My guest this episode is Scott Strand. Scott is Head of Strategy, Operations, and Special Projects at Google, where he founded the Google AI Economy program as well as Google ATLAS. Recently, Scott co-authored two papers, “Mapping Gemini’s Usage in the Economy” and “AI in Science: Early Insights.” Today’s conversation will discuss these two papers, diving into the role AI could play in benefiting society through everyday life, scientific progress, and long-term economic growth.
In this episode:
* Is AI a general purpose technology? (0:40)
* Gemini’s usage in the economy (4:20)
* What will it add up to? (9:44)
* AI and entry level jobs (15:34)
* AI beyond the workplace (19:13)
* AI and the future of science (22:15)
Faster Please! is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.
A lightly edited transcript of our conversation will appear in my Week in Review issue on Saturday. (Another option is using the Substack auto transcript function.)
But here are some edited highlights from the chat:
On the visible impact of AI…
I think you could dig the deepest hole you can think of, and it would be hard to find the impact unless you were really, really pushing for it. Look, there are the basic stats that everyone can see right now. Unemployment rates are at 4.1 percent or something like that, on trend with more or less where they’ve been. In the data that we’ve looked at, it’s not consistent with the story of AI creating either widespread—certainly not widespread—unemployment or even unemployment in specific pockets.
On AI’s usage…
One is that, with past technological innovations, routine tasks are the things that get automated first—the things that are repeatable, where you do them over and over in the same way. That is not what we saw in our data. We saw that most AI usage focuses on non-routine tasks.
The types of things like creative design, hypothesis testing, doing something that’s a little bit out of the ordinary from your job description. Those make up about a third of the tasks in the average American worker’s job basket. It represents about 65 percent of their AI usage. So they’re really over-indexing on using AI for this type of work. And that poses an interesting challenge to the routine task intensity model, where you expect every wave of technological innovation to automate the repetitive stuff. But that’s not quite what we see happening.
The other thing that really surprised us was the amount of blue-collar and manual occupational usage in the data. We came in honestly just expecting to see not very much. That’s certainly what our peer laboratories found. But we saw a pretty significant amount. We see auto technicians using our models to diagnose engine wear and tear, electricians using it to sketch out wiring diagrams and inspect what might be going on.
On micro vs. macro impact…
So, in a micro sense, the productivity benefits are very real and observable. As to when they start to show up in the macro, I don’t know. I think it’s gonna require a lot of things. It’s going to require teaching people to use these tools to their maximum potential. It’s gonna require reorganizing businesses around it. When they first introduced electricity, they needed to retool entire factory floors in order to make electricity available throughout the factory, and only then were they really able to make full use of it. So I think we’re probably gonna see something similar with AI.
On AI’s effect on entry-level jobs…
We found in our data that a lot of the usage was people in high-expertise roles using it for lower-expertise tasks, the types of tasks that entry-level workers often take on. And obviously this is concerning. We can’t confirm or evaluate whether or not it’s dampening entry-level hiring, but I think this is something that we need to look at really closely going forward.
One thing that I find to be a useful frame for thinking about this is David Autor’s expertise framework, where, if you think about the expertise requirements for a given job, they sit on a distribution curve across the number of tasks. Some technologies automate away the low-expertise requirements related to that role, which raises the average expertise level required for that job. This is kind of what happened with accountants when spreadsheets were released. You didn’t really need to have people doing tables of numbers anymore; the computer took care of that. But that freed up a lot of time for accountants to specialize in tax strategy and things of that nature.
On AI’s impact outside the office…
It’s hugely significant. Eighty-six percent of AI interactions are happening outside of work. And we see people using it for really valuable things, things like seeking professional services, navigating legal questions, discussing how to engage with government and civic services. That actually overindexes by 20 times relative to the amount of time people spend there. And about half of that interaction with AI happens on nights and weekends when government offices are closed.
So it raises a really interesting idea that AI may be helping to extend civil society in some way and help people get more out of their governments, whether that’s benefits or just clearing up questions on fines, et cetera. I mean, I’m a parent, I’m a really busy parent, and so I just think of tired parents sitting at the kitchen table at 10 at night, trying to figure out how to fix a broken dishwasher or translate some zoning notice from the city into something that they can do and act on. AI is really providing real value there with a real dollar figure attached to it. But it doesn’t get measured.
On AI in science…
We found that scientists were getting a lot of productivity gains out of this. They’re saving about seven hours per week, which they’re mostly just reinvesting back in more research, which is huge. So it’s not all good news. What’s happening is that scientists are able to generate a lot more hypotheses, which is great, but now there’s a backlog of hypotheses to test, and the bottlenecks on research are starting to shift downstream to things like physical experimentation and clinical validation. So, yeah, lots of exciting stuff, but still lots of work to do.
Hey, my book is on sale everywhere: The Conservative Futurist: How To Create the Sci-Fi World We Were Promised
This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit fasterplease.substack.com/subscribe
