Researchers of the University of Amsterdam launched an experiment: they created a simple social network without advertising and recommendations algorithms, and all users were replaced by AI-Futtas. The project was attended by 500 bots based on the GPT-4O Mini model, everyone set a certain “personality” and political views.
During the five episodes of the experiment, including 10,000 actions, bots began to actively subscribe and interact only with those who shared their views. The authors of “the most sharp and one -sided posts” were most attention.
Although the system did not have advanced algorithms, AI still repeated the typical behavior of people in social networks-the formation of an “echo-camera” and strengthening polarization. Scientists suggest that this is due to the fact that the models are trained on human content.
To reduce the separation, the researchers tried different measures: chronological tape, hiding the number of subscribers, a decrease in viral of content, strengthening opposite views. The effect turned out to be minimal – no more than 6% of the decrease in involvement in politically painted posts. And in some cases, for example, when the biographies are hidden, the situation worsened: radical publications received even more attention.
What is the conclusion of scientists? Allegedly, the structure of social networks, even without algorithms, is inclined to strengthen the separation and push to the extremes.
