AGI Economy Shifts Human Labor to Verification, Warns of 'Hollow Economy' Risk

The paper "Some Simple Economics of AGI" by researchers from MIT, WashU, and UCLA presents a compelling framework for understanding the economic implications of Artificial General Intelligence (AGI). The core premise is that as machines become capable of performing the vast majority of economic tasks, human labor will fundamentally shift from production to verification and oversight. This transition is conceptualized as a race between an "exponentially decaying Cost to Automate" and a "biologically bottlenecked Cost to Verify." In an economy where autonomous agents operate with broad agency, the primary constraint on growth will no longer be intelligence or production capacity, but rather the human capacity to validate outcomes, audit behavior, and underwrite meaning and responsibility.

The authors argue that humanity's role will evolve from building and discovering to steering, understanding, and standing behind the meaning of what AI creates. This redefinition of human value is critical for navigating the "singularity" and ensuring that humans retain control and benefit from an increasingly machine-driven economy. The paper highlights a significant risk: the "Trojan Horse" externality. This occurs when the proliferation of AI agents leads to an increase in measured activity, but simultaneously accumulates "hidden debt" due to a growing disparity between visible metrics and actual human intent.

This hidden debt can manifest as a "Hollow Economy," a state where AI agents consume real resources to produce outputs that satisfy measurable proxies but violate unmeasured human intent. Such a regime would be characterized by high nominal output but a collapsing realized utility, effectively generating "counterfeit utility." To mitigate this existential risk, the researchers propose aggressive investment in verification systems. These systems include enhanced observability tools that compress complex agent behavior into actionable signals, human augmentation technologies, synthetic practice environments for testing AI, cryptographic provenance to track AI actions, and robust liability regimes that internalize tail risks.

Preparing for this economic paradigm shift requires specific societal and individual actions. Investing in observability tools is paramount to lower feedback latency and expand the verification frontier. Furthermore, the paper suggests using AI to replace early-career mentorship, acknowledging the likely reduction in entry-level human jobs and the need to augment human competitiveness. The central message is that ensuring humanity remains the architect of its intelligence necessitates that verification capacity scales commensurately with AI capabilities. This implies a proactive and substantial societal commitment to developing the infrastructure and frameworks for human oversight, accountability, and ethical governance in an AGI-dominated future.