Artificial intelligence cannot currently cost-effectively replace most jobs, the Massachusetts Institute of Technology found in a study that sought to address fears that AI will replace humans in a range of industries.
In one of the first in-depth examinations of the viability of labor-displacing AI, researchers modeled the cost-attractiveness of automating various tasks in the United States, concentrating on jobs that used computer vision—for example, teachers and property appraisers.
They found that only 23 percent of workers, measured in dollar wages, could be effectively replaced. In other cases, because AI-assisted visual recognition is expensive to install and operate, humans have done the job more economically.
AI adoption across industries accelerated last year after OpenAI ChatGPT and other generative tools demonstrated the technology’s potential. Technology companies from Microsoft Corp. and Alphabet Inc. in the US to Baidu Inc. and Alibaba Group Holding Ltd. in China have rolled out new AI services and ramped up development plans — at a pace some industry leaders have warned is recklessly fast. Fears about the impact of artificial intelligence on jobs have long been a major concern.
“‘Machines will steal our jobs’ is a sentiment often expressed in times of rapid technological change. Such anxiety has resurfaced with the creation of large-scale language models,” researchers from MIT’s Laboratory for Computer Science and Artificial Intelligence said in the 45-page paper entitled Beyond AI Exposure. “We believe that only 23 percent of workers’ compensation ‘exposed’ to computer vision AI would be profitable for automation firms, and the reason is the high initial costs of AI systems,” they added.
The cost-benefit ratio of computer vision is most favorable in segments such as retail, transportation and warehousing, all areas where Walmart Inc. and Amazon.com Inc. It is also applicable in the context of health care, the MIT paper states.
The study was funded by the MIT-IBM Watson AI Lab and used online surveys to collect data from about 1,000 visually aided tasks in 800 occupations. Only three percent of such tasks can be economically automated today, but that could rise to 40 percent by 2030 if data costs fall and accuracy improves, researchers say.
