Evidence of AI-Induced Unemployment Starts to Emerge

Microsoft recently released its “Future of Work” report.

I’ll examine that in detail in a future post, but I want to first share emerging evidence of job loss due to AI.  Most people don’t “believe it until they see it,” so I’ll offer this as both proof and hope that educators might take this future of jobs report more seriously.

*Evidence of emerging job loss

There has been an emerging discussion of potential job losses in AI.  The “job loss” debaters generally fall into four camps.

*AI will replace all or nearly all jobs, as most new jobs that are created by AI can also be done by machines.
*AI will replace a good number of current jobs, but new jobs will be created.
*AI can’t do “higher order” thinking jobs; we humans are special.
*AI can do many “tasks,” but it can’t (yet?) do most tasks people do, so their jobs will be protected (though perhaps at a lower wage).

With those thoughts in mind, here are some of the places where we can see at least some job loss due to growing AI abilities.

Google. Google is laying off 5% of its workers—12,000 people—because it can automate its ad sales business. It’s not because revenue is down; it is up, and it is up by billions!

Klarna. Klarna is not laying anyone off, but they are freezing non-engineer hiring, as all but the engineering jobs can be done with AI.

Deloitte. The Deloitte situation is complicated because the headlines will read, “Deloitte uses AI to save jobs.” Yes, Deloitte is using generative AI to eliminate repetitive and time-consuming tasks traditionally assigned to junior staff, such as preparing documents for internal meetings or collecting large amounts of data for client presentations,” but they are trying to use AI to help find new roles for people. Like many, they are saying they “overhired” (sorry, my “oopsie.”) They previously doubled revenue and tripled staffing. They are no longer going to do that, as they can use AI for the work above and to help them maximize the use of the workers they have. There will be fewer jobs in the future.

Translators. There is some emerging evidence of declining job opportunities and wages for translators and a bit for writers. I’m sure we will soon see a much greater effect on translators, as Meta’s new technology enables translation in more than 100 languages with less than 2-second latency, which is less than a translator. This can now be done while wearing Meta’s classes (you’ll hear me in your language, and I’ll hear you in my language). It also means kids won’t want to learn languages.

Customer service and call centers.  Customer service and call center jobs are being eliminated in Australia.  Tools such as air.ai make it easy to replace customer service agents, and some studies show the cost of an AI agent in a call center is less than 5% of the cost of a human agent and leads to more satisfactory resolutions of issues.  This is one concrete area where we will likely see work eliminated over the next year. These are real people.

Amazon. “A factory planning to pump out 10,000 two-legged robots a year is taking shape in Salem, Oregon—the better to help Amazon and other giant companies with dangerous hauling, lifting and moving.” Sadly, Amazon is employing tens of thousands of people every year in such dangerous jobs (if this factory operates for 10 years and never increases its capacity, it will pump out 100,000 robots). But maybe they plan on replacing non-dangerous jobs as well?  LOTS of robots are coming. They can all do factory work.

McDonald’s.  They aren’t unemployed yet, but McDonald’s is partnering with Google to deploy generative AI beginning in 2024, when “thousands” of stores will get hardware and software upgrades. The Verge: At least one outcome will be — according to the company — “hotter, fresher food” for customers. Large orders of fries delivered hot, thanks to large language models? It’s not completely clear what that means, but we can read between the lines: expect more AI-driven automation at a drive-through near you in the coming years. The updated system will help managers “quickly spot and enact solutions to reduce business disruptions.”

The McDonald’s statement also skirts the question of AI replacing human workers, mentioning only that the system should “reduce complexity” for store crews and that it will “power exciting new experiences for crew and customers.” (Maybe the crews will dance instead of work the drive-through.). Wendy’s — another Google Cloud AI customer — held a similar line when it introduced AI ordering tests earlier this year Still, there’s a whiff of “robots replacing human workers” in the air. The generative AI push also coincides with the deployment of a new “bespoke” operating system to unify the experience across the McDonald’s mobile app and store kiosks. Fake Drake drive-through cashiers? Read the full story in The Verge.

On a related note, the Future Skills Organization, an Australian Government initiative, recently released a report arguing that knowledge workers, including those with undergraduate and graduate degrees, are the ones most at risk of AI.

There is a lot of bullshit out there. A lot of these “future of work” reports say things like jobs will be “affected” and "impacted,” and there may be “profound implications.” Terms like these are useful because they both cover the author’s ass (what if this AI stuff is all made up) and they are good corporate-speak that lulls people into thinking they are still going to keep their jobs. Will all of those empowered former drive-through workers get to dance on the tables?

There is a common line that workers will be “kept in the loop,” but as Geoff Hinton quipped

"I've learned one thing about medicine, which is just like other academics: doctors have egos, and saying (that) this stuff is going to replace them is not the right move. The right move is to say, 'It's going to be very good at giving second opinions, but the doctor's still going to be in charge.' And that's clearly the way to sell things, and that's fine."

Bullshitting each other is a uniquely human skill. AI may be able to help us come up with bullshit, but it may never “believe” it.

Broader Considerations

AI doesn’t have to be as good as a human employee/consultant. This is one reason that the argument of the “human add” is not very good. Employees and consultants cost a lot of money. If an AI gets in the ballpark of what an employee or consultant can do, that’s good enough. One user noted on a message board: “The company I work for used to spend money on consultants. But now we just go to ChatGPT and type "Give me a consulting report for a typical software company". If we compare the reports from the "Big Four"] to the output of ChatGPT, the quality is comparable. Close enough that it is no longer sensible to pay for the advice.”  Another user noted: “core problems for the Big Four is that their reports are pretty bad, outdated, meaningless and bereft of insight.”  And see below – AI can do competitor research.

Copilot now, autopilot later. AI is mostly a co-pilot now (ask the attorney who submitted an AI-generated brief without first reviewing it or ask those injured in Teslas while relying on autopilot rather than co-pilot), but AI will improve, and autopilot will start to work. 

One possibility is that time savings and seamless application will hold greater importance than quality improvement for the majority of tasks. Another is that the initial focus will be on augmentation, followed by automation (Huang and Rust, 2018). One way this might take shape is through an augmentation phase where jobs first become more precarious (e.g., writers becoming freelancers) before transitioning to full automation.”(Eloudou, 2023)

There is some (quasi) autopilot now. In some professions, there is barely a difference between AI-only and AI+humans. In radiology, AI+ 1 radiologist are better than AI+2 radiologists.

More than 55,500 Swedish women participated in a study examining the effectiveness of AI in mammography. The researchers found that the current European approach of having two radiologists read every mammogram catches 4% fewer cancers — and generates more false positives — than a single radiologist working with an AI tool known as Insight MMG. And although AI alone isn’t quite as effective as a human-AI pairing, it was as accurate as two radiologists.

While more research is needed, other studies have reached similar conclusions. And while adding a human radiologist may be marginally helpful now, it probably won’t be as the technology improves.

Wage suppression! Even if it doesn’t cause unemployment, it suppresses wages because machines can do an important part of your job. In many cases, it means more people can do it and you are worth less.

The Future

This is going to get worse.  A good example of emerging technologies that will likely create further disruption are agents that can reason and plan, which will allow them to do such things as opposition research on competitor companies.

A new paper in Nature demonstrates that AI’s can carry out autonomous chemical research. Sorry, but when AIs start to act even quasi-autonomously, you just need fewer people.  

According to a new McKinsey report, “Gateway” jobs -- customer support, production supervision, office support, and even coding – that lead to higher pay over the long term are vulnerable to replacement by AI. The report projects the replacement of many of these roles by 2060, but it only looks at generative AI, not other forms of AI that will be developed or quantum computing, which will likely exist by 2060 and probably much sooner.  

In the next 20-25 years, AI can replace approximately 80% of economically valuable jobs.

Skepticism. No matter how quickly this technology continues to advance, there are still skeptics about its evolving capabilities. You can close your eyes and think that, but remember that AI scientists did figure out how to do something that no one thought we ever could: replicate the nuance of human language. And the capabilities of AI improve every day.

AGI is defined as worker replacement. Remember that OpenAI defines AGI as the ability to do human work as well or better than most humans. They say we are getting close to AGI. Many dispute this as an appropriate definition of AGI, but when OpenAI says they have reached AGI, they will be saying they have systems that can do nearly all jobs done by humans as well or better than humans. You can say that isn’t AGI, but you can’t say the system can’t do your job.

“You are Ignoring Job Gains”

From what I can tell, job gains related to AI have mostly been in the areas of data science and machine learning. Customer service agents, factory workers, and translators who upskill in those areas will likely find employment. We’ll see how that plays out.

Also, keep in mind that AI can do more and more jobs, so when new companies are created because of AI, far fewer workers will be required. In fact, who wouldn’t want to start a business that has low labor costs? It seems that investors will push for such things.

At a minimum, most do agree that the rate of job destruction will soon surpass the rate of job creation. Historically, even when the rate of job destruction was much lower (beginning of the Industrial Revolution), there was a period of high unemployment until new jobs were created.

But AI Will Raise Productivity!

Yes, AI will likely raise productivity, but it does not inevitably increase marginal productivity: the amount of productivity each worker gains by serving more customers. In the case of many past productivity gains, and especially in the case of AI, there is often no value to the technology gain; it simply replaces the worker. Darren Acemoglu, an Institute Professor at the Massachusetts Institute of Technology, and Simon Johnson, the Ronald A. Kurtz Professor of Entrepreneurship at MIT Sloan and a former IMF chief economist, explain:

Contrary to popular belief, productivity growth need not translate into higher demand for workers. The standard definition of productivity is “average output per worker”—total output divided by total employment. The hope is that as output per worker grows, so will the willingness of businesses to hire people.

But employers are not motivated to increase hiring based on average output per worker. Rather, what matters to companies is marginal productivity—the additional contribution that one more worker brings by increasing production or by serving more customers. The notion of marginal productivity is distinct from output or revenue per worker; output per worker may increase while marginal productivity remains constant or even declines.

Many new technologies, such as industrial robots, expand the set of tasks performed by machines and algorithms, displacing workers. Automation raises average productivity but does not increase, and in fact may reduce, worker marginal productivity. Over the past four decades, automation has raised productivity and multiplied corporate profits, but it has not led to shared prosperity in industrial countries.

Replacing workers with machines is not the only way to improve economic efficiency—and history has proved this, as we describe in our recent book, Power and Progress. Rather than automating work, some innovations boost how much individuals contribute to production. For example, new software tools that aid car mechanics and enable greater precision can increase worker marginal productivity. This is completely different from installing industrial robots with the goal of replacing people.

For example, self-checkout kiosks in grocery stores bring limited productivity benefits because they merely shift the work of scanning items from employees to customers. When stores introduce self-checkout kiosks, fewer cashiers are employed, but there is no major productivity boost to stimulate the creation of new jobs elsewhere. Groceries do not become much cheaper, there is no expansion in food production, and shoppers do not live differently.

Even nontrivial productivity gains from automation can be offset when they are not accompanied by new tasks. For example, in the American Midwest, the rapid adoption of robots has contributed to mass layoffs and ultimately prolonged regional decline.

The situation is similarly troubling for workers when new technologies focus on surveillance. Increased monitoring of workers may lead to some small improvements in productivity, but its main function is to extract more effort from workers.

The dominant intellectual paradigm in today’s digital tech sector also favors the automation path. A major focus of AI research is to attain human parity in a vast range of cognitive tasks and, more generally, to achieve artificial general intelligence that mimics and surpasses human capabilities. This intellectual focus encourages automation rather than the development of human-complementary technologies.

What Hope is there?

We are in for some rough times ahead, and that is why many of the leading AI scientists and thinkers have been calling for the creation of social welfare funds, with many emphasizing some form of universal basic income.

As educators, we can help students learn AI skills so they can work with AI tools in the workplace and develop broader skills (communication, collaboration, etc.) that will enable them to work with both machines and humans in a way that enables them to take advantage of AI systems. We (with Dr. Sabba Quidwai) cover this in detail in our report.

Eventually, as noted by Anton Korinek, there is a good chance that AI systems will be able to develop these broader skills (and there are aggressive efforts to get them to do so), but if we develop these skills in humans in a way that enables them to collaborate with machines, their odds of remaining employed will improve.

Even in this scenario, will everyone remain employed? Absolutely not, but those with type A personalities who have durable skills and can work with machines have a good chance of remaining employed and will have the best chance of starting their own successful business. For most, the future will be “hussle.”

This is also something we need to start talking about with our students and each other.