The Dynamics of the Machine Heuristic, Misinformation, and Political Informedness

This thesis aimed to answer the question “How do factual- and misinformation purportedly made by artificial intelligence (AI), like bots on social networking sites, affect humans’ cognitive reactions?” To test that, I combined two concepts, the machine heuristic (Sundar, 2008) and political informedness (Kuklinski et al., 2000; Li & Wagner, 2020; van Kessel et al., 2021). I used an experimental design of a 2 (type of agent: human vs. AI) by 2 (type of agent’s informedness: informed vs. misinformed) and measured respondents’ perceptions of accuracy, message credibility, source credibility, objectivity, bias, and their belief accuracy and belief certainty. I measured how participants react to the messages that look generated by an AI compared to the ones by a human, and explored how the different types of agents could lead to differences in belief accuracy and belief certainty of informed and misinformed information. The overall results offer nuance to the concept of the machine heuristic as the type of agent was not related to how accurately people’s answers fit with the best available evidence (belief accuracy) or how confident people were about those answers (belief certainty). No interaction was found between the type of agent and the informedness of the agent. However, higher ratings of the machine heuristic were positively related to accuracy and competence and negatively related to bias. Still, this thesis offers several implications. First, this study demonstrates boundary conditions for the machine heuristic. The machine heuristic might not sometimes make a perceptional difference between human and machine conditions, which was partially observed in recent studies (Araujo et al., 2020, Worcieszak et al., 2021). However, findings did emerge when analyzing only those who had higher ratings of the machine heuristic. Thus, I propose the idea of the evanishing effect, whereby neither humans nor machines are preferred for a task of the machine heuristic. Second, this study adds another resource to be used for the machine heuristic studies in a political context. The finding of no direct relationships between the type of agent and belief accuracy or belief certainty can be a ground for more research in technology and its impacts on politics. Third, this study expands important discussions about the authenticity of communication. As this study did not uphold the machine heuristic, if people’s perceptions about machines (or AI) might be changing, then people might not perceive humans as more authentic sources for information, and messages from humans are not necessarily more authentic than those from capable machine sources. Fourth, the findings of this thesis can be meaningful for news companies to operate their news bot accounts on social media. As the type of agent was not a significant factor in the perception of message credibility, source credibility, and bias. Fifth, this thesis could be used as ground information for social media companies for governing AI and users on their platforms. Among those high in the machine heuristic there was an effects for agent on perceptions of competence and goodwill by the different types of agents. This should be considered when discussing what makes social media platforms more or less authentic public spheres. Lastly, the findings can be useful for the discussion on AI regulations (especially on generative AI). If how people perceive AI spreading information is different from machines in the past, is AI-generated information authentic? What can be regarded as trusted and authentic information on social media? This thesis presents further questions and points to be discussed for AI regulations.