AGI skepticism: Tellers vs. toll booth workers

Artificial general intelligence (AGI) is a hot topic lately, and here’s what it assumes: When machines can do a job, humans get replaced. It’s intuitive. It’s obvious. And for a great many jobs, it’s completely wrong.

If you’re a leader in an organization establishing success criteria for AI projects, the real question isn’t whether AI can replace a human in a job; it’s whether a job is more like a bank teller or a toll booth worker. Those two bellwether occupations tell radically different stories about how automation replaced humans in the recent past, and a whole lot of jobs are going to follow the teller path rather than the toll booth worker path.

As a leader, you’d better be able to tell the difference.

Bank tellers, meet Jevons Paradox

When ATMs debuted in the 1970s, many described them as the poster child for the technology that eliminates jobs. The logic was airtight: Why employ humans to dispense cash and take deposits when a machine can do it 24/7? But a different story emerges about bank teller employment in the US:

1970: 268,300 (per Integrated Public Use Microdata Series USA)
2006: 608,000 (per US Bureau of Labor Statistics)
2024: 347,400 (per US Bureau of Labor Statistics)

Wait . . . what?

The Brookings Institution did a study on the impact of ATMs on bank tellers in 2019 and while ATMs were supposed to kill bank tellers as a career, instead, the number of people employed as bank tellers doubled for a while. Then, over time, as online banking became more popular, teller employment numbers reached a steady state larger than their 1970 level, even as banks rolled out over 500,000 ATMs across the country.

This is a phenomenon economists call Jevons Paradox, named after a 19th-century economist who noticed something weird about coal consumption. William Stanley Jevons observed that when James Watt’s steam engine made coal burning dramatically more efficient, Britain didn’t use less coal; they used more. A lot more. Because suddenly, coal was so efficient that people found a thousand new applications for it.

Banking went through exactly the same pattern with tellers and ATMs.

The International Monetary Fund (IMF) did a study on the teller/ATM phenomenon in 2015 and found that between 1988 and 2004, ATMs cut the number of tellers needed per branch from about 20 down to 13. That’s a massive efficiency gain, with 35% fewer tellers per location. Bank executives saw this and thought, “Great, we can save money.” But here’s where Jevons Paradox kicked in: When each branch became more efficient to operate, banks realized they could open branches everywhere. Urban markets saw branch density increase by 43%.

Suddenly, banking became more accessible, more convenient, more ubiquitous. More branches meant more customers. More customers meant more transactions, and not just the simple ones ATMs handled, but the complex ones that required human expertise. The total volume of banking services exploded precisely because the routine transactions became so efficient to process.

And here’s the part that really matters for leaders thinking about AI implementations: The tellers weren’t doing the same work.

The routine stuff like dispensing cash, taking deposits and checking balances all moved to the machines. What was left were the complicated, nuanced tasks that required human beings and higher-order tasks. Tellers became what the IMF study called “relationship banking team members.” They handled the small business customer with complex needs. They sold financial products. They dealt with the situations that required judgment, empathy and trust.

The technology didn’t eliminate the job — it eliminated the routine parts of the job, made the service so much more efficient and better that demand exploded, and elevated what humans did to higher-value work.

The skills upgraded. The wages went up. The job evolved rather than disappeared.

The toll booth counterexample

Toll booth workers tell a different story, one where automation did replace humans.

It used to be that you’d approach a toll booth and be greeted by a person, who you’d hand physical money in exchange for permission to pass through the toll plaza. The first innovation was “the bucket,” where you’d toss change into a receptacle and the amount was verified mechanically, allowing you to pass. Then, RFID pads placed in your vehicle would get read electronically and associate your passing with an account you previously established with the entity managing the toll plaza. Finally, even the RFID pads were replaced by AI-enhanced video camera license plate readers.

All of that innovation impacted toll booth worker employment in a very different way than ATMs affected bank tellers. According to a 2015 GovTech article:

• Golden Gate Bridge: 28 full-time toll collector jobs eliminated
• Florida Turnpike: 190 collectors and supervisors laid off
• Massachusetts Turnpike: 400 positions removed

Other state and local transportation organizations followed suit, and toll collection has essentially been eliminated as an occupation.

Tellers vs. toll booth workers: What’s the difference?

What’s the difference between these two professions? Why did one not only survive, but thrive, while the other had a different fate?

Bank tellers survived because:

• Automation reduced costs, enabling market expansion
• New branches meant aggregate demand for tellers increased
• Human skills adapted to handle the non-automated, high-value interactions
• The industry grew, creating room for both machines AND humans

Toll booth workers disappeared because:

• Electronic tolling completely replaced the core function
• No market expansion occurred  —  the same roads served the same drivers
• There was no complementary human role to evolve into
• Cost savings went to infrastructure, not service expansion

The AI future: Not a zero-sum game

AI’s future is not a zero-sum game where every single job imaginable becomes uninhabitable by human beings. Many, many jobs more closely resemble bank tellers than toll collectors and this is why full AGI is likely a fantasy. Instead, AI will:

1. Reduce the cost of cognitive tasks — making it more efficient to produce analysis, content, code, designs.
2. Enable expansion — companies that couldn’t afford 10 analysts can now afford the enhanced productivity of 10 AI-augmented analysts doing the work of 50.
3. Create new complementary roles — just as tellers became relationship managers, workers will focus on judgment, creativity, strategy and the human elements AI can’t easily replicate.
4. Increase total demand — when work gets more efficient, we’ll consume more of it, not less.

The bottom line

The AGI replacement narrative assumes a fixed amount of work to be done. But economic history shows that when technology makes something more efficient, it’s not about doing the same amount of work with fewer people — real leaders do vastly more with the same or even more people.

For most jobs, AI isn’t a replacement for humans. Instead, AI improves productivity and expands the scope of what’s economically possible. This then creates new demands for human judgment, creativity and relationship skills as business opportunities that weren’t possible before spark different levels of activity not previously thought possible.

We’re not heading toward a world without workers or with fewer workers, even. We’re heading toward a world where workers accomplish more, and where the economy expands based on that enhanced productivity.

Stop thinking of AI as a pathway to eliminate people. Start thinking about AI as a productivity multiplier that makes new things possible with the people you already have.

This article is published as part of the Foundry Expert Contributor Network.
Want to join?