Apr 10, 2023·edited Apr 10, 2023Liked by Noah Smith

When Ford made cars cheaper to make, that meant more, better-paying car jobs, not fewer. Why? Because back then, lots more people wanted cars than could afford them.

I distrust any "AI economics" that doesn't look at demand elasticity. If it's a service that people would like to have a lot more of, then AI should not reduce but increase employment and wages.

Most people not named Buffett or Musk would like a lot more on-call therapy, personal medicine, secretarial help, legal expertise, custom programming, and favorite genre fiction than they currently get. If AI partially automates those industries, it ought to mean not less but more worker dollars, and likely more and better (if different) jobs.

Which workers? Good question, the winners could easily be different than the current workers. But it should still be a workers' win, unlike the 80s or 90s.

Could AI *fully* automate those and other industries, laying off everybody? Sure; just look at what happened to all America's horses after cars got cheap. But when you're talking "humans, as economically useful as horses," that's no longer a story about scarcity and economics. Full automation of cognitive work would be a society whose big problem for a generation wasn't scarcity allocation by economics, but wealth allocation by politics.

So if it's partial automation, we should look at elasticity and see how many jobs will benefit, not just suffer.

And if it's full automation, humans-as-horses for medical and legal and writing and programming? Well, we should admit that's such a different future that it should be an argument about pensions or UBI, not about "keeping good jobs."

Expand full comment

Having spent a 40 year career in factory automation, I am always amazed at how shallow the jobs-versus-automation arguments are. Why not this: "We can only have more and better things if we automate..." So it's really just a choice. Stay the same or improve.

The precise components inside of our appliances and automobiles - whether internal combustion or EV - could not be manufactured to the precision and volume required for their performance without the automation of machine tools and robots. Do we want these things or not? This isn't just limited to "touch" labor. Many of the products we appreciate could not exist if they were designed by hand. The circuitry inside a microchip would not exist were it not for the design and modeling tools that assist engineers to iterate such complex alternatives. Not to mention that automation makes dangerous jobs safer.

Every form of automation ever invented just amplified the know-how of humans to do more and better things. I would love to see a study on the impact of removing the automation already in place to the benefit of labor!

There may be reasons to consider the social impacts of AI, just like there is for genetic cloning, just like there was for bringing moon rocks back to earth. But job loss isn't one of them.

Expand full comment
Apr 10, 2023Liked by Noah Smith

It turns out that basically all the excuses made for persistent unemployment during the 2010s from automation to skills gap to video games were just distracting from insufficiently expansionary monetary policy, who would’ve guessed that.

I’d love to see your take on NGDP targeting as from the limited amount I’ve seen it looks promising but I haven’t been exposed to any critiques.

Expand full comment

I was fortunate to get a taste of "robots take your job" while entering the infrastructure planning & engineering industry. One software patch in 1998 turned an hours-long task into a simple click of a box. I was of course relieved, but from then on, always aware that I needed to demonstrate value. It was clear much of transportation analysis could be done by software, but there would always be a need for a human to connect dots and interpret mobility.

In my industry, the people most worried about tech advancements are the ones content to just do-a-job rather than show-value.

Expand full comment

Good point about what "automation" really means. No to mention the method of subjective labelling whether job is to be automated. I think it would be better to have scale from 0 to 1 rather than a binary label.

But what is most important is the lack of studies that show how many jobs, of what quality will be created by automation. I wonder why research focuses on destruction part, not creation of jobs. Ultimately what matters is the net effect. It seems the reports are missing this. Why? Is it harder to estimate creation vs destruction? Or maybe destruction is hotter topic that guarantees more interest in study?

Expand full comment

Excellent writing Noah! 100% 👍🏾

Expand full comment

I'm not sure what planet you were on in the 1970s and 1980s, but automation nearly destroyed the Midwest. In the 1950s and 1960s, there were lots of well paying jobs. A decade or two later, the factories were obsolete and the newer ones required a lot fewer workers thanks to automation. Only a small fraction of those well paid workers ever got a job that paid as well. Most of them took a big pay cut and the communities collapsed. Those areas are still in terrible shape.

This happened with dock workers on the coasts as well. Containerization automated ship loading and unloading. Well paid dock work vanished. Places with strong unions made deals that cushioned the blow for a while, but those jobs are gone, automated out of existence. Waterfront communities still have not recovered, and those that have don't provide particularly good jobs.

"There set out, slowly, for a Different World,

At four, on winter mornings, different legs ...

You can't break eggs without making an omelette

- That's what they tell the eggs."

- Randall Jarrell

Expand full comment

The shift to services / knowledge work has stiffened cognitive competition in the American labor market. Will AI make that worse, requiring humans to have even more smarts in order to compete for a dwindling number of jobs? I suspect not, and there is a historical precedent for that optimism.

Prior to the industrial revolution, your economic utility and your physical prowess were correlated. In the 1700's, being a short, nerdy, man made life very hard. However this man is the biggest beneficiary of the steam engine. For the first time, you don't have to be physically strong to be productive. You have to be mentally alert to manage the machine, but the machine itself does the physical labor. Steam shovels absolutely did throw ditch-diggers out of work, but they also provided employment opportunities for men who weren't strong enough to dig ditches. I don't have any studies to back up this intuition, but I suspect many of the economic benefits of the industrial revolution landed on the nerdy, with most of the costs were born by the strong and previously productive.

What lessons for our current resolution? In our economy today, who are the equivalent of physically weak, 17th century nerds? Today, smart people have lots of opportunities, but the cognitively challenged, not so much. Could AI provide an answer to that? Just as steam created opportunities for the physically weaker to be productive, wouldn't AI create opportunities for the cognitively weaker today?


There's a guy that's not a particularly great writer or a particularly great illustrator, but he produces a saleable children's book with the help of AI text and art. To be fair, Ammaar Reshi is certainly no dummy -- using these tools today requires cognitive skills. But that won't always be the case. The first cars required detailed familiarity with their mechanics; your modern car can be driven thousands of miles without you giving its internals a second thought. The same will happen with AI tools, and when it does, the cognitively challenged will have all kinds of opportunities they are currently locked out of. What the steam engine did for physical ability, AI will do for cognitive ability.

Based on the industrial revolution model, I suspect the costs of AI will fall on those who are at the top of the cognitive (and economic) heap. In the industrial revolution, the physically strong were unable to maintain their place against the new technology, largely because they were not politically powerful. I already see evidence that the professional-managerial-class (bureaucrats, middle managers, HR flunkies, business analysts, lawyers) is building a backlash against AI, since it stands to hurt them the most. Considering the political power of the PMC as a group, it remains to be seen if they will be able to achieve what the Luddites could not: limit the new technological development to keep their place at the top of the food chain. The recent "halt AI now!" movement led by Musk and other tech folks may well be less altruistic than it appears.

Expand full comment

Depends how automation is implemented and used. And that depends on whether workers have any say in the implementation -- all very difficult to navigate well and humanely for all concerned, both labor and management.

Expand full comment

I hear you but, then how do you explain the 180 staff writers who were all fired in January at Buzzfeed which stated they would use ChatGPT to write their articles going forward, and have just a few editors to review those articles? Do you not think other companies will surely follow? No one offered those writers job placement, or re-skilling, and there were no jobs created in that process. A few editors got to keep their jobs. Corporations will always find a way to reduce overhead and people are usually first to go. How do you explain Walmart who just announced last week that 65% of its stores will be serviced by automation over the next three years and at the same time as that announcement, they laid off 2,000 employees? There were no jobs created or training provided. Those people are unemployed. If these are not clear, present day, real-world, examples of job displacement from AI/Automation then I don't know what is. When large companies everywhere start doing this, it's a problem. And they will. Because that's what corporations do. They decrease costs and increase profits. A surefire way to do that is with technology like AI and automation. Why wouldn't they do it is the question. Many of these people being fired cannot just go get trained for new technical jobs of the future. There are not enough tech workers for the jobs we have now, which at some point in the past were the jobs of the future. If people didn't choose to train for them then, why would they do it now? I say this not from a place of hating corporations. I've managed multiple businesses. It is what it is. Increase profits, decrease costs. Human workers are always in the line of fire when managing that. If a technology exists that can facilitate that, corporations will do it. Corporations won't keep employees out of the goodness of their hearts. It is what it is and we should prepare for it with comprehensive policy, educational opportunities, and some way to ensure people have the means to survive when displaced. (Some have suggested UBI for that but I haven't researched it enough to know if it'd actually work.)

Expand full comment

Until about 1980, productivity growth translated into shorter standard hours of work. That changed, not (at least not directly) because technology changed but because the balance of power in workplaces changed. That's turning around a bit, but obviously we need continued productivity growth to make shorter hours sustainable.

Expand full comment

All the more ammunition in the argument for universal basic income. There are enough resources for everyone, the problem is our concept of equitable distribution is toxic. It's not that people do not want to work, it's not even that our species has the means to make working optional, it's that we have created a culture afflicted by all the wrong values and many pointless fears. Like the fair of AI. I'm much more afraid of the current Congress, which is full of dishonest shitbirds.

Expand full comment

I agree. But isn't the better question what does AI-driven automation enable? There is a piece in the NYT this morning about AI displacing law jobs. The right question is do we need massive relative cost cuts in producing more lawsuits, legal briefs, etc?

Expand full comment

From the education sector: the elimination of jobs grading 3-page essays on the major themes of The Great Gatsby should have happened decades ago, which would have also eliminated the jobs of those who wrote them for pay. There will be new better jobs in higher ed but not "writing prompts." Ensuring factuality might be one.

Expand full comment

Although prime age employment rates are on the rise, the growing disparity caused by the decoupling of labor and capital returns induced by AI could have negative effects on labor force participation rates among prime age adults. This trend is evident even as employment rates continue to increase or remains near “historical lows”. Essentially these innovations could have deflationary effects on wages. Increased economic return does not equate to increased consumer spending power, as the last 50 or so years have demonstrated. A further decoupling could be apocalyptic.

Expand full comment

The correct way to worry about this is that a given change in technology reduces the market wages for a large number of kinds of people. That's what happened with the technologies that made globalization possible. More winners than losers (or at least the winnings were larger than the losses) but there were losers. With LLM's or AGI in general, there could be more losers or fewer.

I do not think we should claim to know.

Expand full comment