Scholarlink

<p>Recent work in AI ethics has revived Kantian deontology as a potential framework for AI alignment, yet questions remain about whether Kant’s formula of universal law (FUL) can provide concrete action guidance for artificial systems. This paper develops a Kantian-inspired procedural framework that defends the FUL as a viable alignment procedure by clarifying its normative structure and demonstrating how it can be operationalized computationally. Unlike standard reinforcement learning from human feedback, which learns preferences without imposing a prior formal constraint on permissibility, the paper proposes inverse reinforcement learning as a method for enabling AI systems to learn maxim formation and moral constraints from human exemplars, rather than relying on hard-coded rules. Through case studies in algorithmic hiring and content moderation, the paper shows that the FUL, understood as a procedural test, can offer principled and practically applicable guidance for AI alignment without presupposing machine moral agency.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Maxims for machines: operationalizing Kant’s universal law via inverse reinforcement learning

  • 期刊论文

  • 发表日期：

    2026年7月19日

查看全文
  • Oluwaseun Sanwoolu

摘要

Recent work in AI ethics has revived Kantian deontology as a potential framework for AI alignment, yet questions remain about whether Kant’s formula of universal law (FUL) can provide concrete action guidance for artificial systems. This paper develops a Kantian-inspired procedural framework that defends the FUL as a viable alignment procedure by clarifying its normative structure and demonstrating how it can be operationalized computationally. Unlike standard reinforcement learning from human feedback, which learns preferences without imposing a prior formal constraint on permissibility, the paper proposes inverse reinforcement learning as a method for enabling AI systems to learn maxim formation and moral constraints from human exemplars, rather than relying on hard-coded rules. Through case studies in algorithmic hiring and content moderation, the paper shows that the FUL, understood as a procedural test, can offer principled and practically applicable guidance for AI alignment without presupposing machine moral agency.