AI Is Advancing Fast, But Human Labor Isn't Going Anywhere

Dave White posted a fascinating thread about testing OpenAI's new o3's mathematical capabilities.

The results are pretty wild - we're watching AI go from struggling with complex math problems for years, then suddenly being able to solve them well. This is the kind of rapid progression that can keep you up at night (and Dave says he’s losing sleep over it). But we should be careful about the inferences we make from this progress. When things change this fast, it's easy to focus on what's improving rapidly rather than the slower-moving constraints that might matter more.

In this post, I want to explore what this progress means for the economy and labor markets. After all, that's where many people's immediate concerns lie. While some worry about existential risk from AI, the remarkable progress we've seen in AI alignment alongside capability gains suggests this might be less of a concern than labor market disruption - but that's a topic for a different post.

We're About to Find Out How Much Intelligence Actually Matters

There's this implicit model floating around that labor value is dependent on intelligence. Intelligence certainly matters! But I think we're about to see exactly how much. Every mundane thing that separates human labor from what machines can do is a potential source of competitive advantage for human labor. Just like how calculators didn't eliminate mathematicians but highlighted the importance of mathematical intuition and problem formulation, AI might end up revealing which human traits matter most - traits we might currently undervalue. If I had to bet, I would bet that human labor will persist for a long time, perhaps indefinitely.

Why Your Job Isn't Going Anywhere (At Least Not As Fast As You Think)

Look, it’s pretty obvious that these developments will have considerable effects on labor markets in the long run, but there are good reasons to believe it’s going to take longer than you think.

First off, there's a whole mess of institutional friction. AI won't be able to obtain professional licenses. Humans won't trust AI output relative to human outputs, and that's going to be hard to change (for example, you cannot sue an AI if they mess up).

Then there's the really interesting bit: it's not clear our society is currently intelligence limited. There are lots of places in our institutions where more intelligence isn't necessarily going to improve outcomes, at least not in some immediate linear way.

If intelligence becomes cheap, literally every other necessary part of production becomes relatively expensive. Want to know why AI is "making art and writing poems" rather than doing laundry and the dishes? Because our actual bodies are an essential part of our production function.

It’s not hard to imagine the mental equivalents of this — aspects of how we think, problem-solve, collaborate, reach consensus, and so on, that are essential to how humans create value that have been relatively mundane, but which are not easily replicated by AI. It’s shocking to see this progress move as fast as it has, but even with existing capabilities, is there anyone who has been clearly replaced by AI? I haven’t seen a real example (although for sure it will happen). But I have seen many, many anecdotes from people on how incorporating these technologies into their work has made them feel wildly more productive.

A Lot of "Progress" Is Going to Get Wasted

But even if AI can replicate all of what a human can do in a particular field, there’s no guarantee it will actually improve productivity. A lot of it will be "wasted." A lot already is! Consider large firms that have huge budgets for "intelligence" that have bad software. Why? Incentives, organizational structure, market structure can all impact the use of intelligence. Why use AI to improve software when you could use AI to win an internal political battle?

We're Still Waiting to See Real Impact

Despite OpenAI having over a hundred million users and considerable adoption in the software industry, we still don't see dramatic impacts in the demand for software engineers [FRED graph]. Job postings for software engineers are going down, but are fairly stable after a huge run up during the pandemic.

(Source: https://fred.stlouisfed.org/series/IHLIDXUSTPSOFTDEVE)

And, despite seeing exponential growth in AI capabilities on test problems, productivity growth remains relatively flat. This shouldn't surprise us - while pure computational power might grow exponentially, coordinating all the factors needed for actual economic production is much harder. It's one thing to solve a math problem faster; it's another to reorganize a company, retrain workers, redesign processes, and overcome institutional inertia - all while maintaining trust and meeting regulatory requirements.

We've seen this pattern before. The internet and information technology broadly had a similar problem—it didn't really show up in the productivity numbers, growth just continued more or less as it had before. One way to look at that is that despite all the dramatic advances in the technology itself, the actual impact on economic growth follows a more continuous path. Perhaps AI will progress the same way - revolutionary in capability but evolutionary in economic impact.

So What Does This All Mean?

No doubt we are in for a period of lots of change. This will impact people in different ways. But we should still expect people to have jobs in 20 years. There will still be software developers (but maybe fewer). If we are lucky, we will get an increase in productivity that we can use to make lives better.

And maybe the existence of a different class of minds will make us appreciate humans more. Human beings have survived in an extraordinarily adversarial environment for millions of years. We are extraordinarily adaptable. With every increase in AI capabilities, we discover some new subtle capability of people that was there all along. Who knew? This makes me appreciate humanity more, and feel more solidarity. There's something about humanity as the underdog in a vast and dangerous universe that I'm rooting for - maybe AI will too.