AI Waste: The invisible hyper-linear model of Generative AI

In the field of emerging technologies, the issue of the environmental and social impact of the digital domain is becoming a critical area of investigation. Specifically, we aim to initiate a debate on the invisible and hyper-linear consumption model intrinsic to Generative AI, highlighting the implications of a phenomenon that has been little explored so far, but is of growing relevance in the context of a global polycrisis that requires urgent and strenuous rethinking at multiple levels. The goal is to stimulate multidisciplinary reflection on the policies, directives, and design practices necessary to address the waste of resources generated by these systems, aligning with the principles of sustainable digitalization and circularity. Despite circularity being a well-established principle in the design of material artefacts, its transposition into the digital domain at the current state does not have a relevant correspondence. Ongoing research by the authors reveals how the immaterial infrastructure of the digital, often perceived as an “intangible cloud”, embodies a significant consumption of energy, materials, and physical space. Generative AI represents a critical field of study, given its tendency for hypertrophic content production that, if discarded, translates into an unprecedented waste of resources. Generating images is one of the most energy-intensive work of generative AIs: according to a study by Hugging Face and Carnegie Mellon University, generating a single image using an AI model requires the same energy needed to fully charge a smartphone. The article investigates the nature and consequences of AI waste, defining the term and exploring its manifestations. Concrete examples that highlight the severity of the linear production consumption-waste model inherent in these technologies are considered. This “hyper-linear” model, characterised by an actual production cycle that even skips “consumption” to go directly to “waste”, represents a significant challenge to the principles of sustainability and circularity. The research and debate questions focus on the fate of the waste generated by AI: where is it stored, for how long, and with what environmental and social consequences? And even if not stored, what are the possible alternatives that can address the consumption of energy, water, places, and resources related to hypertrophic generation? By critically exploring exemplary cases, we intend to open a discussion on the need to propose innovative solutions for the management and reduction of such waste, including behavioural and systemic design approaches, such as temporary waste databases that avoid its instant oblivion.