Welcome to Quantum Magazine's podcast.

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

I'm Susan Vallett.

Deep neural nets work in AI because an algorithm lets them learn from data.

The problem is you can't apply those learning algorithms

to biological brains, but researchers

are on the hunt for alternatives that just might work.

That's next.

While you're listening to podcasts, remember to check out the other Quantum Magazine podcast, The Joy of X.

Host Steven Stroggatz interviews top-tier scientists and mathematicians, new episodes out now.

Also, tell your friends about this podcast, and give us a like or follow where you listen.

It helps people find the Quantum Magazine podcast.

Today, deep nets rule AI, in part because of an algorithm called back propagation or back

prop. The algorithm enables deep nets to learn from data.

This gives them the ability to classify images, recognize speech, translate languages,

and even make sense of road conditions for self-driving cars, among other things. But real brains

are highly unlikely to be relying on the same

algorithm. Yashua Benjo is a computer scientist at the University of Montreal and the

scientific director of Mila, the Quebec Artificial Intelligence Institute. Brains are able to

generalize and learn better and faster than

state-of-the-art AI systems. And brains can't be doing it with back propagation because it isn't

compatible with their anatomy and physiology. The cortex is a particular problem. Benjo and many

...