As a result, the company spent a lot of time training the new workers hired to replace those who quit. Many of the skills needed were what the researchers called “tacit knowledge,” experiential knowledge that can’t be easily codified but can be absorbed from chat logs by large language models and then mimicked. The company’s bot helped with technical and soft skills, pointing agents to relevant technical documents and suggesting one-liners to calm down irate customers, like “happy to help you fix this ASAP!”
After the bot started helping, the number of issues the team solved per hour increased by 14 percent. In addition, the odds of a worker quitting in any given month dropped by 9 percent, and customer attitudes toward employees also improved. The company also saw a 25 percent decrease in customers requesting to speak with a manager.
But when the researchers broke down the results by skill level, they found that most of the chatbot’s benefits accrued to lower-skilled workers, who saw a 35 percent productivity increase. Higher-skilled workers saw no gains and even saw their customer satisfaction scores drop slightly, suggesting that the bot may have been a distraction.
Meanwhile, the value of that highly-skilled work multiplied as the AI assistant directed less-skilled workers to use the same techniques.
There are reasons to doubt that employers will reward that value of their own free will. Aaron Benanav, historian at Syracuse University and author of the book Automation and the future of workhe sees a historical parallel in Taylorism, a productivity system developed in the late 19th century by a mechanical engineer named Frederick Taylor and later adopted in Henry Ford’s automobile factories.
Using a stopwatch, Taylor divided physical processes into their components to determine the most efficient way to complete them. He paid special attention to the most skilled workers in a trade, says Benanav, “so he could get the least skilled workers to work in the same way.” Now, instead of a demanding engineer with a stopwatch, machine learning tools can collect and disseminate best practices from workers.
That didn’t work out so well for some employees in the Taylor era. His methods were associated with lower earnings for more skilled workers, because companies could pay less skilled employees to do the same type of work, Benanav says. Even if some high-performing workers were still needed, companies needed fewer, and competition between them increased.
“By some accounts, that played a pretty big role in triggering unionization among all these lesser-skilled or medium-skilled workers in the 1930s,” says Benanav. However, some less punitive schemes emerged. One of Taylor’s followers, the mechanical engineer Henry Gantt, yes, the graphic boy—created a system that paid all workers a minimum wage but offered bonuses to those who also achieved additional targets.
Even if employers are incentivized to pay high-performing employees a premium for teaching AI systems, or employees earn it themselves, dividing the loot fairly can be tricky. For one thing, data can be pooled from various workplaces and sent to an AI company that builds a model and sells it to individual companies.
