Research

Atlas Research develops a sustained research programme around AI, human practice and the institutional production of knowledge.

When does AI output become organisational knowledge — and what must remain visible before people can responsibly act on it?

Research programme

Three bounded lines of inquiry

We do not try to cover every question about AI. These three lines are pursued together because knowledge, capability and institutional design change together when AI enters professional practice. Power and governance run through all three.

Research line 01

AI and epistemic status

A fluent answer is not yet evidence, and evidence is not yet organisational knowledge. We study the transitions through which AI-generated material acquires credibility, enters records and begins to shape decisions.

How are source, evidence, interpretation and uncertainty preserved across an AI-supported workflow?
When does repetition or institutional use turn a provisional output into an apparently established fact?
Who has the authority and practical ability to challenge a knowledge claim?

Research line 02

AI and human capability

Expertise includes trained attention, experience, timing and the ability to recognise what matters in a particular situation. We study how AI changes the work and learning through which those capabilities are formed.

Which forms of tacit knowledge and professional judgement make AI useful in practice?
When does assistance widen learning, and when does it remove the experience needed to develop expertise?
How can employee-led AI adoption be governed without suppressing the knowledge it reveals?

Research line 03

Epistemic systems in practice

Governance is also a question of system design. We develop and test research methods and technical structures that keep evidence, interpretation, uncertainty and responsibility connected.

How can an AI-assisted research process make its knowledge claims inspectable?
Which controls support judgement instead of replacing it with procedural compliance?
How can research infrastructure preserve provenance and disagreement over time?

Current foundational work

Foundations of Ontology-driven AI-assisted Research

This work develops a framework in which AI-generated statements do not acquire the status of knowledge without traceable evidence, explicit interpretation and human judgement.

It does not promise that an AI system will never make an error. It is designed to prevent a convincingly formulated output from being treated as knowledge without epistemic control. The framework provides the theoretical basis for IROS, now in active testing within Atlas Labs.

Working concept

The Practice Gap

The Practice Gap is the distance between what can be formally recorded and what people must be able to do in a concrete practice to understand and judge language, rules and situations well.

Explicit knowledge

Documents, data, instructions, prompts, procedures and formal criteria.

Tacit knowledge

Experience, trained perception, timing and recognition of exceptions.

Shared practice

The social setting in which words, rules and actions acquire meaning.

Institutional power

The distribution of authority, recognition, risk and gain.

Ideas we work with

Older ideas, current problems

Philosophy is useful here as a working method, not as intellectual decoration. This is a deliberately selective and provisional set, not a map of philosophy. We use particular ideas where they make a present problem easier to see, question and test; other traditions and voices will enter when the research questions require them.

Tacit knowledge

Michael Polanyi

Helps us ask what happens when organisations preserve the visible products of expertise while overlooking the trained perception, experience and craft that produced them.

Meaning in use

Ludwig Wittgenstein

Helps us distinguish following the words of a rule from knowing how that rule should be applied within a shared professional practice.

Inquiry through practice

John Dewey

Helps us treat research as experimental and revisable work that begins with a concrete problematic situation rather than with a technology in search of a use.

Classification and authority

Michel Foucault

Helps us examine how classifications and norms determine what becomes visible, who may speak with authority and which judgements an institution accepts as true.

Judgement and responsibility

Hannah Arendt

Helps us ask how judgement and responsibility can be preserved when action is distributed across procedures, professional roles and technical systems—and no individual experiences themselves as the author of the outcome.

How we work

Analysis, synthesis, practice and experiment

Philosophical analysis

We clarify concepts, expose assumptions and distinguish technical performance from claims about knowledge, understanding, responsibility or autonomy.

Interdisciplinary synthesis

We connect philosophy with labour research, sociology, organisation studies, law, human-computer interaction and AI research.

Practice research

We study cases and attend to the often hidden work of checking, correcting, repairing and learning around AI.

Atlas Labs

We turn selected questions into limited, transparent and reproducible test situations without treating demonstrations as general proof.

Explore publications Commission research

Black-and-white folded forms with soft shadows.