Highly skilled workers have been training AI — that comes at a cost

The author is David Sarnoff professor of Management of Technology and professor at the MIT Sloan School of Management.

Recent research has fuelled optimism about AI’s potential. In a large call centre study, access to an AI assistant significantly improved agents’ ability to resolve problems, with the biggest gains recorded among newer workers. Other research on coding tools such as GitHub Copilot found that AI helped users, especially junior developers, complete tasks faster.

At first glance, these reports seem to be telling a straightforward story of technological progress. A tool helps people do their jobs better. Customers benefit. Companies benefit. Workers, particularly those early in their careers, gain access to expertise that once took years to accumulate.

But there’s a catch.

AI systems do not generate their capabilities from nowhere. In the call centre case, for example, the model was trained on transcripts of top agents’ conversations, in effect learning to replicate how they asked questions, managed stress, and solved problems. By digitising expertise, companies can scale an individual’s skills across time and location, allowing new hires — even after the expert has left — to perform more like veterans. For organisations and customers, the gains are clear: less experienced agents resolve issues more efficiently and customer satisfaction rises.