The Failed Epistemologies of AI? Making Sense of AI Errors, Failures and their Impacts on Society
Editors
- Prof. Veronica Barassi (veronica.barassi@unisg.ch)
- Dr. Philip Di Salvo (philip.disalvo@unisg.ch)
School of Humanities and Social Sciences
University of St. Gallen
Generative artificial intelligence tools and large language models are gaining a prominent space in our society. Probably for the first time in history, humans have now to relate and interact with technological systems capable of producing and generating new content and knowledge mimicking humans’ imagination, speech, and behaviors in ways that was not possible before. This new state of things brings inevitably profound consequences and potential sea changes for numerous social, scientific, and cultural fields raising epistemological, ethical, political economical and philosophical questions about the epistemologies of AI and the processes of knowledge production of these systems. The race for AI innovation is being framed with reference to the ‘superintelligence’ of our machines, their processing power, their ability to learn and generate knowledge. In public debate, AI technologies are admired for their powers, and feared for their threats. Yet, we are increasingly confronted with the fact that these machines make errors and mistakes, they are fallible and inaccurate, and they are often culturally biased. From Generative AI technologies that ‘hallucinate’ and invent facts to predictive policing technologies that lead to wrongful arrests, our world is quickly coming to terms with the fact that the AI we are building is not only astonishing and incredibly powerful, but often unable to understand the complexity of our human experience and our cultural worlds. Research has shown that AI errors and their problematic outcomes can’t be considered as mere coding glitches, but as the direct expression of the structural inequalities of our societies and they confront us with critical questions about our supposed anthropocentric position as knowledge-creators.
The aim of this special issue is to gather scholars coming from different fields of the social sciences and humanities to investigate how artificial intelligence systems are challenging epistemological assumptions in various societal areas and how the failures of such systems are impacting on knowledge creation and diffusion in their areas of interest. Overall, the special issue aims at overcoming dominant and hyped takes and narratives around AI and its supposed (super)powers, and critically reflect on how we can identify and learn how to coexist with the limitations of AI driven knowledge production.