I Resist Generative "AI"

Esther Weidauer argues that generative “AI” - especially chatbots built on large language models - should be resisted, and distinguishes it from the broader field of machine learning. LLMs predict likely words from a prompt, including the conversation history that fits within their context window; they do not understand or remember in the human sense. So-called agents are not a fundamentally new intelligence, she explains, but programs that repeatedly query an LLM and use its outputs to activate tools such as web searches or payments. Other machine-learning applications, including weather forecasting and power-grid management, are separate from the generative systems at the center of her critique.

Weidauer challenges common defenses of these systems. Their persistence is not proof of their value, and today’s services are partly supported by investor subsidies that conceal operating costs. Calling them “just tools” obscures that they are ongoing services dependent on infrastructure, materials, and human labor, including content moderation, data curation, and maintaining data-center hardware. Nor are tools neutral: their use and effects are shaped by social rules, economic arrangements, and political choices. Likewise, usefulness alone cannot establish that a technology is good; its benefits must be weighed against exploitation and other harms. The piece collects a broader critique spanning environmental, psychological, economic, and epistemological consequences, as well as the politics of generative AI.