AI can't steal human jobs for now, too expensive to deploy for most businesses, shows MIT study

A study conducted by the Massachusetts Institute of Technology (MIT) has revealed that artificial intelligence (AI) is not currently cost-effective for replacing the majority of jobs. The study aimed to address concerns about AI displacing human workers across various industries, focusing on tasks utilizing computer vision, such as those performed by teachers and property appraisers. The researchers found that only 23 per cent of workers, measured in terms of dollar wages, could be effectively supplanted by AI, primarily due to the high upfront costs of AI systems. The research is one of the first in-depth examinations of the feasibility of AI replacing human labour, especially in tasks involving computer vision. The study highlights that AI-assisted visual recognition while showing potential, is expensive to install and operate. In some cases, human workers prove to be more economically efficient in performing tasks. The adoption of AI across industries has accelerated, with the development of technologies like OpenAI’s ChatGPT showcasing the potential of generative tools. Tech companies worldwide, including Microsoft, Alphabet, Baidu, and Alibaba, have rolled out new AI services and ramped up development plans. However, concerns about the impact of AI on jobs have persisted. The MIT study, titled “Beyond AI Exposure,” emphasizes that the sentiment of “machines stealing jobs” has re-emerged with the advent of large language models. The researchers suggest that a mere 23 per cent of worker compensation exposed to AI computer vision would be cost-effective for automation due to the substantial upfront costs involved. Computer vision, a field enabling machines to derive meaningful information from visual inputs, is most favourable in segments like retail, transportation, warehousing, and healthcare, according to the study. While only 3 per cent of visually-assisted tasks can be cost-effectively automated today, the researchers suggest that this figure could rise to 40 per cent by 2030 if data costs fall and accuracy improves. The study, funded by the MIT-IBM Watson AI Lab, utilized online surveys to collect data on about 1,000 visually assisted tasks across 800 occupations. The researchers conclude that more automation is likely in retail and healthcare, while areas like construction, mining, or real estate may see less automation in the near future.