Macromegas #17 - Superforcasting and History

Black-Swan Events, Pandemics, and the Limits of Superforecasting

Hi friends,

Happy Friday!

Superforcasting and History

There are two theoretical limits to (super)forecasting: data availability, and bias (either human or modelling).

(By the way, if you haven’t read Superforecasting, by Tetlock, it is a good read.)

The worst year in our recent(ish) history - An interesting baseline for social resilience planning

In terms of data availability, us humans are subject to what is called the availability bias. Things we can remember or visualise better enjoy a higher weight in our mental prediction framework. This means that the most major Black Swan events, which statistically did not happen in our lifetime nor sometimes even in our recent history as a species, are far less likely to enter in the equation. Even in the rational minds of superforecasters.

That is why I was very interested in this point of view describing the worst year we can remember as a species. This should be our true baseline for this kind of social resilience planning. I simply hope politicians know history better than they know biology…

Read the retrospective here: Why 536 was ‘the worst year to be alive’

This other article looks at the main practical limit to (super)forecasting:

How do you decide on the right plan of action based on an accurate probabilistic prediction?

Most people already struggle with probabilistic (Bayesian) thinking.

But imagine you manage to get your hands on the exact probability of an event happening. How do you plan (invest) to maximise your expected outcomes?

If you are playing poker or blackjack, it might be a reasonably deterministic choice to make. If you are dealing with more complex systems, such as allocating resources in a large organisation, investing in potentially disrupting technologies, or “planning” a country’s economy, the actual steps for you to take are far less straightforward.

This is what this article looks into.

Read it here: The Limits of Applied Superforecasting

Don’t spend too much time trying to predict the future: rely on systems that do well - or at least not irreversibly bad - in all cases.

Thanks for reading,

V