Fortune 500 CEOs and leaders in Silicon Valley suggest that while artificial intelligence (AI) may pose a threat to numerous jobs, the current high costs of AI computer vision technologies make them financially unviable for most US businesses. A recent study focused on tasks involving vision within various human occupations that could be at risk of automation by machines.
The research, led by Neil Thompson at the Massachusetts Institute of Technology, identified 414 vision-related tasks in various US job categories that could potentially be automated using existing AI technology. These tasks ranged from retail store supervisors visually confirming correct price tags to nurse anesthesiologists observing patients for potential warning signs.
The study assessed the economic feasibility of replacing these vision tasks with AI by calculating the costs of training and operating a computer vision model capable of achieving the required accuracy. A comparison was then made between AI costs and the costs associated with human labor, represented as a share of total worker salaries and benefits, as vision tasks typically constitute a small portion of an employee’s responsibilities.
The findings revealed that although 36% of US non-agricultural businesses had at least one worker task suitable for AI computer vision automation, only 8% were cost-effective to automate using AI. Furthermore, only 0.4% of US non-agricultural worker salaries and benefits would be economically justifiable for employers to automate.
The study highlighted that even large US firms with over 5000 employees could cost-effectively automate less than one-tenth of their current vision-related tasks due to the current high costs associated with AI computer vision.
While these findings may provide some reassurance to most US businesses, Gino Gancia at Queen Mary University of London suggests that there may be other AI applications with lower automation costs. He emphasizes the potential impact of AI on job availability and earnings, pointing out that the adoption of generative AI has already affected human freelancers on platforms like Upwork.
Research by Gancia indicates that regions embracing AI more rapidly, such as California, have experienced greater job losses, contributing to increased inequality across firms and workers. Neil Thompson and his colleagues anticipate a significant amount of human work to be automated eventually, contingent on the speed at which the costs of training and developing AI technologies can decrease. They advocate for proactive government programs to benefit displaced workers as automation continues to progress.