Language

English (en)

Date of Award

Spring 5-14-2026

Author's School

College of Arts & Sciences

Author's Department

Philosophy

Degree Name

Bachelor of Arts (A.B.)

Restricted/Unrestricted

Unrestricted

Abstract

The rapid emergence of Agentic AI, which are systems capable of autonomous, multi-step decision-making with minimal human oversight, has prompted growing discourse and concern regarding whether such systems possess a form of agency comparable to that of human beings, and subsequently, what ethical responsibilities ensue if they do. This thesis argues that many of the ethical risks surrounding Agentic AI stem from a fundamental misconception, which is the tendency to describe these systems using the language of agency ("deciding," "acting," "choosing") without examining whether they satisfy the philosophical conditions for genuine rational agency. Drawing on Immanuel Kant's Groundwork of the Metaphysics of Morals, this thesis develops a framework for evaluating Agentic AI against three criteria of Kantian rational agency: acting from a represented understanding of principles, exercising spontaneity, and genuine responsiveness to reasons. Applying this framework, the analysis demonstrates that while Agentic AI systems exhibit sophisticated, rule-governed, and adaptive behavior, they fail to satisfy any of these criteria in the Kantian sense. The analysis reveals that their behavior remains fully heteronomous, determined by externally imposed training, objectives, and system architecture rather than self-authored principles. Instead of utilizing outcome-based frameworks such as utilitarianism, this thesis contends that a Kantian account is better suited to address the deeper conceptual question of who or what is acting, which is important to establish before any evaluation of consequences. Building on this conclusion, the thesis argues that moral responsibility for Agentic AI's behavior is non-transferable and remains distributed among the human agents involved (developers, deployers, and users) rather than diffusing said responsibility to the systems themselves. It further examines the threat that habitual reliance on Agentic AI poses to human rational autonomy, and concludes that recognizing these systems as heteronomous instruments, rather than rational agents, is essential to preserving both moral accountability and human self-governance in an increasingly AI-mediated world.

Mentor

Dr. Anne Margaret Baxley

Download

Share

COinS