Confidence is a presentation signal
Humans often have to judge under time pressure, so clarity and confidence can influence whose answer receives attention. Those qualities may accompany expertise, but they do not establish it. A hesitant answer can be well supported; a polished answer can rest on a false premise.
The practical response is not to distrust confidence. It is to keep confidence in its proper role: communication, not evidence.
Generated fluency widens the gap
Generative systems can produce coherent structure and decisive language when their claims are false or unsupported. NIST calls this risk confabulation. The style of the answer and the status of its evidence must therefore be evaluated separately.
Asking the same system to confirm its own answer may produce another confident explanation. Independent evidence comes from inspecting the underlying source, data or real-world result.
Evidence & context: NIST
A better answer calibrates confidence
| Answer behaviour | What it tells us |
|---|---|
| Clear conclusion with cited evidence | Inspectable, but the evidence still needs checking |
| Explicit assumptions and limits | The boundary of the claim is visible |
| Precise probability without a basis | Precision may be decorative |
| Admits unresolved information | Uncertainty may be handled honestly |
| Changes when challenged | Could reflect correction or simple compliance |
Ask what earns the confidence
Before accepting an answer, extract the claim, identify the source, inspect the evidence, expose assumptions and ask what would change the conclusion. The goal is not permanent doubt. It is confidence proportionate to the available support.
Sources & further reading
- Generative Artificial Intelligence Profile (NIST AI 600-1)
NIST. Risk-management guidance, including confabulation. It does not establish a universal error rate.
- Reproducibility and Replicability in Science
National Academies of Sciences, Engineering, and Medicine. A consensus report on evidence, uncertainty, transparency, reproducibility and replication. Scientific standards need proportionate adaptation outside research settings.
- AI Risk Management and Human-AI Interaction
NIST AI Resource Center. Official guidance on different human and AI decision roles and oversight. Appropriate reliance depends on context, consequence and system evidence.
Examples and exercises are illustrative unless attributed to a source. No independent expert review is claimed.
A correction, a counterexample or an experience worth sharing?
Join the conversation ↗