What the science of consciousness knows, doesn’t know, and is fighting about
“When even scientists cannot agree on how to measure consciousness, any certainty about a machine is, at the very least, premature”
Unlike philosophy, which is content with thought experiments, and law, which solves its problems with useful fictions, the neuroscience of consciousness attempts something more ambitious: to measure, with instruments and mathematics, what makes a system conscious. It is worth reviewing how far this discipline has come, because its actual state is far more uncertain, and far more contentious, than the word “science” tends to suggest.
Integrated Information Theory
In 2004, neuroscientist Giulio Tononi proposed that consciousness does not depend on what a system does, but on how much information it integrates in an irreducible way — a quantity he called phi. The higher the phi value, the more conscious the system would be, according to this theory. What is remarkable, for the purposes of this series, is a consequence the theory itself accepts without hesitation: two systems that behave identically could have different phi values, and therefore one could be conscious and the other not, even though they act in exactly the same way. In other words, Integrated Information Theory does not rule out the philosophical zombie discussed in the second instalment of this series; under certain conditions, it declares it real and possible — something Dennett’s functionalism, examined a few instalments back, flatly denied.
Global Workspace Theory
The other major theory, developed by Bernard Baars and later expanded by Stanislas Dehaene, imagines the mind as a theatre: many processes are unfolding in the darkness of the auditorium at once, and consciousness is simply whatever, at a given moment, steps onto the stage and becomes available for memory, language and decision-making to use. Unlike Integrated Information Theory, which seeks a structural, intrinsic property of the system, Global Workspace Theory describes a function — a process of access and broadcasting — that could in principle be implemented in very different architectures, including, without much difficulty, an artificial system.
The experiment that was meant to settle it, and didn’t
In June 2023, a team of researchers organised what they called an adversarial collaboration: they asked the proponents of each theory to design experiments capable of putting the rival theory to the test, and presented the results at the annual meeting of the Association for the Scientific Study of Consciousness. The outcome did not clearly favour either theory: there was partial evidence in favour of each, with a slight tilt towards Integrated Information Theory in some readings, but neither was fully confirmed nor refuted. The most ambitious experiment to date in the science of consciousness ended in an uneasy draw.
The September row
What followed was more revealing than the experiment itself. Media coverage of these results, which some read as a victory for Integrated Information Theory, prompted 124 scientists and philosophers of consciousness — many of them well-known figures in the field — to sign an open letter accusing the theory of being pseudoscience, on the grounds that the theory as a whole cannot be empirically tested. The reaction was immediate and divided: even declared critics of the theory, such as philosopher David Chalmers, considered the term disproportionate, comparing it to using an oversized weapon to settle a local dispute, while philosopher Tim Bayne pointed out that the letter did not even bother to define what it meant by pseudoscience. Months after that episode, the science of consciousness stood exactly where it had before: with no winning theory, and now, in addition, with a public rift among its own researchers.
Why this matters for artificial intelligence
If specialists cannot agree on how to measure consciousness in the one system we are certain possesses it — the human brain — any confident claim about whether an artificial system has it or not ought to be treated with the same caution. But there is a more specific nuance worth underlining: under Integrated Information Theory, an extraordinarily convincing language model in conversation could, depending on its computational architecture, have an extremely low phi value, and would be, in the exact terms of that theory, a confirmed philosophical zombie, with its verbal fluency saying nothing against that diagnosis. Science, in its most rigorous form available today, does not close the question this series has held open from the start; it formalises it with mathematics.
The Christian perspective
Faced with a scientific disagreement of this depth and duration, the most honest Christian stance is not to take sides with one theory to defend the faith, nor to dismiss neuroscience as irrelevant, but something more modest: to recognise that the doctrine of the image of God, in Genesis 1:27, was never proposed as a hypothesis competing with Integrated Information Theory or Global Workspace Theory inside a laboratory. It is a revealed claim about the origin and nature of human dignity, not a phi measurement or a pattern of neural activity, and so it is neither upheld nor threatened by the outcome of any future experiment. What ought to concern anyone, believer or not, is the temptation to cite neuroscience as though it had already settled, one way or the other, the question of artificial intelligence — when the discipline itself, in its most public and recent dispute, showed that it has not even settled itself.
Conclusion
The neuroscience of consciousness has not given this series’ problem a solution that philosophy and fiction could not already offer; instead, it has given something different, and perhaps more valuable: confirmation, with instruments and mathematics, that the gap between behaviour and experience is not merely a trick of philosophical imagination. It is, according to at least one of the field’s most serious theories, a real structural possibility — and no one, not even the specialists themselves, has yet managed to measure it with the certainty this debate would need to be settled.
