On AI Governance: Taming the Chimera
K.A. Taipale *
Abstract:
Generative AI systems are increasingly employed to mediate human decision-making and provide essential services, yet they remain at their core epistemically unstable stochastic engines that simulate coherence and presence without possessing cognition or personhood. In addition to the inherent instabilities of their statistical architecture, AI systems are further shaped by opaque design and deployment choices that embed normative values throughout the system and introduce additional epistemic risk.
As a result, consequential human decision making and action are increasingly delegated to machines whose epistemic foundations differ fundamentally from human reasoning. This delegation is accelerating regardless of specific technical architectures. This monograph argues that AI governance should be understood as the problem of structuring the epistemic conditions under which such delegation is warranted or can be socially justified. It develops an analytic framework for evaluating those conditions and locating accountability in the technical, institutional, and civic architectures that exercise control over these conditions.
Drawing on law, philosophy, STS, and systems architecture, this monograph examines how generative simulations acquire social and institutional force, how they induce unwarranted reliance, and how they are engineered and shaped. It argues that governance must be reoriented toward structuring the conditions under which reliance is warranted—not as machine minds to be aligned or technology to be regulated, but as socio-technical infrastructures to be governed and held accountable.
Part I diagnoses the nature of generative simulation, its persuasive force, and its inherent instabilities. Part II analyzes the mechanisms through which simulation is engineered, sculpted, and politically shaped, showing how design interventions often compound inherent epistemic risks. Part III develops an analytic framework—a reliance calculus—for evaluating the conditions under which reliance on machine-mediated cognition can occur. The framework evaluates situational seduction, epistemic conditions, and the context and stakes of deployment in order to locate responsibility and accountability across design, deployment, and use.
Rather than prescribing specific regulatory structures or policies, this monograph offers an agnostic methodology that diverse actors—developers, regulators, institutions, courts, insurers, and users—can apply within their own contexts and circumstances to judge whether reliance is warranted. It argues that those who exercise control over the epistemic conditions of reliance bear a corresponding duty to act non-arbitrarily with respect to those conditions, and that accountability attaches to the fulfillment or failure of that duty. In this sense, taming the chimera means holding architecture—not illusion—responsible for reliance under conditions of uncertainty.
Authors Note: This monograph is written for those interested in shaping AI governance, whether through scholarship, policy, legal doctrine, institutional practice, systems design and development, or deployment decisions. While it draws on insights from diverse fields, the aim is not to narrowly address specialists within those fields but rather to offer an analytic frame that enables conversation, understanding, and coordination across disciplinary and professional boundaries.