Abstract

Studying the epistemology of current AI approaches, and how they relate to human sense making and cognition, is a worthwhile endeavour, as it allows observers to step back and gain a "meta" perspective on the work researchers undertake in the field. Current AI models, particularly large language models (LLMs), are predominantly grounded in positivist epistemology, treating knowledge as an external, objective entity derived from statistical patterns in data. This paradigm is appropriate for applications in natural sciences. However, it fails to capture "facts-in-the-conscience", the subjective, meaning-laden experiences central to human and social sciences. In contrast, phenomenology, that brings in hermeneutics and constructivism, provide a more fitting foundation for the development of AI applications in human and social sciences, recognizing knowledge as an intentional process, shaped by human interaction with the world, as well as by community consensus. Phenomenology highlights the subjective lived experience and intentionality necessary for hermeneutic meaning-making, while constructivism emphasizes the human active role in learning and the social negotiation of new knowledge within communities of practice. Of course, this requires human intervention and cooperation with AI algorithms, a practice already started: we underline here its epistemic conceptual foundation. This paper argues for a formal paradigm shift in AI applications in human and social sciences, enabling recursive interaction between AI algorithms and human cognition, thus integrating higher order cybernetics through human sense making. Such a shift makes AI not merely a tool for knowledge retrieval but a co-participant in epistemic evolution, supporting trustworthy, context-sensitive, and meaning-aware AI systems within socio-technical frameworks. Socio-technical research reinforces this perspective by illustrating how technological systems evolve through reciprocal adaptation with human practices, allowing AI to facilitate dynamic, participatory knowledge formation, rather than static data processing. Epistemic AI approaches involving humans can leverage autopoietic concepts - like human structural coupling, and social consensus domain - to cure inherent vulnerabilities of AI alone.

This paper further elaborates relevance and rigor of differing AI approaches via their relation to human processes of understanding and cognition, for the benefit of future researchers..

Download

Share

COinS