Optimists love the ATM story: banks installed ATMs, but teller employment grew because cheaper branches meant more branches. But is this the right analogy for AI? A careful look at automation history reveals a more nuanced β and less reassuring β picture.
Five Famous Automation Transitions
1. ATMs and Bank Tellers (1970β2010)
| Period | ATMs Installed | Teller Employment | What Happened |
|---|---|---|---|
| 1970 | ~1,000 | 300,000 | ATMs begin deployment |
| 1990 | 80,000 | 500,000 | Cheaper branches β more branches β more tellers |
| 2000 | 325,000 | 530,000 | Peak teller employment |
| 2010 | 425,000 | 560,000 | Still growing (pre-mobile banking) |
| 2024 | 470,000 | 450,000 | Finally declining β mobile banking + AI |
The lesson: ATMs didn't kill tellers immediately. But the story isn't over β teller employment is now declining as digital banking + AI finally completes what ATMs started. The full displacement just took 50 years.
2. Agricultural Mechanization (1900β1970)
| Year | Farm Workers (millions) | % of Workforce | What Happened |
|---|---|---|---|
| 1900 | 10.9 | 38% | Pre-mechanization |
| 1940 | 9.0 | 17% | Tractors replace horses and hand labor |
| 1970 | 3.5 | 4% | Full mechanization |
| 2024 | 2.6 | 1.3% | Precision agriculture beginning |
The lesson: Agriculture is the clearest case of automation destroying an entire job category. 8+ million jobs eliminated over 70 years. Workers migrated to manufacturing β but that option required physically moving and entire communities were devastated.
3. Manufacturing Automation (1980β2020)
- U.S. manufacturing output doubled from 1980 to 2020
- Manufacturing employment dropped 37% β from 19.6M to 12.3M jobs
- Entire regions (Rust Belt) experienced decades of economic decline
- Workers were told to "retrain" β most couldn't or didn't
- Political consequences: communities feeling left behind fueled populist movements
4. Self-Checkout (2000βpresent)
- Cashier employment has been remarkably stable at ~3.3M despite widespread self-checkout
- BUT: cashier wages have stagnated (real decline when adjusted for inflation)
- Hours per cashier have decreased β more part-time, fewer full-time positions
- Self-checkout is now being combined with AI vision systems β the next phase may be different
5. Telephone Operators (1920β1980)
| Year | Operators | What Happened |
|---|---|---|
| 1920 | 200,000 | Peak employment β every call hand-connected |
| 1940 | 350,000 | Growing demand outpaced automation |
| 1960 | 250,000 | Direct-dial becoming standard |
| 1980 | 80,000 | Mostly eliminated |
| 2024 | ~5,000 | Effectively extinct |
What the Patterns Tell Us
| Historical Pattern | Applies to AI? | Why / Why Not |
|---|---|---|
| Automation creates new jobs in the same sector | Partially | AI creates some roles (prompt engineers, AI trainers) but far fewer than it displaces |
| Workers migrate to adjacent sectors | Harder | AI affects all knowledge sectors simultaneously; fewer "adjacent" safe havens |
| Displacement takes decades | Unlikely | AI adoption speed is 10β100x faster than mechanical automation |
| Productivity gains create demand | Maybe | Depends on whether companies hire more workers or pocket the savings |
| Government programs ease transition | Uncertain | No programs exist at the scale needed; political will unclear |
The Uncomfortable Conclusion
History shows that automation always displaced workers in the short-to-medium term, even when it eventually created more jobs. The optimistic long-run outcome (more jobs, higher wages) required:
- Decades of adjustment time β AI isn't giving us that
- New sectors to absorb workers β less clear where displaced knowledge workers go
- Massive public investment β the GI Bill, interstate highways, public universities
- Political stability β communities that felt abandoned turned to extremism
The ATM story is comforting but potentially misleading. The manufacturing story β where automation brought immense national wealth but devastated specific communities for generations β may be the better analogy. The question is whether we learn from that history or repeat it.