The Artificial in ‘Artificial Intelligence’: How Imagination Shapes AI Regulation

Review

This paper makes a sharp and timely intervention into AI regulation by arguing that legal and policy debate is not merely shaped by technical facts, but by the metaphors through which AI is rendered thinkable. Using cognitive linguistics as its interpretive frame, it examines three dominant terms in contemporary AI discourse (“intelligence,” “black box,” and “hallucination”) and shows how each carries hidden assumptions that steer regulatory attention, allocate responsibility, and create doctrinal path dependence.

The paper’s central contribution is conceptual. It treats metaphor not as decorative rhetoric, but as governance infrastructure. That is an important move. In practice, AI regulation often inherits its categories from public discourse, vendor language, and adjacent legal domains. By showing how “intelligence” encourages anthropomorphism, how “black box” falsely localizes causality inside a bounded technical object, and how “hallucination” naturalizes foreseeable design failures, the paper exposes how language can quietly distort accountability. This is especially valuable for public-interest technology work, where naming errors routinely become governance errors.

The analysis of “hallucination” is particularly original. Reframing these outputs as “deferential hazards” usefully shifts the discussion from pseudo-pathology to design choices, interaction patterns, and liability. Likewise, the critique of chain-of-thought as pseudo-explainability is excellent: it identifies a real risk that plausible narratives may be mistaken for causal explanations in administrative, judicial, or compliance settings. The proposal for AI Integration Assessments is also one of the paper’s most policy-relevant ideas, because it moves beyond model-centric transparency toward organizational accountability.

Methodologically, this is a doctrinal-conceptual paper rather than an empirical one, and on those terms it is largely persuasive. The authors clearly state that they are using cognitive linguistics and conceptual metaphor theory to analyze how legal meaning is structured. The paper is well-read, interdisciplinary, and internally coherent. However, its evidence base is mostly interpretive. It draws on legal scholarship, policy documents, court cases, and AI literature, but does not systematically code regulatory texts, test metaphor prevalence across jurisdictions, or demonstrate causal effects on legal outcomes. That does not invalidate the argument, but it does mean some of the stronger claims about regulatory path dependence remain more plausible than proven.

There are also some gaps. First, the paper provides its clearest account at diagnosis and somewhat thinner at comparative validation. It would benefit from a clearer account of why these three metaphors were selected over others such as “alignment,” “safety,” “agent,” “reasoning,” or “guardrails,” which also shape current AI governance. Second, the proposed term “deferential hazards” is analytically better than “hallucinations,” but it may struggle to travel outside expert circles; the paper acknowledges this, but the practical politics of adoption deserve fuller treatment. Third, while the argument is highly relevant to high-impact public systems, it does not engage enough with Digital Public Infrastructure contexts, where metaphors can harden into procurement standards, welfare system design, and citizen-facing legitimacy claims. That is fertile ground the paper could address more explicitly.

The paper is novel because it links legal theory, AI governance, and cognitive linguistics in a way that cuts through the usual hype/fear binary. It does not ask only what AI does, but how language pre-sorts what regulators think AI is. That is an unusually productive question. Its likely impact is highest among legal scholars, governance researchers, and policymakers designing accountability frameworks. For practitioners, its practical value lies in warning against lazy anthropomorphism and model fetishism. The core lesson is simple but potent: bad metaphors produce bad governance.

Overall, this is an intellectually strong and policy-relevant paper. It is not an empirical study of regulation in action, but as a conceptual intervention it is rigorous, distinctive, and genuinely useful.

Key Insight

AI regulation is being shaped not only by technical architectures but by metaphors that silently define where risk, responsibility, and accountability are presumed to sit.