Conversations with a large language model improve attitudes toward Muslims and Islam without harming attitudes toward Jews or Christians

Can a brief conversation with a large language model (LLM) measurably improve attitudes toward a stigmatized religious minority? We investigate this question in the context of Americans’ attitudes towards Muslims and Islam. We also ask whether doing so worsens attitudes towards Jews and Judaism, given the current high-salience conflict between Muslims and Jews in the Middle East. To answer these questions, N=604 participants had a three-round dialogue with an LLM that either (i) corrected participants’ misconceptions about Islam and Muslims (treatment) or (ii) discussed sports (control). As expected, we find that the treatment significantly improved attitudes towards Islam and Muslims relative to the control. Furthermore, we do not find that improving attitudes towards Islam and Muslims comes at the cost of worsening attitudes towards Jews or Christians. On the contrary, we find some evidence that the treatment also improves attitudes towards Jews, Judaism, and Christianity. Thus, customized LLM chatbots may help address religious prejudice.