Can a one-time event have a subjective probability?

Yes. The interpretation concerns uncertainty in belief, not only repeated trials.

Must two rational people agree?

Not necessarily if their information or assumptions differ; accounts of rationality impose differing constraints.

Is this subjective logic?

No. Subjective logic is a specific formalism with additional opinion components and operators.

AI Glossary / Statistics and probability / Beginner

Subjective probability

Definition

Subjective probability expresses an individual's or agent's degree of belief that a proposition is true, given the information and assumptions available to that agent. It is a probabilistic judgment rather than a claim that probability must be defined only by repeated-event frequencies. A coherent numerical judgment follows probability rules, but coherence alone does not guarantee good evidence, accurate forecasting, or agreement between people with different information.

What it means

A project manager might assign a probability to meeting a deadline even though the exact project will not be repeated indefinitely. The number expresses uncertainty about that defined event. Historical frequencies, expert knowledge, and models can inform the judgment without making the interpretation purely frequentist.

Subjective does not mean uninformed guessing or permission to ignore evidence. It means the probability is attached to an agent's state of belief. Two agents may differ because they have different evidence or assumptions, and each should explain the basis of the estimate. Philosophical accounts differ about the additional constraints rational beliefs should satisfy beyond mathematical consistency; this entry uses the broad degree-of-belief meaning.

How it works

Define the event precisely, including its time horizon and resolution criteria. Identify the information available at the moment of assessment. Elicit or calculate a probability while checking consistency with related events, such as the complement and mutually exclusive alternatives.

When new evidence arrives, a Bayesian update provides one formal way to revise beliefs. Specify a prior probability and the probability of the evidence under the event and its alternative. Bayes' rule then gives the updated probability, provided the conditioning event has positive probability and the model is appropriate.

Document assumptions and examine sensitivity to plausible alternatives. A precise decimal should not imply more knowledge than the evidence supports. Where judgments will recur, record forecasts before outcomes are known and evaluate calibration and usefulness across suitable collections of cases.

Separate probability from preference. A desirable outcome is not therefore likely, and a highly likely outcome is not automatically worth acting on. Decisions also depend on consequences, alternatives, constraints, and the value of obtaining more information. Probability assessment and decision evaluation are connected but distinct tasks.

Three practical examples

1. A project team estimates the chance of finishing by a specified date using progress evidence, dependencies, and documented assumptions.

2. An expert assesses the probability that a technical fault explains an observed symptom and updates the judgment when a diagnostic result arrives.

3. An AI planning system represents uncertainty about a future event while keeping the probability separate from the reward assigned to that event.

Think of it this way

Think of a clearly labeled confidence dial tied to a particular question. Evidence can move the dial, and related dials must remain mathematically consistent. The dial is neither a wish nor a guarantee; its usefulness depends on the evidence and reasoning behind its position.

Worked example

Consider an original project example. Let D mean that a release misses its planned date. Before a review, a manager assigns P(D) = 0.30. Let W mean that the review finds a specified warning sign. The manager's model assigns P(W given D) = 0.80 and P(W given not D) = 0.20.

The joint probability of delay and warning is 0.30 × 0.80 = 0.24. The joint probability of no delay and warning is 0.70 × 0.20 = 0.14. Therefore P(W) = 0.24 + 0.14 = 0.38.

After observing W, Bayes' rule gives P(D given W) = 0.24 / 0.38, about 0.6316, or 63.2%. It is not 80%: that number described the warning probability conditional on delay, a different direction of conditioning.

For sensitivity, suppose the prior delay probability had been 0.10 while the same likelihood assumptions were retained. The posterior would be 0.08 / (0.08 + 0.18), about 0.3077. The two results show how prior judgments and evidence assumptions affect the update. These are stipulated beliefs, not measured project frequencies or a recommended forecast for any real release.

Illustrative case study

In a fictional company, Birch Delivery asks team leads for probabilities of completing a migration by Friday. Early estimates vary because some leads interpret completion as technical deployment while others include customer acceptance. The team agrees on one resolution rule, records each estimate and its evidence, and updates after a readiness review. It later compares forecasts with outcomes across similar projects. The process makes disagreement inspectable and encourages learning without pretending that one numerical estimate eliminates uncertainty.

Common misconceptions and mistakes

Subjective probability is not automatically arbitrary, and frequentist data can inform subjective judgments. The interpretation concerns what the probability represents, not a ban on empirical evidence.

Mathematical coherence does not establish empirical accuracy. An internally consistent set of probabilities can still rely on misleading assumptions.

Conditional probabilities cannot generally be reversed: P(evidence given event) differs from P(event given evidence). Nor does one failed high-probability forecast prove that the assessment was unreasonable; unlikely outcomes occur. Repeated, well-defined forecasts and their evidence are more informative for evaluating judgment quality than hindsight about a single case.

Business use

Subjective probabilities make uncertainty explicit in planning, scenario assessment, and expert judgment when repeated-event data are incomplete or the decision is unique. They can help teams compare assumptions and identify where additional evidence would matter.

Businesses should define events consistently, record forecasts before resolution, and distinguish an assessor's belief from an established base rate. Use sensitivity analysis when plausible judgments differ substantially. Avoid presenting an AI-generated confidence statement as a calibrated probability merely because it is numeric. The process that produced the number and its observed reliability are part of the evidence needed for use.

Check your understanding

Answer: Someone assigns probability 0.7 to a clearly defined event.

What probability must they assign to its exact complement for coherence?

Answer: It must be 0.3. This consistency requirement does not prove the original 0.7 estimate is accurate.

Related terms

Learn next

Study Bayes' rule and the distinction between conditional probabilities. Then learn probability elicitation and evaluation of repeated forecasts.

Sources

Howell's author glossary, Subjective probability: https://www.uvm.edu/~statdhtx/methods8/Glossary/Glossary.html . Alan Hájek, Interpretations of Probability, subjective interpretation section: https://plato.stanford.edu/entries/probability-interpret/ . The entry presents the broad degree-of-belief interpretation rather than resolving philosophical disputes. The project update and fictional case are original teaching examples.

Previous: Subjective logic Next: Subjectivity