First, we have made some changes here, which you may or may not notice:
- Emails are now being sent from paulkedrosky.com, not from kedrosky.org. I'm hoping it will solve some longstanding email issues.
- All existing subscribers are now on the free tier of my regular email sends. There will be a paid tier shortly with searchable archives and ungated access to more notes.
- I've also changed around a lot of software in the background, which shouldn't matter. Let me know via the Contact page (in the footer) if I've broken anything.
- This is still highly experimental as I add various things.
Now, on with some thoughts.
01. Summarizing My AI Risk-Reward Views
I'm writing and talking about AI more than I'd like, and the topic mostly exhausts me now. Having said that, my general views haven't changed:
- Short-term vs long-term risks
I'm more concerned about short-term economic risk than about long-term existential risks, but you can worry about both at once, contrary to what some imply. The specific form of short-term risk is the destabilizing impact of job destruction coming much faster than job creation, which now seems almost inevitable. More on this in the future, but the risks are not just at the regional level, but also at the sovereign level.
- Models for Thinking About Automation
People are generally using the wrong mental models for thinking about what's going on. It's worth noting that there is only a small sample of periods with wide automation, so every analogy is fraught, making the recent waves of factory and office automation even less helpful. A better metaphor, insofar as there is one, is the Industrial Revolution, with wide-ranging, multi-decadal impacts across a host of fields.
- Tech job creation
The appropriate near-term economic concern should be that the pace of job disruption outpaces that of job creation, which seems already to be happening. Historically, the phenomenon I am most concerned about is something like Engel's pause, the multi-decade period during and after the Industrial Revolution when laborers became poorer, even if countries and regions became wealthier.
There are, to be clear, immense opportunities for human flourishing here, but they won't come from arm-waving about how technology "always" creates more jobs than it destroys. Or by name-calling critics, or suggesting malign motives to nuanced opinions about the complex role of technology in modern societies. As recent research by David Autor, Daron Acemoglu, and others has made clear, technology's jobs-related impacts are better thought of as a series of path-dependent choices than as some universal economic constant.