Intellectual Furnishings

Critical AI studies with a historical perspective

Our core concept of “assistance” takes the agential relationship between human and AI to be an unequal but distributed phenomenon, circumventing the fruitless discourse of the machinic replacement of human beings. The assisting device acts as a moderator and facilitator of human ideas and knowledge production, and it is aimed at “creativity,” understood here as the productive outcome of a shared interaction rather than as the characteristic of a single participant.

Two foci elucidate these concepts from different historical and conceptual angles. While the “Deep History of AI” investigates the assistive function of “intellectual furnishings” like desks and writing cabinets from the Baroque to the present, the focus on prompts as the interface between human and machine examines the change of language use while becoming both a medium of assistance and an agent in the creative interaction with the technology.

The range of the disciplinary and methodological backgrounds together with a close collaboration with our partners from the history of science, media studies, and computer sciences aims to contributing to Critical AI Studies, an innovative field which just starts to emerge.

Focus 1: A Deep History of AI

A comprehensive history of AI with a focus on the interaction between scholars and their individual ’thinking devices’ or ‘intellectual furnishings’ like card indexes, tailor-made desks and other writing devices still needs to be written. Assisted Thinking complements the standard history in cybernetics and information science with a media-archaeological approach that focuses on the use of intellectual furnishings and scholarly practices. It excavates the roots and concepts of writing assistance by building a genealogy of thinking machines from 1700 to the present, discussing tools like Leibniz’ Zettelschrank, or the writing desks built by the Roentgens in Neuwied in the 18th century as early AI devices.

Niklas Luhmann repeatedly referred to his Zettelkasten as a “communication partner” which, in line with actor-media theory, possessed both agency and a certain autonomy. What is still missing is a deep historical investigation of the concept of the “scholarly machine”, placing it in the context of an innovative theory of assistance that includes the digital world and artificial intelligence. When Luhmann and other users of intellectual furnishings refer to their tools as “thinking” objects, they mostly speak on a metaphorical level. In the digital age, however, intellectual furnishings, especially those using methods of machine learning, have truly become learned machines.

Focus 2: Promptology (PhD project)

If assistance can take many forms in the interaction with writing environments, one of the most important significance of current LLMs, which increasingly serve as scholarly machines, is that they operate on the basis of natural language. Serving as a link to focus 1, this second focus follows the outsized role of language as the medium of assistance in modern language models.

For the very first time in computing history, machines do not work with a strictly formalized syntactical language such as a programming language like LISP that was specifically developed for AI use but the vernacular of everyday human speech itself. This has not only opened up LLMs to practically everyone with access to an Internet connection, doing away with expertise knowledge as a prerequisite for their use, as now natural-language “prompts” are the only instruction needed; it has also changed the status of language itself, both in its technical conceptualization and as a part of complex human-machine assemblages. Thus, language in language models operates both on the back-end and the front-end, both in the dataset LLMs are trained on as well as the interaction with these LLMs through natural-language prompts all the while appearing to change the use of language outside of these models. From both perspectives, this means a shift in our understanding of linguistic expression.

However, as the deep history of AI of focus 1 intends to prove, the interaction between humans and machines has always been a two-way street: As much as humans build machines to do their assistive bidding, the machines in turn force humans to fall in line with their ways of working. If the interface of human-machine interaction becomes natural language in prompt engineering, LLMs in turn train their users to follow a specific language use to make the machine work. This focus aims to contribute to a framework for assessing the feedback effects of language use and language generation a “promptology”. This focus will result in a dissertation examining the different functions of language at the interface between thinking machines of the present and the past and their users, facilitating both assistance and inspiration in a mutual manner.

Duration and Funding

The project has been funded with the generous support of The NOMIS Foundation, Zurich.

The project will start work in spring 2025 with a duration of three years. More information on events, publications and the related PhD project (in collaboration with the University of Berkeley, Cal.) will follow soon on this website.