Tyler Cowen – The #1 Bottleneck to AI Progress is Humans

This is my fourth interview with the pre-eminent infovore Tyler Cowen – and yet I’m always hearing new stuff from him. We talked at the Progress Conference 2024 about why he thinks AI won't drive explosive economic growth, the real bottlenecks on progress, him now writing for AIs instead of humans, and the difficult relationship between being cultured and fostering growth – among many other things.

Thanks to the Roots of Progress Institute (with special thanks to Jason Crawford and Heike Larson) for such a wonderful conference, and to @freethink for the videography.

Sponsors

Links

Timestamps
(00:00:00) Economic Growth and AI
(00:15:45) Founder Mode and increasing variance
(00:30:19) Effective Altruism and Progress Studies
(00:33:53) What AI changes for Tyler
(00:45:45) The slow diffusion of innovation
(00:50:41) Stalin's library
(00:53:07) DC vs SF vs EU

[meisherenow]

Stanford grad here: the problem with the fancy universities is that, while they select for intelligence and industry, they also select for docility, timidity, and eagerness to please. These traits are easily exploited by meme plagues, funneling intelligence into rationalization of orthodoxy rather than exploration and innovation.

[KeyStageMaths]

17:50 Is this ad meant to sound like a hostage video?

[AIForHumansShow]

Again, thanks Dwarkesh for these convos. Just listened to the audio it's so good and you're doing stuff no one else is doing.

[bwalker]

Cowen doing a good job of throwing the appropriate amount of cold water. Diffusion of new tech and truly robust implementation will simply hit roadblocks (some that can't be foreseen). None of this suggests that even the wildest of AI predictions might ultimately come true - it's just a matter of timing.

[tblb]

Remind me when was the last time an economist predicted the future correctly. They are very smart after the fact.

[Penrose]

The live audience format is great Dwarkesh, love the energy bro

[TheVistastube]

Whoa! You got Tyler to do ad reads? That’s crazy!

[rjalaskan]

22:49 Tyler's answer here reads like a complete cop out to me.

[CompletedReview]

we're not just talking intelligence, we're talking intelligence + agency.

[scottcorbitt]

Also, the reason a lot of technologies are adopted slowly *is* an intelligence bottleneck - people can’t figure out where to apply it right away. But if the innovation is intelligence itself, you don’t need to figure it out, you can just ask it!
@dwarkesh I feel like you honestly asked all the right questions though, and I hope he updates a bit on this! Especially with the China analogy - yes, China was playing catch-up growth because they could copy a lot from a more advanced society…just like we would be able to “copy” a lot from AI!

[dewijones]

You are only 24?! You are insanely knowledgeable - especially for your age. Love your podcast

[alulim]

Had to stop 40 mins in because it didn't feel like he was answering the questions posed. Also seems to trust his intuition greatly, but cannot give compelling reasons for that.

[philipherr]

That’s crazy. A lot of my biggest writing works have been for the AI’s in the past couple years so it’s nice to see I’m not alone.

[jeffbruno]

A more optimistic view for AI economic growth:
1) AI can improve and make new iterations on its own, being different in kind to past historical examples. The steam engine couldn't invent the next steam engine, so the GDP gains were made off a single iteration.
2) If one region of the world adopts AI with fewer regulations and has the economic growth to show for it, other countries will be more likely to follow suit.

[sakchais]

Tyler is a very wise / humble man. He's got a lot figured out.

[Blaccastro]

The idea of making content for ai for some reason make sense to me

[hyperedwin]

I started writing for the AIs in the year 2002. I also attended the first conference for AGI in 2008.

[kohlkohlbrenner]

30 years for existing companies to collapse in the face of ai native companies doesn't make sense to me

[alexmolyneux]

AI isn't equivalent to adding general population though. The main lack of consensus is that progress is the result of the extreme minority pushing forward. If we make AI at the extreme minority capability then it should add a lot of growth.

[luketheduke]

Love this podcast. Tyler Cowen I can’t take seriously though