AI is driving experts away from online communities, research finds

Monthly questions on Stack Overflow have dropped nearly 76% since ChatGPT launched in 2022, but the more troubling trend is who's leaving.

A new University of Auckland study finds it's disproportionately the platform's most skilled, highest-reputation contributors abandoning ship, and generative AI appears to be accelerating their exit.

Business school researcher Dr Kenny Ching tracked 24,304 Stack Overflow contributors over 17 months, finding that high-reputation users began withdrawing at an accelerating rate starting in 2022, right as generative AI tools went mainstream.

Less established users tended to leave first, but it's the departure rate among veteran experts that kept climbing over time, closing the gap with newer users who left earlier.

Ching refers to this as "signal compression"; the similarity between AI-generated answers and those of experts makes the former appear good enough and causes expertise to lose its value.

"As AI-generated content becomes more common, people might feel their expertise and effort no longer stand out or are valued," he said.

It appears clear from Ching's figures that AI, rather than Stack Overflow's gatekeeper nature in content moderation, was the primary factor that hurt the site’s top contributors.

According to Ching, the same phenomenon could extend to other areas of life as well, including the education system, office settings in corporations, and scientific circles.

"This isn't just about coding platforms," he said, arguing, "but the incentives to learn deeply may disappear everywhere where AI creates an adequate substitute."

With this pipeline dried up, further iterations of models might be forced to rely on other sources of data, such as discussions in Slack groups, Discord servers, or repetitive queries made directly to chatbots by the users, thus creating a lot of uncertainties regarding the quality of the training process.