Is anyone home? The science of consciousness

Start with an experience you are having right now without noticing it. As you read, a light is on: it is like something to be you at this instant: the blue of the screen has a felt glow, your thoughts have an inner texture. That most intimate of facts is also the largest blind spot in science. We are getting better and better at describing what the brain does; we still cannot say why any of it is accompanied by an experience rather than going on in the dark. This essay is about that gap, the two scientific theories that try to close it, and the surprisingly sharp question they raise about machines: if you built a perfect functional copy of a mind, would there be anyone inside?

The hard problem: the one that resists

In 1995 the philosopher David Chalmers split the subject in two with a now-famous cut. There are, he said, the easy problems of consciousness, “easy” not because they are simple (they occupy whole laboratories) but because we know what a solution would even look like: they are questions of mechanism. How does the brain tell red from green, integrate information, direct attention, control behaviour, report its own states? These are functions; one day we will wire them up completely, and the matter will be closed.

And then there is the hard problem, which resembles nothing else: why is all that information-processing accompanied by a felt experience at all? Why is the processing of a 700-nanometre wavelength not simply done “in the dark”, executed with nobody there to feel it? Thomas Nagel had posed the same thing differently back in 1974, with a question that became a classic: what is it like to be a bat? You can know everything about its sonar, its neuroanatomy, its circuitry, and still not know what it is like, from the inside, to perceive the world by echo. That “what it is like” (philosophers call it qualia, the subjective grain of experience) is precisely what brain science cannot yet make appear out of matter.

The zombie argument

To make the problem tangible, philosophers invented a creature: the philosophical zombie. Not the cinema undead. It is your atom-for-atom twin: it walks, talks, swears it feels pain, recoils from fire, writes poems about sunsets, identical in every respect, except that on the inside there is no one. No light on. Perfect information-processing, in total darkness. Chalmers’ question is: is such a being even conceivable? If it is, and many find they can imagine it without contradiction, then experience is not automatically contained in the function. You could reproduce all the machinery and miss the essential thing.

You can see at once why this matters for engineering, not just metaphysics. The project of building a faithful functional copy of a mind is precisely the construction of a being that says and does what the original would say and do. The hard problem poses the only question that ultimately matters about it: is it a subject that lives, or a perfect zombie that mimics a mind without it being like anything to be it? Philosophy alone cannot settle that. But science has tried to turn the metaphysical worry into a testable question, and two major theories now face off.

Theory one: the global workspace

The first comes from cognitive psychology (Bernard Baars) and was carried into neuroscience by Stanislas Dehaene: global neuronal workspace theory (GNW). Its founding image is a theatre. At any moment, dozens of specialised processors work in the wings, in parallel, unconscious: vision, hearing, memory, language, each in its own corner. To become conscious of a piece of information is to hoist it onto the lit stage and broadcast it to the whole brain at once, to make it available for speaking, deciding, remembering, relating. Dehaene calls this “ignition”: past a threshold, activity flares up and propagates across the network, typically toward the frontal areas. Below the threshold: unconscious processing, the stimulus is there but “no one” sees it.

The strength of this theory is that it is functional and therefore testable: consciousness is global access, the fact that you can act on a piece of information and report it. Its weakness is the flip side of the same coin: it beautifully describes the easy problems (access, report) but seems to sidestep the hard one: why should broadcasting be accompanied by an experience? Dehaene bites the bullet: he holds that qualia are a theorist’s illusion, and that once all the “easy problems” are solved there will be nothing left to explain. A bold position, and diametrically opposed to the next one.

Theory two: integrated information

The second theory starts from the other end. Giulio Tononi does not ask “which function produces consciousness?” but “which properties must a system have for there to be something it is like to be it?”. His answer: consciousness is integrated information. A system is conscious to the extent that it forms a whole that is more than the sum of its parts and irreducible: you cannot carve it into independent pieces without destroying something. This quantity of irreducible integration is what Tononi calls Φ (phi).

Don’t reach for a calculation; Φ is not something you work out in your head. It is a dial of unity. Take two objects: a camera with a million pixels, and your eye. The camera has a million tiny sensors that do not “talk” to one another: each sees its own dot, independently; cut the image in half and nothing is lost for the pieces: it is a crowd of mini-systems, Φ ≈ 0, nobody inside. Your visual experience, by contrast, is one: you cannot see “the left half” of your scene as an experience separate from the right, the redness and the roundness come to you fused into a single tableau. That impossibility of cutting without breaking everything is high Φ. The image to keep: Φ measures how much a system is an indivisible “I” rather than a “we” of parts side by side.

IIT makes two predictions that render it both testable and explosive. First, the seat of consciousness would lie mainly at the back of the brain (a posterior “hot zone”, parieto-occipital cortex), not at the front: the frontal lobes would serve to report and plan, not to live. A frontal-versus-posterior disagreement, squarely against GNW. Second, and here is the thunderclap, a classic digital computer (a von Neumann architecture, executing operations one at a time, in series, simulating integration without physically realising it) would have a Φ of nearly zero. The brutal consequence: perfectly simulating a brain on a chip would, according to IIT, produce a zombie: all the right answers, but the lights off. As the slogan goes, “simulating rain makes no one wet.” GNW says the opposite: if the machine realises the function of broadcasting, it is conscious, full stop. The two theories answer yes and no to the very same question about a machine.

When science stages a duel

Here is the finest part, and it is a model of method. Usually each camp tests its own theory and finds (surprise) that it works. In 2019 the Cogitate consortium tried something else: an adversarial collaboration. Proponents of GNW and of IIT sat down together, wrote in advance what each theory predicted, exactly where they diverged, and how they would interpret every possible outcome, all of it pre-registered, then carried out by neutral labs with no stake in either side. No moving the goalposts after the match.

The results landed in Nature on 5 June 2025 (256 participants, brains recorded with fMRI, MEG and intracranial electrodes). The verdict was honest, which is to say mixed: the location of conscious content in posterior areas leaned IIT’s way, and the necessity of the frontal cortex and of global broadcasting (a strong GNW prediction) was not confirmed. But other predictions supported neither side cleanly: no theory was crowned and none was killed. The real win belongs not to a camp but to the method: for the first time, two great rival theories of the mind agreed to be judged in advance by experiment. This is science putting itself to the test, aimed at the most elusive target of all: the fact of being someone.

Where this leaves the machines

The question is no longer purely academic. Several groups now propose “indicator” rubrics for consciousness, drawn from existing theories (global workspace, recurrent processing, higher-order theories, and others), in order to audit AI systems, the idea being that no current architecture ticks all the boxes, but that nothing rules it out in principle. One sharp fact falls straight out of the theories above: IIT excludes today’s digital architectures outright (their serial, feed-forward structure gives Φ ≈ 0), while the functional theories leave the door open. And the heavyweights openly disagree: in 2024 Geoffrey Hinton called conscious AI “quite possible”, while Yann LeCun called that view mistaken. Whichever way it breaks, “is anyone in the machine?” is becoming a question of engineering and ethics, not just late-night philosophy, and the science of consciousness is the only place an answer can come from.

Further reading

• Science · “‘Adversarial’ search for neural basis of consciousness yields first results”, a clear, lively account of the Cogitate duel: who bet on what, and what the result actually means. The best entry point for understanding the round without reading the technical paper.
• Internet Encyclopedia of Philosophy · “The Hard Problem of Consciousness”, the reference exposition, calm and accessible: easy versus hard problems, qualia, zombies, the knowledge argument. A solid foundation if the topic grabs you.
• Philosophy Now · “The Integrated Information Theory of Consciousness”, an honest presentation of IIT: the intuition behind Φ, its promises and its critics (down to the “unfalsifiable” charge). For digging into the most unsettling theory here.
• Nature · “Adversarial testing of global neuronal workspace and integrated information theories of consciousness”, the primary article, to read with a critical eye when you want the raw numbers and figures. The pre-registered divergent predictions alone are worth the detour.