Suspension of Judgment in Artificial Intelligence - Uncovering Uncertainty in Data-Based and Logic-Based Systems

Abstract

This thesis demonstrates how suspension of judgment can be integrated into systems of artificial intelligence (AI). Suspension of judgment is a crucial epistemological phenomenon that allows humans to remain neutral and to refrain from forming definitive opinions in unclear situations. A successful implementation of suspended judgment into AI systems presents a promising approach to mitigating erroneous outputs, particularly in high-stakes domains. Consequently, this research aims to analyze various AI systems to identify and enhance their ability to react neutrally when confronted with uncertain or conflicting information. By exploring the nature of suspension and its epistemological norms, this thesis provides a philosophical analysis of fruitful implementations of neutrality in AI systems. It introduces various case studies of different AI frameworks, covering both logic-based and data-based systems, and critically assesses their current and potential capabilities to suspend judgment. The analysis reveals that while some architectures inherently possess mechanisms to communicate neutrality, others lack an appropriate capacity to respond constructively in the face of uncertain or conflicting information. As a solution, fundamental modifications are proposed for incorporating the option to suspend into existing AI architectures. The thesis contributes significantly to the fields of epistemology, philosophy of mind, and artificial intelligence, providing a deeper understanding of the epistemic possibilities of AI systems. The findings have practical implications for the development of more robust and reliable AI systems, potentially capable of acknowledging and expressing uncertainties. Such advancements are essential for enhancing transparency, trustworthiness, and effective human-AI interaction.