Superforecasting

Selected notes from an unfinished working draft published in 2019 about Philip Tetlock and Dan Gardner's Superforecasting.

This is a great book about how to look at predictions and how to do them better. It draws on Tetlock's work and the Good Judgment Project. The question that interested me was how business owners could make better forecasts.

Measure and revise

The book argues that predictions improve when we state them clearly, measure their accuracy and learn from the result. Vague forecasts can be stretched to fit almost any outcome. Numerical probabilities and tools such as Brier scores force more precise thinking. Without a feedback loop, forecasters have little basis for improvement.

Look outside the case

Tetlock contrasts the “inside view”, which focuses on a particular case, with the “outside view”, which starts from how often similar events occur. The notes also contrast ideological “hedgehogs” with more eclectic “foxes”. The latter draw on different perspectives, question their first answer and combine useful information rather than treating one big idea as a complete explanation. This connects with Thinking, Fast and Slow and the work of Sherman Kent on precise estimates.

Update without overreacting

A forecast is a judgment based on the information available at the time. New information should change it, but the book warns against both ignoring new evidence and overreacting to noise. The notes emphasise a growth mindset, deliberate practice and the willingness to admit error. They also warn about hindsight bias: once we know the outcome, it is easy to misremember what we expected.

Use teams carefully

Different perspectives can improve a forecast when people share information and challenge assumptions. Groupthink can instead multiply mistakes. The Good Judgment Project results discussed in the book suggest that well-run teams can be more accurate than isolated individuals. For leaders, the tension is to deliberate openly and then act decisively while staying willing to revise the plan.

The book is also clear about limits. Some questions cannot be forecast far ahead with useful precision. Where prediction runs out, planning for surprise matters. The ten commandments for superforecasters offered a concise process reminder alongside these longer notes.