Antifragility, explained.
Antifragility describes a response that benefits from variation within a relevant range. It differs from robustness, which resists change without necessarily gaining from it.
Why it happens
The input average is not enough to determine the output average when the response is curved. A convex response can gain from spread; a concave one can lose from it. Benefits depend on the response and on surviving the shocks.
Taleb connects antifragility with benefiting from variation through a convex response. Robustness is a different property: resistance to change.
Read the result
Hold the average shock at zero and change its spread. Comparing all three curves under the same inputs isolates response shape. The invented quadratic payoffs illustrate convexity rather than measuring a real system's resilience.
A worked example
A portfolio of small experiments
A team can stop unsuccessful trials at a limited cost while expanding an unusually successful one.
Greater variation can create more valuable successes if downside stays bounded and the team has enough reserve to continue.
That asymmetry is useful only while the losses remain manageable and the successes can actually scale.
OPTIONAL DEEPER DETAILGo deeper: inside the model
Inside this model
Two equally likely inputs are −s and +s. Payoffs are 100 − kx²/100 (fragile), 100 (robust), and 100 + kx²/100 (convex). The graph plots their equally weighted means across shock sizes. All three receive the same input distribution.
Where this idea is useful
A practical use
Compare a system whose failures accelerate under load with a portfolio of small experiments whose losses are capped and successes can expand.
A common misconception
“Any stress makes a system stronger.”
Excessive shocks can destroy a system or change its response. Benefiting from some variation does not imply benefiting from every shock.
What this explanation leaves out
- These quadratic response functions are invented teaching examples, not calibrated systems. The benefit depends on the response remaining convex over the relevant range; extreme shocks may change it.
How do I distinguish antifragile from robust?
Ask what happens to the payoff when uncertainty increases under comparable average inputs. A robust response remains relatively unchanged; an antifragile response improves over the range being considered.
Which costs are capped, which benefits can expand, and where would a shock become destructive?
Associated thinkers
Further reading
Explore the original research or the teaching reference behind this experiment.