The Problem of Induction
The philosophical "turkey problem" — why past stability, however long it has lasted, is never proof against a future black swan.
Taleb grounds the book's argument in the classical philosophical problem of induction — the observation, traceable to David Hume, that no finite number of confirming past observations can ever logically prove a general rule will continue to hold in the future, however large that number of past observations is. He illustrates this with what has become the book's most-cited example: a turkey fed and cared for every day for a thousand days, with each additional day of good treatment increasing its statistical confidence that this is simply how life reliably works — until the day before Thanksgiving, when the turkey's entire confident model of the world is violently and permanently falsified.
The turkey problem's uncomfortable implication is that the turkey's confidence was actually highest exactly when its risk was greatest — a thousand days of confirming evidence had made the turkey more certain of its safety at precisely the moment that safety was about to end, not because the turkey reasoned poorly from the data available to it, but because the data available to it was itself fundamentally silent about the one event that mattered most, echoing the silent evidence problem from earlier in this course.
Taleb applies the turkey problem directly to financial risk management: a strategy or asset class that has performed reliably and safely for even several decades of historical data provides no logical guarantee it will continue to do so, and — the book's more unsettling twist — an extended period of stability can actually increase certain kinds of underlying risk, as participants gradually take on more leverage and complacency precisely because the historical track record looks so reassuring, a dynamic this Book Club's Principles for Navigating Big Debt Crises course documents concretely in how long-term debt cycles build during periods that feel safe and stable right up until they don't.
- The problem of induction: no finite number of confirming past observations can logically prove a pattern will continue into the future.
- The turkey problem illustrates that confidence built from a long run of confirming data can be highest exactly when true risk is greatest.
- Applied to finance, a long historical track record of stability provides no logical guarantee of future stability, and can even mask rising underlying risk as participants grow complacent.