Researchers at the University of Amsterdam modeled a social network entirely populated by AI-powered chatbots to test whether it's possible to make online spaces less toxic. The experiment showed that even with platform intervention, social networks remain polarized and full of extreme content. This is reported by Futurism, writes UNN.
Details
Social networks have long turned into echo chambers of angry content, where users often fall into a trap of outrage. To investigate whether this dynamic can be changed, scientists from the University of Amsterdam created a completely artificial platform where all users were GPT-4o-based chatbots.
The experiment aimed to test six strategies for combating toxicity: chronological news feed, amplification of diverse viewpoints, hiding social statistics, and removing user biographies. However, none of them yielded significant results. Some interventions even exacerbated the problem: the chronological feed reduced inequality of attention but brought the most extreme content to the forefront.
"Toxic content forms network structures that inversely affect post visibility, creating a vicious cycle of toxicity," explained Associate Professor Petter Törnberg. He added that even the use of artificial intelligence is not an "ideal solution" because it reflects human behavior with all its biases and limitations.
The experiment also showed "extreme inequality of attention," where a small fraction of posts receives the most visibility. Combined with generative AI, which is increasingly used to create content that seeks to maximize attention, this could make polarization and disinformation even more widespread.
Törnberg emphasizes that current social media models may be doomed: even when attempting to intervene algorithmically, platforms remain hotbeds of polarization and extremism.
"It's hard for me to imagine how traditional social networks will survive in such conditions," the researcher concluded.
Ukraine launched the first state service with artificial intelligence19.08.25, 12:37