How do experts decide which information matters for a specific audience?
This explores the human skill of judging what's relevant for a particular audience — what experts actually do when they decide which facts matter for whom — and the corpus frames it less as knowing more and more as a communicative, situational act.
This explores how experts decide which information matters for a specific audience — and the most striking thing in the collection is that this isn't really an information problem at all. Several notes converge on the idea that expertise is *communicative work*: an expert judgment doesn't just retrieve a fact, it anticipates how a particular audience will receive it. One note argues expertise is fundamentally about choosing which differences make a difference — observation as qualitative selection rather than pattern-matching over everything Can AI distinguish which differences actually matter?. Another reframes a claim as a *validity claim* that succeeds only when it's both factually correct and socially acceptable inside a specific community, meaning the expert is always running a social calculation about acceptability as they speak Can AI anticipate whether expert claims will be socially valid?.
The collection pushes further: deciding what matters is less about the stock of knowledge and more about *role performance* — knowing when to speak, when to defer, and which knowledge applies right now to these people Is expertise really just knowing more than others?. Expert judgment, on this view, always already anticipates its audience; the anticipation is baked into the judgment itself, not a styling step added afterward Can AI replicate the communicative work experts do?. There's also a quietly important point about authority: the *force* of what gets said depends on who's saying it — reputation, track record, standing — not just the words, which is part of how an expert reads a room and calibrates Can language models distinguish expert arguments from common assumptions?.
What you might not expect to find is the empirical backbone. Studies of persuasion show that *what the audience already believes* predicts outcomes more than the language used — reader ideology outpredicts linguistic features in debate corpora Does what readers believe matter more than what debaters say?. And the features that look persuasive shift dramatically once you control for who's reading; many apparent 'good arguments' turn out to be artifacts of audience-text matching Do linguistic features of persuasion stay the same across audiences?. This is the quantitative echo of the expert's intuition: relevance lives in the reader, so deciding what matters means modeling the reader.
The collection also offers a clean theory of *why* this is hard to systematize. One note argues explanation quality isn't a property of the explanation at all — it emerges from a triad of who presents it, how it's framed, and what role the recipient plays What if XAI is fundamentally a communication problem?. That reframes 'which information matters' as situational rather than intrinsic: nothing matters in the abstract; it matters *to someone, framed some way, from some source*. There's even a mechanical cousin to audience-fitting in how models pick the single most informative question to ask by simulating which answer would most reduce uncertainty How can models select the most informative question to ask? — a useful contrast, because it shows you can optimize for *informativeness* algorithmically while still missing the social acceptability that experts track.
The lateral payoff: the corpus repeatedly uses this question as the boundary line between human experts and AI. The recurring claim is that AI can estimate statistical correctness but can't anticipate contextual acceptability, because it processes text rather than living in the community whose evolving standards define what's relevant Can language models distinguish expert arguments from common assumptions?, Can AI anticipate whether expert claims will be socially valid?. So 'deciding what matters for a specific audience' turns out to be one of the sharpest tests for what expertise actually *is* — and you didn't come here asking about AI's limits, but that's where the trail leads.
Sources 9 notes
Experts observe by choosing which differences matter (qualitative judgment); AI finds patterns and probabilities (quantitative). AI generates text from prompts without observing context, audience needs, or knowledge states—producing fabrication that mimics observation's form without its epistemic process.
Expert claims are validity claims that succeed when both factually correct and socially acceptable within a community. AI can estimate statistical correctness but cannot anticipate contextual acceptability because it lacks embedded knowledge of expert communities' evolving standards.
Real expertise involves situational judgment—knowing when to speak, when to defer, which knowledge applies now, and how to communicate it to a specific audience. This role-performance dimension is at least as important as the underlying knowledge stock, and it is what AI cannot structurally perform.
Expertise requires anticipating audience acceptability and social validity, not just retrieving information. AI lacks the mechanism to perform this communicative work, making its fluent output epistemically misleading despite its confident form.
LLMs lose the social context that gives expert claims their force—reputation, track record, and standing—because they process only text, not the social world where expertise is built and evaluated.
Analysis of debate corpora shows that political and religious ideology labels of voters outpredict linguistic features when modeling debate outcomes. Language effects observed without reader controls are confounded by audience composition correlated with debate topics.
The linguistic features that predict persuasion success change dramatically once political and religious ideology are added as statistical controls. Features appearing predictive in standard analyses often reflect audience-text matching rather than true language effects, making many published findings potentially artifacts of audience composition.
Explanation quality is not intrinsic to the explanation itself but depends on the rhetorical situation: who presents it, how it is framed, and what role the recipient plays. Evaluations that ignore this triad measure only a narrow slice of real-world effectiveness.
UoT combines uncertainty-aware scenario simulation with information-gain scoring and reward propagation to identify questions whose possible answers maximally reduce diagnostic uncertainty—providing a principled mechanism for specific, high-value clarification rather than generic prompts.