Model Convo: Nick Krosse
On AI's Microcosmic History, Jevons Paradox, and Gesamtkunstwerk
This week’s convo is with Nick Krosse, the Technology and Innovation Policy Director at the American Action Forum. You can find him on social media at x.com/nkrosse.
What was your path into AI, and what are you working on now?
I’ve been working in tech policy for the better part of the last twelve years, and as my mentor and friend Adam Thierer likes to say (I’m paraphrasing), you won’t be working on the same issues two years from now. It’s a fast-paced, dynamic issue area where policy discussions are always playing catch up with technology.
I started out primarily focused on telecom issues when I was a graduate fellow at the Mercatus Center, then worked on social media policy for most of my time at Stand Together, adding AI to my portfolio as it increasingly dominated the tech policy conversation.
It has been a natural evolution given both where industry and policymakers’ attention have shifted, but many of the issues across these technologies are the same (e.g., competition, privacy, liability, infrastructure deployment). Indeed, it feels like we’ve been speedrunning many of the same issues that policymakers have had with social media over the past decade with AI (child safety, intermediary liability, etc).
My new role at American Action Forum allows me to work on all of the above, but one theme you might be able to draw out from my current and upcoming work is trying to find the right balance on local, state, and federal regulation of technology. Technology policy inherently deals with economic activity that crosses state and local borders, where a unified federal framework will provide businesses large and small more regulatory certainty than a patchwork will.
But technology policy also confronts many regulatory questions in areas where state and local governments are best situated to govern, such as in insurance and zoning. The objective of federalism isn’t to provide any one level of government preference over the other for all issues, but rather to find the right balance between letting state and local governments regulate where they have the requisite local knowledge to do so most effectively while not creating barriers to interstate commerce.
What works of art have most shaped your views on AI?
Georges Seurat’s “Un dimanche après-midi à l’Île de la Grande Jatte” (“A Sunday Afternoon on the Island of La Grande Jatte”) is maybe the best painting for understanding what large language models (LLMs) do. The painting is probably the most famous example of “pointillism,” a painting technique that uses color theory to form a cognizable picture from individual dots of a single color.
Seurat aligned the dots in such a way that when you stood far enough back, your eye blends them to form the picture. Similarly, LLMs are both human-constructed but also emergent phenomena. Any individual token, or routine or tool, doesn’t make sense outside of the full context of the prompt or model, but looked at from the right distance, it can be a powerful tool for expressing human thought and creativity.
What’s your most contrarian take on AI?
I don’t know how contrarian this is (especially given all of the research being done on the topic that supports my take), but I think a lot of people—including the companies building the foundational models—fundamentally misunderstand the potential effects that AI will have on the labor market.
Many commentators seem to perceive a sort of binary: either AI is a bubble that will fail to produce any productivity gains, or it will replace every white-collar job and leave us all unemployed slaves to the machine. But automation typically targets tasks, not jobs. The effects that will have on the labor market are far more ambiguous and mean that there may not actually be a strict tradeoff between productivity gains and employment levels.
So far, the available research suggests the effects have been net positive for the labor market. Some roles may be entirely automatable, but many even entry-level roles contain at least one task that is not easily automated. Workers may instead experience a shift away from tasks that can easily be automated toward those that cannot. Jevons’ paradox may further mean that more automation increases demand for labor, as the automation of simple tasks both reduces the cost of completing them while increasing worker productivity.
There are a lot of potential pathways that the labor market could evolve with this new general purpose technology, and while considering how to deal with coming changes in the job market is a good idea, the tales of a jobs apocalypse (and the policies that flow from that assumption) seem to be a misdirection of concern and effort.
What are you reading, watching, or listening to now?
My favorite daily listen is NPR’s The Indicator podcast. I don’t always agree with the take their reporters have on a given issue, but they are some of the best at telling a quick story through numbers (or, showing rather than just telling). I aspire to be as good as them at it!
Project Hail Mary was such a delight to see earlier this year. Ryland Grace is proof that humans are capable of extraordinary things even in the most extreme circumstances.
Go-to emerging tech music track?
Anything classical. The great classical composers, the ones anyone today who is not a classical music student has heard of, were all innovators. Beethoven transformed the symphony from a trifling piece to a vehicle for composers to express their utmost creativity. Richard Wagner built his own opera house to give himself complete control over his artistic vision, his Gesamtkunstwerk (“comprehensive artwork”). Maurice Ravel was expelled twice from his musical studies for his avant-garde style but is now considered one of the most innovative composers of the twentieth century. Igor Stravinsky’s “Le sacre du printemps” (“The Rite of Spring”) caused a riot at its premiere. Any of these mirrors a story you can read about in the history of Silicon Valley.
This interview has been edited for length and clarity with hyperlinks added. It reflects only the views of its subject.