic (Block, 2011c; Brinck and Kirkeby-Hinrup, 2017; Brown et al., 2019; Gennaro, 2004, 2013; Kirkeby-Hinrup, 2022; Rosenthal, 2011, 2012; Weisberg, 2010, 2011), contains cold cases. In one early publication, Hakwan Lau and David Rosenthal proposed the cases of rare Charles Bonnet syndrome (Lau and Rosenthal, 2011) as evidence of higher-order misrepresentation (higher-order thoughts that do not have a target first-order state). In “normal” Charles Bonnet syndrome, subjects have visual hallucinations. However, in the rare cases, subjects had damage to the primary visual cortex (Ashwin and Tsaloumas, 2007; Duggal and Pierri, 2002), which Lau and Rosenthal hypothesize underpins the generation of first-order states. The upshot is that there (ex hypothesis) cannot be any first-order visual states corresponding to the visual hallucinations in the rare cases of Charles Bonnet syndrome. In sum, supposedly, the cases rare Charles bonnet syndrome demonstrates that higher-order representations without target first-order states are sufficient for conscious experience. In his assessment of the underlying argument supporting this conclusion, (Kirkeby-Hinrup 2014) found an equivocation in one of the premises. Furthermore, neither of the possible resolutions to the equivocation allowed the argument to persist. Consequently, the cases of rare Charles Bonnet syndrome cannot be counted when assessing and comparing the empirical support for higher-order theories (the group of theories for which the possibility of higher-order misrepresentation is relevant).

Each of these two examples found underlying issues with proposals of empirical support. Importantly, the identified issues were not identical. With respect to D'Aloisio-Montilla's deployment of aphantasics' performance on retro-cue tasks to argue in favor of overflow, the issue concerned the interpretation of the empirical paradigms. With respect to Lau and Rosenthal's deployment of the cases of rare Charles Bonnet syndrome to argue in favor of higher-order misrepresentation, the issue concerned the way they structured the inference from empirical data to theoretical claims. Together, these two examples go some way to demonstrate the heterogeneity of issues that may obtain in the application of empirical evidence to theoretical argument. Furthermore, these examples demonstrate the importance of assessing cold cases, since each case resulted in the discovery of an issue that meant we should not count the empirical support in question when assessing and comparing theories.

On a final note, one should not be blind to other positive side-effects of allocating attention to cold cases. While this activity is justified alone by its role in validating evidence, there are also positive upshots in cases where issues are found. Firstly, it prompts proponents of the evidence to reconsider evidence and reconstrue their argument, which constitutes progress. Secondly, when considering collections of cases where issues have been identified it may be possible to extract patterns that may inform us about the viability of certain ways of constructing arguments (e.g., arguments of type X often involve equivocations). Finally, merely getting more eyes on a case may yield interesting outcomes. This means that diverting attention from the hot topics may reveal untapped and fruitful novel perspectives or areas of research; things that would have gone unnoticed without more people attending to a given cold case.

The first alternative we consider is Joint Determinant Theory (“JDT”). At the core of this approach is the view that consciousness is fundamentally pluralistic, and theories need not worry about being right because their proposed processes may all play a role in conscious experience (He, 2023). According to JDT, different conscious contents such as emotions, thoughts, volitions, and perceptions do not rely on one underlying consciousness-generating mechanism. Instead, integrated conscious experience supposedly is the result of the interaction of a variety of content-specific NCCs, each having their own underlying neural substrate.

Consequently, on this view the search for a universal NCC that underpins all conscious states is misguided. Instead, JDT proposes that conscious experience can be jointly created by distinct neural mechanisms. Because of this, a pluralistic approach to the study of consciousness is recommended by He. A natural consequence of this is that we should study content-specific NCCs separately, and that they may require separate explanatory models. As there is less controversy concerning the studying of local NCCs, this would allow us to build a “stronger empirical foundation” (He, 2023, p. 9). This same fact would also allow researchers to converge around clear explananda, as is not the case currently.

However, the approach does not offer the immediate gratification of theoretical accounts, being rather a tentative framework for the continued study of consciousness. As (He 2023, p. 10) notes: “An implication of this pluralistic view is that there may not be a ‘Eureka’ moment that explains all of consciousness and the work ahead may take longer than some may expect.” However, He is cautiously optimistic about whether studying phenomena separately will eventually yield the neurobiological principles of consciousness and proposes JDT as a better foundation for future endeavors to discover them.

(Mitchell 2002, 2003) proposes the Integrative Pluralism (IP) framework that aims to resolve conflicts between theories using multiple levels of analysis. Mitchell questions whether supposedly competing theories really are about the same thing (see Section 2 on Convergence above). Perhaps thinking they are about the same thing is what makes them appear mutually exclusive. As an alternative, IP suggests that theories provide complementary and compatible explanations.

Mitchell does not see non-commensurability as a problem. Even same-level theories need not conflict—even though they can—due to their different ways of abstracting over the phenomenon. Instead, competing theories can be seen to address different aspects or features of the same phenomenon. Theoretical pluralism is, according to IP, justified and so two different same-level models can provide two different perspectives on the same thing.

By integrating different models, a better, more unified, understanding of the phenomenon is possible according to IP. Consequently, a plurality of theories can give a better explanation of a phenomenon by being integrated. So, on any level of description, there should be allowed a plurality of theories, but when multiple levels are integrated it will turn out that only one “series” of theories will be compatible and thus most plausibly considered tentatively true.

The construct first-approach, argues that the way the theory-driven approach bridges the gap between theory and empiricism, for example by ways of formulating NCCs, is inherently problematic (Fazekas et al., 2024). The proposed alternative is instead to focus on the theoretical constructs themselves.

Extracting constructs is done by “deconstructing” existing theories of consciousness. This involves stripping away excessive theoretical infusion to identify “clean” theoretical constructs. The next step is to map these constructs—including the characteristics of their corresponding lower-level empirical constructs—onto a so-called “construct space” (see Fazekas et al., 2024, Figures 1 and 3). As for the latter, these features are the properties at the neural level which signals the presence of the higher-level construct. Mapping these onto a shared space reveals their relationships, for instance if constructs inhabit overlapping regions. This may allow for empirical investigation, for example, if they are partially defined in light of the same neural activity patterns.

Moreover, by highlighting similarities between theories, the approach is hopeful that, over time, more sophisticated theoretical frameworks might emerge that build on insights gained from this “bottom-up” investigation of consciousness. To elaborate, the space might reveal regions of the construct space that correlate strongly with the presence of consciousness. These regions in turn may be characterized in high-level terms, thereby unlocking a way to guide the study of high-level constructs using the construct space. In any case, building studies on constructs themselves, instead of proceeding from a theory eliminates the risk of disagreements around what needs explaining (see Section 2), since the lower-level empirical constructs supposedly are theory neutral. This approach has also been utilized in a recent study to look at a proposed taxonomy of mental states by their contents (Van den Driessche et al., 2025).

The Multiple Generator Hypothesis (Kirkeby-Hinrup et al., Forthcoming) proposes a distinction between principles and generators. Principles denote ways consciousness can be generated. A Generator instantiates a principle in some physical system. Hypotheses about possible generators can be abstracted from extant hypotheses about the neural correlates of consciousness (Block, 2005; Boly et al., 2017; Fazekas et al., 2024; Ferrante et al., 2023; Fink, 2016; Francken et al., 2022; Koch et al., 2016; Lepauvre and Melloni, 2021; Overgaard et al., 2020; Overgaard and Overgaard, 2010). Similarly, hypotheses about principles can be extracted by analyzing extant theories in terms of sufficiency claims. To illustrate, according to higher-order theory (Rosenthal, 2008, 2012) a higher-order thought is necessary and sufficient for conscious experience (sometimes called “subjective appearance”). Abstracted into the MGH framework, this suggests the following principle: some kinds of metacognitive relation are sufficient for consciousness.

From this, the MGH suggests that—pace what appears to be the tacit assumption in the field—perhaps neither principles nor generators need be mutually exclusive. In other words, in case the hypothesized above principle is actually true (if certain kinds of metacognitive relations in fact are sufficient for consciousness) this does not entail that other principles may not also be true. Similarly, in case we find NCCs (a generator) corresponding to the hypothesis of proponents of Global Workspace theory (Mashour et al., 2020) this does not entail that there are not also other NCCs (generators) in the brain that generate conscious states independently of the workspace generator. Consequently, the MGH suggests that one or more principles and/or generators exist. Importantly, the MGH differs from the JDT account discussed above regarding the interaction and independence of generators. On the JDT, multiple NCCs (that also may vary depending on content) co-operate to produce a single conscious state, whereas on the MGH generators are independent of each other, and can each produce a conscious state.

Finally, the MGH remains neutral on whether the mapping between principles and generators is one-to-one or one-to-many. To elaborate, it is left open whether a single principle can be instantiated in physically dissimilar generators (i.e., whether the principle is multiple realizable). Perhaps it depends on the principle, and some principles require specific physical features of generators, whereas other principles are more flexible with respect to how they can be instantiated.

(Storm et al. 2024) point out that much research within consciousness studies attempts to explain different aspects of consciousness. And many theories seemingly converge—on the neural level—motivating attempts to integrate them in the hope that such a pluralistic methodology can further our understanding of consciousness. That is, the theories need not contradict one another, rather they could, in fact, be compatible and even complement each other concerning the fundamental neural mechanisms and processes. Accordingly, on Storm et al.'s view, consciousness is to be seen as a phenomenon that can include many aspects or subtypes—making the many competing theories in the field, diverging in their focus, strength rather than a weakness.

By stepwise investigating how different theories are similar, and how they differ, on multiple levels of organization, Storm et al. offers a proof of concept concerning the possibility to identify and explore a fruitful integration of theories of consciousness. This methodology is claimed to offer convergence between several theories which offers hopes for integrative future work regarding our understanding of consciousness. However, Storm et al. acknowledge that there still are many gaps remaining—theoretical as well as empirical.