The model is the museum: generative AI and the expropriation of...
Most current discussions around ‘AI’ frame these technologies as tools or assistants at the service of human actors. While convenient, these metaphors might obscure the fact that the effective training of Large Language Models and other kinds of machine learning systems require intensive data scraping and processing only available to the largest tech companies in the world. With that in mind, this essay seeks to examine the effects of generative AI in the field of arts, culture, and creativity by comparing these systems to social institutions. We contend that, similar to the modern museum, neural networks enable protocols of governance and dispersion that amplify the purchase of certain cultural signals.
Generative AI and the Modern Museum
Operating at scale, these technologies function much like an archive, which Michel Foucault examined as a locus for the accumulation and exercise of power. We argue that, if left unchecked, AI models might provoke a significant skewing of whole socio-cultural milieus according to their statistical rationality, toward what Hito Steyerl described as ‘mean images.’ We demonstrate how this skewing takes place in art projects such as Nora Al-Badri’s Babylonian Visions, which seems to reproduce a kind of essentialization typical of a colonial episteme, disconnecting visual patterns from historical circumstances and ways of doing. In conclusion, we propose that a comparison between AI models and the modern museum may shed new light on issues of data extractivism, cultural expropriation, and the assimilation of otherness through stereotypification and homogenization as accomplished by algorithmic means.
The Influence of Museums on Artistic Movements
As the story goes, Cubism owes its existence to Pablo Picasso’s visit to the Trocadéro Ethnographic Museum in Paris, in 1907. Later in the same year, the artist would complete the revolutionary Les Demoiselles D'avignon, marking a definitive departure from the pictorial paradigm of the Renaissance. The characters’ features, designed after African masks, more than demonstrate the impact the museum collection had on the artist. They underscore the influence of purported ‘primitive’ esthetics in shaping European modern art.
Though Picasso had other documented contacts with African art, the museum best encapsulates this open trade of cultural references. Museums played a pivotal role in modernity’s civilizing project as powerful epistemic infrastructures. In rendering things visible, museums stage common knowledge. Tony Bennett describes their capacity as an exhibitionary complex to ‘organize and co-ordinate an order of things and to produce a place for the people in relation to that order’. That means to say: while extracting objects from ordinary circulation, the museum turns them into components of classification and discourse.
The Archival Constitution of Generative AI Systems
In their capacity to redistribute objects as information at scale, museums offer a compelling lens to examine the impact of generative artificial intelligence in political economies of representation. The archival constitution of generative AI systems has reignited debates from the mid-2000s, when digital media facilitated a surge in creative practices rooted in detournement and appropriation.
Training generative AI systems involves distilling datasets into their constitutive patterns and reestablishing these patterns as the network’s parameters (‘weights’). No singular element survives the training process as-is; rather, they are collectively decomposed into the conditions for the model to operate as an information system. That means to say that, even though individual items are not stored within the model itself, they still exist therein as a set of distinct statistical probabilities. In that sense, training effectively produces the model as an archive, as famously described by Michel Foucault: a system that governs the orderly accumulation of statements and their appearance as unique events.
AI Systems and Cultural Appropriation
To begin with, it is crucial to acknowledge that generative AI platforms are liable as powerful cultural intermediates. Nowadays, much like the largest public institutions, neural network models are able to amplify the purchase of certain cultural signals. The more pervasive a model is, the more its protocols of governance and dispersion may affect media ecologies at large.
