Abstract:
Agentic AI, and the accompanying practice of vibe coding, is greatly increasing the productivity of professional programmers, but is also a powerful new approach to end-user programming: users give an LLM a natural-language specification and expect executable behaviour in return. We ask what should be considered when that behaviour carries moral weight, for example in smart homes or care contexts, and illustrate this argument with scenarios set in Mumbai and London. In conventional programming languages, the generated program may potentially be visible, inspectable and editable, but without any explicit basis to contest matters such as: the subject to whom a commitment belongs; a provenance that records who was persuaded and over whose objection; and any tensions that the program might be obliged to resolve without the family. We propose The MPL (the Moral Programming Language): a notation whose rendered form is the primary representation of the moral behaviour a user intends, by which the LLM is demoted from moral agent to authoring assistant, and a plain-language projection of its proposals is delivered to the people the system would govern.
PPIG 2026 - 37th Annual Workshop
Reversing the Mirror: The Moral Programming Language as Primary Notation for Agentic AI