Implementations of the Mind Guild

Transforming consciousness from a philosophical problem into an engineering discipline

We build inspectable AI systems that turn theories of consciousness into computational mechanisms and falsifiable experiments. We do not assume these systems are conscious—we test which mechanisms produce flexible, mind-like capabilities.

Explore the current build Try a product Read the research
Our vision

Grounded in cognitive and brain science, we identify, simulate, and test the mechanisms that let systems model self and others, integrate perception and memory, imagine possible futures, cooperate, plan, and respond safely to novelty—a foundation for more flexible, trustworthy AI in human-machine teaming that data and compute alone cannot deliver. -- The Mind Guild Collective

Implementation dashboard

From theory to a build that can fail.

Simulated, Situated and Structurally Coherent Qualia is our first reference architecture: a five-layer implementation built as a reproducible vertical slice, with each claim mapped to code, a test, a metric, or an explicit deferment.

01 · Architecture In specification

Five-layer architecture

One shared object-centric representation, operated on by five interacting functions in a perception–action–feedback loop.

  1. 01Object bindingIdentity, stability, SFA and mutual information
  2. 02Self/world modelSource attribution and forward model T
  3. 03Pattern completionAssociative completion and compositional recall
  4. 04Action and suppressionCandidate evaluation and behavioral regulation
  5. 05Coherence regulationPrecision, valence, goals and top-down feedback
Inspect the Layer 1 diagram →
02 · Current build In development

Reference vertical slice

A paper-aligned MuJoCo scene family is being designed to exercise the developmental path from stable object slots to self/world attribution, action, imagination and coherence regulation.

  • Paper-to-code contractsClaims, mechanisms and interfaces become executable requirements.
  • Embodied task familyMultiple objects, action-dependent observability, continuous physics and goal flexibility.
  • Reproducible evidenceFrozen configurations, seeds, interventions, traces and explicit failure accounting.
Watch the Layer 1 working session →
03 · Falsification Protocol design

What would change our minds

Success thresholds and failure conditions are specified before held-out results are viewed. Negative and inconclusive results remain first-class outputs.

  • No active-exploration advantageA yoked passive observer acquires an equivalent self/world representation.
  • No goal-sensitive decompositionChanging the agent’s goal does not reorganize the relevant representation.
  • No architectural advantageTargeted ablations match the full system on composition, valence or imagination tests.
Inspect the claims and evidence →
Products · Evidence · Participation

Try it. Inspect it. Help build it.

The public surface of Mind Guild connects working tools, scientific claims, implementation artifacts and scoped collaboration.

Try a product

Train a Memory Palace

Build ordered loci, create a Person–Action–Object vocabulary, encode randomized card decks and score exact recall.

Launch trainer All products
Participate

Contribute to a testable theory of mind

Explore the code and technical work, or propose a scoped collaboration across AI, cognitive neuroscience, mathematics, robotics or experimental design.

Explore GitHub Propose a collaboration