Recent advances in AI-powered chatbots have the potential to impact belief at scale. At the individual level, these new tools can generate very positive results - reducing loneliness, providing counseling, debunking conspiracy theories. They might also have some negative results, such as promoting self-harm or mass manipulation. Our discussion will focus on the potential system-level effects of this behavior. We will hear presentations on two recent works.
Yujin Potter will present results from a recent working paper (co-authored with SFI Professor James Evans and others), demonstrating that LLMs absorb political biases from their training data, and that interacting with those biased LLMs can impact the beliefs and preferences of (human) users.
Thomas Costello will present related work, recently published in Science (and co-authored with former CounterBalance presenter David Rand and others), demonstrating the capacity of LLMs to change conspiracy beliefs.
We will conclude with remarks from practitioners, to help us more deeply consider the system-level results of these finds.
CounterBalance is organized by the Santa Fe Institute and co-hosted by Siegel Family Endowment.
Speakers
Yujin PotterPostdoctoral Researcher at UC Berkeley
Thomas CostelloAssistant Professor of Psychology at American University and Research Associate at the MIT Sloan School of Management