Nassim Taleb’s recent article demonstrates why traditional forecasting methods systematically underestimate extreme events, with profound implications for international affairs professionals.
An international affairs professional must understand how forecasting organizations work in order to provide context to their leaders when they inevitably solicit forecasts and sometimes ground decisions in forecasted probabilities. For that and other reasons, I’ve spent the last several posts examining the Good Judgment Project’s (GJP’s) forecaster training and optimization techniques.
Now for a critique of forecasting itself, which, again, will help you contextualize your expertise. My past blogs included skeptics’ views of which factors actually contributed to GJP winning IARPA forecasting tournaments, but those critiques were not digs at forecasting as a whole.
Nassim Nicholas Taleb and Pasquale Cirillo in their late 2025 article titled The Regress of Uncertainty and the Forecasting Paradox argue forecasting as a practice gives a false sense of the future’s predictability. According to them, forecasters treat uncertainty as if they know its boundaries. This “infinite regress of doubt” isn’t just a philosophical concern. It materially impacts your work.
The forecasting paradox
Taleb and Cirillo’s central claim—what they call the Forecasting Paradox—states that the risk in the future is “larger than the risk observed in the past because the future must also contain our uncertainty about the parameters of the past.” This means that the future is structurally more extreme than the past. Not because the future is inherently wilder, but because forecasts typically don’t account for the uncertainty about past uncertainty, which compounds into “fatter-tailed” probability distributions.
Most of us intuitively think about probability like a bell-shaped curve. Most outcomes cluster near the middle, extremes are rare, and get nearly vanish as you move further out. A “fat-tailed” distribution works differently. Extreme events are rare, but meaningfully possible. The far ends don’t fade to nothing. They stay stubbornly probable.
Taleb and Cirillo wrote, “our ignorance about our ignorance structurally reshapes predictive distributions.” Extreme events are far more likely than historical data suggests.
When an international affairs professional reviews historical cases, like 30 years of diplomatic crises between two rivals, they’re identifying patterns: how often crises escalated, what conditions mattered, which factors proved decisive. That’s their understanding of the past.
These forecasts often simply project those patterns forward, without also projecting the compounded uncertainty at each earlier step in an attempt to capture whether they read the past correctly. Instead, international affairs practitioners often halt this regress or “chain of nested doubts” after stating the confidence interval. They use phrases like “moderate confidence” as if they know them with certainty. Taleb and Cirillo argue this is a profound mistake.
Collapse the exploding tree of counterfactuals
When international affairs professionals brief policymakers, they typically say “Based on historical patterns, prior national goals, or foreign perceptions, we assess X with Y confidence.” This assigned confidence captures variation in historical cases, not the uncertainty about their—or their colleagues’—interpretation of those past cases.
What’s an international affairs professional to do? In scenario planning, each possible future branches into sub-futures, which branch again. X country could escalate, or de-escalate. If escalate, regional actors could intervene or stay neutral. If intervene, the intervention could succeed or fail. Within three steps you have dozens of scenarios. By ten steps, it’s unmanageable.
Practitioners build their assessment on what they think they know—relevant historical cases, expert consensus, observable patterns. But they’re actually uncertain about all those inputs. Sometimes the historical analogy is more relevant than they thought, sometimes less. Sometimes expert consensus is prescient, sometimes it’s groupthink. Sometimes observable patterns hold, sometimes they break.
When a practitioner accounts for all the ways the foundation could be shakier than they realize, Taleb and Cirillo advise that they shouldn’t just add a bit more uncertainty around their central estimate. Extreme outcomes that seemed dismissible become scenarios a practitioner must take seriously. You should research more wildcards. Policymakers should better prepare for the extremes, rather than assuming the steady state.
The solution: More humility
So what’s Taleb and Cirillo’s solution? They provide a mathematical bridge from the philosophical concept of epistemic uncertainty to the practical necessity of using heavy-tailed distributions for any realistic modeling, forecasting, or risk management endeavor. In more simple terms:
1. Continue to explicitly acknowledge your doubt, but also acknowledge the fat tails. When you write “we assess with moderate confidence,” still explain what could make that confidence misplaced. But also test your reasoning against extreme alternatives and concisely include that outcome in your analysis as well.
2. Default to robustness over precision. When uncertainty is deep and nested, the right response isn’t more granular probability estimates. It’s preparing for the worst and then moving forward confidently. Researchers do this by exploring more wildcards. Policymakers do this by making decisions were intended outcomes can survive even if they’re badly wrong.
Be real with yourself. Remember what international affairs assessments are often built on:
- Historical cases where the similarities to current situations are debatable.
- Expert opinions that disagree and could all be wrong.
- Assumptions about relationships and incentives that we can’t verify.
- Patterns we’ve observed with hindsight that might not hold moving forward.
And because organizations put probability ranges on those assessments, they get to call it rigorous analysis. I jest.
Shutdown
We’ve all heard that the future is like the past—until it’s not. History repeats itself…or history doesn’t repeat itself but rather rhymes. Where actually the future isn’t more volatile because conditions changed—it’s more volatile because your forecast must contain your ignorance about your historical judgment.
So what can you do differently? Stop trying for precision or optimization, and make decisions with positive outcomes that survive even if the future unfolds in the fat tails.
Taleb and Cirillo’s math confirms what experience teaches: confidence is seductive, humility is wise, the future always has more surprises than the past suggests, but when we get there, we’ll forget it all because of hindsight bias. Several winning forecasts don’t improve our ability to predict the future. According to the authors of The Regress of Uncertainty and the Forecasting Paradox, forecasts make us more vulnerable and unprepared for the extreme events, which are more likely than we think.
Does that mean training won’t help? No. It means that you can’t stop learning and you need to have a plan for the worst-case scenario. Your objections to signing up for the International Affairs Professional Development Course probably won’t make sense tomorrow.
Join the Practitioner’s Network email list to avoid missing the next several posts on how to position yourself to thrive in the extreme events, what Taleb calls being “antifragile.”
