Consciousness: The Hard Problem

In 1995, Australian philosopher David Chalmers presented a distinction at the “Toward a Science of Consciousness” conference in Tucson, Arizona, that reshaped the entire field of philosophy of mind. He divided the problems of consciousness into two categories: the “easy” problems and the “hard” problem. The easy problems, despite their name, are enormously complex: explaining how the brain discriminates between stimuli, integrates information from different modalities, controls behavior, and directs attention. They are “easy” because, in principle, they can be explained in terms of computational and neural mechanisms.

The hard problem is of a different nature: why do physical processes in the brain generate subjective experience? Why is there something that "it's like" to see red, feel pain, or taste chocolate? A philosophical zombie, identical to us in every neural and behavioral process but with no inner experience, seems logically conceivable. If it is conceivable, then subjective experience does not necessarily follow from physical processes, and no mechanistic explanation will be sufficient.

The easy problems of consciousness involve explaining the functions and abilities associated with conscious experience. How does the visual system process wavelengths and classify them as "red"? How does attention select one stimulus among many? How does working memory keep information active? These questions are difficult in practice, but methodologically "easy": we can investigate them using neuroimaging, computational models, brain lesions, and electrophysiology.

The hard problem is in a different category. Even if we fully explain how the brain processes the color red (photoreceptors, V1, V4, integration), one question remains: why is this processing accompanied by an EXPERIENCE of red? Why doesn't it happen "in the dark," with no one inside? Science can, in principle, map every neuron involved in perceiving red, but the map does not contain the experience. It's like describing every water molecule without the description being wet.

Daniel Dennett, Patricia and Paul Churchland represent a radical response to the hard problem: it does not exist. Eliminative physicalism argues that consciousness, qualia, and subjective experience, as we intuitively understand them, are mistaken concepts from a "folk psychology" that will be eliminated by advances in neuroscience, just as "phlogiston" was eliminated by chemistry and "élan vital" by molecular biology.

For Dennett, what we call qualia are actually complex functional dispositions. When I say “I see red,” I’m making a verbal report about a brain state, but there is no “inner screen” where red appears to an “inner spectator.” That is the Cartesian theater: an illusion. What exists are distributed processes that generate verbal behavior and stimulus discrimination. When neuroscience completes its work, the “hard problem” will dissolve as a pseudoproblem.

Giulio Tononi, an Italian neuroscientist at the University of Wisconsin, proposed Integrated Information Theory (IIT) as an attempt to formalize consciousness mathematically. The central idea: consciousness is information integrated in an irreducibly unified way. A system is conscious to the extent that it integrates information that cannot be reduced to its parts. Tononi defines a quantity, Phi, that measures a system’s degree of information integration.

IIT makes bold predictions. According to the theory, consciousness exists in degrees: a thermocouple would have a tiny Phi (almost zero), while a human brain would have a huge Phi. A classic digital computer, no matter how powerful, would have a low Phi because its transistors operate in a modular, decomposed way. This means that, according to IIT, a perfect simulation of the brain in silicon would NOT be conscious, because its architecture doesn't integrate information in the same way. A controversial but testable position.

In 1988, Bernard Baars proposed Global Workspace Theory (GWT): consciousness emerges when information processed by specialized modules (vision, hearing, memory, emotion) is "broadcast" to a global workspace accessible to all modules simultaneously. This space acts like a "spotlight" that illuminates selected information, making it available to the whole brain and enabling behavioral flexibility and integration.

Stanislas Dehaene and Jean-Pierre Changeux developed the neuroscientific version of GWT, identifying the global workspace with frontoparietal networks that “broadcast” information via long-distance connections. When a stimulus crosses the threshold for access to the workspace, a neural “ignition” occurs: massive, sustained activation visible on EEG. Subliminal stimuli activate local modules but do not “light up” the workspace: they are processed but do not become conscious. The theory explains the difference between unconscious and conscious processing well, but its critics argue that it describes what consciousness does, not what it is.

Contemporary artificial intelligence brilliantly solves the "easy problems": pattern recognition, language processing, stimulus classification, coherent text generation, and reasoning about complex problems. Language models like GPT and Claude produce responses that, in many contexts, are indistinguishable from human responses. But are they conscious? Is there something it is "like" to be an LLM processing tokens?

Bruzzo points out that AI makes Chalmers's distinction brutally clear: we can create systems that process and generate coherent text, recognize emotions, and compose music, but the question of whether they FEEL anything remains completely open. Without solving the hard problem for humans, we can't even formulate the question for machines. The hard problem remains human. Or perhaps not even human: it remains unsolved for any entity, biological or artificial.