When We Trust Someone, What Exactly Are We Trusting?

Geoffrey Chen

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

The same person can be exceptionally reliable in one matter and a poor person to rely on in another.

A friend may never deliberately deceive you, yet routinely omit qualifications when repeating news. A doctor may possess deep knowledge of a disease but cannot decide for a patient which side effects are worth accepting. A colleague may calculate a budget accurately while underestimating a risk because they are responsible for the project. A relative may sincerely want the best for you, but benevolent intention does not guarantee an understanding of your situation.

Everyday language compresses these differences into global declarations: “I trust him” or “I don't trust her”. Trust starts to look like a light that must be switched entirely on or entirely off for each person.

That compression hides the difficult part. In trusting someone, we may be relying simultaneously on their competence, honesty, attention, goodwill, role, awareness of their limits and willingness to accept responsibility after an error. These qualities do not always travel together. Trust without an object and a scope can slide from reasonable reliance into devotion to a personality. Conversely, one failure can become a reason to reject an entire person.

Trust is more than “I estimate that you are probably right”

It is tempting to reduce trust to a probability judgement. We observe someone's record, estimate how likely they are to tell the truth or perform successfully, then decide whether to rely on them. That is part of trust, but it does not exhaust it.

If I believe a solid chair is unlikely to collapse, I rely on it. I would not ordinarily say the chair has been morally trustworthy. Interpersonal trust also involves a form of entrustment: I allow your action or judgement to enter my plans and therefore make myself vulnerable. You can not only fail; you can let me down.

Because of this relational dimension, the same accurate information can mean something different when offered in different ways. A doctor who says, “On the current evidence, this is my recommendation, but there are two important uncertainties”, takes a different kind of responsibility from someone who presents the same conclusion with unqualified certainty. The first communicates a conclusion together with its limits and grounds. The second asks the listener to trust an image of authority whose internal structure remains hidden.

An influential model in organisational research distinguishes three antecedents of perceived trustworthiness: ability, benevolence and integrity. The model developed by Mayer, Davis and Schoorman was designed for organisational settings and should not be treated as a complete theory of every intimate relationship. Its distinctions remain valuable. A person may be able to do something, disposed to care about another's interests, and committed to principles that make conduct predictable. When one element is absent, the character of reasonable trust changes.

Five questions concealed inside “Is this person credible?”

When another person's statement will shape our judgement, a vague question about credibility can be separated into at least five more precise ones.

First, do they have relevant competence? Someone who repairs pianos does not thereby understand epidemiology. An outstanding scientist is not automatically skilled at communicating risk. A title is evidence within a domain, not a passport across all domains.

Second, are they adequately informed on this occasion? Competence can be stable while attention and access to information fluctuate. A generally reliable person may not have read the full document, may have confused a date, or may have heard only one side.

Third, are they honestly representing both what they know and what they do not know? Honesty is not merely the absence of a literal lie. Withholding a decisive qualification, presenting conjecture as fact or knowingly causing a listener to infer more than one can support also damages epistemic trust.

Fourth, how are their interests related to mine in this matter? A conflict of interest does not prove that a claim is false. It changes the degree and kind of checking that is appropriate. A salesperson may genuinely know the product. Because the result of the sale affects them, independent support matters more.

Fifth, if the result is wrong, do they permit questions, correction and accountability? Someone who never concedes error places nearly all the risk on those who trust them, even if their historical success rate is high. Correctability is not an optional service added after trust; it helps constitute a trustworthy relationship.

These distinctions show why trust should be scoped. “I trust this accountant's interpretation of the tax rule” does not entail “I trust the accountant to decide what kind of life I should pursue”. “I trust my friend to describe her own pain” does not require accepting every explanation she offers of other people's motives. Respect for a person can coexist with reservation about a particular judgement.

Excessive suspicion can also create error

The obvious danger in trust is gullibility. Less visible is the error of withholding credibility where it is due.

A speaker's accent, social identity, style of expression and status can all affect how hearers distribute credibility. One person may receive more trust than the evidence warrants because they speak fluently, confidently and in a recognisably professional style. Another may be discounted because they are anxious, unfamiliar with dominant terminology or a member of a group burdened by prejudice, even when describing something they know intimately. Work in the epistemology of testimony describes this as a form of epistemic injustice: a person can be wronged specifically in their capacity as a knower. The philosophical literature on testimony and epistemic justice reminds us that a listener is not only defending against error. By distributing credibility badly, the listener can impoverish shared knowledge.

“Be more vigilant” cannot therefore mean “believe people less”. If suspicion falls selectively on people of lower status, non-native speakers or those uncomfortable with public performance, social prejudice has merely been dressed as rational caution. Genuine epistemic vigilance checks both other people's statements and the cues we ourselves use to judge them.

Two substitutes for evidence deserve particular care. One is confidence treated as competence. Firm expression may arise from knowledge, but also from personality, training or failure to appreciate complexity. The other is intimacy treated as accuracy. Knowing someone well gives relevant evidence about character; it cannot supply technical competence in the matter at hand.

AI turns “Whom do I trust?” into “Which chain am I relying on?”

When a person receives an answer from an AI system, trust becomes even more layered. The user encounters a unified speaking interface, but the answer may involve training data, retrieved sources, system design, the prompt, automated filters and the user's own framing of the question. Saying simply “AI is trustworthy” or “AI is untrustworthy” is about as precise as declaring a hospital building trustworthy.

Better questions concern the chain. On which tasks does the system perform reliably? Can it expose its sources? Do those sources support the sentence currently being made? How does it communicate uncertainty? Where values are involved, who is making the decision? If an error occurs, can the user detect and repair it?

An AI can be highly capable at organising language without possessing benevolence and responsibility in the interpersonal sense. It may be designed to assist, but it does not understand a commitment to this particular person as a friend or professional does. An interface that says “I understand” has not thereby achieved relational understanding. Users may reasonably rely on a function without extending functional reliability into global trust in a coherent personality.

This is not a reason to expel AI from a knowledge life. It is a reason to unfold the anthropomorphic phrase “trusting it” into inspectable tasks, sources and responsibilities.

Good trust preserves judgement instead of cancelling it

No one can audit every relationship before acting. One purpose of trust is to make cooperation possible without absolute certainty. Necessity, however, does not make blindness inevitable.

Mature trust changes with domain, consequence and new evidence. It allows another person to say “I don't know” and does not interpret every reservation as betrayal. It also permits the trusting person to ask questions without treating inquiry as an insult. The more important the relationship, the less it should depend on assumptions that cannot be spoken.

When we say that we trust someone, the most valuable object of trust may not be perpetual correctness. It may be a more specific set of dispositions: to describe accurately what one knows; to acknowledge a boundary; to hear counterevidence; and, when an outcome affects another person, not to place all responsibility back on the person who chose to trust.

The opposite of trust is not only suspicion, and the completion of trust is not obedience. Good trust is a scoped entrustment. I allow your knowledge to enter my world while preserving the ability to see where it came from, where it ends and how it can be corrected.

A person's world cannot be built without others. But their entry into it should not require a blank cheque. Trust capable of sustaining a knowledge life is not confidence that another person has every answer. It is confidence that between us there remains room to offer reasons, acknowledge error and revise together.

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.