Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A computational framework for human values

Nardine Osman, Mark d'Inverno

TL;DR

This work introduces a formal, taxonomy-based framework for representing human values and grounding them in computable properties to support value-aligned AI. By modeling values as abstract concepts connected to leaf property nodes and organizing them into context-sensitive taxonomies with coherence-enforcing aggregation, the approach enables explicit reasoning about value importance and alignment. The framework also addresses how values evolve across contexts and how individuals and collectives can hold and aggregate values, using bottom-up or top-down implementations and a quantitative alignment measure $\mathcal{A}(e,\mathcal{V}_c)$ that weights property satisfaction by importance. Through the uHelp running example, the paper demonstrates practical grounding, context adaptation, and alignment computation, illustrating potential applications in domains such as healthcare and participatory design. Overall, the framework provides a formal, interdisciplinary foundation for designing AI systems whose behavior is provably aligned with human values, while outlining directions for future research and deployment.

Abstract

In the diverse array of work investigating the nature of human values from psychology, philosophy and social sciences, there is a clear consensus that values guide behaviour. More recently, a recognition that values provide a means to engineer ethical AI has emerged. Indeed, Stuart Russell proposed shifting AI's focus away from simply ``intelligence'' towards intelligence ``provably aligned with human values''. This challenge -- the value alignment problem -- with others including an AI's learning of human values, aggregating individual values to groups, and designing computational mechanisms to reason over values, has energised a sustained research effort. Despite this, no formal, computational definition of values has yet been proposed. We address this through a formal conceptual framework rooted in the social sciences, that provides a foundation for the systematic, integrated and interdisciplinary investigation into how human values can support designing ethical AI.

A computational framework for human values

TL;DR

This work introduces a formal, taxonomy-based framework for representing human values and grounding them in computable properties to support value-aligned AI. By modeling values as abstract concepts connected to leaf property nodes and organizing them into context-sensitive taxonomies with coherence-enforcing aggregation, the approach enables explicit reasoning about value importance and alignment. The framework also addresses how values evolve across contexts and how individuals and collectives can hold and aggregate values, using bottom-up or top-down implementations and a quantitative alignment measure that weights property satisfaction by importance. Through the uHelp running example, the paper demonstrates practical grounding, context adaptation, and alignment computation, illustrating potential applications in domains such as healthcare and participatory design. Overall, the framework provides a formal, interdisciplinary foundation for designing AI systems whose behavior is provably aligned with human values, while outlining directions for future research and deployment.

Abstract

In the diverse array of work investigating the nature of human values from psychology, philosophy and social sciences, there is a clear consensus that values guide behaviour. More recently, a recognition that values provide a means to engineer ethical AI has emerged. Indeed, Stuart Russell proposed shifting AI's focus away from simply ``intelligence'' towards intelligence ``provably aligned with human values''. This challenge -- the value alignment problem -- with others including an AI's learning of human values, aggregating individual values to groups, and designing computational mechanisms to reason over values, has energised a sustained research effort. Despite this, no formal, computational definition of values has yet been proposed. We address this through a formal conceptual framework rooted in the social sciences, that provides a foundation for the systematic, integrated and interdisciplinary investigation into how human values can support designing ethical AI.
Paper Structure (21 sections, 14 equations, 4 figures)

This paper contains 21 sections, 14 equations, 4 figures.

Table of Contents

  1. Introduction
  2. A Formal Model for Value Representation
  3. What are values? A computational approach
  4. Implementation Choices
  5. Specifying property nodes.
  6. Choosing the codomain of value importance.
  7. Ensuring the coherence of value importance.
  8. The Running uHelp Example
  9. How do values change with context? Context-based value taxonomies
  10. Implementation Choices
  11. Constructing Context-Based Taxonomies.
  12. Visualising Taxonomies.
  13. The Running uHelp Example
  14. Who holds values? Individuals vs collectives
  15. Implementation Choices
  16. ...and 6 more sections

Figures (4)

  • Figure 1: Abstractions for the value fairness: numbers indicate node importance, specified (black) and deduced (gray)
  • Figure 2: Value concepts with more than one parent node
  • Figure 3: Different context-based value taxonomies for fairness in uHelp
  • Figure 4: Individual and collective value taxonomies for fairness in uHelp's community of single mothers

Theorems & Definitions (3)

  • Definition 1: Value taxonomy
  • Definition 2: Coherence of value importance
  • Definition 3: Context-based value taxonomy