Written with Alistair Alexander for presentation at After AI Symposium (17 May 2024)
AI can be seen as the ultimate and most extreme manifestation of extractive technological accelerationism. The rapid development of AI infrastructure to meet the exponential scaling of Large Language Models is already causing power shortages in the US energy grid (Halper 2024). Key figures in AI claim that its astronomical power demands will require a future energy “breakthrough” (Reuters 2024). According to Scientific American, large-scale energy from nuclear fusion is not likely to materialise “before around 2050 (the cautious might add on another decade)” (Ball 2024). Meanwhile, the resource inputs for manufacture of the specialised AI GPU processors are not disclosed by manufacturers, but they are also necessarily prodigious. The dynamics of industry hype, speculation, and financialisation exist independently of the material requirements and ecological consequences of AI technologies.
In this presentation we offer strategies to assess AI’s ecological and social impact – both positive and negative – beyond the current hype cycle. We present ecological frameworks to help to methodically assess the true potential of AI – or lack thereof – in a context of climate and ecological crises and (our hoped for) capacity for planetary regeneration. Like all technology, AI depends on natural resources and relatively stable ecological circumstances. We use the concept of “after” as part of a long series of post-bubble and post-collapse states where aspirations and promises of AI hit boundaries of resource requirements to manufacture AI, the energy to power AI, and the ecological consequences of these processes, including associated acceleration of GHG emissions, and associated climate change impacts. Intersecting vectors of ecological, social, and economic crises will all be accelerated with incautious approaches to AI where the supposed “unintended consequences” have yet to be accounted for in the new proposals. In this context, the speculation that widespread AGI can exist in the context of the polycrisis is pure fiction. The fact that the tech world has made this narrative as dominant as it is, reveals a deep denial and dismissal of the ecological context in tech discourses, a consequence of ecological illiteracy.
EcoLabs 