Superforecasting

The verdict

Superforecasting is the most important book on prediction written in the twenty-first century. Philip Tetlock, who spent decades documenting the failures of expert prediction (his earlier work showed that the average political pundit predicts about as well as a dart-throwing chimpanzee), discovered through the Good Judgment Project that some people are genuinely, consistently, measurably better at predicting the future. They are not psychics. They are not geniuses. They are people who think in a specific, trainable way.

The book’s great contribution is moving the conversation from “prediction is impossible” (the Taleb position) to “prediction is difficult but improvable” (the Tetlock position). Both positions contain truth. Black swans are real, and overconfident prediction is dangerous. But within the realm of forecastable events — elections, economic trends, geopolitical developments — some people dramatically outperform others, and the reasons are identifiable.

The Good Judgment Project

Tetlock organized a massive forecasting tournament sponsored by the Intelligence Advanced Research Projects Activity (IARPA). Thousands of volunteers made predictions about hundreds of geopolitical events — would North Korea test a nuclear weapon, would Greece exit the eurozone, would the Arctic sea ice reach a certain level. Their predictions were tracked, scored, and analyzed.

A small subset of forecasters — about two percent — consistently performed far above the rest. Tetlock called them superforecasters. Their accuracy was not just better than average — it was better than intelligence analysts with access to classified information. And their accuracy persisted over time, which ruled out luck.

The question was: what do these people do differently?

The cognitive profile of superforecasters

Superforecasters share a cluster of thinking habits that Tetlock documents in detail.

Probabilistic thinking. Where most people think in categories (it will happen / it will not happen), superforecasters think in precise probabilities. They distinguish between 60% likely and 65% likely, and this granularity matters. Over hundreds of predictions, the person who calibrates at this level of precision dramatically outperforms the person who rounds to the nearest quarter.

Updating. Superforecasters treat their forecasts as living documents. When new information arrives, they adjust — sometimes by a fraction of a percent, sometimes substantially. They do not anchor to their original estimate or change it only when forced. They are perpetually fine-tuning.

Aggregation. Rather than committing to a single analytical framework, superforecasters draw on multiple perspectives and aggregate them. They are foxes, not hedgehogs — they know many things rather than one big thing, and they synthesize across perspectives rather than filtering everything through a single lens.

Intellectual humility. Superforecasters are confident enough to make specific predictions but humble enough to acknowledge that they could be wrong. They actively seek out disconfirming evidence and view changed minds as progress, not failure.

Growth mindset about forecasting. Superforecasters treat forecasting as a skill to be improved, not a talent you either have or lack. They study their errors, look for patterns in their mistakes, and deliberately practice the specific skills that make them more accurate.

The practice of good forecasting

Tetlock distills specific practices that anyone can adopt. Start with the base rate — how often does this type of event occur? Then adjust based on specific factors that make this case different from the average. This two-step process (outside view first, inside view second) consistently outperforms starting from the inside view alone.

Break complex questions into components. Instead of asking “will this company succeed?” ask a series of more tractable sub-questions about market size, competitive dynamics, team quality, and timing. Aggregate the answers to the sub-questions into an overall estimate.

Seek out disagreement. Superforecasters actively look for smart people who disagree with them and try to understand why. They do not treat disagreement as a threat to be defeated but as information to be incorporated.

Keep score. Track your predictions, compare them to outcomes, and analyze your errors. Most people never do this, which means they never learn from their forecasting mistakes. The simple act of keeping score transforms forecasting from opinion to practice.

The limits

Tetlock is clear about what superforecasting cannot do. It works best for questions with defined time horizons and measurable outcomes within a few months to a couple of years. It does not work for long-range predictions about complex systems (will AI transform the economy in twenty years?), and it does not eliminate uncertainty — it manages it.

The book also does not address the question of what to do with better forecasts. Knowing that something is 72% likely is only useful if you have a framework for making decisions under that level of uncertainty. Tetlock focuses on the prediction side and leaves the decision-making side mostly to other books.

Read this if…

You make decisions based on predictions about uncertain events — in investing, business strategy, policy, or personal life. The book is also essential if you consume predictions from experts and pundits and want a framework for evaluating whose predictions to take seriously. If you read only one chapter, make it the one on the cognitive profile of superforecasters.

Skip this if…

You want a philosophy of uncertainty rather than a practical method for reducing it. If your interest is in the fundamental limits of prediction in complex systems, Nassim Taleb’s work is more relevant. Tetlock and Taleb are complementary, not contradictory — one focuses on what can be predicted, the other on what cannot.

Start here

Read Chapters 1-2 for the backstory and the key findings. Then jump to Chapter 5 (supersmart?), Chapter 7 (supernewsjunkies?), and Chapter 10 (the ten commandments for aspiring superforecasters). These chapters contain the actionable core of the book.