"One of our challenges for the future is to describe to our markets and to our high-tech products what is it that we really want. In AI, there's a common goal of maximizing expected utility. We've spent decades on the 'expected' and the 'maximizing' parts, but have taken the 'utility' as a given. Until the public has the power to say what it is it really wants, the markets will choose poorly." - Peter Norvig, Director of Research, Google (2016)

Playlist: https://goo.gl/Xpxp5m
Long-form: https://goo.gl/tJT2ML

CONNECT:
- If you enjoyed this video, please subscribe to this channel.
- AI Podcast: https://lexfridman.com/ai/
- Show your support: https://www.patreon.com/lexfridman
- LinkedIn: https://www.linkedin.com/in/lexfridman
- Twitter: https://twitter.com/lexfridman
- Facebook: https://www.facebook.com/lexfridman
- Instagram: https://www.instagram.com/lexfridman
- Slack: https://deep-mit-slack.herokuapp.com

48 years ago, 2 venerable scribes, by the names of Jagger and Richards, wrote, you can't always get what you want, but you get what you need.

But turns out, Mick and Keith actually got this exactly backwards.

In today's market-driven economy, you can't always get what you need, but you always get what you want.

So we need better health care, clean water, food security for all, climate change mitigation, social equality.

But what do we get from the best and brightest of today's technologists?

Of course the breakdown is these aren't the things that we really want.

So 1 of our challenges for the future is to be able to better describe to our markets, to our high-tech products, what is it that we really want?

Now, in economics, in game theory, and artificial intelligence, there's a common goal of maximizing expected utility.

And we've spent decades on the expected part.

That's statistics, it's probability theory, it's machine learning.

And we spend more decades on the maximizing part.

That's the theory of algorithms, as it applies to all these fields.

But mostly, we leave the utility part unstated.

We just take that as a given, and saying that's what it is that we're trying to maximize, but we never question what it is.

We haven't developed the tools to let the public better describe what they want.

So don't just think about data science, information visualization, statistical inference, probabilistic reasoning.

Think also about utility science, desire visualization, and ethical calculus.

Build systems that protect the desires of the few, as well as the many.

Design algorithms with payoffs that ensure a sustainable future, as well as a near-term return on investment.

We have the power to do many things, but until the public has the power to better say what it is it really wants, the market will always choose poorly.


Speaker 0: 48 years ago, 2 venerable scribes, by the names of Jagger and Richards, wrote, you can't always get what you want, but you get what you need.

Speaker 1: But turns out, Mick and Keith actually got this exactly backwards. In today's market-driven economy, you can't always get what you need, but you always get what you want. So we need better health care, clean water, food security for all, climate change mitigation, social equality. But what do we get from the best and brightest of today's technologists? Angry birds, plants versus zombies, cat

Speaker 0: videos, photos of your food

Speaker 1: that will self-destruct in 5 seconds. And why do we get these? Because these are the things we want. Of course the breakdown is these aren't the things that we really want. So 1 of our challenges for the future is to be able to better describe to our markets, to our high-tech products, what is it that we really want?

Now, in economics, in game theory, and artificial intelligence, there's a common goal of maximizing expected utility. And we've spent decades on the expected part. That's statistics, it's probability theory, it's machine learning. And we spend more decades on the maximizing part. That's the theory of algorithms, as it applies to all these fields.

But mostly, we leave the utility part unstated. We just take that as a given, and saying that's what it is that we're trying to maximize, but we never question what it is. We haven't developed the tools to let the public better describe what they want. So don't just think about data science, information visualization, statistical inference, probabilistic reasoning. Think also about utility science, desire visualization, and ethical calculus.

Build systems that protect the desires of the few, as well as the many. Design algorithms with payoffs that ensure a sustainable future, as well as a near-term return on investment. We have the power to do many things, but until the public has the power to better say what it is it really wants, the market will always choose poorly. You

See all Lex Fridman transcripts on Youtube

Peter Norvig: Utility in AI