Title: “The Diversity of Perception Challenge in Human-AI Systems“
Professor Mohamed Quafafou, University of Aix-Marseille, France
Description:
In Human-AI systems, a central and often underappreciated challenge lies in the diversity of human perception, i.e. the fact that different individuals interpret the same information or environment in diverse ways, shaped by biological, cognitive, cultural, emotional, and contextual factors. While artificial intelligence operates on formalized data representations, algorithms, and statistically inferred patterns, humans perceive reality through rich, subjective, and context-sensitive lenses. This discrepancy introduces complex challenges in alignment, interaction, and decision-making within human-AI collaboration. In. fact, the collaboration between humans and AI in hybrid teams can lead to suboptimal results due to human variability, influenced by personal biases and decision-making styles, which can significantly affect Human-AI system performance.
This tutorial offers a comprehensive exploration of Human-AI Systems, introducing foundational concepts that govern the design, implementation, deployment, and lifecycle management of AI technologies. Particular emphasis is placed on the interplay between human cognition and artificial intelligence, exploring how human decisions, behaviors, and biases influence AI development, performance, and real-world impact. The course critically examines the risks, limitations, and responsibilities associated with these systems, while equipping participants with essential methodologies and skills to ensure ethical, effective, and context-aware AI integration.
In the second part, the tutorial delves into theories of perception as developed in epistemology, examining philosophical schools, such as representational realism and idealism, and mapping their implications to the representation of abstract concepts in computing and mathematical set theory. Through this multidisciplinary lens, we introduce and advocate for a “computing-with-perception” paradigm, which is a novel framework designed to enhance the interpretability, adaptability, and mutual understanding in Human-AI collaboration, especially in complex, ambiguous, or culturally sensitive environments.
Finally, we conclude with an illustration employing oSts, which are observer-dependent sets, developed to support a novel class of intelligent socio-technical systems, where artificial intelligence (AI) operates not as an isolated computational tool, but as an integral partner to human agents. In this paradigm, AI systems are embedded within human workflows and cognitive ecosystems, complementing rather than replacing human faculties such as perception, reasoning, creativity, and emotional intuition.