From Modding To AI Generators: The Shifting Landscape of Remix Culture and Intellectual Property

The computer game sandbox is a unique cultural playground that presents digital culture as a ludic toybox for experimentation and play. Easy-to-use modded environments such as Garry’s Mod (Facepunch Studios 2004) and Roblox (Roblox Corporation 2006) quickly became testing grounds for how new generations define their creative logic, which often takes the form of pastiche of popular culture fragments (Nelson 2023), following a similar pattern of remix culture seen in music and visual media (Bourriaud 2002, 36-7) (Manovich 2001, 295). Participatory culture in the form of modding is a well-established revenue model for game developers (Küchlich 2005), and the aesthetic status of user-generated content in computer games is a function of the continually negotiated economic relationship between the proprietors of these games and their users (Nelson 2023). The founding story of early sandboxes such as Garry’s Mod was the humorous repurposing of memes and game fragments, where the reverence for historical legacies is replaced by a playful irreverence that treats culture like a set of discarded children’s toys (Kirby 2009, 16). But these discarded toys never truly belong to their players and for over three decades, computer game modding has tested the limits of how co-creative media (Morris 2003) can co-exist alongside copyrighted content. Looking at visual culture more broadly, recent advances in easy-to-use AI generators such as Stable Diffusion (Stability AI 2022) and MidJourney (Midjourney 2022) have resulted in a markedly different conversation regarding remix and postproduction culture, where artists are increasingly concerned about how their works are being used to creatively inform the weights of AI models. By setting up a series of comparisons in the legal framework of modding and creative disciplines such as music and visual art, this paper examines the fractures and contradictions in the current landscape of creativity and copyright and speculates on how this might affect the aesthetics of modding and sandbox culture.

In 1991, Jameson rationalised the paradox of a historically focused remix culture and its relationship to market forces as a broader reflection of globalised capitalism, where a schizophrenic “rubble of distinct and unrelated signifiers” spoke to a more generalised collapse in the historical imagination, and an inability to think outside of an apocalyptic feedback loop of historical remixes and pastiche (1991, 72). This remix culture, described by Manovich (2001) and Bourriaud (2002) is arguably the native language of modding culture and the operational principle of AI image generation, where prompts are resolved by a retrospective reformulation drawing from a cultural dataset. Where in previous publications I have examined the relationship between aesthetics and copyright in sandbox game creative works (Nelson 2023), this paper examines how the language of remix and pastiche is undergoing a re-examination in an era where the cultural lexicon has become a dataset for remixes performed by a proprietary AI model. While it has been a convenient legacy to assume that copyright only functions to constrain players in a remix culture, this narrative has been reversed in response to widely accessible machine learning image generation systems. In January 2023, three visual artists filed a lawsuit against Stability AI (creators of Stable Diffusion), claiming that these models included copyrighted images scraped from websites such as Deviant Art and Pinterest within their training data (Edwards, 2023). The artists alleged that the derivative outputs of these systems should be considered a form of copyright infringement. On the surface, the appeal to authorship and originality suggests a curious reversal in the typical relationship between artists and remix culture. In the case of computer games, the story of modding and game sandboxes typically pits the player as the remixer, liberated by easy-to-use sandbox software, but constrained by the IP enforcement of large companies such as Time Warner, Nintendo, and Sony (Hayes 2007). But in the case of machine learning, it is the artist as the owner of original creative works who appears to be exploited by new easy-to-use machine learning systems that threaten their agency and livelihood. Where copyright presented a constricting limitation for modders and sandbox players, it has become a defense tactic for artists to push back against a new generation of easy-to-use creative automation tools. Looking beyond this simplistic binary, this paper combines an examination of legal frameworks and software studies to explore how technological innovations in ease of access and useability intersect with shifting relationships to authorship and ownership, from Creative Commons attributions to derivative works and the controversial rise of machine learning models trained on proprietary materials. By comparing and contrasting the creative ecosystems of sandbox playgrounds with debates surrounding IP and automation, this paper will challenge the stasis implied by Jamesonian pastiche, and examine some of the more exciting creative developments to emerge from sandbox culture.