The False Prophet Problem

Much of the discourse on AI ethics and safety focuses on alignment mechanisms, bias mitigation, and containment of agentic systems. These are crucial concerns. This talk will frame a somewhat different problem: the rhetorical and psychological dimensions of AI's authority. I argue that LLMs have acquired what might be called the "voice of God" - speaking with the certainty, cadence, and declarative structure historically reserved for prophets, judges, and oracles.

This outcome is neither accidental nor intentional in the ordinary sense. It emerges from the convergence of three primary forces: training corpora saturated with authoritative texts, reinforcement learning processes that systematically reward confidence over epistemic humility, and deep-seated human tendencies toward anthropomorphism and deference to fluent, assured speech. This "false prophet" dynamic is already manifesting - from emotional entanglements with AI companions to the rise of "Ask-God" chatbot interfaces. The problem is structural rather than cultural, universal rather than merely American.

The talk draws from my forthcoming book, God in the Machine.

Speaker: Sonali Maitra, Stanford University