AI as a Social Tech: Forget Sci-Fi, Says Scholar

Political scientist Henry Farell (Blavatnik School of Government, Oxford) has reignited debates about AI’s societal role with his talk and March 2025 paper titled *Large AI Models are Cultural and Social Technologies*. Farell argues that modern AI systems function as 'Lossy Information Aggregation Tools'—systems that compress vast datasets into simplified representations, akin to historical tools like state bureaucracies (as described by James C. Scott) or markets (per Friedrich Hayek). This framing highlights AI’s limitations: its outputs are inherently incomplete and distorted, a feature, not a bug, that shapes its real-world consequences.

Farell critiques the dominant AI narratives—particularly the 'AI 2027' camp predicting rapid AGI development and the 'AI as a normal technology' perspective—as overly simplistic and sci-fi-tinged. He dismisses AGI forecasting as speculative distraction, urging focus on current risks, such as AI’s role in critical infrastructure. Citing a statistical physicist colleague, he argues current AI models are fundamentally 'Markovian'—unable to break free from their training data patterns, despite surprising emergent capabilities. Farell calls for greater collaboration between social scientists and STEM researchers to ground AI governance in empirical reality rather than futurist tropes.