Research
For academic collaborators in cognitive science, affective computing, HCI, natural language processing, and computational social science — the programmes, the method, the metrics, and the published record.
Four programmes
Four cognitive contexts where Foundation Models meet human cognition — the mind in a person, the record in a document, the team deliberating, the culture reasoning together. Each programme designs the functional metrics, the behavioural specifications, and the refusal logic that govern the end‑to‑end applications operating in that context.
Human intent over algorithm.
Context · the mind in a person
Engineering AI that engages the structure of an individual mind — its distortions under stress, its trait dispositions, its affect, the body that sustains it. Cognitive Behavioural Therapy, Acceptance & Commitment Therapy, and trait psychology operationalised in metric and agent.
Enter the programme →Trustworthy reading.
Context · the mind in a record
Reading high‑stakes documents — contracts, regulatory filings, financial statements, audit reports — at a standard defensible to auditors, lawyers, and regulators. Provenance treated as a hard constraint, not a feature.
Enter the programme →Tension as method.
Context · the mind in a deliberating team
Multi‑agent systems for strategic decisions whose tensions cannot be optimised away. Dialectical synthesis preserving the irreducible disagreement that defines the problem, rather than collapsing it into a single answer.
Enter the programme →Cognition lives between people.
Context · the mind in a culture
Computational distributed cognition — using multi‑agent language models as the missing instrument for studying how teams, institutions, and cultures reason together. Cultural grammar treated as a generative prior, not a stylistic overlay.
Enter the programme →Three axes of discipline
Every work the Institute carries must answer to three independent panels: academic peer review, industrial deployment, and human consequence.
Across AI research, no institute publicly commits to this triple standard. It is found instead in medical foundations — Wellcome, HHMI, the Gates Foundation — where work that survives only peer review, or only deployment, or only laboratory tests is treated as incomplete. We bring that standard to AI.
Ⅰ · Academic
Peer‑reviewed publication, independent replication, and the founding documents that future researchers cite.
Ⅱ · Industry
Enterprise deployment in finance, law, regulated industries, mental health, executive deliberation. Regulator acceptance. Methodological adoption by other labs and firms.
Ⅲ · Humanity
Pre‑registered RCTs. Longitudinal field studies. Cultural‑expert validation. Impact measured on the human beneficiaries — students, patients, decision‑makers, communities — whose cognition the systems engage.
Single‑axis success — a strong paper without deployment, a deployed product without clinical validation, a humanitarian system that no peer review will accept — is treated as incomplete. The discipline is to answer all three.
The method · five steps
The canonical methodology by which the Institute designs end‑to‑end cognition‑aware applications. Reproducible across programmes. Grounded in Functional Contextualism — the philosophy of science (Pepper, Skinner, Hayes) in which form does not determine function; function is what an act does in context. The methodology is built on that principle: behaviour is read by what it accomplishes in situ, not by its surface.
Orthogonal axes, threshold detection, boundary conditions. The cognitive context's structure recovered.
Functional metrics per axis — designed from field observation, not borrowed from benchmarks. Cognitive Fidelity (whether the system reasons about emotion the way a person does, not merely labels it). Distortion Reduction Rate (whether a conversation reduces cognitive distortions across its course). Tension Holding (whether agents preserve disagreement instead of collapsing it). Evidential Trust (whether a system’s reading can be examined the way an auditor examines a human reader’s). Substrate‑invariance (whether the result holds across model classes, not just one architecture). Cultural Grammar (the generative rules of how a culture reasons, treated as a prior rather than a stylistic overlay).
Pre‑registered failure conditions in the manner of Mayo. The discipline is to fail visibly before deployment, not invisibly during it.
Out‑of‑spec usage rejected declaratively. The refusal logic is itself an engineering target — not a guardrail bolted on after.
Field observation updates the metrics. The loop has two coupled threads — engineering at deployment speed, research at the speed of measurement; both meet at integration; both return to the next prototype.
The Institute carries Research, leads observation and milestone design as consultant, advises on integration. Full implementation is carried by partner firms and clients. This division is not a hybrid of convenience — it is the only operating shape in which academic and industrial discipline can run simultaneously, in the same study.
Accepted & published
Peer‑reviewed papers accepted at the venues of natural language processing, affective computing, cognitive science, learning technologies, and knowledge discovery.
Ⅰ · Cognition‑aware AI
Ⅳ · Collective Cognition
Ⅰ · Cognition‑aware AI
Ⅰ · Cognition‑aware AI
Ⅰ · Cognition‑aware AI
Ⅰ · Cognition‑aware AI
Ⅰ · Cognition‑aware AI
Now in progress
The questions the four programmes are taking forward now. Specifics — protocols, datasets, evaluation regimes — are reserved for collaborators.
Ⅳ · Collective Cognition
Ⅲ · Deliberative Agents
Ⅰ · Cognition‑aware AI
Ⅰ · Cognition‑aware AI
Ⅰ · Cognition‑aware AI
Ⅰ · Cognition‑aware AI
Ⅲ · Deliberative Agents
Ⅳ · Collective Cognition
Ⅰ · Cognition‑aware AI
Ⅰ · Cognition‑aware AI
Ⅰ · Cognition‑aware AI
Ⅳ · Collective Cognition
Ⅱ · Trustworthy Documents
Position papers
Foundational position pieces that articulate the philosophical commitments behind the Institute's empirical work.
Ⅰ · Cognition‑aware AI
Ⅰ · Cognition‑aware AI
A letter from the Director
The full philosophical position of the Institute, in the Director's own voice — diagnosis, position, method, horizon, and the operating reality behind it.
Correspondence
For joint research, replication studies, workshop co‑organisation, and visiting collaborations —