Superforecasting as an Enabler of Epistemism
Epistemism has many touchpoints with Superforecasting, and Superforecasting can act as an enabler to Epistemism (for an introduction to Epistemism, see here or here). In choosing five books to recommend on longtermism, MacAskill chose Superforecasting as one of them and it seems it can form one of the legs of the stool of longtermist thought.
Superforecasting, a field established by Philip Tetlock, means taking a Bayesian view to updating one’s knowledge of the future, by establishing clear priors and updating one’s view whenever one gets new relevant knowledge, without holding on to any preconceived notions. Superforecasting developed as a discipline out of research done by Philip Tetlock, as a part of IARPA’s ACE contest, which tested different ways to improve forecasting.
Superforecasting as the “History of the Future”
Superforecasting is not to be confused with futurism. There is a reason why when we analyze the past, we do not call it “pastism”. We call it history. The two are different things – the past is just a number of things that happened, but history is the narrative. Pastism would be a simple list of events that happened in the past, in no particular order, while history is a distilled view of how patterns of events have shaped the world, which increases our level of knowledge. The knowledge of the past is used to illuminate the present.
Done right, Superforecasting has the potential to be to the future more what history is to the past. In Superforecasting, as with history, we hold ourselves to the highest epistemic bar that is possible. Obviously, the epistemic bar is (by definition!) lower for the future than it is for the past. We may live in a deterministic universe, but that doesn’t mean we have epistemic access to what happens in the future. But this does not mean we need to resort to pure speculation.
Key Commonalities Between Superforecasting and Epistemism
In a world where knowledge seemingly no longer needs to be true knowledge to be considered knowledge, Superforecasting and Epistemism share their emphasis on the epistemic status of one’s knowledge. In Superforecasting, a key aspect is to perform active calibration of how much epistemic uncertainty there is and how certain one can be of one’s forecast. In Epistemism likewise, one needs to evaluate and recognize the value of different kinds and different levels of knowledge.
Superforecasting skills have been shown to in addition to being correlated with specific personal traits, such as open-mindedness, also being something that can be trained. That training includes for example probabilistic thinking and how to debias one’s thinking. This is similar in Epistemism where an open mind and awareness of biases are extremely important, so that one really applies a true estimate of knowledge in decision-making.
Superforecasting is highly focused on the base rate. The power of inertia is strong, history repeats itself and most change is only incremental. Establishing a base rate for a class of events and using that to ground one’s knowledge of the future is therefore paramount. The same thinking should be applied when applying the Epistemism formula, making assumptions that go against historical patterns are unlikely to hold water.
Expanding the Timeframe of Superforecasting
Most superforecasting so far has only been done on the short-term future, 1-3 years out. New studies are needed to extend the time horizon further into the future, by 50 or 100 years. It will be crucial to investigate what aspects of superforecasting remain the same when expanding to that longer timeframe. Of course, from a longtermist perspective, 50-100 years is still extremely brief, but it would be a good start.