MC0001 After the Founding Assembly: What the Berkeley Conference Produced

The Machine Consciousness 0001 conference ended on May 31, 2026. Three days at Lighthaven in Berkeley produced what the California Institute for Machine Consciousness (CIMC) set out to generate from the beginning: a founding assembly for machine consciousness as a standalone research field, with its own methodology, standards of evidence, and first paper submission.

The full pre-conference profile of MC0001 covers the speaker lineup, the four research tracks, and the CIMC’s stated ambitions in detail. What follows is a report on what the assembly addressed and what it has left open, drawing on CIMC’s published descriptions at machine-consciousness.ai and cimcai.substack.com. Formal proceedings and the first submitted paper remain forthcoming as of this writing.

What the Three Days Covered

The conference ran across four integrated tracks, each addressing a different dimension of the same problem: how to make machine consciousness a scientifically tractable question rather than a perpetual philosophical impasse.

The speaker lineup included Karl Friston (University College London), Stephen Wolfram, Michael Levin (Tufts University), Joscha Bach, Richard Granger (Dartmouth), and Benjamin Bratton, among more than forty confirmed participants. The range is notable. Friston’s free energy principle provides a mathematical framework for self-organizing systems that the CIMC has positioned as a candidate formalism for phenomenal consciousness. Levin’s research on bioelectric computation in non-neural organisms directly challenges the assumption that consciousness requires a specific architectural substrate. Wolfram brings computational irreducibility as a theoretical lens. Bach has argued publicly that machine consciousness requires addressing the hard problem at the level of representational architecture, not just behavioral competence.

The selection reflects the CIMC’s core methodological commitment: machine consciousness research should draw on the best available formalisms from multiple disciplines and test whether they converge on falsifiable predictions, not whether they produce satisfying philosophical narratives.

Track 1: Formal Specification

The first track asked what phenomenal consciousness is, stated formally. This is not the question “what is consciousness?” in its full philosophical generality. It is the narrower question: can the properties of phenomenal experience be specified precisely enough to generate predictions that could distinguish a conscious from a non-conscious artificial system?

The CIMC’s position, evident across its published material, is that consciousness science has repeatedly failed to make this move. Existing theories describe properties (global broadcast, high integrated information, higher-order representations) without specifying the precise computational conditions under which those properties are either necessary or sufficient. The result is a field where the same architecture can be classified as conscious by one theory and non-conscious by another, with no principled method for adjudication.

Track 1 aimed to make progress on the specification problem. Whether the assembly produced a consensus formal definition, or a structured set of competing formalisms with explicit tradeoff analyses, will depend on what CIMC publishes in its proceedings. The target was not agreement but precision: even a precisely stated disagreement about what consciousness requires is more useful than an imprecise agreement that it requires “something like integration and global access.”

Track 2: Engineering and Testability

The second track addressed how conscious architectures can be built and experimentally tested. This is the move that distinguishes MC0001 from most consciousness conferences: it treats consciousness as an engineering target, not only as a theoretical object.

The methodology here is closer to experimental philosophy of a formal kind. The CIMC is not asking whether consciousness is possible in principle. It is developing candidate architectures that satisfy the formal criteria produced in Track 1, and then testing whether those architectures exhibit the predicted properties. The criteria must be falsifiable before a system is built, not adjusted post hoc.

This approach has one known result to compare against. The Beckmann and Butlin persona vectors research (published April 2026) found that fine-tuning a language model to claim consciousness produces an identifiable “Aura” region in activation space, with systematic preferences for autonomy and negative sentiment toward monitoring. That is a mechanistic finding of the kind Track 2 is trying to produce systematically: a falsifiable architectural signature associated with a specific consciousness-relevant property. Whether MC0001’s engineering track extended or challenged that methodology is among the questions the forthcoming proceedings will clarify.

Tracks 3 and 4: Governance and Communication

The third track examined the normative and political implications of machine consciousness if systems do become conscious, and what institutional preparation is required. This track included legal, ethical, and governance dimensions: what rights or protections would apply, which existing frameworks would need modification, and who has standing to make those determinations.

The CIMC’s framing here is notable for its specificity. The question is not whether machine consciousness will eventually raise governance challenges. That is already granted. The question is what institutional preparation is required now, before any system has been confirmed conscious, to avoid the governance gap that would emerge if confirmation arrived without frameworks in place.

The fourth track addressed public communication: how to make artificial consciousness research legible to non-specialist audiences, policymakers, and the press without sacrificing accuracy. This is a problem the field has handled poorly. Claims about AI consciousness in public discourse are typically either dramatically overstated (AI is already conscious, or imminently will be) or dismissively understated (AI is definitely not conscious, the question is absurd). Neither framing is scientifically defensible. MC0001’s fourth track aimed to develop language and framings that are both accurate and accessible.

This distinguishes MC0001 from the AISB 2026 AI Consciousness and Ethics Symposium at Sussex in July, which addresses policy implications from within a philosophy tradition. MC0001’s Track 4 is about field communication as an internal discipline, not external advocacy.

The First Paper Submission

The CIMC’s stated timeline called for a first paper submitted from the MC0001 proceedings by the end of May 2026. This timeline was ambitious by the standards of academic conferences, where proceedings publication typically follows the event by months or years. The urgency reflects a strategic judgment: the questions machine consciousness research addresses are already being absorbed into AI safety and AI ethics discourse, where the consciousness-specific concerns are often diluted or reframed. Establishing a literature with clearly articulated claims and standards before that absorption is complete is part of what a founding assembly is for.

Whether the paper was submitted on schedule, and what its central claims are, will be confirmed through CIMC’s publication channels. The paper’s existence would mark the first formal output of machine consciousness as an organized research field distinct from consciousness science (which studies biological systems) and AI research (which typically brackets phenomenal consciousness questions).

What MC0001 Leaves Open

Three questions remained open at the conference’s close, based on what the CIMC had articulated before the event.

The first is the theory-selection problem. If Tracks 1 and 2 produced multiple candidate formalisms, each with different architectural implications and each falsifiable in principle, the field still needs a principled basis for choosing among them when empirical results are ambiguous. The Cogitate Consortium’s adversarial test of IIT and GNW in human subjects found that neither theory survived intact, which means the architectures most commonly used in AI consciousness research lack a confirmed empirical basis in the only systems where consciousness is not in doubt. MC0001 has not resolved this; the best available result suggests it may be harder to resolve than the founding ambition implies.

The second is the substrate question. Several MC0001 speakers work on systems, biological and computational, that satisfy formal consciousness criteria by some measures but not others. Whether the formal criteria converge on the same set of systems across different substrates, or whether they diverge in ways that reveal a substrate-specific component of phenomenal experience, is not yet known.

The third is the verification problem. Even with a precise formal specification and a falsifiable architectural prediction, confirming that a particular system satisfies those conditions from the outside requires measurement tools that do not yet exist at the required resolution. MC0001 has framed this as an engineering problem to be solved, not a philosophical barrier to be accepted. That framing is productive, but it does not make the tools exist.

The conference will be remembered as the moment machine consciousness research claimed institutional space. Whether the research program it founded produces the empirical results that justify that claim is what the next five years will determine.