生成式人工智能纠正疫苗错误信念:针对外向性与伪科学信念的启示
一项针对1435名美国成年人的组间实验显示,针对外向性定制的AI生成纠正信息显著减少了疫苗错误信念,效果与高质量通用信息相当;但针对伪科学信念的信息无效甚至适得其反。研究提示AI辅助信息生成需人工监督。
要点
- 针对外向性的AI纠正信息显著减少疫苗错误信念,效果与高质量通用信息相当。
- 针对伪科学信念的AI信息无效,在伪科学信念强的人中甚至强化负面态度。
- AI生成纠正信息需人工监督,未来需优化人机协作。
核心结论
- AI辅助信息生成在针对外向性等人格特质时可有效纠正疫苗错误信念,但针对根深蒂固的信念可能无效甚至适得其反,强调人工监督的必要性。
研究设计与样本
- 组间实验,1435名美国成年人随机分配到对照组、通用纠正信息组、外向性靶向纠正信息组或伪科学信念靶向纠正信息组。
- AI用于生成靶向纠正信息,通用和对照信息来自现实世界示例。
- 参与者在接触纠正信息前后对疫苗错误信念陈述的同意程度进行评分。
主要发现
- 外向性靶向信息显著减少疫苗错误信念,效果与高质量通用信息相当,尤其在外向性水平较高的参与者中。
- 这些效果未延伸到一般疫苗接种态度。
- 伪科学信念靶向信息无效,在某些情况下适得其反,强化了伪科学信念强的个体的负面态度。
局限性
- 原文未报告局限性。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
BACKGROUND: Misinformation about vaccines is a significant barrier to public health, fueling hesitancy and resistance. Generative AI offers a scalable tool for assisting public health communicators in crafting targeted correction messages tailored to audience characteristics. This study investigates the effectiveness of AI-generated messages targeting extraversion and pseudoscientific beliefs compared to high-quality generic and non-vaccine-related messages. METHOD: In a between-subjects experiment, 1435 U.S. adults were randomly assigned to one of four conditions: control, generic correction, extraversion-targeting correction, or pseudoscientific-belief-targeting correction. Participants rated their agreement with vaccine misbelief statements before and after exposure to a correction message. AI was used to generate the targeted correction messages, while the generic and control messages were sourced from real-world examples. RESULTS: Extraversion-targeting messages significantly reduced vaccine misbeliefs, performing comparably to high-quality generic messages, particularly among participants with higher extraversion levels. However, these effects did not extend to general vaccination attitudes. Pseudoscientific-belief-targeting messages were ineffective and, in some cases, backfired, reinforcing negative attitudes among individuals with strong pseudoscientific beliefs. CONCLUSION: This study demonstrates the potential of AI-assisted message generation for crafting effective correction messages, particularly when targeting personality traits like extraversion. However, the findings suggest that certain AI-generated messages may be less effective or even counterproductive when targeting entrenched beliefs, underscoring the need for human oversight in refining AI-generated messages. Future research should explore additional audience characteristics and optimize human-AI collaboration to enhance the effectiveness of AI-generated correction messages in public health communication.
引用
- Lu H. Generative AI for vaccine misbelief correction: Insights from targeting extraversion and pseudoscientific beliefs. Vaccine. 2025. PMID: 40086038. doi:10.1016/j.vaccine.2025.127018
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。