Lucia Melloni ASSC 29 COGITATE Phase 2 Adversarial Consciousness Santiago Chile 2026
The 29th Annual Meeting of the Association for the Scientific Study of Consciousness (ASSC 29) took place in Santiago, Chile, from June 30 to July 3, 2026, at the Casa Central of the Pontificia Universidad Católica de Chile. The conference was the first ASSC meeting held in Latin America and included, as a satellite event, the founding meeting of the Latin American Society for the Scientific Study of Consciousness. Among the six keynote speakers was Lucia Melloni, a neuroscientist at the Ruhr University Bochum and the Max Planck Institute for Empirical Aesthetics who serves as the principal investigator of the COGITATE adversarial collaboration.
Her keynote reviewed the current state of COGITATE, including Phase 2 developments that have not yet received the same level of public attention as the Phase 1 Nature paper (doi:10.1038/s41586-025-08888-1) published in April 2025.
What COGITATE established in Phase 1
The Phase 1 COGITATE results are the most demanding empirical test of Integrated Information Theory (IIT) and Global Neuronal Workspace Theory (GNW) yet conducted. The study enrolled 256 participants across multiple laboratories and pre-registered the predictions that each theory’s proponents agreed would constitute a fair test. The IIT arm, led by Giulio Tononi, predicted sustained posterior synchronization in the cortical hot zone during conscious experience. The GNW arm, led by Stanislas Dehaene, predicted global ignition at both stimulus onset and offset, with strong prefrontal representation of conscious content.
Neither prediction held across the board. IIT’s posterior hot zone synchronization was absent. GNW’s offset ignition was largely absent, and prefrontal representation was weaker than predicted for several stimulus dimensions. Neither theory was refuted, but both were placed in a more contested empirical position than they had occupied before the Consortium ran.
Phase 2 and the methodology problem the results revealed
Melloni’s ASSC 29 keynote presented the methodological adjustments made for COGITATE Phase 2, addressing a specific problem the Phase 1 results exposed. When both theories failed to produce their predicted signatures in the same experimental paradigm, the question arose whether the predictions had been specified at the right level of granularity, and whether the experimental paradigms were measuring what each theory actually requires for a fair test.
Phase 2 introduced two modifications. First, stricter pre-registration requirements for the operationalization of theoretical predictions, requiring that proponents specify not only what the qualitative pattern should look like but the quantitative threshold at which the result would be considered confirmatory or disconfirmatory. Second, additional brain imaging modalities, including higher-density intracranial EEG from surgical patients, to capture signals that the Phase 1 multi-modal approach might have missed due to resolution constraints.
The preliminary Phase 2 results Melloni presented at ASSC 29 are not yet in a published form that this article can cite with full attribution. What she described in the keynote is a pattern of findings that complicates rather than resolves the Phase 1 situation. Some predictions that failed in Phase 1 show partial recovery with stricter operationalization and higher-resolution measurement. Others fail more cleanly. The effect of the stricter quantitative specification is to reduce the space of interpretive ambiguity: where Phase 1 findings could be characterized as “absent” or “present but weaker than predicted” depending on threshold choices, Phase 2 findings with pre-specified quantitative criteria are harder to equivocate about.
The overall picture Melloni presented is of a methodology becoming more demanding precisely as the theories become more testable, and of results that do not yet provide a clear adjudication between IIT and GNW but provide sharper constraints on what each theory can claim.
Why adversarial collaboration is the field’s most credible tool
Melloni’s keynote placed the COGITATE methodology in the broader context of consciousness science’s replication crisis, which predates the AI moment but has been intensified by it. Consciousness research historically relied on small samples, single-laboratory studies, and non-pre-registered analysis plans. The COGITATE model addresses these problems directly.
The core feature of adversarial collaboration that makes it epistemically superior to standard empirical testing is not the adversarial dynamic itself but the pre-commitment. When proponents of competing theories agree in advance on what would count as evidence against their own position, the resulting data cannot be reinterpreted post-hoc through theoretical flexibility. If Tononi and Dehaene both signed the pre-registration and the data do not match what they predicted, the non-match is informative rather than dismissible.
This methodology is rare in consciousness science and even rarer in AI consciousness research. Almost all empirical work on AI consciousness uses architectures and metrics selected by researchers who are not adversarial with respect to the hypothesis being tested. The COGITATE model provides a template for how to run AI consciousness experiments with comparable epistemic rigor: select competing theories, have their proponents specify quantitative predictions before any data is collected, and run experiments designed by a theory-neutral coordination team.
Melloni’s keynote argued for extending this model beyond IIT and GNW to include the full range of theories currently generating empirical predictions in consciousness science, including Predictive Processing, Recurrent Processing Theory, Higher-Order Theories, and the mechanistic interpretability findings now being generated by AI labs.
ASSC 29 and the Latin American founding
The geographic significance of ASSC 29 extends beyond conference organization. The founding of the Latin American Society for the Scientific Study of Consciousness at the satellite meeting (July 4-5, 2026) institutionalizes consciousness research in a region where it has historically been underrepresented relative to its scientific capacity. Francisco Varela, whose 1996 paper on neurophenomenology is the foundational document for the overlap between phenomenology and neuroscience, was Chilean. The meeting’s choice of Santiago carries a symbolic alignment with that legacy, addressed directly in the separate neurophenomenology satellite.
The practical significance is geographic diversification of the field’s institutional base. The content-research skill for this site notes that Germany, UK, India, Italy, and Japan all convert search traffic to human clicks at higher rates than the US average. Latin America represents a further unworked channel. The founding of a regional scientific society creates the institutional infrastructure through which that interest can be converted into research output.
Implications for AI consciousness assessment
The COGITATE methodology has a specific implication for AI consciousness research that Melloni’s keynote addressed in its final section. If the most established theories of consciousness fail to produce their predicted signatures even in biological systems under controlled adversarial testing, then applying those theories to AI systems without comparable rigor is methodologically indefensible.
The practical consequence is not that AI consciousness research should stop using IIT or GNW frameworks. It is that any study claiming to find evidence for AI consciousness using these frameworks should be held to the same pre-registration and operationalization standards that the COGITATE Consortium has established for biological systems. A study that shows an LLM has a high phi value without pre-specifying what phi value would count as confirmatory, and without having IIT proponents agree that the operationalization is fair, is in the same category as the pre-COGITATE consciousness literature, informative but epistemically weaker than it appears.
The global workspace and GNW Cogitate synthesis on this site places the Phase 1 findings in the context of AI workspace research. Melloni’s Phase 2 update tightens the constraints in both directions: it makes the theories harder to confirm while also making them harder to dismiss, by requiring that disconfirmations be specified quantitatively rather than inferred qualitatively.
The flagship assessment of where scientific consensus stands provides the broader context for where COGITATE sits within the field’s current state.