Will AI Shrink Total Jobs? Demand vs Productivity
A recent V2EX thread asked a question many people feel but few unpack cleanly: if we push AI hard, will total jobs in society shrink?
The OP’s answer is blunt: no. Not because of the slogan “AI creates new jobs,” but because of a near-accounting claim—jobs follow demand, not productivity.
Here is the distilled argument, plus where it helps and where it breaks.
Nail the definition first
Productivity, in the post, means output per unit of time.
Raise productivity and one worker makes more stuff in the same hour. That immediately implies a counterintuitive point:
Higher productivity is not the direct cause of more jobs.
If one worker used to make 10 units per hour and you needed 10 workers, and later one worker makes 100, you only need one worker for the same demand. Headcount falls.
So why did employment rise for so long as productivity rose?
The OP’s move: don’t confuse sequence with causation. Jobs rise because demand rises.
One table is enough
| Output per worker | Demand | Jobs |
|---|---|---|
| 1 unit | 10 | 10 |
| 1 unit | 100 | 100 |
| 10 units | 100 | 10 |
| 5 units | 100 | 20 |
Same productivity, 10× demand → 10× jobs. Same demand, 10× productivity → jobs cut to 1/10.
Roughly:
Jobs ≈ total demand / output per worker
Jobs grow only when productivity growth fails to catch demand growth. Hence the punchline:
Productivity growth cannot keep up with demand growth.
Why does demand run so fast?
Three engines in the post:
- Population — more people, more basics.
- Desire — from one village phone to latest phones plus chargers, tablets, laptops, ergonomic chairs… desire rarely saturates.
- Social pressure — when peers consume a lifestyle, opting out is costly.
Then the leap—a bold conjecture:
Human demand growth forever outruns productivity growth.
If jobs track demand, and demand forever wins, AI will not shrink total jobs.
The OP admits the whole proof hangs on that “forever.”
The sharpest cuts in the comments
1. Desire ≠ effective demand
Can people afford their desires? If desire automatically became purchasing power, chronic deflation and overcapacity would be hard to explain. Money can be printed; distribution still decides whose demand counts.
2. Demand is a function of price
Everyone may want a trip to the Moon, but price rations absurd wants. With fixed total capacity, unlimited nominal desire does not become unlimited payable demand.
A useful amendment also showed up: higher productivity can create demand by lowering prices.
3. “Forever” struggles with multi-decade counterexamples
Japan’s lost decades and post-pandemic deflation look like stretches where demand growth lagged productivity growth. Saying “people are still alive / unemployment didn’t explode” explains social survival, not that the mechanism held.
4. Aggregates can hold while structure still hurts
Even if total employment doesn’t collapse, entry-level collapse, thinner apprenticeship paths, and a falling labor share can still hurt badly. Total jobs, job structure, and income distribution are three different layers. This thread mostly covers the first.
5. A longer-horizon compromise
Pull the timeline out and you may get a boring equilibrium: AI does AI work, humans do human work, and humans keep some jobs simply because they remain cheaper than AI.
A more useful reading for builders
You don’t need to pick “AI ends all work” or “demand always saves us.” Watch three things:
- Which layer of effective demand your product creates — time saved, status, social belonging, entertainment, or new “wasteful” desires.
- Tools that only cut cost without expanding demand squeeze peers; tools that unlock new payable scenes expand the pie.
- When debating AI and jobs, separate totals / structure / distribution. Mixing them produces noise.
Keep the table. At minimum it stops you from translating “productivity ↑” directly into “jobs ↑” or “jobs ↓.” The real fight is whether demand—and whose demand—can keep up.
Comments