Tuoya Saren (Emory University School of Law) has posted Who Speaks for the Algorithm? Attribution in Environmental Decision making on SSRN. Here is the abstract:
Artificial intelligence is rapidly reshaping environmental decision-making, yet administrative law has failed to articulate who, if anyone, is legally responsible for algorithmically generated analysis. Federal agencies increasingly rely on AI tools tomodel environmental impacts, project emissions, and synthesize massiveadministrative records under the National Environmental Policy Act (NEPA). As algorithmic systems move from peripheral support to the analytic core of Environmental Impact Statements, a foundational legal question emerges: whenanagency relies on artificial intelligence, whose reasoning is the law reviewing? Recent Supreme Court decisions—including Loper Bright Enterprises v. Raimondo and Seven County Infrastructure Coalition v. Eagle County—haveintensified this problem. By rejecting automatic deference and reaf irming that deference must be earned through reasoned explanation, the Court has placedrenewed emphasis on identifiable agency judgment. Yet existing administrativedoctrine presupposes human authorship and of ers no framework for determiningwhether algorithmic analysis can be attributed to the agency itself. As a result, environmental review risks becoming scientifically sophisticated yet legallyunowned—producing decisions that are transparent but not accountable, disclosedbut not attributable. This Article identifies a doctrinal vacuum at the intersection of administrative law, environmental governance, and artificial intelligence. It argues that courts andagencies have conflated two distinct inquiries: attribution and reliability. Beforecourts can assess whether AI-assisted environmental analysis is scientifically sound, they must first determine whether that analysis is legally attributable to the agency as institutional judgment. Attribution, this Article contends, is the threshold conditionof judicial review. To fill this gap, the Article proposes the Attribution and Adoption Test (AAT)—a governance framework for determining when algorithmically generatedenvironmental analysis may be treated as the agency’s own reasoning under theAdministrative Procedure Act and NEPA. The AAT operationalizes attributionthrough three dimensions: Institutional Control and Adoption, Epistemic Integrity, and Public Traceability. Rather than mandating technical audits or disclosure of proprietary code, the framework focuses on whether the administrative recorddemonstrates a transparent institutional chain of reasoning suf icient to support judicial review. By reconceptualizing attribution as a distinct procedural inquiry, this Articlerestores doctrinal coherence to environmental review in the algorithmic era. It of ers courts a principled basis for calibrating remedies, agencies a clear complianceroadmap, and administrative law a mechanism to preserve accountability without abandoning technological innovation.