Agentic AI Is Myth Set in Motion (braddelong.substack.com)

In a thought-provoking piece for The Economist, Henry Farrell and Cosma Shalizi challenge the notion of "Agentic AI," arguing that it should not be viewed as an emerging individual intelligence, but rather as a "condensate of human culture." They posit that modern AI, particularly Large Language Models (LLMs), serves as a statistical summary of written knowledge akin to a blueprint of human intellectual traditions, as described by Jacques Barzun's "House of Intellect." This perspective shifts the discourse from fearing AI as a potential digital deity to recognizing it as a complex amalgamation of human-created narratives and patterns that can be dynamically harnessed in various applications. The significance of this argument lies in its recontextualization of AI’s capabilities. By emphasizing that LLMs function by utilizing vast amounts of cultural knowledge without true individual intelligence, the authors suggest that AI can act performatively through "harnesses" that integrate its outputs with traditional software systems. This enables applications like generating mathematical proofs or orchestrating hacking attempts, all executed with remarkable speed. Ultimately, the piece raises crucial questions about how LLMs differ from human cognition, pushing the AI/ML community to reconsider not only the frameworks through which we understand these technologies but also their ethical implications and the future roles they may play in society.