Nostophiliac AI: Artificial Collective Memories, Large Datasets and AI Hallucinations

Abstract This article explores the concept of Artificial Collective Memories as an interplay between collective memory and large datasets employed by ai algorithms. A collective memory represents “the recollection of events shared by a group” (Roediger, 2015). It includes lived experiences and human interactions. Contrarily, large datasets used by ai algorithms, such as Large Language Models ( llm s) or text-to-image models, are amassed by scraping internet data representing human communication. Those large datasets and the vast number of digitised representations of memories they encapsulate are references to the collective memories they represent, which we refer to as “Artificial Collective Memories”. Using our audio-visual project Nostophiliac ai as the starting point of this discourse and with references to our projects The ai Oracle and ai Reconstructing Realities , we are exploring areas of convergence between artificial intelligence and collective memories. With this article, we aim to enhance our understanding of ai by examining it through the lens of collective memory and, inversely, to obtain fresh insights on human collective memory through observations of generative ai ’s applications.