Cloud Hosting

In a paper that has been submitted to the journal Management Science, Thompsons team concluded that its not enough for AI systems to be good at tasks now performed by people. The system must be good enough to justify the cost of installing it and redesigning the way the job is done. Thompson said that for now, there are a lot of places where ... humans are a more cost-efficient way to do that.

The researchers focused on AI applications that use machine vision, the ability of computers to understand visual data such as pictures and video. For instance, many smartphones have a machine vision feature that can identify objects that have been photographed by the phones camera.

The researchers chose machine vision because its a well-established technology thats already used for many tasks, like reading labels or inspecting manufactured goods for defects.

Over a three-year period beginning in 2020, the researchers identified 420 commercial and industrial tasks where an AI system with machine vision might be capable of supplanting human workers. They conducted online surveys of people with expert knowledge of how each task was performed, to figure out how much a business might lower its costs by substituting machine vision for human labor.

The researchers found that in most cases, the visual aspect of a job was just a small part of a workers responsibilities. For instance, bakers must visually inspect the loaves of bread theyre producing. But bakers spend only about 6 percent of their time doing this. Shifting this job to AI might enable the bakery to get by with fewer workers. But the cost of todays machine vision systems is much too high to justify cutting labor costs by just 6 percent. Only if AI becomes much cheaper would the tradeoff make sense.

Applying the same measurements to all 420 tasks, Thompson and his colleagues found that installing machine vision in place of human labor would lead to lower overall costs only 23 percent of the time. In the great majority of cases, human workers were cheaper.

Of course, these results apply only to the current state of the art. Thompson noted that AI systems keep getting better and cheaper. Over time, theyll be cost-effective at a much larger number of jobs.

In addition, the study only looks at machine vision applications. It doesnt include research into the impacts of the latest generative AI systems, such as ChatGPT. Thompson said his group is now working on ways to assess the potential impact of these systems. But he noted that this might take quite a while, because figuring out the potential cost savings for thousands of specific tasks requires a great deal of research.

While the new research paper has yet to go through peer review, other academics think Thompsons team is on the right track.

Christoph Riedl, associate professor of supply chain and information management systems at Northeastern University, called the study absolutely plausible. Riedl noted that it fits the pattern of other major innovations that only achieved their full potential many years after they were first invented. When we started using electricity to run factories instead of steam engines, switching to electricity didnt make factories more productive, said Riedl. We had to learn how to use this new technology to gain any benefits from it.

Joseph Fuller, a professor at Harvard Business School, where he studies the future of work, said that the latest AI innovations will have dramatic impacts on some jobs. For instance, he said generative AI is already so good at creating computer software that it will reduce demand for human programmers. Its relatively easy to teach an AI how to write software just feed it lots of examples created by humans.

But teaching an AI how to manufacture a complex object is far more difficult. I cant unleash it on some database on the accuracy of manufacturing processes, said Fuller. There is no such data, so I have nothing to train it on.

Hiawatha Bray can be reached at hiawatha.bray@globe.com. Follow him @GlobeTechLab.

Link:

AI not yet ready to replace human workers, says MIT report - The Boston Globe