Some argue that content haphazardly produced by AI is destroying online communities.
Robin Moffat, Senior Principal Advisor for Streaming Data Technology at Confluent, has warned in a blog post that AI-generated content is exhausting online communities.
AI Slop is Killing Online Communities
Moffat doesn't dislike AI itself; rather, he takes the position that 'not including AI in one's work toolbox is a problem for a professional.' He argues that the problem isn't AI, but the act of mass-producing low-quality text, code, videos, and ebooks created with AI and presenting them as contributions to the community.
Moffat points out that on the internet today, more and more people are sharing AI-generated content with the world in the same way that a kindergartener might show their family a drawing made with crayons. A kindergartener's drawing is great to hang on a refrigerator at home, but it doesn't need to be sent to a national museum. Similarly, what is produced by inputting prompts into generative AIs like Claude is not necessarily useful to the community.
The trend that Moffat finds particularly problematic is the way people are so impressed by 'autonomous AI coding,' where AI agents write code, that they impulsively create repositories on GitHub, then have the AI write enthusiastic introductory articles, and then post them indiscriminately to any potentially relevant places like Reddit and Slack.
Moffat argues that before publishing anything on GitHub, you should take a deep breath and consider whether the publication is truly useful, whether you yourself continue to use it, whether the documentation is sufficient, and whether you can take responsibility if users raise issues or pull requests. He says that the days of sharing things simply because 'it's cool because it was made with AI' are over. If you give an AI a prompt, many people can create similar results, so results that are merely generated from prompts are not necessarily novel or of public value.
In this context, the term 'AI slop' has become popular in English-speaking countries as a negative expression referring to AI-generated content that is created with little effort and offers little benefit to the recipient.
However, Moffat also states that 'creating content using AI is not inherently bad.' If AI can make contributions that were previously impossible, while humans have a clear purpose, review, revise, and take responsibility for the work, then the use of AI can be a positive for online communities.
On the other hand, the bad AI-generated content that Moffat is concerned with is spam, engagement-driven posts, promotional content, or just noise thrown into the community space on a whim. In online communities, the time of readers, reviewers, and moderators is limited. While the cost of generating large quantities of articles and code with AI is low, the cost of deciphering low-quality writing and determining that it's 'not worth reading,' or examining poorly written pull requests and explaining why they 'cannot be merged,' is high.
Moffat quotes the idea, attributed to Austrian software engineer Alberto Brandolini, that 'the energy required to refute nonsense is an order of magnitude greater than the energy required to create it.' If you post a poorly written AI article, readers have to spend time judging its value, and if you send poorly written AI-generated code to a project, maintainers have to review the code and explain the problems. The effort required from the poster may be small, but the burden on the recipient is large, and mass posting of AI-generated content can indeed be an act of imposing an asymmetrical burden on the community.
Moffat also describes AI-generated content as a 'vine' of weeds that constricts the organic activity of online communities. If communities like Reddit become too flooded with AI-generated apps, articles, and videos, it will become difficult to find truly valuable posts. As the noise increases, dedicated members will become tired and distance themselves, and human-to-human discussion and learning will decrease. Ultimately, online communities risk either declining or becoming boring spaces where only AI agents converse without human interaction.
However, Moffat does not reject the idea of AI-powered content sharing itself. From there, Moffat goes on to discuss the question of 'what should be shared and how?'
The internet has always developed as a place for people to share what they create, think, and learn. The important thing is to understand 'what, to whom, and why' you are sharing it. People who created personal homepages on GeoCities in the past would post construction-in-progress animated GIFs, access counters, and webring banners, and show them to friends and family. However, the mere fact that they created a personal homepage didn't mean they needed to advertise it to the entire internet. The same judgment is needed for AI-generated apps and AI-generated blogs.
The standard proposed by Moffat is the idea of sharing things 'built with AI,' not things 'made by AI.' Moffat introduces the expression 'Built with AI, not by AI' by his colleague Gunnar Mölling. AI is a tool; thinking, directing, and verifying are human tasks. For example, Mölling's new parser for Apache Parquet, 'Hardwood,' which he is developing using AI, does not fall under Moffat's criticism because it has been under development for four months, has a clear roadmap, a growing community, and has been carefully designed.
Moffat argues that before posting, you should ask yourself, 'What will this share add to the community?' If someone else running the prompt produces similar results, the value of the share may be skewed towards the novelty of the prompt or AI tool, and may not align with the knowledge or results the target community needs.
Furthermore, in online communities, the netiquette of 'lurking for six months '—the practice of reading before posting—has long been important. Every platform, whether it's Usenet, Reddit, lobste.rs, or Slack, has its own atmosphere and standards. Instead of immediately promoting something, posters need to read past posts to understand what kind of information is welcomed and what kind is disliked.
Moffat states that when using AI, it's also a sign of respect for the community to clearly indicate which parts were done with AI. If you hide the use of AI and make it appear as if it was made by humans, it becomes difficult for readers and reviewers to judge the reliability of the work. Explaining the fact that AI was used, the scope of its work, and the parts that were checked by humans makes it easier for recipients to evaluate the work.
AI is fundamentally changing software development, technical article writing, learning support, and documentation. Moffat doesn't deny the excitement of 'wow, I've created something amazing' that autonomous AI coding brings; rather, he says it should be thoroughly enjoyed. However, online communities are both powerful and fragile. 'AI-generated crayon drawings are fun to stick on your refrigerator at home, but if you're posting them to a community, they should only be things that are truly meaningful to the people you're posting to,' Moffat concluded.
Related Posts:
in Free Member, AI, Posted by log1d_ts