Theories of Consciousness: The Leading Scientific Explanations

Why We Need Theories of Consciousness

Consciousness is a phenomenon that everyone experiences but no one fully understands. We know that brains produce conscious experience, but we do not know how or why. A theory of consciousness should, at minimum, explain why certain physical processes give rise to subjective experience while others do not, predict which systems are conscious, and ideally provide a way to test those predictions.

The stakes are high. Without a good theory, we cannot determine whether animals suffer, whether AI systems might develop inner lives, or what happens to consciousness under anesthesia, in coma states, or in neurological conditions that alter awareness. A complete theory would transform medicine, ethics, law, and our understanding of our place in the natural world.

Integrated Information Theory (IIT)

Integrated Information Theory, developed by neuroscientist Giulio Tononi beginning in 2004, takes a mathematical approach to consciousness. IIT starts from the phenomenology of experience, the properties that consciousness has from the inside, and works backward to identify what physical properties a system must have to generate those experiences.

According to IIT, consciousness is identical to a specific type of information processing called integrated information, quantified by a measure called phi. A system has high phi when it generates information that is simultaneously highly differentiated (each state is distinct from every other state) and highly integrated (the information generated by the whole system exceeds what is generated by its parts independently). The theory predicts that consciousness is a fundamental property of any system with high phi, regardless of what the system is made of.

IIT has several notable implications. It suggests that consciousness comes in degrees, that it is present even in simple systems with non-zero phi, and that the specific quality of a conscious experience is determined by the geometry of the information structure. Controversially, IIT also predicts that standard digital computers have very low phi regardless of what software they run, because their components interact in a feedforward, modular way that does not generate much integrated information.

Global Workspace Theory (GWT)

Global Workspace Theory, originally proposed by Bernard Baars in 1988 and significantly developed by Stanislas Dehaene, approaches consciousness from a cognitive and neural perspective. GWT draws an analogy between consciousness and a theater: specialized unconscious processors (the "audience") compete for access to a limited-capacity global workspace (the "stage"), and the information that makes it onto the stage is broadcast to all processors simultaneously, making it available for reasoning, memory, and action.

In neural terms, GWT identifies consciousness with "ignition," a sudden, widespread activation of prefrontal and parietal brain regions that occurs when sensory information crosses a threshold and is broadcast globally. This ignition pattern has been observed experimentally and distinguishes conscious perception from unconscious processing. Subliminal stimuli can be processed locally but never trigger the global broadcast.

Unlike IIT, GWT is primarily a functional theory. It defines consciousness in terms of what it does (broadcast information globally) rather than what it intrinsically is. This makes GWT more compatible with the possibility of machine consciousness: if an artificial system achieved a genuine global workspace architecture, GWT would predict it could be conscious, regardless of whether it was made of neurons or transistors.

Higher-Order Theories

Higher-order theories of consciousness hold that a mental state becomes conscious when the system forms a representation of that state, a thought about the thought. You are conscious of seeing red not simply because you have a visual representation of red, but because you have a higher-order representation that you are having that visual experience.

Several variants exist. Higher-order thought (HOT) theories, associated with David Rosenthal, require an explicit thought about the first-order state. Higher-order perception (HOP) theories suggest a perceptual-like monitoring of first-order states. Both share the core idea that consciousness arises from a specific relationship between levels of representation within a cognitive system.

Higher-order theories make testable predictions about the neural basis of consciousness. They predict that damage to the brain regions responsible for higher-order representations (likely in prefrontal cortex) should disrupt consciousness even if first-order sensory processing remains intact. Some evidence supports this prediction, though the interpretation is debated.

Predictive Processing and Active Inference

Predictive processing frameworks, developed by researchers including Karl Friston, Andy Clark, and Anil Seth, propose that the brain is fundamentally a prediction machine. Rather than passively receiving and processing sensory input, the brain constantly generates predictions about what it expects to perceive and then compares those predictions with actual sensory data. Consciousness, on this account, may arise from the brain ongoing effort to model its own states and predict its own activity.

Anil Seth has extended this framework with his "beast machine" theory, arguing that consciousness is rooted in the brain predictions about the body own internal states (interoception). Your experience of being alive, of being a self that exists in a body, is itself a prediction generated by the brain. This grounds consciousness in embodiment and suggests that disembodied AI systems, which do not maintain an internal physiological state, may lack a key ingredient for consciousness.

Attention Schema Theory

Michael Graziano Attention Schema Theory (AST) offers yet another perspective. AST proposes that consciousness is the brain simplified model of its own attention processes. Just as the brain constructs a body schema (an internal model of the body position and state), it constructs an attention schema (a simplified model of how attention is being deployed). This attention schema, Graziano argues, is what we experience as consciousness.

AST has interesting implications for machine consciousness because it suggests that consciousness is essentially a modeling trick, an internal representation that could, in principle, be built into any sufficiently sophisticated information-processing system. If consciousness is just a certain kind of self-model, then building a machine with the right kind of self-model might produce machine consciousness.

Comparing the Theories

These theories are not all mutually exclusive, and some researchers attempt to combine elements from multiple frameworks. However, they do make different predictions about several key issues.

On the question of machine consciousness, IIT is the most skeptical (predicting that digital computers have low phi), while GWT and AST are more open to the possibility. On the question of animal consciousness, IIT is the most generous (predicting consciousness even in simple systems), while higher-order theories are more restrictive (requiring metacognitive abilities that simpler animals may lack).

On the question of what explains the subjective quality of consciousness (qualia), IIT offers the most developed account (linking qualia to the geometry of integrated information), while GWT focuses more on the access and functional aspects of consciousness without explaining why there is a subjective feel at all.

The Templeton Foundation adversarial collaboration program has funded direct experimental comparisons between IIT and GWT, with results published in 2023 and subsequent rounds of testing ongoing. These experiments are helping to narrow the field, though no single theory has emerged as the clear winner.

Quantum Theories of Consciousness

A more speculative class of theories proposes that quantum mechanical processes play a role in consciousness. The most well-known of these is Orchestrated Objective Reduction (Orch OR), proposed by physicist Roger Penrose and anesthesiologist Stuart Hameroff. Orch OR suggests that consciousness arises from quantum computations occurring in microtubules within neurons, and that these quantum processes give rise to conscious experience through a specific type of quantum state reduction.

Quantum consciousness theories are controversial within the mainstream consciousness research community. The main objection is that the brain operates at temperatures where quantum coherence, the preservation of delicate quantum states, would be expected to break down almost instantly. However, research in quantum biology has shown that some biological systems (like photosynthetic complexes) do maintain quantum coherence at biological temperatures, which has somewhat softened this objection.

If quantum processes are indeed essential to consciousness, this would have significant implications for machine consciousness. Classical digital computers do not perform quantum computations, so they could never be conscious under a quantum theory. Quantum computers, however, might be candidates, though simply performing quantum computations would not automatically produce consciousness any more than performing classical computations does.

The Road Ahead

The field of consciousness science is at a pivotal moment. For the first time, theories are being tested empirically rather than debated purely in philosophical terms. Advances in neuroscience provide increasingly precise data about the neural correlates of consciousness. Advances in AI provide new systems against which theories can be tested. And the growing urgency of the AI consciousness question is driving funding and attention toward a field that was, until recently, considered too speculative for mainstream science.

The coming decades will likely see significant progress in narrowing the field of viable theories. This progress will not only deepen our understanding of the most intimate aspect of our existence, but will also provide the tools we need to navigate the ethical and practical challenges of increasingly sophisticated AI systems.