Navigating Uncertainty : Comparing AI and Human Language in the Expression of Confidence

Sammanfattning

This paper investigates how the language of Artificial Intelligence (AI) could affect user trust and perception. By comparing the language patterns of AI-generated responses to natural human language, this study examines linguistic differences, particularly in the expression of uncertainty. Research questions about the characteristics of AI-generated responses, particularly the absence of uncertainty markers, and how these features potentially could influence users' acceptance of information, will be addressed. This paper is rooted in theory about uncertainty in language and uses a qualitative comparative analysis of AI language patterns by focusing on hallucinations from the AI models ChatGPT and Copilot. Results demonstrate that human language expresses significantly more markers of uncertainty, and that the lack of these markers in AI responses could influence and mislead users into believing information is accurate, even when it is not. The findings suggest that it may be worth exploring whether the level of certainty in AIgenerated speech should better align with its actual reliability. This research contributes to understanding the importance of maintaining a critical perspective when engaging with AI.