Cartography of Generative AI (cartography-of-generative-ai.net)

This piece maps generative AI as a global, layered industry rather than a purely technical novelty: models are trained by statistically imitating vast, scraped internet datasets to predict and reassemble words, pixels and sounds; their user-facing fluency is then honed through fine‑tuning and moderation performed by low‑paid micro‑workers in the Global South (documented in places like Lebanon, Uganda and Kenya). Cultural producers supply much of the raw creative material and simultaneously face displacement as automated services outpace human labor. Start-ups (OpenAI, DeepMind, Anthropic) and media narratives about existential AI risk concentrate influence and capital, shaping calls for self‑regulation even as governments begin formal rules (EU AI law, early 2024). The infrastructure and materiality behind these models are equally consequential: GPU‑centric training runs on data‑centre fleets supplied by near‑monopolies (notably Nvidia) and fabs like TSMC using ASML lithography. Producing chips and batteries demands large quantities of copper, gold, lithium and cobalt—linked to environmental damage, water depletion and human-rights abuses in Chile, Peru, Brazil and the Congo. The result is an opaque, extractive supply chain that externalizes social and ecological costs while consolidating power among a few tech and semiconductor firms—raising technical, ethical and regulatory challenges for the AI/ML community around provenance, labor standards, accountability and sustainable compute.