‘Quality decays exponentially following AI arrival’: Research shows experts and contributors leaving online communities amidst silent ‘knowledge reset’ Rahimnoorali11@gmail.com (Rahim Amir) | usagoldmines.com

• Generative AI may have unintentionally pushed away some of the highest-quality ‘expert’ contributors to sites such as Stack Overflow as users increasingly adopted tools trained on their feedback instead
• The issue stems from said users feeling their expertise and effort are unrewarded, with AI often offering the same solutions at a faster pace
• The move is not limited to online coding communities, threatening to spill over into other areas such as classrooms, corporate workplaces, and scientific communities

Research from the University of Auckland on Stack Overflow’s demise over the last few years points to an increasingly worrying trend in the software community: the best, or highest-skill, contributors are leaving in droves.

AI, which arguably bridges the gap between most entry-level and mid-range coders and some of the best in the business, might actually be accelerating the latter’s exit from online communities, as they feel their efforts are no longer as valued as they once were.

Stack Overflow has seen a nearly 76% decline in monthly questions posted since ChatGPT’s advent in 2022, indicating that both new and existing users are abandoning the site.

A much broader problem than just Stack Overflow?

Stack Overflow’s problems and the reason for its decline were multi-faceted; however, many users felt that the site and some of its most talented contributors engaged in a certain degree of hubris.

This, coupled with heavy-handed moderation that many called ‘self-righteous,’ meant that users finding a viable option would inevitably leave the platform.

ChatGPT and its AI alternatives became considerably more pliable and, over time, doubled as search engines for many coders with routine, repeatable queries, even as AI increasingly handled questions such as syntax issues better than before.

This, in turn, reduced the number of questions asked on the platform and, despite a generative AI ban enacted soon after ChatGPT went online, led to a loss of answerers that may prove impossible to replace in the long term.

The issue may no longer be confined to online coding communities; researchers indicate it could spill over into other areas such as classrooms, offices, and other research communities, where low-effort answers are harder to discern from those of subject-matter experts thanks to ever-evolving, retrained AI models.

“If everybody can create a good quality response or output using AI, some people may think, ‘Why should I make an effort to share my expertise and participate?”, publisher of the study Dr Kenny Ching explained.

Ching termed this ‘signal compression’ as expert and non-expert solutions became harder to separate, even as it became less rewarding to be a subject-matter expert on topics that AI could also easily weigh in on.

The question that does come to mind here, however, is a simpler one: if AI was trained on user-contributed data and an increasingly small amount of it exists on platforms such as Stack Overflow, where does the upcoming knowledge reset take us in terms of AI capabilities?

While future AI models will not get “dumber,” so to speak, they might turn to different avenues for training, such as Slack chats, Discord servers, or even users who currently ask them the same coding-related questions they once did on Stack Overflow.

Whether this replaces experts who no longer wish to contribute or simply makes AI more prone to errors over time, thanks to how its feedback loop functions, is an interesting question in a society that finds it increasingly hard to discern between AI and human answers.