Does artificial intelligence bias perceptions of environmental challenges?

Abstract Artificial intelligence (AI) is reshaping how humans obtain information about environmental challenges. Yet the outputs of AI chatbots contain biases that affect how humans view these challenges. Here, we use qualitative and quantitative content analysis to identify bias in AI chatbot characterizations of the issues, causes, consequences, and solutions to environmental challenges. By manually coding an original dataset of 1512 chatbot responses across multiple environmental challenges and chatbots, we identify a number of overlapping areas of bias. Most notably, chatbots are prone to proposing incremental solutions to environmental challenges that draw heavily on past experience and avoid more radical changes to existing economic, social, and political systems. We also find that chatbots are reluctant to assign accountability to investors and avoid associating environmental challenges with broader social justice issues. These findings present new dimensions of bias in AI and auger towards a more critical treatment of AI’s hidden environmental impacts.