Interactional Expertise
Harry Collins' discovery that sustained linguistic immersion in a domain confers genuine expertise — short of the embodied kind — and why a well-structured CLN knowledge base is the highest-density interactional expertise delivery mechanism yet built.
Collins' discovery — the blind gravitational wave physicist
In the 1970s, sociologist of science Harry Collins began embedding himself in the gravitational wave physics community — not to do physics, but to study how physicists communicated and reached consensus. He spent years in their labs, attending their conferences, reading their papers, and — crucially — talking with them at length about what they were doing and why.
By the late 1990s he had been doing this for two decades. He then submitted to an unusual test. He was given a questionnaire about gravitational wave physics designed to discriminate experts from non-experts. His answers, blind-reviewed by the physicists themselves, were consistently rated as those of a genuine domain expert — despite the fact that he had never touched a gravitational wave detector, had no physics training, and could not actually do the work.
Collins named what he had acquired: interactional expertise. He had, through long immersion in the community's language and concerns, developed the ability to speak as a physicist without being one. Not a parody of physics — genuine navigable knowledge of the terrain, the disputes, what counts as good evidence, what questions are currently alive, where the hard problems sit.
Interactional expertise — the ability to converse meaningfully within a specialist domain, acquired through linguistic immersion, without possessing the embodied practical skills required to contribute to that domain directly.
Distinct from contributory expertise — the full tacit-plus-explicit knowledge that enables original contribution to the practice — and from interactional imposture — the superficial fluency that sounds plausible but cannot withstand substantive dialogue.
The finding was philosophically significant because it fractured what had been a clean binary. Michael Polanyi had argued that language alone cannot transmit tacit knowledge; you have to do the thing. Collins agreed — but showed that language, if rich enough and sustained long enough, can confer a third kind of knowledge that is neither book-learning nor craft skill. Something real, genuinely useful, and not previously named.
The expertise ladder — Collins' full taxonomy
Collins and Evans (2007) mapped expertise as a hierarchy in which interactional expertise sits at a specific, well-defined level — below contributory expertise but well above mere acquaintance with a field.
What interactional expertise actually enables
The value of interactional expertise is not that it lets people pretend to be what they are not. It is that it enables a specific and genuinely important set of capabilities that neither lay people nor contributory experts can perform equally well.
The cross-domain translation capability is particularly significant. Many of the hardest coordination problems in organisations, research, and policy are not technical problems within a single domain — they are interface problems between domains. The person who holds interactional expertise in two adjacent fields can often navigate these interfaces more effectively than a contributory expert in either.
How interactional expertise is acquired — and what CLN changes
In Collins' original account, interactional expertise required years of direct immersion in the community: attending seminars, having informal conversations, participating in the social world of the domain. It was slow, expensive in time and access, and largely unavailable to anyone who could not physically be present in the community.
This scarcity was the entire reason the concept seemed notable. Most people never develop interactional expertise in anything beyond their immediate professional circle — not because they lack the ability, but because the immersion is inaccessible.
CLN changes this equation in one specific and significant way: a comprehensive CLN knowledge base — one with full operator coverage, honest ? gaps, rich * rationale chains, explicit ~ tensions, and well-placed ! risks — transmits much of what Collins says requires conversational immersion.
This is because what Collins was actually describing when he said "language and conversation" was not prose or text per se — it was semantically structured knowledge. The structure of a CLN knowledge base — its explicit operator typing, its visible disputes and tensions, its marked gaps — mirrors the structure of expert discourse more closely than any prose document.
Prose transmits what the author chose to say. CLN transmits the semantic shape of the domain — which things are definitions, which are risks, which are unresolved, which are contested, which are exceptions, which are the rationale chains that only experts know to look for. A reader of CLN can orient within a domain the way Collins' physicists oriented — by knowing the landscape's topology, not just a list of its contents.
The difference is between a map and a catalogue. Both contain geographic information. Only one tells you where you are.
CLN as interactional expertise delivery mechanism
The claim is precise and falsifiable: a reader who works through a comprehensive CLN knowledge base for a domain should be able to pass Collins' discriminator test — to be rated by domain practitioners as having a genuine understanding of the field — without possessing contributory expertise.
The mechanisms by which CLN produces this effect are direct mappings onto what Collins identified as the sources of interactional expertise:
Where CLN falls short of full interactional expertise
The claim that CLN produces interactional expertise is strong — but it has limits. Collins identified aspects of interactional expertise that emerge specifically from participation in dialogue, not from reading even very well-structured documents.
The feel for epistemic register
In real discourse, experts convey uncertainty through tone, hedging patterns, and the way they qualify claims — not just through the content of what they say. A physicist who says "this is well-established" and one who says "this is the current consensus but I have doubts" may write identical declarative sentences. CLN's ~ and ? operators capture much of this — but not all. The fine grain of expert epistemic culture requires immersion that no document fully provides.
Knowing how experts argue, not just what they argue
Interactional expertise includes knowing which kinds of evidence are taken seriously in a field, which methodological moves are considered rigorous, and which rhetorical gambits are recognised as signs of weakness. CLN documents capture conclusions with high fidelity. They are less good at transmitting the meta-level norms that govern how conclusions are reached and what makes them credible.
Social and political texture
Every expert community has interpersonal dynamics — grudges, alliances, paradigm loyalties, generational conflicts. These shape which claims get a hearing and which are dismissed. CLN's @ and ! operators can surface some of this. The full social texture requires the kind of embedded observation that produced Collins' original expertise.
The practical upshot: CLN produces a fast, high-density, portable version of interactional expertise — sufficient for most operational and cross-functional purposes. It falls short of the full interactional expertise that only prolonged direct participation provides. The gap matters most in high-stakes situations where the social and political texture of a domain are as important as its technical content: scientific policy, clinical governance, negotiating with deeply specialised counterparties.
Interactional expertise vs interactional imposture
Collins drew a sharp line between interactional expertise — genuine navigable knowledge of a domain — and what he called interactional imposture: the superficial fluency that sounds plausible from the outside but cannot withstand substantive engagement. This distinction matters practically because imposture is common and easily confused with the real thing.
The CLN-based interactional expertise path is specifically designed to block these imposture failure modes. A reader of a CLN document that includes ~ lines cannot come away believing the field is settled. A reader of a document with ? lines cannot believe they know the open questions without knowing them. A reader of a document with ! lines has been explicitly confronted with what practitioners are afraid of.
The structural honesty of CLN is, in this sense, an imposture prophylactic. It is much harder to come away from a rich CLN knowledge base with the confident ignorance of an impostor than it is to come away from a prose document with the same defect.
AI systems and interactional expertise — the CLN case
The most consequential application of the interactional expertise concept in 2026 is to AI language models. Large language models are trained on enormous corpora of natural language — books, articles, conversations, code. The question of what kind of knowledge they thereby acquire is essentially Collins' question, applied to a new kind of learner.
The answer, following Collins' framework, is that language models acquire something close to interactional expertise — but systematically warped by the biases of their training data. They have been immersed in language about many domains. They can converse with practitioners in ways that feel substantive. But their "interactional expertise" is not anchored in the social reality of any specific expert community — it is an average over all discourse about a topic, which produces a kind of generic interactional fluency that fails precisely where Collins' genuine IE succeeds: in identifying what is contested within a specific community, who the authoritative voices are, what the community is currently afraid of.
This is one of the strongest CLN value propositions for teams deploying AI assistants: a CLN knowledge base does not just help human readers develop interactional expertise. It confers organisationally-grounded interactional expertise on any AI system that has access to it — replacing the LLM's generic discourse-average with this community's specific, operator-typed knowledge structure.
Practical use cases — who benefits most
Designing CLN documents for interactional expertise
Not all CLN documents produce interactional expertise in their readers. Documents that consist only of . procedures and → verification steps produce operational competency — the ability to execute a task — not interactional expertise. For a CLN document to confer IE, it needs specific structural properties.
The last point deserves emphasis: there is a real tension between a CLN document designed for operational use and one designed to produce interactional expertise. The operator is executing under time pressure and needs the procedure clean. The new entrant seeking IE needs the disputes, rationale chains, and open questions. The best CLN knowledge bases serve both purposes by structuring them as distinct layers within the same document — operational flow in the main body, with IE-producing annotations accessible via * children and linked ~ / ? sections.