AI Too Costly to Replace Most Jobs: MIT Study

MIT study debunks the myth of AI replacing humans

Artificial intelligence cannot now replace the majority of occupations in cost-effective ways, according to research conducted by the Massachusetts Institute of Technology to address concerns about AI replacing humans in a wide range of industries

In one of the earliest in-depth investigations on the practicality of AI displacing labor, MIT study researchers calculated the cost-effectiveness of automating numerous tasks in the United States, focusing on jobs that required computer vision, such as teachers and property appraisers. They discovered that AI is too costly to replace jobs, and only 23% of workers, assessed in terms of dollar salaries, could be efficiently replaced. In other circumstances, people performed the task more efficiently because AI-assisted visual identification is costly to install and maintain.

The adoption of AI across businesses surged last year after OpenAI's ChatGPT and other generative tools demonstrated the technology's capabilities. Tech companies ranging from Microsoft Corp. and Alphabet Inc. in the United States to Baidu Inc. and Alibaba Group Holding Ltd. in China launched new AI services and increased development plans at a rate that some industry executives said was too rapid. Concerns regarding AI's influence on jobs have long persisted.

"'Machines will steal our employment' is a common concern stated during periods of fast technological progress. "Such anxiety has resurfaced with the creation of large language models," the researchers from MIT's Computer Science and Artificial Intelligence Laboratory said in the 45-page report 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 branch of artificial intelligence that allows machines to extract meaningful information from digital photographs and other visual inputs. Its most common applications include object identification systems for self-driving cars and photo categorization on smartphones.

Computer vision has the best cost-benefit ratio in categories such as retail, transportation, and warehousing, all of which are dominated by Walmart Inc. and Amazon.com Inc. According to MIT research, it is also practical in the healthcare environment. According to the authors, a more aggressive AI rollout, mainly through AI-as-a-service subscription options, might scale up and make additional applications more viable.

The MIT-IBM Watson AI Lab supported the study, which employed online surveys to collect data on about 1,000 visually-assisted activities from 800 jobs. Only 3% of such jobs can be automated cost-effectively today, but the researchers predict that figure will climb to 40% by 2030 if data prices decrease and accuracy improves.

The sophistication of ChatGPT and rivals such as Google's Bard has reignited concerns about AI taking employment, as the new chatbots demonstrate competency in activities that people were previously capable of completing. The International Monetary Fund stated this week that over 40% of global employment will be affected and that authorities must carefully balance AI's promise with the negative consequences.

Many debates at the World Economic Forum in Davos last week centered on AI's potential to replace workers. Mustafa Suleyman, co-founder of Inflection AI and Google's DeepMind, described AI systems as "fundamentally labor-replacing tools."