Phew! AI Won't Take Your Job Just Yet, MIT Data Finds — But There's A Caveat

With fears of job displacement by artificial intelligence spreading far and wide, workers everywhere are trying to predict whether AI will take their jobs, and what to do about it.

Yet, researchers at the Massachusetts Institute of Technology (MIT) believe the rate of transformation will be slower than expected, as it's still not economically feasible for companies to replace employees with an AI system.

The researchers developed a model to calculate the cost of building an AI system that would replace a worker's tasks. They used it to understand whether it would make sense for the company to replace that worker with a machine. 

The conclusion: human labor is still cheaper in most cases.

Job Displacement By AI Is A Valid Concern

Disruptions to the labor markets rank high amongst concerns stemming from the rapid development of AI.

In May of last year, a survey commissioned by Microsoft Corp. MSFT found 49% of workers were worried that AI would take their jobs. That fear reached 60% of workers who are already using AI to complement their daily tasks, as per a separate survey from December.

Their worries are not unfounded.

All through 2023, news reports and financial analyses consistently predicted that the impact of AI in the workplace would be substantial. A Goldman Sachs report from last year put at least 300 million full-time jobs at risk of automation by generative AI.

Earlier this month, the International Monetary Fund warned that at least 40% of jobs globally will be affected by AI, worsening income inequality and social tensions.

AI Won't Replace Jobs So Quickly, Study Finds

A new paper by MIT’s Computer Science & Artificial Intelligence Laboratory argued most concerns about AI taking people's jobs focus on the technological capabilities of artificial intelligence.

These predictions tend to assume that if a job can be automated, it will be, but fail to take into consideration the economic feasibility of going through with that automation.

While workplace automation continues on a steady course, the researchers argued at its current level, AI technology continues to be more expensive for employers than human labor in at least three-quarters of all positions at risk of being replaced.

Looking at the costs of generating and maintaining an AI system, the group of researchers led by Neil Thompson wrote only 23% of positions that could be replaced with AI are attractive to automate from a financial standpoint.

A key question for the researchers was not whether AI would transform the workplace, but how fast this transformation would occur. A slower shift would allow for less disruption to people's lives, their job security and by extension to the overall economy.

"AI job displacement will be substantial, but also gradual," wrote the researchers. This meant there was a window to mitigate the negative impacts of AI in the workplace with policy and worker retraining.

Another key issue in analyzing the impacts of AI in the workplace was learning to distinguish between full-task automation, which resulted in job losses, with partial automation, which increased productivity.

The MIT model was not comprehensive. It used computer vision — the branch of AI that can recognize and understand the world visually — as a case to understand how AI could impact the workplace.

The researchers chose to work with computer vision, instead of large-language models such as ChatGPT, because it's an older technology and there's more data available to calculate its performance and economic feasibility.

AI-as-a-Service Could Change The Game

The realization building and maintaining AI systems is currently more expensive than hiring human labor can be made clear with the simple example of a bakery.

About 6% of a baker's tasks involve visually inspecting ingredients to make sure that they haven't gone bad. If a small bakery with five bakers making $48,000 each a year would automate these tasks, they would save $14,000 per year.

That figure, argue the researchers, is by far smaller than the cost of developing and maintaining an AI system able to perform the same level of quality control.

Yet in this case, it is assumed that the small bakery should be the one in charge of making that development.

If a tech company develops an AI system that can recognize ingredients which have gone bad, and can market it for less than $14,000 a year, then the tables would have turned for the same bakery.

AI-as-a-service, says Thompson, could "democratize access to AI technologies, allowing smaller businesses and organizations to benefit from AI without the need for extensive in-house resources."

The development and mass scaling of AI-as-a-Service solutions could greatly impact the speed at which AI technology replaces jobs.

Below are some ETFs to look at:

• Global X Robotics and Artificial Intelligence ETF BOTZ is down 0.4% on Tuesday at the time of this writing.
• Roundhill Generative AI And Technology ETF CHAT is up 1%.
• iShares Robotics and Artificial Intel Mltsctr ETF IRBO is up 0.1%.

Photo: Made with Dall-E.