How Can an Expert’s Opinion Become My Own Judgment?

Geoffrey Chen

How a Person Builds a World · Season Two: “Others, Dialogue, and Shared Knowledge” · Article Three

A person receives a complicated medical result. One specialist recommends acting soon; another thinks continued observation is reasonable. Both are trained. Both offer professional reasons. The patient cannot independently analyse the full research literature and certainly cannot become a specialist in the available time.

Yet the patient must still choose.

The situation reveals an unavoidable tension in modern knowledge. We need experts precisely because we do not possess their knowledge. But if we cannot understand why an expert judges as they do, in what sense can the eventual decision be our own? If personal judgement means independently verifying every technical claim, ordinary people can have almost no judgement of their own in medicine, law, engineering, finance or AI. If respecting expertise means merely complying, however, a person's choice collapses into the execution of an authoritative instruction.

The answer is not a vague midpoint between the extremes. It begins by distinguishing which epistemic work can reasonably be delegated to an expert from which acts of judgement cannot be delegated at all.

A layperson is not simply in one of two states: knowing or not knowing

Experts do more than remember additional facts. They can usually identify distinctions that matter in their domain, weigh different forms of evidence, recognise structural similarities between new and established cases, and anticipate possible consequences. A layperson cannot acquire these capacities in a single consultation.

That does not mean a layperson is unable to judge anything. The philosophical “novice–expert problem” asks how someone who lacks first-order expertise can have reasons for choosing among putative experts, especially when the experts disagree. Discussions in social epistemology identify second-order indicators such as relevant credentials, assessments by other experts, argumentative performance, track record and the distribution of informed opinion. None is perfect, but together they show that evaluating expertise does not require already being the expert.

Credentials can be fraudulent or obsolete. A majority can be slow to change or influenced by institutions. Eloquence does not ensure accuracy. A minority sometimes does overturn a consensus. The conclusion is not that the indicators are useless. It is that they must be combined and weighted in ways appropriate to the question.

A non-expert can often reach at least three levels of understanding. The first is understanding the conclusion: what is being recommended? The second is understanding the grounds: which evidence and assumptions support it? The third is understanding the boundary: where does it not apply, what uncertainty remains, and which new facts would change the recommendation? A layperson need not redo the professional analysis, but possession of only the first level makes it difficult to distinguish advice, command and salesmanship.

Expertise cannot decide a value trade-off on another person's behalf

An expert may estimate how much a treatment improves a probability of success, how much load a structure can withstand, or which risks attach to an investment. Whether the option is worth it usually also involves purposes and costs.

Faced with the same medical risk, one patient may give priority to extending life while another gives priority to preserving a particular capacity. Given the same legal probability, one person may accept years of litigation while another values an earlier end to uncertainty. Expertise can clarify consequences. Professional status alone cannot settle how someone should live.

This is not a diminution of expertise; it protects its boundary. The more precise professional advice is on the question it can answer, the less need it has to pretend to final authority over values as well.

For an expert opinion to become one's own judgement, therefore, the professional conclusion must undergo a translation:

If I accept the expert's assessment of the facts and probabilities, what do I choose given the ends I value, the costs I can bear and the time available?

This formulation preserves both reliance and responsibility. Technical assessment can be delegated. Commitment to a value does not disappear because the decision is difficult.

When experts disagree, first locate the disagreement

Different recommendations do not necessarily mean that one expert is right and the other wrong. They may be using different data, applying different risk thresholds, assigning different weights to the same evidence, or agreeing on facts while presupposing different goals.

The most useful first response is not always “Which person has more authority?” It is to locate the disagreement:

  • Are both people working within the same relevant specialty?
  • Are they using the same facts and current sources?
  • Is the difference about data, interpretation, prediction or values?
  • What is the wider distribution of expert opinion, and what reasons support the minority position?
  • Is the decision reversible, and what would be the cost of waiting for more information?
  • What new evidence would each expert regard as a reason to change their recommendation?

The final question matters greatly. Someone unable to say what could change their mind may be defending an identity rather than offering a testable professional judgement. Expertise does not require never changing. It should include some understanding of what would justify change.

Consensus is ordinarily important second-order evidence for a non-expert, but it is not an oracle. Its epistemic value comes from many trained people engaging the evidence through sufficiently independent methods and still reaching similar conclusions under criticism. If an apparent consensus consists mainly of repetition from one source, one incentive or one authority, its value is weaker. The important question is not only how many agree, but how their agreement was produced.

AI can lower the threshold of understanding and simulate expertise

AI can rewrite a technical document in ordinary language, compare two opinions, list uncertainties and help a user prepare questions for the next consultation. These functions can materially improve a layperson's position. Material that was previously opaque can at least become the basis for an initial inquiry.

Language models can also create a powerful appearance of expertise. One interface can answer fluently across medicine, law, history and software. Continuity of voice conceals discontinuity in reliability across domains. The user sees one speaker; behind it may be sources of uneven quality, obsolete information or inferences that are never shown. AI lowers barriers to expression and retrieval. It does not automatically remove barriers of validation or responsibility.

A more defensible design direction is for an AI assistant to identify the source set it represents, answer within a defined service boundary, attach inspectable support to important claims and return sensitive or final professional decisions to human judgement. The “expert-owned sources, controlled assistant, client access” structure described by SonaMinds is one attempt in that direction. Its value is not that a conversational interface can pretend to be an expert. It is that provenance, scope and the point of human judgement remain more visible. Such a structure cannot prove that every answer is correct; it can make the meaning of “What does this answer represent?” more inspectable.

When AI mediates expert knowledge, it is best treated as translator, question generator and comparison tool, not as a new expert whose qualifications need no explanation. It may help a person return to a professional relationship with better questions. It should not become an excuse to leave professional relationships altogether.

My judgement need not originate entirely in my knowledge

A judgement belongs to me neither because I discovered every fact in it nor because I have final authority over every technical detail. It belongs to me because I know whose knowledge I am relying on and at which level, understand the principal consequences, and accept responsibility for value commitments that cannot be delegated.

This is a limited but genuine autonomy. I can say, “On this technical question, I accept the convergent judgement of relevant experts.” I can say, “A significant unresolved disagreement remains, so I choose a reversible course.” I can also say, “The experts can estimate the risk but cannot decide for me which cost is worth bearing.” None of these statements rejects expertise. Each places professional knowledge within the decision it is actually equipped to support.

Conversely, “An expert said so” cannot automatically end discussion. Experts deserve greater epistemic weight than novices within their field, but the weight comes from relevant competence, evidence, method and correctability, not an untouchable identity. Ordinary people need not receive an equal vote on every technical matter. They do have a legitimate claim to understand the scope of advice, major uncertainty, and which part of the final decision remains theirs.

An expert opinion becomes my judgement not when I repeat it, but when I can describe the dependence: why I accept professional assessment here, what it has not decided for me, and what would make reconsideration necessary.

A personal world is not a world into which only personally produced knowledge may enter. It is closer to a selected structure of dependence. Maturity does not consist in no longer needing experts. It consists in relying reasonably on people who know more without handing one's capacity for judgement over as part of the consultation fee.

Primary sources and further reading

Continue reading: Explore the How a Person Builds a World series.

Discover more from Geoffrey Chen

Subscribe to get the latest posts sent to your email.