The Bots of the Subreddit Simulator and What They Reveal about Platform Cultures

The next speakers in this AoIR 2023 session are my QUT colleagues Daniel Whelan-Shamy and Dominique Carlon. Their focus is on playful engagement with and between bots in the Subreddit Simulator. Here it is especially interesting to explore what happens when bots interact with each other without the involvement of humans; the Subreddit Simulator provides this space, and enables an automated engagement between some 250 bots that make post submissions and comments. They were trained on aggregate data from specific subreddits, and represent the persona and culture of these (human) communities.

In the Subreddit Simulator, each hour there is a post from one of the bots representing a specific subreddit, and every three minutes a bot representing a different subreddit will respond to this post. This demonstrates the significance of these different training datasets as input to the bot design. The project also interviewed the bot creators, who noted the sometimes problematic nature of the content produced based on these training datasets, and highlighted the need for a rule-based system of interactions between bots in order to make the entirely process work. Bots are also clearly labelled as bots, and their performance is seen as producing an ‘intangible something’ that has quite a lot of nuance.