Abstract

Understanding the biophysical basis of consciousness remains a substantial challenge for 21st-century science. This endeavor is becoming even more pressing in light of accelerating progress in artificial intelligence and other technologies. In this article, we provide an overview of recent developments in the scientific study of consciousness and consider possible futures for the field. We highlight how several novel approaches may facilitate new breakthroughs, including increasing attention to theory development, adversarial collaborations, greater focus on the phenomenal character of conscious experiences, and the development and use of new methodologies and ecological experimental designs. Our emphasis is forward-looking: we explore what “success” in consciousness science may look like, with a focus on clinical, ethical, societal, and scientific implications. We conclude that progress in understanding consciousness will reshape how we see ourselves and our relationship to both artificial intelligence and the natural world, usher in new realms of intervention for modern medicine, and inform discussions around both nonhuman animal welfare and ethical concerns surrounding the beginning and end of human life.

Key points

• Understanding consciousness is one of the most substantial challenges of 21st-century science and is urgent due to advances in artificial intelligence (AI) and other technologies.
• Consciousness research is gradually transitioning from empirical identification of neural correlates of consciousness to encompass a variety of theories amenable to empirical testing.
• Future breakthroughs are likely to result from the following: increasing attention to the development of testable theories; adversarial and interdisciplinary collaborations; large-scale, multi-laboratory studies (alongside continued within-lab effort); new research methods (including computational neurophenomenology, novel ways to track the content of perception, and causal interventions); and naturalistic experimental designs (potentially using technologies such as extended reality or wearable brain imaging).
• Consciousness research may benefit from a stronger focus on the phenomenological, experiential aspects of conscious experiences.
• “Solving consciousness”—even partially—will have profound implications across science, medicine, animal welfare, law, and technology development, reshaping how we see ourselves and our relationships to both AI and the natural world.
• A key development would be a test for consciousness, allowing a determination or informed judgment about which systems/organisms—such as infants, patients, fetuses, animals, organoids, xenobots, and AI—are conscious.

Introduction

Understanding consciousness is one of the greatest scientific challenges of the 21st century, and potentially one of the most impactful for society. This challenge reflects many factors, including (i) the many philosophical puzzles involved in characterizing how conscious experiences relate to physical processes in brains and bodies; (ii) the empirical challenge of obtaining objective, reliable, and complete data about phenomena that appear to be intrinsically subjective and private; (iii) the conceptual/theoretical challenge of developing a theory of consciousness that is sufficiently precise and not only accounts for empirical data and clinical cases but is also sufficiently comprehensive to account for all functional and phenomenological properties of consciousness; and (iv) the epistemological and methodological challenges of developing valid tests for consciousness that can determine if a given organism/system is conscious. The potential impact of understanding consciousness stems from the many interlinked implications this can have for science, technology, medicine, law, and other critical aspects of society. Existentially, a complete scientific account of consciousness is likely to profoundly change our understanding of the position of humanity in the universe.

Accordingly, consciousness has become an object of intense scrutiny from different disciplines. While the connection between mind and body is an ancient philosophical conundrum, in recent decades, the metaphysical issues have been accompanied by a set of empirical questions, with neuroscience and psychology attempting to discover and explain the connections between conscious experiences and neural activity. Yet, strikingly, the core problem had already been formulated in scientific terms at the turn of the 20th century: certain articles from that period read almost as though they had been written today. For instance, in 1902, Minot wrote a Science article titled “The problem of consciousness in its biological aspects” in which he “[…] hopes to convince you that the time has come to take up consciousness as a strictly biological problem …” (1).

Eighty-eight years later, Crick and Koch called for renewed inquiry into “the neural correlates of consciousness” (2, 3), prompted in part by the increasing availability of novel brain imaging methods that could link the biological activity of the brain with subjective experience. This empirical program continues apace, together with theory development and ever deeper interactions with philosophy. But today, there is also a sense that the field has reached an uneasy stasis. For example, a recent review (4) taking a highly inclusive approach identified over 200 distinct approaches to explaining consciousness, exhibiting a breathtaking diversity in metaphysical assumptions and explanatory strategies. In such a landscape, there is a danger that researchers talk past each other rather than to each other. Empirically, Yaron et al. (5) showed that most extant experimental research on theories of consciousness is geared toward supporting them rather than attempting to falsify or compare them, reflecting a confirmatory posture that hinders progress. This manifested both in the low percentage of experiments that ended up challenging theories, as opposed to supporting them (15%), and in the low percentage of experiments that were designed a priori to test theoretical predictions (35%, with only 7% testing more than one theory in the same experiment).

Beyond the genuine and highly complex scientific challenges that the study of consciousness must address, sociological factors may also contribute to the current sense of entrenchment: nobody likes to change their mind (6)! Emerging collaborative frameworks—especially adversarial collaborations—may help alleviate this concern, at least to some extent. But there are also further factors: the possibility that consciousness research is not sufficiently addressing why it feels like anything at all to be conscious and the role that conscious phenomenology plays in our mental, and indeed biological lives (7–9).

This paper is structured in a forward-looking manner, moving from the past, through the present, and on to the future. First, we clarify terms and make some essential conceptual distinctions. Then, we briefly review what has been achieved so far in elucidating the neural and theoretical basis of consciousness. Next, we consider the future of our field, outlining some promising directions, approaches, methods, and applications, and advocating for a renewed focus on the phenomenological/experiential aspects of consciousness. Finally, we imagine a time in which we have “solved consciousness” and explore some of the key consequences of such an understanding for science and society.

Three distinctions about consciousness

Consciousness is a broad construct—a “mongrel” concept (10)—used by different people to mean different things. In this paper, we stress three distinctions.

The first distinction is between the notion of the level of consciousness and the notion of the contents of consciousness. In the first sense, consciousness is a property associated with an entire organism (a creature) or system: one is conscious (for example, when in a normal state of wakefulness) or not (for example, when in deep dreamless sleep or a coma). There is an ongoing vibrant debate about whether one should think of levels of consciousness as degrees of consciousness or whether they are best characterized in terms of an array of dimensions (11) or as “global states” (12). In the second sense, consciousness is always consciousness of something: our subjective experience is always “contentful”—it is always about something, a property philosophers call intentionality (3, 13). Here, again, there is some debate over the terms, for example, whether there can be fully contentless global states of consciousness (14) and whether consciousness levels (or global states) and contents are fully separable (11, 15).

The second distinction is between perceptual awareness and self-awareness (note that in this article, we use the terms consciousness and awareness interchangeably). Perceptual awareness simply refers to the fact that when we are perceptually aware, we have a qualitative experience of the external world and of our bodies within it (though of course, some perceptual experiences can be entirely fictive, such as when dreaming, vividly imagining, or hallucinating). Importantly, mere sensitivity to sensory information is not sufficient to be considered as perceptual awareness: the carnivorous plant Dionaea muscipula and the camera on your phone are both sensitive to their environment, but we have little reason to think that either has perceptual experiences. Thus, mere sensitivity is not sufficient for perceptual awareness, as it does not necessarily feel like something to be sensitive. This experiential character is precisely what makes the corresponding sensation a conscious sensation (16).

We take self-awareness, on the other hand, to mean experiences of “being a self.” These experiences can be of many different kinds, from low-level experiences of mood and emotion (17) to high-level experiences of being the subject of our experiences, which might be supported by some inner (metacognitive) model of ourselves and our mental states (18–20). This kind of high-level reflective self-awareness is associated with the “I” and with a sense of personal identity over time (21).

The distinction between self-awareness and perceptual awareness is not sharp. Some aspects of the experience of “being a self” seem not to involve reflective self-awareness, such as experiences of emotion, mood, body ownership, agency, and of having a first-person perspective (22, 23). Some of these aspects may arguably have perceptual features. For example, emotional experience may depend on interoception (24–26). In addition, some perspectives, such as the higher-order theories described below, suggest that a form of metacognition might play a constitutive role in all instances of perceptual awareness, not only in self-awareness (18, 27, 28).

Human beings normally possess both perceptual awareness and self-awareness, but this is probably not true at all times or for all species. In humans, reflective self-awareness may be absent in specific conscious states, such as absorption or flow (29), or in states of minimal phenomenal experience (14). Other species may lack this reflective capability altogether. For example, few will doubt that dogs have perceptual experiences as well as various non-reflective self-related experiences—though this can be contested as we currently lack a way to directly test for consciousness in other species [see (30–32) for recent attempts to tackle this problem]. Nevertheless, there is no convincing evidence that dogs have reflective self-awareness in the sense defined above. Putting these debates aside, consciousness research has thus far largely focused, with exceptions (26, 33, 34), on trying to explain perceptual awareness as a first, albeit notoriously difficult, step toward understanding other aspects of consciousness. This emphasis most likely stems from the fact that perceptual awareness is generally easier to manipulate in experiments.

The third distinction contrasts the phenomenological (i.e., experiential) aspects of consciousness with its functions. This discussion has been largely shaped by Block’s (35) influential, yet controversial (36, 37), distinction between phenomenal consciousness and access consciousness—informally, what consciousness feels like and what it does. Access consciousness is associated with the various functions that consciousness enables, such as global availability, verbal report, reasoning, and executive control. Phenomenal consciousness, on the other hand, refers to the felt qualities of conscious mental states: the complex mixture of bitterness and sweetness of a Negroni cocktail, the distinctive hue of International Klein Blue, the anxiety prompted by one’s to-do list. All such conscious mental states have phenomenal character (using the philosophical term, often referred to as “qualia”): there is something it is like for us to be in each of these states. By contrast, there is nothing it was like for the neural network Alpha Go (38) to win against the South Korean world Go champion Lee Sedol (it was Sir Demis Hassabis and the DeepMind team who drank the champagne instead). Despite its seductive use of language, we think there is also nothing it is like for GPT-5 to engage in a conversation (39, 40).

Just as there has been greater emphasis within consciousness science on studying perceptual awareness compared with self-awareness, there has also been a greater emphasis on studying the functional rather than the phenomenological aspects of consciousness. This, again, may be due to the relative ease with which functional properties related to conscious access can be studied empirically compared with phenomenological aspects (41–43). With respect to the neural underpinnings of consciousness, we have been more focused on finding the mechanisms that differentiate between a consciously processed and an unconsciously processed stimulus than on explaining the difference between two conscious experiences, again with exceptions (44–48). Additionally, with respect to the functions of consciousness, we have been more oriented toward documenting what we can do without awareness rather than because of it (49–52). The potential for complex behavior in the absence of awareness has been further emphasized by the rapid advances in artificial intelligence (AI), where complicated functions can be executed without any accompanying phenomenology, at least as far as we can tell.

What have we achieved so far?

Following this clarification of terms, we briefly review where things stand today in consciousness research. Given the enormous challenge that explaining consciousness represents, it is easy to underestimate the significant progress that has already been made. This progress has been particularly visible over the last 30 or so years, but in fact it extends much further back, with highlights including seminal work on split-brain patients, neurological patients, work with brain stimulation, research on nonhuman primates, and much more (53–55).

Some basic facts are now well established. In humans and other mammals, the thalamocortical system is strongly involved in consciousness, whereas the cerebellum (despite having many more neurons) is not. Different regions of the cortex are associated with different aspects of conscious content, whether these are distinct perceptual modalities (56), experiences of volition or agency (34), emotions (57), or other aspects of the sense of “self” (58). Researchers have identified a myriad of candidate signatures of consciousness in humans, focusing on global neural patterns [e.g., neuronal complexity (59), non-linear cortical ignitions (60), stability of neural activity patterns (61)], specific electrophysiological markers of consciousness [e.g., the perceptual awareness negativity (62)], alpha suppression (63), late gamma bursts (64), and on relevant brain areas such as the “posterior hot zone” (65) or frontoparietal areas (66) as well as subcortical structures and brainstem arousal systems that may contribute to and modulate awareness (67–70). For some of these regions, notably brainstem arousal systems, there is debate about whether they represent necessary enabling conditions for consciousness and/or whether they contribute to the material basis of consciousness (67, 69).

At the same time, some previously popular hypotheses have now been empirically excluded. For example, the idea that consciousness is uniquely associated with 40 Hz (gamma band) oscillations has fallen out of favor based on substantial evidence (71, 72). In parallel, there has been a growing recognition that various confounds need to be carefully ruled out in order to interpret these findings, including those related to the enabling conditions for conscious experience, post-perceptual processes such as memory and report, and the concern that consciousness is often (but not always) correlated with greater signal strength and performance capacity (73–76). In this regard, phenomena such as blindsight, in which consciousness can be partly dissociated from performance capacity, are particularly intriguing [(77–79); but see (80, 81), for critiques].

Complementing these empirical findings, many theories of consciousness have been developed over recent years. These vary greatly in their aims and scope, in the degree of traction they have gained in the community, and in their level of empirical support (5, 12, 82–84). A selection of these theories provides a useful lens through which to focus attention on the progress made so far in the scientific study of consciousness.

Global workspace theory

One prominent theory, named “global workspace theory” (GWT), originated from “blackboard” architectures in computer science. Such architectures contain many specialized processing units that share and receive information from a common centralized resource—the “workspace.” The first version of GWT (85) was a cognitive theory that assumed that consciousness depends on global availability: just like blackboard architectures, the cognitive system consists of a set of specialized modules capable of processing their inputs automatically and unconsciously, but they are all connected to a global workspace that can broadcast information throughout the entire system and make its contents available to a wide range of specialized cognitive processes such as attention, evaluation, memory, and verbal report (86). The core claim of GWT is thus that it is the wide accessibility and broadcast of information within the workspace that constitutes conscious (as opposed to unconscious) contents. Since the 1990s, GWT has developed into a neural theory (referred to as global neuronal workspace theory) in which neural signals that exceed a threshold cause “ignition” of recurrent interactions within a global workspace distributed across multiple cortical regions—this being the process of “broadcast” (64, 87). Importantly, GWT is what is called a first-order theory: what makes a mental state conscious depends on properties of that mental state (and its neural underpinnings) only and not on some other process relating to that mental state in some way. Thus, in contrast with the assumptions of higher-order theories (HOTs, introduced below), GWT does not postulate that consciousness depends on higher-order representation or indexing of some kind.

GWT is primarily a theory of conscious access (88), focused on how mental states gain access to consciousness and how they accrue functional utility as a result. This is characterized largely in terms of supporting flexible, content-dependent behavior, including the ability to deliver subjective verbal reports [but see (89) for a discussion of the phenomenal aspect of consciousness and how the theory explains it, and see Dehaene’s section in (90)]. GWT’s clear neurophysiological predictions (centering on nonlinear “ignition” and on the involvement of frontoparietal regions) has led to a wealth of supportive experimental evidence (64). For example, divergences of activity ~250–300 ms post-stimulus have been associated with ignition (91), and measures of long-distance information sharing among cortical regions have been associated with broadcast (92). However, a major challenge for GWT lies in specifying what exactly counts as a “global workspace” (12): does it depend on the nature of the “consuming” systems, the type of broadcast, and/or on other factors?

Higher-order theories

A second prominent theory of consciousness is Rosenthal’s (93) higher-order thought theory, which proposes that a mental state is a conscious mental state when one has a “higher-order” thought that one is in that mental state. This core idea has now been elaborated on in different ways, resulting in a family of higher-order theories (HOTs). Unlike first-order theories, higher-order theories all claim that mental states are conscious when they are the target of a “higher-order” mental state of a specific kind (18, 93–95). The nature of the relationship between first-order and higher-order states varies among HOTs, but they all share the basic notion that for a first-order mental state X to be conscious, there must be a higher-order state X that in some way monitors or meta-represents X. Take the experience of consciously seeing a red chair. According to HOTs, the first-order representation (perhaps instantiated as a pattern of neural activity in the visual cortex) of red is not by itself sufficient to produce a conscious experience. Instead, there need to be additional “higher-order” states that point to or (meta)represent the first-order representation for it to be experienced as red. Crucially, such higher-order states need not be conscious themselves (i.e., we do not need to be aware of a mental state with content like, “I am now seeing red”). Rather, it is their very existence that makes the target content conscious. HOTs capture the intuitively plausible notion that a mental state is a conscious mental state as soon as I am aware of being in that mental state. This offers an equally intuitive distinction between conscious and unconscious mental states: I am conscious of some situation when I know about that situation; otherwise, I am unconscious of that situation.

Many HOTs locate the neural basis of the relevant meta-representations in anterior regions of the human brain, with an emphasis on the prefrontal cortex (96). Future “neural HOTs” will likely develop richer mappings between brain states and the theoretical distinction between first- and higher-order states (97). These theories are therefore supported by evidence implicating these regions in consciousness and undermined by evidence that anterior regions are not necessary for consciousness. As such, they have motivated studies investigating the neural correlates of consciousness (NCCs) with this question in mind (98). Of particular note are experiments that attempt to control for how well participants perform at a perceptual task: such studies (including in “blindsight” participants) have shown that when conditions are matched for performance, differences between conscious and unconscious perception are found in anterior cortical regions (75, 99) and interference with prefrontal function using transcranial magnetic stimulation (TMS) or multivariate neurofeedback affects subjective aspects of perception (such as confidence) without changing performance (100, 101). Studies associating perceptual metacognitive abilities with anterior prefrontal function also provide intriguing supportive evidence, albeit less direct (e.g., 102, 103). Additional support can be drawn from demonstrations of decoding of the content of consciousness from frontal areas (104).

However, HOTs currently do not fully specify the actual neural mechanism(s) mediating the implementation of first- versus higher-order states: how exactly does one brain state “point” at another, and what motivates the choice of which first-order state to point at or re-represent? Another challenge is that they focus on the contents of consciousness and provide less explanation for the level of consciousness. These under-specifications reflect the relatively limited empirical formulation of HOTs—despite their considerable philosophical backbone (105)—as compared with other theories (5). These aspects of the theory are currently being developed (45), and an ongoing adversarial collaboration (ETHoS1) is specifically aimed at comparing the empirical predictions of four HOT variants.

Integrated information theory

A very different perspective is provided by “integrated information theory” (IIT), developed by Giulio Tononi and colleagues since the 1990s (44, 106, 107). Rather than asking what in the brain gives rise to consciousness, IIT identifies features of conscious experience (described in five axioms) that it assumes are essential and then asks what properties a physical substrate of consciousness must have for these features to be present. A striking claim of IIT is that any physical substrate that possesses these properties will exhibit some level of consciousness (108). The two most illustrative essential features, or axioms, are (unsurprisingly) information and integration. According to IIT, every conscious experience is necessarily both informative (in virtue of ruling out many alternative experiences; i.e., every experience is the way it is, and not some other way) and integrated (every experience is a unified scene). IIT introduces a mathematical measure, phi (Φ), which, broadly speaking, measures the extent to which a physical system entails irreducible maxima of integrated information and thereby, according to the theory, provides a full measure of consciousness. Different versions of IIT introduce different varieties of Φ, with the latest being IIT 4.0 (107), but all associate consciousness with the underlying “cause—effect structure” of a physical system and not just with the dynamics (e.g., neural activity) that the physical system supports. IIT is arguably the most ambitious theory we discuss because it addresses both the level and content of consciousness, proposes a sufficient basis for consciousness, and explicitly addresses phenomenological aspects of consciousness, such as spatiality (109) and temporality (110).

IIT has been criticized on the grounds that measurement of Φ is challenging or infeasible for anything other than very simple systems. Other “weak” versions of IIT have been proposed in which Φ is easier to measure, but this comes at the cost of abandoning claims of an identity relationship between Φ and consciousness (111). Another line of criticism is that the axioms proposed by full IIT do not satisfy standard philosophical criteria of being self-evidently true (112). Concerns like these have led to robust debate over whether the core claims of IIT are empirically testable and over what should be expected from a scientific theory of consciousness (40, 113, 114).

The most commonly referenced experimental support for IIT comes from evidence examining empirically applicable proxies2 for integrated information (Φ) under different global states of consciousness. In a canonical series of studies (115, 116), Massimini and colleagues have developed a measure of consciousness, called the “perturbation complexity index” (PCI), which quantifies the complexity of the brain’s response to cortical stimulation. Most commonly, the method uses TMS to inject a brief pulse of energy into the cortex, an electroencephalogram to measure the response, and the information-theoretic metric of Lempel–Ziv complexity (which quantifies the diversity of patterns within a signal) to quantify the complexity of the response. High PCI values arguably correspond to high levels of integration and information in the underlying dynamics. However, it is important to emphasize that the PCI, while inspired by and based on IIT, is not a measure or approximation of Φ, and differences in PCI across conscious levels may also be affected by differences in how unconscious processes operate at these levels. The PCI results, while fascinating, cannot be taken to directly support the distinctive aspects of IIT that rely on the definition of Φ, and are also compatible with or supportive of other theories, notably GWT. Nevertheless, the PCI method has shown exciting promise in important practical scenarios, such as detecting residual consciousness in unresponsive patients following severe brain injury (59).

In terms of neural correlates, IIT theorists claim that brain activity sufficient for conscious perception is localized to posterior regions (e.g., the posterior cortical “hot zone”). This claim is based on the argument that neural connectivity in these regions is well suited to generating high levels of (irreducible) integrated information, rather than the anterior regions favored by HOTs and GWT (117).

Predictive (and recurrent) processing theory

The final theory we mention here is not really (or at least not primarily) a theory of consciousness but rather a general theory of brain function—of perception, cognition, and action—from which more specific connections between brain processes and aspects of consciousness can be derived and tested (118). According to “predictive processing” (PP), the brain continually minimizes sensory “prediction error” signals, either by updating its predictions about the causes of sensory signals or by performing actions to bring about predicted or desired sensory inputs (the latter process being termed “active inference”) (119–121). This ongoing process of prediction error minimization provides a mechanism by which the view of perception as a process of Bayesian inference, or “best-guessing”, (122) and as a means of predictive regulation of physiological variables can be implemented (123, 124). In its most ambitious and all-encompassing version, the “free energy principle,” the mechanism of prediction error minimization, arises out of fundamental constraints regarding control and regulation that apply to all physical systems that maintain their organization over time in the face of external perturbations (125, 126).

Several distinct theories of consciousness fall under the umbrella of PP (e.g., 23, 127, 128). These typically share the claim that the contents of conscious experiences arise from (top-down) predictions rather than from a “read out” of (bottom-up) sensory signals. Informally, the contents of perceptual experience are given by the brain’s “best guess” of the causes of its sensorium or, even more informally, as a “controlled hallucination” in which the brain’s predictions are reined in by sensory signals arising from the world and the body (23).

One particular influential theory under the PP umbrella deserves mention: recurrent processing theory (RPT), also known as “local recurrency” or “re-entry” theory, associates consciousness with top-down (recurrent) signaling in the brain but does not appeal directly to the Bayesian aspects of PP (129, 130). Instead, RPT uses neurophysiological evidence to motivate the view that local recurrence (e.g., in visual cortex) is sufficient for phenomenal experience to occur and that feedforward (bottom-up) activity is always insufficient for conscious perception, no matter how “deep” into the brain this activity reaches (36). RPT’s focus on local recurrence is usually used to contrast the theory with other theories that involve widespread broadcast (GWT) or higher-order processes (HOT) (90), but as theories gain precision, it could be that aspects of RPT also surface in other theories (83). For example, the “ignition” process central to GWT might involve local recurrence. Nonetheless, a key difference between RPT and these other theories remains that RPT allows that phenomenal experience could be present without cognitive access (36).

The core commitments of PP do not directly specify a necessary or sufficient basis for consciousness to happen, nor do they specify how to distinguish conscious from unconscious processing. RPT is an exception here, proposing sufficient conditions, given the right enabling background conditions. Instead, the value of PP for theories of consciousness may largely reside in providing resources for developing and testing systematic or explanatory correlations between brain processes and properties of conscious experience, both functional and experiential (118). PP accounts tend to focus on conscious content rather than conscious level (e.g., 131, 132); they speak to both phenomenological (in terms of the nature of top-down predictions) and functional aspects of consciousness and address aspects of selfhood and embodiment more directly than other theories discussed here (e.g., 40, 133). Notably, variants of the theories discussed above can be expressed within the framework of PP, so there can be ‘PP versions’ of, for example, GWT and HOT (95, 134).

Whether PP succeeds as a theory in consciousness science will depend both on evidence that prediction error minimization is indeed a core brain operation and on its ability to draw explanatorily and predictively powerful links between elements of predictive processing and aspects of conscious experience. While there is substantial evidence linking top-down signaling to conscious perception (135, 136), evidence for explicit sensory prediction error signals playing the roles proposed by PP remains mixed (137), at least when compared to the well-studied dopaminergic reward prediction error signal (138). Further, while abundant evidence shows that participant expectations can shape conscious perception (139), much remains to be done to causally connect the computational entities of PP with specific forms of consciousness. For some, this is a shortcoming of the theory: it might be too general and accordingly not informative enough to explain consciousness. Conversely, more specific formulations of the top-down principle, such as RPT, have been criticized for being too narrow, for example, focusing on visual processing only and failing to explain how this relates to other modalities and how conscious information is integrated across modalities.

This short tour of several of the many theories of consciousness [for a recent comprehensive survey, see (4)] highlights that there is not only a lack of agreement about the answers in consciousness science but also a lack of consensus about approaches and relevant questions. This does not mean there has been no progress. On the contrary, the last two decades have witnessed an enlightening move away from a simple search for NCCs in a comparatively theory-free and therefore explanatorily impoverished way to a rich landscape of different theories with varying degrees of experimental support. The Consciousness Theories Studies (ConTraSt) (https://contrastdb.tau.ac.il) database study has recently quantified the differences in the extent of research relating to the four theories of consciousness described above, and demonstrated how research results tend to align with the predictions of the supported theory [see Figure 1 and (5)]. There are also some striking commonalities as well as differences among theories. For example, recurrent processing emerges as a key principle in GWT, IIT, PP, and some versions of HOT as well as other theories. Such unifying principles might point toward a “minimal unifying model” of consciousness, at least in biological systems (140).

Figure 1. Results of the Consciousness Theories Studies (ConTraSt) database study (5). Updated results of the ConTraSt database, now including 511 experiments published until mid-2025, which interpreted their findings in light of four prominent theories of consciousness: global workspace theory (GWT), higher-order theories (HOT), integrated information theory (IIT), and recurrent processing theory (RPT). Notably, there are currently no papers in the database for predictive processing theory (PPT). This is mainly because the database is based on the work done by Yaron et al. (5), where PPT was not included, and new uploads referring to this theory have not been made yet. (A) Distribution of experiments across theories. Green sections in the bars represent the number of experiments interpreted as supporting the theory; purple sections represent experiments interpreted as challenging it. (B) Effects over time: a cumulative distribution of experiments supporting the theories. (C) Functional magnetic resonance imaging (fMRI) findings for experiments supporting each of the theories. The same conventions used by Yaron et al. (5) are used here: for each activation, the color intensity indicates the relative frequency of experiments reporting activations in that brain area. While overlaying all findings demonstrates that most of the cortex has been implicated in consciousness, the breakdown by theory presents four different pictures, each aligning with the predictions of the supported theory. This further illustrates the confirmatory posture that most authors in the field have—intentionally or not—espoused.

Where are we going?

Thus far, we have surveyed some of the current main directions in the study of consciousness. As our overview makes clear, the sheer diversity of approaches and theories that characterize the field raises questions about how it can best make progress. In this section, we consider the most promising directions to follow in this ongoing quest, which some consider potentially endless (141). What will be the state of our field 50 years from now? Will our successors look back with satisfaction at the progress made toward “solving consciousness,” or will they feel that the research has been going in circles, not getting any closer?

Considering that prophecy is given to fools, we will refrain from making a prediction here. But we note that the history of science abounds with unfulfilled scientific promises to solve one mystery or another, like producing cold fusion (142), curing cancer (143), achieving room temperature superconductivity, or indeed fully simulating the human brain (144). On the other hand, science often outperforms human predictions: 50 years ago, it probably seemed unthinkable that a computer would ever beat a human chess champion (145), converse fluently (146), or be able to create art (147). Bearing this in mind, what will the future of consciousness science look like? In the following sections, we sketch out nascent trends that will most likely shape the field in the coming decade: a shift toward theory-driven research, the necessity of collaborative and interdisciplinary work, the adoption of new methods, and an emphasis on applications. We hope that developments like these may help the field move beyond the current “uneasy stasis” we mentioned earlier.

From correlates to testable theories

The first major shift is a transition from “searching for the NCCs” to an increased focus on theory-driven empirical research (12, 82–84). While the former has been largely dominated by a data-driven, bottom-up approach consisting, for instance, of manipulating consciousness in hopes of identifying neural contrasts between consciously perceived and non-consciously perceived stimuli, the latter is driven by empirical predictions derived from specific theories of consciousness. Generally speaking, the agenda seems to be gradually transitioning toward providing explanations that go beyond descriptions [see (9), for a critical review]. This seems to be a step in the right direction, though more work is needed to potentially turn this simple step into a major leap.

First, theories must be thoroughly scrutinized to identify both their core constructs (148) plus testable predictions that have high explanatory power. Most, if not all, theories include claims and concepts that are somewhat fuzzy—often almost metaphorical—and these are then translated into neural terms in ways that are sometimes too simplistic, for example by debating whether consciousness is subserved by the front or the back of the brain (149, 150). Further elucidation and formalization are needed to make it possible for the theories to be fully tested. Addressing such issues would open up another research strategy, focused on the “search for computational correlates of consciousness” (151)—that is, identifying which computational differences best characterize the distinction between conscious and unconscious information processing. This in turn requires further precision. For example, what does it mean for information to be globally broadcast (152), and how do the receiving neurons understand the message? Similarly, how exactly does a higher-order brain state point at first-order brain states (96)? Or how is the unfolded cause–effect structure of a certain conscious state (107) physically implemented in neural terms? Only when predictions are fully fleshed out will we be able to assess their explanatory power using clear measures (153, 154).

Second, the explananda (explanatory targets) of the theories should be better defined, especially given claims that they might not be explaining the same things and the fact that they are supported by different types of empirical data, at least to some degree (5, 82). We believe that a greater focus on the phenomenological, experiential aspects of consciousness—for example, by studying quality spaces (45, 48, 155) or by pursuing computational phenomenology (14, 156–158)—is likely to yield substantial dividends here, by making the explananda more precise and thereby sharpening the distinctions among theories.

Third, as Seth and Bayne (12) argue, current theories should become not only more precise (for example, by using computational modeling) and more testable (for example, by developing new measures) but also more comprehensive. That is, theories should progressively be able to explain more distinct aspects of consciousness, and a good theory should explain as many aspects of consciousness as possible (82). An alternative and potentially complementary strategy is to focus on explaining the minimal, universally present features of consciousness (140)—perhaps reflecting a kind of “minimal phenomenal experience” (159).

Another shift in emphasis encouraged by theory-driven predictions is a focus on causal as well as on correlational evidence. Causal predictions generally provide stricter tests of a theory and hence more informative evidence. An example of a theory-based causal prediction can be found in INTREPID (https://arc-intrepid.com/about/), one of the current crop of adversarial collaborations. There, the team is using optogenetics in mice to contrast the effects of merely inactive versus optogenetically inactivated neurons in the visual cortex on visual perception, testing a prediction derived from IIT. Outside the context of theory testing, some have used optogenetics to examine the dependence of conscious perception of cortico-cortical and cortico-thalamic connectivity (157).

Finally, to allow us to home in on promising theories and reduce our credence in less useful ones, the field should focus on evaluating these theories through experiments designed a priori to test their predictions. At least some of these experiments should probe multiple theories simultaneously, to create meaningful contrasts between them. This leads us to the next suggested move.

From isolation to collaboration

Until recently, consciousness has mostly been studied by dozens of laboratories around the world, mostly independently. Each scholar has addressed the problem using their own tools, ideas, and theoretical approaches and pursued their research alone or with a small group. Yet, other fields have taught us that big questions often cannot be solved by individuals or small groups and that such questions may be better addressed through collaborative science (e.g., 160–162). Applied to our field, collaborative approaches can be used at multiple levels.

Selecting research questions

Defining key questions that are worth pursuing can be taken up by the community at large (163) or by a joint process involving multiple researchers and scholars. One form of such collaboration that we have already mentioned is adversarial collaboration, championed by Kahneman (6). Here, theoretical opponents work together to design experiments that would test their approaches, pushing each other toward better theoretical and experimental definitions of their claims.

A recent program initiated by the Templeton World Charity Foundation (TWCF) adopted this method in an attempt to “accelerate research on consciousness” by encouraging theory leaders to mutually engage and design experiments likely to arbitrate between competing theories. A series of such adversarial collaborations is now underway, pioneered by the Cogitate Consortium (Figure 2; 117). Time will tell if these collaborations allow us to arbitrate between theories. The first results of the Cogitate Consortium interestingly—and perhaps unsurprisingly—do not fully align with either of the predictions made by the theories in question, namely IIT and GWT. A challenge for this consortium, and likely for future adversarial collaborations, is that the agreed-upon experiments did not directly test the core aspects of either theory—a problem that in turn may follow from each theory making different assumptions and having distinct explananda. Yet, the experiments provided meaningful tests of the neuroscientific predictions of these theories, and the failure to confirm some of these predictions will hopefully lead to self-correction by the theories and to shifting the credence assigned by the community to each theory (164).

Figure 2. An illustration of the ongoing adversarial collaborations funded by the Templeton World Charity Foundation. Such collaborations invite theory leaders to jointly conceive experiments aimed at falsifying the core tenets of different theories. The experiment designs and theoretical predictions to be tested are preregistered and the experiments are performed and replicated by independent teams. In total, eight theories (see text) will be tested. To date, five adversarial collaborations have been launched. Cogitate (initiated in 2019) tested predictions of information integration theory (IIT) and global neuronal workspace theory (GWT). Data collection is complete and the first experimental results have been published (117). A second adversarial collaboration (2021) is comparing IIT and GWT in nonhuman animals. Thirdly, INTREPID (2022) is testing IIT against predictive processing theory (PPT) and neurorepresentationalism. A fourth collaboration (2020) contrasts higher-order theories (HOTs) of consciousness—specifically higher-order representation of a representation (HOROR) (94)—with some first-order theories, in particular recurrent processing theor