Problem of Induction

The problem of induction is the limit of learning future certainty from past observations. In fooled-by-randomness, Taleb uses the classic swan problem to show why repeated confirmation can create dangerous overconfidence: seeing many white swans does not prove that all swans are white.

Market Version

Markets often reward strategies that are implicitly built on the statement "this has not happened before." That is not evidence of safety. It may mean only that the sample is incomplete.

Examples:

• A carry trade works for years until a devaluation.
• A credit strategy has low losses until the cycle turns.
• A model looks stable until correlations shift.
• A trader survives many calm periods and concludes calm is normal.

Practical Lesson

Past data is useful, but it should train suspicion as well as confidence. The question is not only "what has happened?" but:

• What has not appeared in this sample?
• What would break the model?
• Is the historical period long enough and varied enough?
• Am I confusing absence of evidence with evidence of absence?

Connections

• epistemic-humility - Induction limits confidence.
• skewness-and-asymmetry - Rare events can dominate outcomes despite sparse historical evidence.
• survivorship-bias - The past sample may exclude failures.
• bubble-detection - Bubbles often depend on believing a recent pattern has become permanent.

Sources

• fooled-by-randomness - Taleb's induction, swans, Popper, skepticism, and rare-event framing.