Generative AI as recombinant appropriation: multimodal AI, deepfakes, speech synthesis, and the voice cloning of Slavoj Žižek and Werner Herzog in The Infinite Conversation (2022)

E-mail: e.voigts@tu-braunschweig.deAbstract The article considers generative AI applications as cases of automated recombinant appropriation. Recombining existing data, generative AI is an automated appropriation machinery. Recombinant appropriation is precisely what occurs when large language models (LLMs) ‘adopt’ and ‘adapt’ material available to them in big data reservoirs, stringing tokens together in acts of purely statistical ordering. Considering questions of scale, originality, authorship, and accountability, the article discusses the ‘post-artificial’ creativity and meaning-making in AI appropriation. Intersemiotic or multimodal AI systems are of interest to adaptation studies as they integrate and process various kinds of data modalities, not just language and text, but also audio, images, or video. The article argues for the importance of the critical practice of AI appropriation art, focusing as an example on the AI-powered speech synthesis technologies employed in Giacomo Miceli’s The Infinite Conversation (2022) and Jens Hollerup Helle’s (non-AI) conversation remixes on YouTube (The Talk, 2023–). Miceli’s use of large LLMs and text-to-speech technologies to create The Infinite Conversation from a recombination of readymade content comments critically on deepfakes and, more generally, the detrimental implications of everyday appropriations by generative AI. AI-automated recombinant appropriation should be welcome only if it is critically incisive and inventive, exposing the appropriative extraction of human labour and data as well as misuses such as harmful deepfakes.