Michael Gingras

Comparative Advantage Can’t Save Us From the Transition

I’ve been reflecting on Noahpinion’s blog post arguing that, even if AI becomes superhuman at everything, humans will still have jobs. Why? Comparative advantages.

The argument goes like this: even if AI is better than humans at all tasks (i.e. has competitive or absolute advantage), humans can still do valuable work thanks to comparative advantage. That is, AI will be so good — and compute so scarce — that we’ll want to use it only on the highest-value tasks. As a result, lower-value tasks will fall to humans by default, keeping us economically relevant.

I think this framing is helpful in expanding how we think about AI and labor. It challenges the overly simplistic “AI replaces humans” narrative and points to subtler dynamics of resource allocation, opportunity cost, and tradeoffs. But the argument also carries some huge assumptions — and glosses over the very real turbulence that will occur on the path to this supposed stable equilibrium.

Let’s break it down.

Compute Is Not (Yet) That Scarce

The whole comparative advantage argument hinges on the idea that AI compute is limited and will always be allocated to the most valuable uses. This constraint is what keeps humans in the loop.

But today, compute doesn’t feel particularly scarce — at least not in the way the post implies. You can spin up vast amounts of general-purpose compute on AWS. Yes, frontier models require specialized GPUs and data center infrastructure, but those are growing rapidly. For most economic tasks, the constraint is not “we don’t have enough compute,” but “we haven’t figured out how to use AI productively and reliably in this domain.”

To believe that humans will retain economic value thanks to comparative advantage, you have to buy into a future where compute becomes so valuable — and demand for AI so infinite — that every marginal gigaflop is precious. That’s not obviously wrong, but it’s not obviously right either. Diminishing returns, satiation, and other practical bottlenecks (e.g. data, integration, regulation) may cap the value of AI long before we hit a compute-scarce singularity.

The “$300 Haircut” Is Trickle-Down Economics in Disguise

Another issue is the idea that even if AI creates massive inequality, it’ll be okay because the ultra-rich will shower money on human-provided services. If Sam Altman becomes a quadrillionaire, maybe he’ll pay $300 for a haircut, and that money will “trickle down” to the barber.

But let’s be real: this is just a high-tech version of trickle-down economics — the belief that concentrated wealth at the top will eventually benefit the rest of society through consumption. History isn’t kind to that assumption.

There’s no guarantee that the super-wealthy will spend meaningfully on human labor. They might instead double down on capital — buying robot chefs, AI tutors, and virtual therapists, all owned and operated by other capital holders. Meanwhile, land, energy, and goods could inflate to levels that human wage workers — even those “saved” by comparative advantage — simply can’t afford. The net result? Jobs may persist, but the quality of life could stagnate or even fall for most people.

So while comparative advantage might mathematically justify human employment, it doesn’t guarantee a fair or fulfilling economy.

Comparative Advantage Doesn’t Help During the Transition

This is, in my view, the most important blind spot in the original post.

Even if comparative advantage eventually guarantees that some human labor survives in the long run, it says nothing about the disruption that will happen in the short and medium term. On the path to an AI-dominated economy, we’re likely to experience waves of displacement and job churn — and comparative advantage doesn’t prevent any of that.

It’s not hard to imagine:

AI becomes good enough to replace radiologists, writers, customer service reps, and junior developers.
Millions of workers are displaced.
The labor market can’t keep up with the need to reskill millions of workers.
Inequality soars as AI productivity flows to a narrow set of owners.
Cultural dislocation, political backlash, and social instability rise.

And during this period, the “safety net” of comparative advantage hasn’t kicked in — because the opportunity cost of AI hasn’t risen high enough yet. The theory might eventually reallocate humans to lower-opportunity-cost tasks, but the pain happens first.

Ownership Is Everything

Underlying all of this is a question the original post mostly skirts: who owns the AI?

If humans own the AI infrastructure — through governments, cooperatives, open-source models, or broadly distributed capital — then comparative advantage becomes a mechanism for creating shared prosperity.

But if AI is owned by a small number of private actors, then comparative advantage becomes cold comfort. It just means we’re still allowed to work while the new feudal lords reap 99% of the surplus.

The real question isn’t will humans still work? It’s what kind of society will we live in? Will labor exist as dignified participation in a flourishing economy — or as the luxury service class to a handful of AI capitalists?

The Right Conversation

I appreciate the original post for surfacing an important and under-discussed point: humans don’t need to be better than AI to remain economically relevant. But comparative advantage is not a magic shield. It doesn’t prevent displacement. It doesn’t guarantee fairness. And it doesn’t ensure that humans will want to work — or be well-rewarded when they do.

The real questions we should be asking are:

How do we own and govern AI collectively?
How do we ensure the gains from AI are widely distributed?
How do we manage the transition — not just the endpoint?

Because even if economic theory tells us that human jobs will survive, we still have to decide whether the society they exist in is one we actually want to live in.