Skip to main content

Value Awareness

To integrate value awareness into the GNW, we must first understand what awareness entails. The term "aware" is inherently ambiguous, but it generally relates to explicitly recognizing, perceiving, or knowing something. For instance, situation awareness [Endsley, 2011] describes a mental state where all relevant elements of a situation and their interrelationships are understood, enabling prediction (or at least the formation of reasonable expectations) about future events or appropriate actions.

Within VALAWAI, we focus specifically on value-aware AI, which we define as:

"An AI system that is capable of capturing the relevant (human) value system, understanding that value system, abiding by it, and explaining its own behavior and that of others in terms of that value system."

Achieving value awareness requires, we argue, four key technical capabilities:

  1. Value Acquisition: The AI must be able to capture the relevant value system, reflecting the stakeholder's value preferences, which are often context-dependent. For example, the values of equality or fairness can have different meanings and importance across various application domains. This acquisition process may involve learning from data, explicit input from users, or a combination of both.

  2. Value Representation and Reasoning: The AI needs to represent values in a formal and computationally tractable manner, enabling reasoning about them. This includes the ability to detect value conflicts, analyze the impact of adopting certain values, and perform other logical operations on value representations.
    This capability is essential for understanding the complex interplay of values within a given context.

  3. Value Alignment: Mechanisms are required to assess whether a given behavior aligns with the preferred values. This involves comparing actions or decisions against the represented values and determining the degree of congruence or conflict.
    This alignment process may involve quantitative measures or qualitative assessments.

  4. Value-Based Explainability: For transparency and to enhance human understanding of AI behavior, the AI must be able to explain how its actions (or the actions of others) relate to the underlying value system. This explainability component clarifies the connection between behavior and values, supporting both value-aligned decision-making and the value-aligned design of AI systems.

These four capabilities enable value-aware AI to achieve a range of functionalities, including:

  1. Value Compliance Monitoring: Monitoring compliance with values by human or AI systems, potentially providing feedback or supporting certification processes.

  2. Implicit Value Deconstruction: Inferring the values implicitly used by a human or AI system based on observed behavior, and providing feedback to make them aware of these values.

  3. Value-Driven Action Advisement: Advising a course of action to a human or AI system that is compatible with a given value system.

  4. Autonomous Value-Aligned Action: Autonomously performing actions that are compatible with a given value system.

Furthermore, value-aware AI can be applied not only to behaviors but also to the motivators of behavior, such as norms. This leads to additional capabilities:

  1. Norm-Value Alignment Analysis: Analyzing the alignment of norms with values, providing feedback on norms, or supporting norm certification processes.

  2. Implicit Value Deconstruction from Norms: Inferring the values implicitly adhered to by norms.

  3. Value-Driven Norm Advisement: Advising on norms to adopt in a given situation, ensuring compatibility with a given value system.

  4. Autonomous Value-Aligned Norm Selection: Autonomously selecting norms that are compatible with a given value system.

Finally, value-aware AI can also play a crucial role in helping humans understand and evolve their own value systems:

  1. Personal Value System Evolution Support: Helping individuals decide how to evolve their personal value systems.

  2. Collective Value System Deliberation Support: Facilitating collective decision-making regarding shared or agreed-upon value systems.

It is crucial to emphasize that deciding on value systems remains a strictly human prerogative.
AI cannot make decisions about value systems but can only adopt and operate according to the value system(s) provided by its stakeholders.

These four technical capabilities—value acquisition, value representation and reasoning, value alignment, and value-based explainability—form the foundation for developing truly value-aware AI.
In the following sections, we will illustrate how these capabilities can be integrated into the GNW architecture.