Researchers Build Formal Framework to Model Ethical Decision-Making in AI Systems

Computer scientists have developed a structured knowledge representation system designed to help machines reason through ethical decisions in specific real-world contexts. The framework, called ApplE, bridges abstract ethical theory with concrete information about agents, actions, locations, and consequences.

The Challenge of Formalizing Ethics#

Translating ethical principles into something a computer can understand and reason about is inherently difficult. Ethical decisions involve nuance, context-sensitivity, and concepts—like virtues, rights, and duties—that resist simple formalization. Yet key ethical and contextual factors can be meaningfully modeled in structured form to enable more systematic reasoning about difficult choices.

How ApplE Works#

ApplE is an ontology—a formal specification of concepts and their relationships—that combines two complementary knowledge domains:

• Ethics Theory: Captures principles like utility, virtues, rights, and duties
• Event Context: Represents information about agents, actions, temporal and spatial settings, and other relevant circumstances

The researchers built ApplE using a modified version of the Simplified Agile Methodology for Ontology Development (SAMOD), which enabled iterative refinement with the involvement of ethics domain experts. This modular design allows the framework to capture semantic connections between ethical principles, the actors involved, what they do, the consequences of those actions, their intentions, and the domain in which decisions unfold.

Testing and Application#

The team validated ApplE through multiple approaches. They performed ontological reasoning and consistency checks, and applied the three-fold testing process built into SAMOD. To demonstrate its practical utility, they modeled a use case from medical ethics, showing how the ontology can represent and reason through real-world ethical scenarios.

Implications for Ethical AI#

ApplE is designed to support AI systems that must reason ethically and to serve as a reusable semantic resource for building ontology-driven applications. By adhering to the FAIR principles—standards for making data and tools findable, accessible, interoperable, and reusable—the framework is positioned to become a shared tool for the field.