Really interesting point that if AI handles more “mechanical work,” the scarce skills shift toward design, judgment, and interpretation.
How do you think graduate training should adapt to that? In particular, what “manual” skills do you still think students need to practice themselves (even if AI can do them faster) because they’re essential for developing good judgment?
It’s a great question! Definitely game theory because it disciplines our thinking around what questions are interesting and how best to interpret estimates carefully. Definitely also stats/econometrics so you can tell good analyses from bad. And then tons of substantive stuff in your area—lots of on the ground knowledge
Agree; the contexts and disciplines that can identify problem spaces more robustly and across developmental states are going to become -- for those seriously enough to actually want to pursue deeper solutions and evolve or change systems -- increasingly indispensable. There is a need and hunger here that is not yet as apparent or overton-window friendly as it may become ahead.
Reading this article made me think about the debate now going on as to whether AI will eliminate lots of jobs, create new ones, or just allow people to become more productive. I think the answer depends upon the amount of work that needs to get done and the number of people in any given field who are willing and able to do the work. In the domains the author describes, there is clearly an almost infinite amount of useful work to do and only a relatively small pool of humans available to do it. In this scenario, no one loses a job. The humans just become almost infinitely more productive. I think it's a very different matter when we are talking about the type of mundane white collar work that is quite finite in nature and for which there are a lot of people capable of doing it. Most people do their work in a circumscribed business environment in which producing an unlimited amount of data would likely only produce marginal benefit if any for the company. So, no upside to doing it. Still an open question if brand new knowledge oriented jobs that we cannot even imagine right now will emerge. However, it's hard to imagine what those might be.
Andy, it's incredible to me that after proposing replacing swarms of human researchers with swarms of AIs you don't list mass human unemployment as one of the risks of getting this wrong. What are your thoughts on that?
As an economics Ph.D. student, I am so happy to see a professor bringing this up.
When I entered my Ph.D. program, I learned about the various problems with the current research system, such as p-hacking and data faking. At one point, I even told my advisor that I want to work on making an institution to check papers closely for those problems (was told that it is cool, but might lead to a very different career path).
Then, AI changed our lives. I see it as a new way to check research. We could now feed AI a data package (or even just a paper without a data package), and AI can reproduce it, tell us about the cleaning decisions the author made, and what happens if we change controls in table 1. But at the same time, we would need new systems rather than simply software. This article is excellent at explaining what this new system could be.
This is awesome and these are the types of things we all need to be exploring with AI in the white collar space. It reminds me of a post I just wrote in relation to the idea that IF humans can replicate and exponentially increase their own work, why wouldn't we do it? That's as much an exciting prospect as it is a scary one for purpose. https://substack.com/home/post/p-180379698
Really interesting point that if AI handles more “mechanical work,” the scarce skills shift toward design, judgment, and interpretation.
How do you think graduate training should adapt to that? In particular, what “manual” skills do you still think students need to practice themselves (even if AI can do them faster) because they’re essential for developing good judgment?
It’s a great question! Definitely game theory because it disciplines our thinking around what questions are interesting and how best to interpret estimates carefully. Definitely also stats/econometrics so you can tell good analyses from bad. And then tons of substantive stuff in your area—lots of on the ground knowledge
Glad I wrote a book about how to think rather than how to drive a computer.
Agree; the contexts and disciplines that can identify problem spaces more robustly and across developmental states are going to become -- for those seriously enough to actually want to pursue deeper solutions and evolve or change systems -- increasingly indispensable. There is a need and hunger here that is not yet as apparent or overton-window friendly as it may become ahead.
Reading this article made me think about the debate now going on as to whether AI will eliminate lots of jobs, create new ones, or just allow people to become more productive. I think the answer depends upon the amount of work that needs to get done and the number of people in any given field who are willing and able to do the work. In the domains the author describes, there is clearly an almost infinite amount of useful work to do and only a relatively small pool of humans available to do it. In this scenario, no one loses a job. The humans just become almost infinitely more productive. I think it's a very different matter when we are talking about the type of mundane white collar work that is quite finite in nature and for which there are a lot of people capable of doing it. Most people do their work in a circumscribed business environment in which producing an unlimited amount of data would likely only produce marginal benefit if any for the company. So, no upside to doing it. Still an open question if brand new knowledge oriented jobs that we cannot even imagine right now will emerge. However, it's hard to imagine what those might be.
What does the tech stack look like for you?
Andy, it's incredible to me that after proposing replacing swarms of human researchers with swarms of AIs you don't list mass human unemployment as one of the risks of getting this wrong. What are your thoughts on that?
Really terrific piece, and one that is making me rethink some of my current work setups/arrangements. Thank you for the public service.
As an economics Ph.D. student, I am so happy to see a professor bringing this up.
When I entered my Ph.D. program, I learned about the various problems with the current research system, such as p-hacking and data faking. At one point, I even told my advisor that I want to work on making an institution to check papers closely for those problems (was told that it is cool, but might lead to a very different career path).
Then, AI changed our lives. I see it as a new way to check research. We could now feed AI a data package (or even just a paper without a data package), and AI can reproduce it, tell us about the cleaning decisions the author made, and what happens if we change controls in table 1. But at the same time, we would need new systems rather than simply software. This article is excellent at explaining what this new system could be.
This is awesome and these are the types of things we all need to be exploring with AI in the white collar space. It reminds me of a post I just wrote in relation to the idea that IF humans can replicate and exponentially increase their own work, why wouldn't we do it? That's as much an exciting prospect as it is a scary one for purpose. https://substack.com/home/post/p-180379698
Interesting things to consider re future of education: