AI has great capabilities, but remains an expensive option for some tasks
Yuichiro Kayano/Moment RF/Getty Images
Artificial intelligence may find its way into many people's jobs, but AI computer vision technology is still cheap enough to be valuable to most U.S. companies today, according to Fortune 500 CEOs and Silicon Valley leaders. Not. This finding comes from a study of human tasks, particularly those involving vision, that are at risk of machine automation.
In this study, researchers focused on whether vision tasks involved in various human jobs are economically worth replacing with existing AI computer vision. “There are a lot of tasks that you can imagine applying AI to, but you wouldn't actually want to do because of the cost,” he says. neil thompson Co-authored by researchers at Massachusetts Institute of Technology. Published today as a working paper.
Thompson and colleagues identified 414 visual tasks in U.S. occupations that could potentially be automated with existing AI technology. These jobs include those of retail store supervisors who visually check to see if products have the correct price tag, or who notice dilated pupils or changes in cheek color that can be warning signs of potential problems. Includes a trained nurse anesthetist who monitors the patient's care.
The researchers calculated the cost of training and operating an AI computer vision model that can handle these tasks with the required precision. We then compared the cost of AI to the cost of human labor. The latter is expressed as a percentage of the employee's total salary and benefits. This is because visual tasks typically make up a small portion of a given employee's job duties.
As a result, while 36 percent of U.S. non-agricultural businesses have at least one employee task that can be automated with AI computer vision, there are few tasks that would be cost-effective to automate using AI. was found to be only 8%.
They also concluded that only 0.4 percent of U.S. non-farm worker pay and benefits is actually cost-effective for employers to automate.
The current cost of AI computer vision means that even large U.S. companies with 5,000 or more employees (over 99.9 percent of all U.S. companies) can cost-effectively automate less than one-tenth of their existing vision tasks. It means that.
While such a finding that AI computer vision would be too costly for most U.S. companies “may sound like a reassuring finding,” “there may be other consequences.” [AI] “Applications with low automation costs.” Gino Gancia at Queen Mary University of London.
rush to Adopts “generation AI” The ability to create new content is already having a negative impact on the number of jobs available and the incomes of human freelancers on online platforms such as: upwork. Gancia's research also shows that regions of the U.S. with industries that are more advanced in adopting AI, such as California, are already experiencing greater job losses.
“In general, we find that new technologies spread unevenly,” Gancia says. “As a result, automation and AI are likely to contribute to increasing inequality between businesses and workers.”
Thompson and his colleagues expect a significant amount of human work to be automated in the long term. But that depends on how quickly the costs of training and developing AI technology come down.
“There is going to be a significant amount of automation going forward, so governments need to start preparing for it,” Thompson said. “But there is plenty of time to put in place real programs that can bring benefits. [displaced] Worker. “
topic:
Source: www.newscientist.com
