How to Survive Artificial Intelligence

Summary

Rejects the three dominant AI narratives (doom, salvation, transhumanist enhancement) as ways of avoiding a catastrophe already underway. Griziotti treats AI as a relational reality rather than “AI vs. the human.” It is a web of people, algorithms, hardware, capital, and politics, analyzed through Barad’s diffraction, Simondon, and his own concept of neurocapitalism. He argues that AI is the operative core of a “technofascist megamachine”: hardware-centered, ecologically devastating, financially circular, and already deployed in surveillance and war. The epilogue looks to Gen Z revolts and to “fragile, partial, interstitial” alternatives: federated platforms, data cooperatives, mesh networks, small local models, and decolonial datasets.

Why it’s useful here

• Foundational text for the ai thread. It places AI inside the same relations of extraction and control that the rest of the blog addresses.
• Its three directions match this blog’s work directly.
  • Seizing infrastructures from monopoly: Mastodon, data co-ops, Indigenous data sovereignty.
  • Breaking ecological devastation: lightweight local models on community devices, mesh networks, situated AIs on small datasets that federate while staying autonomous.
  • Incorporating subaltern knowledge: decolonial datasets, Masakhane.
• The question it leaves open is this blog’s question: “How can these practices scale without reproducing centralization?” This is where IAD and polycentricity come in.
• Hardware is central again. AI’s concentration in GPUs and data centers resembles the mainframe era. That makes community-owned compute a political question, not just a technical one.
• Models critical use. Griziotti co-wrote the essay with a chatbot and reports on it honestly (“the work of discernment is irreplaceable”). That is relevant to how this project itself uses AI.

Key ideas

• Neurocapitalism: emotions, cognition, and affect become both raw material for value and variables of social control
• Meta-technics vs. meta-automation: the human capacity to invent new techniques vs. ML as the “automation of automation” (after Pasquinelli)
• “Everything is a hallucination” (after Colin Fraser): LLM output is always probabilistic, so the problem is using a probabilistic tool for deterministic tasks
• World-technology (after Caronia): LLMs are not a prosthesis but an ecosystem that reshapes knowledge and epistemic authority
• Comfort as control (after Mhalla): “The true opposite of democracy today is not dictatorship; it is comfort”
• Computational necropolitics: automated decisions about whom to surveil, exclude, and kill (Lavender, Where’s Daddy?, Palantir)
• Molecular strategy: accumulate partial victories and points of friction rather than storming a “Winter Palace”

Related

• griziotti-2018-neurocapitalism.md
• doctorow-2025-enshittification.md
• ostrom-2010-beyond-markets-and-states.md (polycentricity as an answer to “scale without centralization”)
• bookchin-1971-post-scarcity-anarchism.md
• illich-1973-tools-for-conviviality.md
• Local copy and related sources (Mhalla, Caronia, Griziotti on neurocapitalism): ~/data-commons/reference/