Skip to main content Skip to secondary navigation
Main content start

Emmanuel Candès: How to increase certainty in predictive modeling

Today’s predictive algorithms carry too much uncertainty says one mathematician, who is working to bring confidence to the models that, increasingly, rule our lives.

Anyone who’s ever made weekend plans based on the weather forecast knows that prediction – about anything – is a tough business.

But predictive models are increasingly used to make life-changing decisions everywhere from health and finance to justice and national elections. As the consequences have grown, so has the weight of uncertainty, says today’s guest, mathematician and statistician Emmanuel Candès.

Candès knows this paradigm all too well. He is an expert in identifying flaws in today’s highly sophisticated computer models. He says the secret to better prediction rests in building models that don’t try to be right every time, but instead offer a high degree of certainty about things of real consequence.

In that regard, the old scientific maxim holds, he says. Correlation does not equal causation. The statistician’s job, therefore, is helping to sort through the noise to find the nuggets of truth in the things that really matter, as Candès tell listeners to this episode of Stanford Engineering’s The Future of Everything podcast with host Russ AltmanListen and subscribe here.

Embed Code

Related Departments