According to a study led by researchers from the Massachusetts Institute of Technology, human labor can complete certain tasks more cost-effectively than computer vision systems. The study, conducted in collaboration with scientists from IBM and the Productivity Institute, involved surveying workers to identify the necessary capabilities for AI systems to perform their duties. By comparing the costs of implementing AI systems with human salaries, the researchers determined that, at current expenses, US businesses are unlikely to automate the majority of vision tasks with ‘AI Exposure.’ Only about 23 percent of worker wages for vision-related tasks are deemed economically feasible to automate.
The high costs associated with training, deploying, and maintaining machines equipped with sensors and cameras running AI algorithms often outweigh the benefits, especially for tasks that are highly specific. For instance, the report mentions quality control assessments in industries like a bakery, where a computer vision system could be trained to identify spoiled ingredients. However, considering that only a small fraction of a baker’s job involves inspecting food quality, the cost-effectiveness of AI implementation is questionable. The study references data from the US Department of Labor’s Bureau of Labor Statistics O*NET, which estimates that quality inspections represent merely six percent of a baker’s responsibilities. Given the relatively low cost of employing human workers for such tasks, AI may not offer a compelling economic advantage in this scenario.
Neil Thompson, a co-author of the study and principal investigator at MIT Computer Science & Artificial Intelligence Laboratory, highlighted a gradual adoption of AI across various sectors, challenging the notion of rapid job displacement driven by AI technologies. The research team examined 420 vision-related tasks, gathering input from 5-9 workers for each task to assess the feasibility of automation.
While the findings may provide reassurance to certain professions like bakers, concerns persist regarding the potential impact of generative AI on knowledge workers. Unlike computer vision systems that require specialized hardware, large language models (LLMs) can run on standard laptops and offer versatility in handling various tasks. With the ability to be easily customized with specific data, LLMs have the potential to perform a wide range of functions beyond vision-related duties.
The debate surrounding the impact of AI on job security remains unsettled, with diverging opinions on whether AI will create new employment opportunities or lead to the obsolescence of certain roles. The future implications of AI technology on the workforce continue to be a subject of ongoing discussion and analysis.
