Erik Brynjolfsson- Economics and AI
Metadata
- URL:
- Publisher: Before AGI
- Published Date: 2024-07-11
Notes
- Reminds me of a lot of econ stuff I've forgotten about
- Elasticity of demand
- Complements vs Substitutes
- The key arguments here are that:
- The right way to pursue AI is as a complement to human labor as opposed to a substitute.
- Because complements are what really increase productivity.
- Example: a car is much more useful than something that walks at the same pace as humans
- Takeaways:
- Maybe what we should be planning for is how to take advantage of the impending surplus in human capital
- Inspired:
Highlights
In the research that I've been looking at, we find that the technology has been leading to bigger and bigger gaps between companies. The leaders are pulling away the top 10%, the top 20% of firms are pulling away. Whether you measure it by profits, sales, return on investment, they're adopting these technologies and getting a lot of benefits from it. And the median firm or the lagging firms are falling further behind.
Note: This is like Rome. City states become countries when the best path up isn’t better but bigger
how electricity took 30 to 40 years from its invention and early adoption in factories in the 1880s till about the 1920s, before you saw really big productivity gains. And what happened in between was that at first, people were just using electricity to replace steam engines, but basically doing all their work the same way as before. It took a generation before they reinvented the factory and started having, instead of one big central power source, a bunch of smaller motors, each electrically powered.
Note: I guess electricity was not as obviously superior.
But this does make the point about gap between idea and implementation and acceptance
If you make a machine that replicates what human workers do, then it tends to be a substitute, and that leads to a concentration of wealth and power. If you make a machine that complements humans, that is, to do things differently than humans and make the human skills more valuable and do other things that the humans can't do, well, that tends to make labor more valuable instead of less valuable.
So Workulix takes the task based approach that I described earlier, goes through any organization's tasks up to 200,000 individual tasks for company. And each company gets a unique sort of fingerprint of where AI can help it the most, but in particular, where AI can augment their workforce and make them more effective,