Artifical Intelligence for Science Communication: The Impact of Visual and Lingustic Representation on Credibility and Persuasiveness 

Abstract

Recent developments such as declining interest in news and the increasing use of social media as a primary source of information have, through a complex interplay of algorithms, selective consumption, and misinformation, led to a decrease in perceived credibility, growing hostility, and societal polarization. This presents major challenges for science communication. At the same time, emerging technologies like artificial intelligence (AI) offer promising new opportunities in this field. However, studies so far have mostly implemented AI in the form of simple chatbots, meaning virtual agents that communicate via text and lack any human-like visual representation. Yet, studies suggest that a more human-like appearance could positively influence perceived credibility and persuasiveness. In this context, the present thesis investigates both the general suitability and potential risks of AI-driven virtual agents for science communication, as well as the impact of their visual (with vs. without a human-like model) and linguistic (text-based vs. spoken) representation on users’ perceptions. For this purpose, an application is being developed in which an intelligent virtual agent can be used to discuss a scientific topic. The qualitative and quantitative evaluation of this application indicates that AI has the potential to promote open discourse and provide new perspectives on scientific topics, but should be designed and used with caution. The effect of the specific form of representation, in turn, proves to be highly individual and can only be generalized to a limited extend.