AGI-2025

We’re inviting submissions for a workshop at AGI-2025 that challenges foundational assumptions about AGI and alignment using a formal functional model.  
Purpose.This workshop invites submissions of 2-page briefs about any model of intelligence of your choice, to explore whether a functional model of intelligence can be used to very simply visualize whether those models are complete and self-consistent, as well as what it means for them to be aligned. Most AGI debates still orbit elegant but brittle Axiomatic Models of Intelligence (AMI). This workshop asks whether progress now hinges on an explicit Functional Model of Intelligence (FMI)—a minimal set of functions that any system must implement to achieve open-domain problem-solving. We seek short briefs that push the field toward a convergent functional core rather than an ever-expanding zoo of incompatible definitions.
Motivation.

1. Imagine you’re a brilliant AI programmer who figures out how to use cutting-edge AI to become 10X better than anyone else.
2. As good as you are, can you solve a problem you don’t understand?
3. Would it surprise you to learn that even the world’s leading AI researchers don’t agree on how to define what “safe” or “aligned” AI really means—or how to recognize when an AI becomes AGI and escapes meaningful human control?
4. Three documents have just been released that attempt to change that:

• The Structural Threshold of AGI : a model that defines the functional point at which an AI crosses into general intelligence. (https://drive.google.com/file/d/1bIPfxGeFx3NOyzxptyd6Rno1bZmZd4KX/view?usp=drive_link)
• Toward a Complete Definition of AI Alignment : a model that defines what it would take for an AI to remain stably aligned across all future contexts. ( https://drive.google.com/file/d/1AhKM4Y3tg4e6W_t9_wm9wwNKC5a7ZYZs/view?usp=sharing )
• A Preregistered Global Coherence Collapse Experiment : a public experiment designed to test whether the world has already crossed the point where such alignment is even possible without a structural phase-change in collective intelligence. (https://drive.google.com/file/d/1kXH-X5Mia66zG4a7NhE2RBJlZ4FgN8E9/view?usp=sharing)

Together, they offer a structural hypothesis that spans alignment, epistemology, and collective intelligence.

1. You don’t need to read them all yourself—ask your favorite AI to summarize them. Is that better than making no assessment at all?
2. These models weren’t produced by any major lab. They came from an independent researcher on a small island—working alone, self-funded, and without institutional support. If that disqualifies the ideas, what does it say about the filters we use to decide which ideas are even worth testing?
3. Does that make the ideas less likely to be taken seriously? Or does it show exactly why we’re structurally incapable of noticing the few ideas that might actually matter?
4. Even if these models are 95% wrong, they are the only known attempt to define both AGI and alignment in ways that are formal, testable, and falsifiable. The preregistration proposes a global experiment to evaluate their claims.
5. The cost of running that experiment? Less than what top labs spend every few days training commercial chatbots. The upside? If even 5% of the model is correct, it may be the only path left to prevent catastrophic misalignment.
6. So what does it say about our institutions—and our alignment strategies—if we won’t even test the only falsifiable model, not because it’s been disproven, but because it came from the “wrong kind of person” in the “wrong kind of place”?
7. Have any major labs publicly tested these models? If not, what does that tell you?
8. Are they solving for safety, or racing for market share—while ignoring the only open invitation to test whether alignment is structurally possible at all?

This workshop introduces the model, unpacks its implications, and invites your participation in testing it. Whether you're focused on AI, epistemology, systems thinking, governance, or collective intelligence, this is a chance to engage with a structural hypothesis that may already be shaping our collective trajectory. If alignment matters—not just for AI, but for humanity—it may be time to consider the possibility that we've been missing the one model we needed most.

1 — Key Definitions: your brief mustengage one or more of these.

Term

Working definition to adopt or critique

Intelligence

The capacity to achieve a targeted outcome in the domain of cognition across open problem domains.

AMI (Axiomatic Model of Intelligence)

Hypothetical minimal set of axioms whose satisfaction guarantees such capacity.

FMI (Functional Model of Intelligence)

Hypothetical minimal set of functions whose joint execution guarantees such capacity.

FMI Specifications

Formal requirements an FMI must satisfy (e.g., recursive self-correction, causal world-modeling).

FMI Architecture

Any proposed structural organization that could satisfy those specifications.

Candidate Implementation

An AGI system (individual) or a Decentralized Collective Intelligence (group) that claims to realize an FMI specification or architecture—explicitly or implicitly.

  2 — Questions your brief should answer

1. Divergence vs. convergence: Are the number of AMIs, FMIs, architectures, and implementations increasing, or do you see evidence of convergence toward a single coherent account?

2. Practical necessity: Without such convergence, how can we inject more intelligence into high-stakes processes like AI alignment, planetary risk governance, or collective reasoning itself?

3. AI-discoverable models: Under what complexity and transparency constraints could an AI that discovers its own FMI communicate that model in human-comprehensible form—and what if it cannot but can still use that model to improve itself ?

4. Evaluation design: Propose at least one multi-shot, open-domain diagnostic task that tests learning and generalization , not merely one-shot performance.

3 — Required brief structure (≤ 2 pages + refs)

1. Statement of scope : Which definition(s) above you adopt or revise.

2. Model description : AMI, FMI, or architecture being advanced.

3. Convergence analysis : Evidence for divergence or pathways to unify.

4. Evaluation plan : Visual or mathematical tests you will run using the workshop’s conceptual-space tools.

5. Anticipated impact : How the model helps insert actionable intelligence into real-world alignment problems.

4 — Submission & Publication

• Upload via EasyChair ( specify “ Morning Session” in title ). https://easychair.org/conferences2/submissions?a=34995586

• Deadline : July 24, 2025 .

• Presentation : 3-minute lightning talk + live coherence diagnosis.

• Date and Schedule: The workshop will be held 9:00 am to 12:00 pm local time in Reykjavik, Iceland where the AGI-2025 conference is being held. The workshop program is here: https://agi-conf.org/2025/workshops/

• https://easychair.org/conferences2/submissions?a=34995586

• Archiving : Accepted briefs are intended for the special issue of a journal to be decided, and will be cross-linked in an open repository for post-workshop comparison and iterative refinement.
   

5 — Who should submit

Researchers, theorists, and practitioners in any domain—AI, philosophy, systems theory, education, governance, or design—are encouraged to submit. We especially welcome submissions from those outside mainstream AI research whose work touches on how intelligence is modeled, expressed, or tested across systems. Whether you study cognition, coherence, adaptation, or meaning itself, your insights may be critical to evaluating or refining a model that claims to define the threshold of general intelligence. No coding required—only the ability to express testable functional claims and the willingness to challenge assumptions that may be breaking the world.

The future of alignment may not hinge on consensus among AI labs—but on whether we can build the cognitive infrastructure to think clearly across silos. This workshop is for anyone who sees that problem—and is ready to test whether a solution has already arrived, unnoticed.