Superforecasting

We talk to David Spiegelhalter, Professor of the Public Understanding of Risk, about the science of forecasting. Who or what are the superforecasters? How can they help governments make better decisions? And will intelligent machines ever be able to outdo the humans at seeing into the future?  From Cummings to coronavirus, a conversation about the knowns, unknowns and what lies beyond that.


Talking Points: 


Tetlock discovered that some people make better predictions than others.

  • Some of the qualities that make this possible are deeply human, such as doggedness, determinedness, and openness to new information, but others are mathematical. 
  • Superforecasters are highly numerate: they have a sense of magnitude.


Good superforecasters isolate themselves emotionally from the problem: you have to be cold about it. 

  • Think about George Soros shorting the pound. 


There’s a difference between having more superforecasting and more superforecasters. 

  • How do you integrate people like this into existing institutions?
  • These people are often disruptive. 
  • Probabilistic information is finely grained: what does this mean for political decision making?


Superforecasters aren’t decision makers: they give you the odds. 

  • But they are better than the betting markets.
  • Betting markets reflect what people would like to happen rather than what they should think will happen. They aren’t cold enough.


Tetlock’s book places a huge emphasis on human characteristics. 

  • Algorithms can do superforecasting only in repetitive, data rich restrictive problems
  • Tetlockian problems are much more complex. 
  • People often make a category error when they think about what AI can do. 


Mentioned in this Episode: 


Further Learning: 


And as ever, recommended reading curated by our friends at the LRB can be found here: lrb.co.uk/talking


 

See acast.com/privacy for privacy and opt-out information.