The researchers looked at the cost-effectiveness of automating tasks, focusing on roles that could use computer vision — a type of AI that derives information from images and video.

The study found that just under a quarter of the wages paid for vision tasks would be worth automating. In some cases, workers are still more economic because AI-assisted visual recognition technology is expensive to install and operate.

"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," the researchers said in the paper.

The study is one of the first attempts to estimate which tasks are economic for US companies to automate. While other research has sought to identify the tasks and roles most exposed to AI automation, those studies have largely neglected the economic implications of installing the technology.

"'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 MIT researchers said.

They noted that previous predictions and studies were notably vague about the timeline and extent of automation "because they do not directly consider the technical feasibility or economic viability of AI systems, but instead use measures of similarity between tasks and AI capabilities to indicate exposure."