Welcome to Quantum Magazine's podcast.

Each episode we bring you stories about developments in science and mathematics.

I'm Susan Vallett.

When we design a skyscraper, we expect it will perform to specification.

The tower should support so much weight.

It should be able to withstand an earthquake of a certain strength.

It may slightly sway in strong winds, but it will remain standing.

But with one of the most important technologies of the modern world, we're effectively building

blind. We play with different designs and tinker with different setups,

but until we take it out for a test run, we don't really know what it can do or where it will fail.

What technology is that?

The neural network, which underpins today's most advanced artificial intelligence systems.

Increasingly, neural networks are moving into the core areas of society.

They determine what we learn of the world through our social media feeds.

They help doctors diagnose illnesses.

They even influence whether you'll spend time in jail if you're convicted of a crime.

Boris Hannan is a mathematician at Texas A&M University and a visiting scientist at Facebook AI research.

He says in the past year or so, we've just started to get the language that will help explain some of the success of deep learning.

Hanning compares the situation with neural networks to the development of the steam engine.

The first steam engines were incredibly inefficient. All they could do is pump water. They just were not

engineered very well because the idea was so raw. But then 50 years later, there were already

railroads all across England. The steam engine was good enough to power a locomotive.

...