Originally published . AI polished for a better reading experience: .
This week I attended a conference with The Economist editor Daniel Franklin. Titled “The World in 2016,” the talk set out expectations for the year across twelve groups of issues. One of the newer trends he highlighted was the rise of so-called superforecasters.
I immediately wanted to know more. I love forecasting — partly because of my interest in economics and politics — but I had never thought there was a settled term for people who do this well. (In my case it is mostly for fun; still, who knows.) So I looked the word up.
A superforecaster is someone who excels at predicting the future. Often they are amateurs rather than professional analysts. Many were first identified through a contest run by IARPA (Intelligence Advanced Research Projects Activity) and challenged against CIA analysts who had access to sensitive information, while the superforecasters were limited to public data. Using the Brier score — a proper scoring rule for the accuracy of probabilistic forecasts — the amateur superforecasters scored about 30% better than the CIA analysts. Remarkable.
Superforecasters are not geniuses. They tend to score high on IQ tests (above roughly 80% of other respondents) without sitting at the absolute top. They are strong on elections, conflicts, political situations, and similar questions. To forecast, they use mathematical models and build probabilistic scenarios with different likelihoods for each path. That is close to what I learned in decision-making during my MBA — and there is real pleasure in the craft. You start with a feeling, gather data, structure and correlate it, and if a trend appears you assign a probability that it continues. For the far future you may stay mostly qualitative; for the near future you cannot do without consistent data.
What can we already predict from available numbers? One example: the current generation in Spain is likely to pay for pensions twice — first for today’s retirees, then for their own retirement. The data and trend are clear. By the time I retire, there will be fewer than two people of working age to fund each pension. Even under full employment, that is not enough. Those forecasts are relatively easy. Harder ones appear when the paradigm itself shifts — for instance with the rise of artificial intelligence and talk of a singularity, a threshold beyond which ordinary prediction breaks down. At the ultra-short horizon, big data already shapes forecasting of demand, traffic, climate, epidemics, and more. The combination of new information sources and better algorithms is enormous. We live at a moment when forecasting is becoming ubiquitous — and superforecasters are people who naturally digest data and propose algorithms that make sense of it.
In 2015, Philip E. Tetlock and Dan Gardner published Superforecasting: The Art and Science of Prediction. I hope to read it soon; by the critics’ accounts it looks excellent. According to The Wall Street Journal, Superforecasting is
“The most important book on decision making since Daniel Kahneman’s Thinking, Fast and Slow.”
Harvard Business Review paired it with Jordan Ellenberg’s work on mathematical thinking.
For anyone curious about what The Economist expected for 2016, here is a short film:
And below, a worthwhile Financial Times piece on the history of forecasting:
Comments
Loading…
Log in or register to comment.