Humans still cheaper than AI in vast majority of jobs, MIT finds

Artificial intelligence can’t replace the majority of jobs right now in cost-effective ways, the Massachusetts Institute of Technology found in a study that sought to address fears about AI replacing humans in a swath of industries.

In one of the first in-depth probes of the viability of AI displacing labor, researchers modeled the cost attractiveness of automating various tasks in the US, concentrating on jobs where computer vision was employed — for instance, teachers and property appraisers. They found only 23% of workers, measured in terms of dollar wages, could be effectively supplanted. In other cases, because AI-assisted visual recognition is expensive to install and operate, humans did the job more economically.

The adoption of AI across industries accelerated last year after OpenAI’s ChatGPT and other generative tools showed the technology’s potential. Tech firms from Microsoft Corp. and Alphabet Inc. in the US to Baidu Inc. and Alibaba Group Holding Ltd. in China rolled out new AI services and ramped up development plans — at a pace that some industry leaders cautioned was recklessly fast. Fears about AI’s impact on jobs have long been a central concern.

“‘Machines will steal our jobs’ is a sentiment frequently expressed during times of rapid technological change. Such anxiety has re-emerged with the creation of large language models,” the researchers from MIT’s Computer Science and Artificial Intelligence Laboratory said in the 45-page paper titled Beyond AI Exposure. “We find that only 23% of worker compensation ‘exposed’ to AI computer vision would be cost-effective for firms to automate because of the large upfront costs of AI systems.”

Computer vision is a field of AI that enables machines to derive meaningful information from digital images and other visual inputs, with its most ubiquitous applications showing up in object detection systems for autonomous driving or in helping categorize photos on smartphones.

The cost-benefit ratio of computer vision is most favorable in segments like retail, transportation and warehousing, all areas where Walmart Inc. and Amazon.com Inc. are prominent. It’s also feasible in the health-care context, MIT’s paper said. A more aggressive AI rollout, especially via AI-as-a-service subscription offerings, could scale up other uses and make them more viable, the authors said.

The study was funded by the MIT-IBM Watson AI Lab and used online surveys to collect data on about 1,000 visually-assisted tasks across 800 occupations. Only 3% of such tasks can be automated cost-effectively today, but that could rise to 40% by 2030 if data costs fall and accuracy improves, the researchers said.

The sophistication of ChatGPT and rivals like Google’s Bard has rekindled concern about AI plundering jobs, as the new chatbots show proficiency in tasks previously only humans were capable of performing. The International Monetary Fund said last week that almost 40% of jobs globally would be impacted and that policymakers would need to carefully balance AI’s potential with the negative fallout.

At the World Economic Forum at Davos last week, many discussions focused on AI displacing the workforce. The co-founder of Inflection AI and Google’s DeepMind, Mustafa Suleyman, said that AI systems are “fundamentally labor-replacing tools.”

One case study in the paper looked at a hypothetical bakery. Bakers visually inspect ingredients for quality control on a daily basis, but that comprises only 6% of their duties, the researchers said. The saving in time and wages from implementing cameras and an AI system is still far from the cost of such a technological upgrade, they concluded.

​“Our study examines the usage of computer vision across the economy, examining its applicability to each occupation across nearly every industry and sector,” said Neil Thompson, director of the FutureTech Research Project at the MIT Computer Science and Artificial Intelligence Lab. “We show that there will be more automation in retail and health-care, and less in areas like construction, mining or real estate,” he said via email.