AI Technologies , Large Language Models

AI Automation Won't Steal All Jobs, for Now

Study Looks at Impact of AI on Tasks Requiring Visual Processing
AI Automation Won't Steal All Jobs, for Now
Human eyes are still more cost-effective than computer eyes, MIT researchers concluded. (Image: Shutterstock)

Artificial intelligence may not steal our jobs just yet, but only because humans are currently cheaper to employ.

See Also: Business Rewards vs. Security Risks of Generative AI: Executive Panel

Analysts expect AI to affect 40% of global employment, automate 25% of the labor market in the next few years and drive nearly half of all work by 2055. OpenAI's ChatGPT alone could affect 80% of jobs in the United States.

But many of the human jobs that could be replaced with AI are not "economically beneficial" to automate - at least for now - found the researchers at MIT's Computer Science and Artificial Intelligence Laboratory.

The study moves beyond AI exposure - which looks at what might be automated technologically - and accounts for the economics of the adoption technology to know what is attractive to automate, MIT CSAIL research scientist and study co-author Neil C. Thompson told Information Security Media Group.

The researchers surveyed workers on 1,000 visually assisted tasks across 800 occupations to understand the characteristics an AI system would need to fully replace their jobs, analyzed the cost of building such a model and determined whether non-farm businesses in the U.S. would pay upfront and bear the operating expenses for such a system.

An example of visual analysis used in the report is checking the quality of food at a bakery, which comprises about 6% of a baker's duties. A small bakery with five bakers making $48,000 each per year has potential labor savings of $14,000 per year from automating this task. But that amount is "far less" than the cost of developing, deploying and maintaining a computer vision system, so it wouldn't be economical to automate the task.

The researchers also considered self-hosted, self-service AI systems that can be fine-tuned for specific needs and need not be built from scratch - such as the models OpenAI sells. The conclusion still held true. Even if the system costs just $1,000, it wouldn't make economic sense for the business to automate, the researchers concluded.

"Even if we consider the impact of computer vision just within vision tasks, we find that the rate of job loss is lower than that already experienced in the economy,” the report said. "Even with rapid decreases in cost of 20% per year, it would still take decades for computer vision tasks to become economically efficient for firms."

Many such tasks are not attractive targets for automation because their AI replacements are too costly. Many companies lack the critical mass of workers doing those tasks to justify the cost of creating and deploying a system to automate them, Thompson told Information Security Media Group.

The cost-benefit ratio of computer vision - a field of AI that helps computers derive information from images, videos and other inputs - allows only 23% of workers to be effectively supplanted by AI, the report says.

AI systems will be cost-effective if deployed at scale, said Martin Fleming, a fellow at The Productivity Institute and a co-author of the report. "Most systems are cost-effective to deploy when single systems can be used across entire sectors or the whole economy. Deployment at scales is required." Only about one-quarter of tasks that could be replaced with an AI vision model are economical to automate if the system is applicable only at the level of an individual firm, he said.

Fleming said that the cost-effectiveness of AI models will likely play an important role in the proliferation of the technology. Over time, changes in the cost of AI systems or the scale at which they are deployed have the potential to increase automation. Scale can be gained either by large firms focusing on applications or through the formation of AI as a service, he said.

With AI as a service, system development costs could be amortized by deploying the system across many firms, making many more applications economically attractive, Fleming said. "But such a profound industry transformation would likely require industry collaborations or policy initiatives to enable data sharing across companies."

"In contrast to what is often said about rapid, AI-driven job displacement, change will occur much more slowly and, in many cases, not at all."

That does not mean the job market will stagnate. "All jobs will change," Fleming said. "Workers who are willing to redesign their jobs, learn new skills and find new ways of working will prosper. In the post-pandemic era, we've already seen nearly three-quarters of workers change jobs. Substantial labor market transformation is already underway."

The study does not consider cases in which AI can create brand-new jobs or augment human labor instead of replacing it. "We see that it is important to split apart the impact of AI on job automation, rather including task augmentation, which is likely to have much different effects on the economy," Thompson said.

MIT-IBM Watson AI Lab, which could potentially benefit from a narrative about AI not being a threat to the economy, funded the study. Thompson said that the researchers did not "experience any pressure to shape the results, nor is it even clear which type of result IBM would have wanted. So we did not experience our work together as a conflict."

About the Author

Rashmi Ramesh

Rashmi Ramesh

Assistant Editor, Global News Desk, ISMG

Ramesh has seven years of experience writing and editing stories on finance, enterprise and consumer technology, and diversity and inclusion. She has previously worked at formerly News Corp-owned TechCircle, business daily The Economic Times and The New Indian Express.

Around the Network

Our website uses cookies. Cookies enable us to provide the best experience possible and help us understand how visitors use our website. By browsing, you agree to our use of cookies.