Some tasks can be done more cheaply by human labor than by computer vision systems, according to a study led by researchers at the Massachusetts Institute of Technology.
Scientists from the institute, IBM, and the Productivity Institute surveyed workers to find out what capabilities computers need to do their jobs, and how they can build and install such AI systems. We added up the costs incurred and compared this to human salaries.
“Given today's costs, U.S. companies choose not to automate most vision tasks with ‘AI exposure,’ and we found that only 23 of those tasks will not be automated. [percent] “The proportion of workers' wages paid for visual tasks is attractive to automate,” the researchers concluded. [PDF].
In other words, computer vision systems are too expensive to replace employees in more than three-quarters of the jobs considered.
Machines with sensors and cameras that run AI algorithms can be expensive to train, deploy, and maintain, and are not always worth it if they only perform specific chores. One of the examples cited in this paper considers a quality assurance assessment that inspects items for defects and defective disposal.
For example, a bakery can deploy and train a computer vision system to check whether the ingredients it uses are spoiled.
The report cites O*NET data from the U.S. Department of Labor's Bureau of Labor Statistics, which estimates that only 6% of bakery jobs involve checking food quality. . If a small business employs her five bakers and pays each of them approximately $48,000 per year, each employee is paid $2,280 per year to inspect ingredients. Multiply this by 5 and you get $14,400.
Estimates suggest that AI cannot outweigh the costs.
“This points to a more gradual integration of AI into various sectors and is in contrast to the often hypothesized rapid displacement of jobs by AI,” said co-author of the study and MIT Computer Science Professor. Neil Thompson, Principal Researcher at the Institute for Artificial Intelligence, explained: statement.
The team analyzed 420 visual tasks and surveyed five to nine workers for each task, Thompson said. register.
So, are our jobs safe?necessarily
While this paper is good news for bakeries, there are widespread and understandable concerns that generative AI will replace knowledge workers. That's because the large-scale language model (LLM) that handles writing tasks can be run on an off-the-shelf laptop and doesn't require a fancy camera.
LLM can be easily fine-tuned with custom data to perform many more common tasks.
Opinions are divided on whether AI will displace people's jobs. Some believe this technology will introduce new types of jobs, while others believe it will make certain roles obsolete. The jury is still out. ®
