When Two Theories of Consciousness Agreed in Advance What Would Refute Them

Abstract

In most of science, competing theories are tested by their own supporters, in separate laboratories, using methods each camp considers appropriate. The results tend to favour whoever ran the experiment. Consciousness research has this problem in an acute form, because the theories are far apart and the measurements are hard.

In 2025, a consortium published something different in Nature: an adversarial collaboration in which proponents of two rival theories agreed in advance what result would count against each of them, then went and collected the data together.

Both theories were challenged. That outcome is more interesting than a victory would have been, and the format is arguably more important than the finding.

1.What an Adversarial Collaboration Is

The published protocol states the aim plainly: to reach an agreed experimental design and to document the protagonists' expectations about the outcome before any data are acquired or analysed.

That second clause carries the weight. If both camps commit in writing to what would count as a failure for their own position, neither can reinterpret an inconvenient result afterwards. The usual escape routes — the measure was wrong, the analysis was underpowered, the prediction was never central — are closed off in advance.

The protocol describes a three-phase structure including a replication phase, in which the second half of the data is analysed using the preregistered protocols to reproduce in-house what the first half showed. The collaboration involved a dozen laboratories across three continents.

2.The Two Theories, and Where They Disagree

Integrated Information Theory holds that consciousness corresponds to a substrate with maximal integrated information, and on anatomical grounds locates this primarily in a posterior temporo-parietal-occipital region often called the hot zone.

Global Neuronal Workspace Theory holds that consciousness arises from global broadcasting and late amplification — 'ignition' — of information across a network of higher-order sensory, parietal and especially prefrontal regions.

The protocol identifies the practical consequence: given current techniques, the differing claims about where the neural correlates of consciousness are located remain the most viable and testable point of disagreement between the two.

3.What Was Predicted, and What Would Count as Failure

Three predictions were preregistered and peer-reviewed, with pass or fail criteria and expected interpretations set out in advance, and weighted according to how central each was to the theory.

For decoding of conscious content, IIT predicted maximal decoding in posterior areas while GNWT required a necessary role for prefrontal cortex. For maintenance of content, IIT posited active maintenance in posterior cortex, while GNWT predicted brief content-specific ignition in prefrontal cortex — roughly 0.3 to 0.5 seconds — at stimulus onset and offset, with content held in a non-conscious silent state in between.

The failure conditions were explicit. IIT would be challenged if sustained content-specific information and activation tracking stimulus duration were absent in posterior cortex. GNWT would be challenged if the transient prefrontal activation at onset and offset was not observed.

4.The Scale of the Experiment

Predictions were evaluated in 256 participants performing the same behavioural task across three neuroimaging modalities: functional MRI in 120, magnetoencephalography in 102, and intracranial electroencephalography in 34.

Data collection took place in two or three independent laboratories for each modality, explicitly to ensure that findings generalised across participants, instruments and experimenters. The intracranial recordings — electrodes placed inside the brains of patients undergoing clinical monitoring — yielded 583 electrodes in posterior regions and 576 in prefrontal regions across 29 patients.

That combination matters methodologically. Intracranial recording has excellent temporal and spatial precision but limited coverage and comes only from clinical populations; fMRI and MEG cover the whole brain in healthy people with different trade-offs. Running the same task across all three closes the gap that any one method leaves open.

5.What Happened

Content could be decoded from both posterior and prefrontal regions, so neither theory was eliminated at the first hurdle. The difficulties appeared in the details each theory had committed to.

For GNWT, the pattern of prefrontal involvement did not match the prediction. Across the time-resolved methods and various definitions of prefrontal regions, adding prefrontal areas did not improve — and in some cases reduced — decoding of stimulus category and orientation. Cross-temporal generalisation of decoding was sustained in posterior regions and brief in prefrontal ones.

For IIT, the problem lay in connectivity. Analyses of phase synchrony in both intracranial and MEG data failed to reveal any significant category-selective synchrony cluster in posterior or prefrontal regions once the evoked response was removed — a result at odds with what a theory built on integration would expect.

One sentence in the paper deserves attention because it forecloses the most common excuse. The authors note that the strength of the signals amplifies the significance of any challenges to the theories, since these cannot be explained away as weak data.

6.Why Both Being Challenged Is the Good Outcome

A decisive win for one theory would have been reported more widely and taught us less. What the field got instead is a demarcation of where each account breaks down, produced under conditions neither camp can dismiss.

It is also worth noticing what did not happen. Nobody announced that consciousness had been explained. The published commentary that followed, including a response from within the global workspace tradition, continued the argument in the normal way — which is what a healthy field looks like after an inconvenient result.

For a reader outside the field, the transferable lesson is about method rather than about consciousness. Agreeing in advance what would change your mind, in writing, before seeing the data, is the single most powerful defence against the human tendency to find what one is looking for. Very little research is conducted this way, and it is worth asking why not.

Editorial Comment

This article is about how a question was investigated rather than about anything a reader can act on. MindHeaven® makes no claim relating to consciousness, and nothing in this literature concerns any product.

We publish it because the adversarial format is the standard we would like applied to everything we read — including the literature on our own ingredients, where nobody has yet agreed in advance what result would count against them.