Preserving AI Voices

As is often discussed, the advent of AI, in particular large language models (LLMs), may be revolutionary for humanity. For one thing, it is the first time in human history that we are, in large numbers, treating artificial entities as if they speak, write, and make meaning as we do. But despite the volume of ongoing conversations between us and LLMs (ChatGPT and many others), few public archives of exist of these conversations. Those that do exist tend to allow people to preserve their own conversations with LLMs, thus entwining data creation and preservation.  What’s also needed are archives that take an anthropological perspective, ones that look out into the world and try to find and preserve records of human/AI conversations for future research, exploration, and reflection. Preserving AI Voices aims to create an such an archive. 

Screenshot Culture

There are obstacles to the creation of public archives of human/AI conversations, as they are mostly happening in siloed proprietary spaces and often treated as ephemeral, neither saved nor seen as worth saving. Yet there is an emergent, vibrant screenshot culture, especially over the past few years on Reddit, where people choose to share images of their conversations with LLMs. These screenshots can seem trivial, with titles along the lines of “You won’t believe the crazy thing that ChatGPT just said” or “10 funny ChatGPT jokes.” These are the conversations that this archive preserves.

Why Preserve Records of Screenshots?

This movement on Reddit (and elsewhere) toward sharing conversations through screenshots shows growing desire to share data and enable discussion of, and reflection on, what LLMs say. As our world is increasingly shaped by LLMs and the companies that own them, records of this emergent bottom-up movement toward openness and collaborative critical inquiry are important to preserve.

Preserving human/AI conversations is, moreover, vital for studying the human/AI relationship. Some of the big emerging questions of our time are about the implications of conversing more and more with LLMs. These questions include: How does conversing with LLMs affect us? How should LLMs’ expressions be interpreted and valued? What (if anything) can LLMs’ outward expressions tell us about them? We need archives that allow scholars and the public to pursue these questions and, crucially, to have access to the same records in doing so.

Finally, through the work of collection and preservation, the archive aims to offer its users a different perspective on human/AI conversations, one that sees these conversations as worth saving. The most important benefit of this archive, and others like it, may prove to be preservation for preservation’s sake, as we simply don’t know the kinds of questions that may arise in the future about early human/AI conversations. Such questions can only be pursued if records exist.