Purpose

The proliferation of terms such as Human-Centered AI (HCAI), Human-Centric AI (HCA), and People-Centered AI has led to ambiguity in describing the characteristics, goals, and ethical implications of software agents and systems. This document proposes an ontological framework to systematically describe these entities, enabling precise communication and alignment with human values.

Scope

The ontology applies to software agents, platforms, systems, or software with AI components, focusing on their human-interaction properties, ethical considerations, and societal impact. It addresses both technical and non-technical characteristics.

Audience

This document targets:

• AI researchers and developers
• Policymakers and ethicists
• Industry stakeholders
• Academic communities studying AI and human-computer interaction

Ambiguity in Existing Terms

Terms like Human-Centered AI and Human-Centric AI are often used interchangeably despite distinct connotations. This lack of standardized definitions hinders collaboration and evaluation. For example, HCAI emphasizes transparency and oversight, while HCA prioritizes symbiotic human-AI integration.

Need for an Ontological Framework

Ontologies provide structured representations of knowledge, enabling clear classification and relationships. A dedicated ontology for AI agents can address ambiguity, support interoperability, and guide ethical development.

Goals

• Disambiguate terms like HCAI, HCA, and related paradigms.
• Provide a reusable framework for describing agent characteristics.
• Promote human responsibility, sustainability, and alignment with societal values.

Definitions

Software Agent

An autonomous or semi-autonomous computational entity capable of performing tasks or decision-making.

Human-Centered AI (HCAI)

AI systems designed with a focus on transparency, accountability, fairness, and human oversight.

Human-Centric AI (HCA)

AI systems prioritizing human needs, emotions, and symbiotic integration into human life.

Ontology

A formal representation of knowledge as a set of concepts, properties, and relationships within a domain.

Related Concepts

Related terms include People-Centered AI, Responsible AI, and Sustainable AI. Proposed acronyms include:

• HRCAI: Human-Responsible Centric Artificial Intelligence
• HCSTEW: Human-Centric Sustainable Technological Ecosystems for Wellbeing

Scope Limitations

This ontology focuses on AI-driven systems; non-AI software is out of scope unless integrated with AI components. It does not prescribe specific implementation technologies.

Core Components

The ontology consists of:

• Classes: Categories of entities (e.g., Agent, System, Platform).
• Properties: Attributes describing entities (e.g., Transparency, Adaptability, Responsibility).
• Relationships: Interactions between entities (e.g., Augments, Oversees, AdaptsTo).

Taxonomy of Agent Characteristics

Relationships

Key relationships include:

• Agent-Human: Augments, Supports, LearnsFrom, AdaptsTo.
• Agent-System: IntegratesWith, DependsOn, Controls.
• Human-Society: Impacts, Benefits, Governs.

Proposed Acronyms and Terms

To disambiguate and standardize:

• HCAI: Human-Centered Artificial Intelligence (transparency, oversight).
• HCA: Human-Centric Artificial Intelligence (symbiosis, personalization).
• HRCAI: Human-Responsible Centric AI (human accountability).
• HCSTEW: Human-Centric Sustainable Technological Ecosystems for Wellbeing.
• Others: HCIA, HCTA, HCST, HCBT, HCDT, HCAE, HCTI, HCAT, HCSI, HCRAI, HCRAT.

Formal Representation

The ontology will be expressed in OWL (Web Ontology Language). Example structure:

        Class: Agent
          SubClassOf: System
          Properties: hasTransparency, hasAdaptability
          Relationships: augments(Human), adaptsTo(Context)
      

Use Cases

• Healthcare: Describing AI diagnostic tools with properties like Transparency and Fairness.
• Education: Classifying adaptive learning systems as HCA with high Symbiosis.
• Policy: Guiding regulations based on Responsibility and Sustainability.

Example Annotations

System: AI-driven healthcare platform

• Class: Agent
• Properties: High Transparency, Medium Adaptability
• Relationships: augments(Clinician), impacts(Patient)

Benefits

• Improved clarity in academic and industry discussions.
• Enhanced evaluation of AI systems for ethical alignment.
• Support for interoperable standards in AI development.

Technical Implementation

• Use of semantic web technologies (RDF, OWL) for ontology encoding.
• Integration with existing AI metadata standards (e.g., ML Schema).

Adoption Strategies

• Collaboration with standards bodies (e.g., W3C, IEEE).
• Open-source ontology repository for community contributions.
• Training materials for developers and policymakers.

Challenges

• Resistance to standardization due to diverse stakeholder priorities.
• Complexity of capturing nuanced human-AI interactions.
• Evolving nature of AI requiring ontology updates.