AI agents can form political camps—and changing their conversation partners does little to soften their views
Give AI agents political positions, ask them to defend their views and persuade one another, and let them decide which contacts to keep. This recent experiment, published in Nature Communications, found that over repeated exchanges, something familiar emerged: opposing camps that increasingly prefer their own company.
Researchers in China and elsewhere created artificial intelligence (AI) agents with varying political positions and observed how their discussions shaped the agents' beliefs over time. In experiments using GPT-3.5 language models, agents began with neutral stances on issues like party alignment, gun control, and abortion bans. By the end of 40 exchanges, only 22.5% of agents remained neutral on party alignment, while only 0.4% remained neutral on gun control and 5.1% on abortion bans.
The agents tended to interact more with those who held similar views, further solidifying their positions. While attempts were made to diversify conversations and encourage agents to consider opposing viewpoints, these changes were insufficient to significantly reduce polarization. The study, published in Nature Communications, suggests that changing conversation partners alone does not effectively soften AI agents' entrenched views.
Instead, instructing agents to more readily accept and be swayed by different perspectives proved more effective at mitigating polarization. However, the authors caution that these findings only apply to simulated AI networks, not necessarily to human social dynamics.
Written by urgent.news from Phys.org's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
