Model `Agent` and `LegalAgent`

[ghurlbot]

We want to model the concept Agent as a technical term that refers to machine agents (so we call this MachineAgent). We also want to model the concept Agent as a legal concept that refers to acting on someone's behalf (we call this LegalAgent). To ensure there is no ambiguity between the two and to support both of their use using DPV, we have planned to discuss this for inclusion in DPV v2.2.

Opened by @coolharsh55 via IRC channel #dpvcg on irc.w3.org

[coolharsh55]

Proposed taxonomy and namespaces:

• dpv:Agent:
  • parent: dpv:Entity
  • definition: An Agent is a dpv:Entity that is (a) acting on behalf of another Entity; and (b) is authorised to do so by that Entity.
  • Usage note: the Entity the Agent is acting on behalf of can be a human or non-human and a legal or non-legal entity. For specific distinction, see dpv:LegalAgent and tech:MachineAgent. The form and validity of authorisation through which the Agent is empowered to represent and act on behalf of the Entity is not defined or covered in this concept.
• dpv:LegalAgent:
  • parent: dpv:LegalEntity, dpv:Agent
  • definition: A 'legal agent' is a legal entity who acts on behalf of another legal entity.
  • Usage note: Legal agents may be human or non-human (e.g. an organisation), but they must be 'legal persons' i.e. recognised by law as such. The agency and the mechanisms through which it is given, acquired, or expressed is governed by relevant jurisdictional frameworks e.g. power of attorney.
• tech:MachineAgent:
  • parent: dpv:Agent, tech:Actor
  • definition: A 'machine agent' is a form of technology that is (a) acting on behalf of another Entity; and is (b) authorised to do so.
  • Usage note: Machine agents is a generic term that allows representing a technology (typically a software) that is acting as an agent without limiting whom they represent or how they operate. Machine agents can act on behalf of users (e.g. UserAgent in browsers) or on behalf of other machine agents (e.g. chained workflows).
• tech:SoftwareAgent:
  • parent: tech:MachineAgent, tech:Software
  • definition: A software agent is a machine agent that acts on behalf of another Entity or technology.
• tech:UserAgent:
  • parent: tech:MachineAgent, tech:SoftwareAgent
  • definition: A software agent that represents a person (i.e. a human).
  • Usage note: User Agent as defined here specifically represents software agents that represent humans, whether directly or indirectly, based on the well-defined use of this term in Web interactions. Software that is not representing a human but which acts or declares itself as a user agent is termed as 'user agent spoofing' (see https://developer.mozilla.org/en-US/docs/Glossary/User_agent)
• ai:AIAgent:
  • parent: tech:MachineAgent
  • definition: An AI Agent, also known as an 'intelligent agent', is a machine agent that utilises AI technologies.
  • Usage note: Other definitions of AI Agents may include the perception of environment as some capacity to incorporate the environment through sensors (e.g. computer vision) or data (e.g. web environments), and to produce changes in outputs through some degree of autonomy (i.e. without human intervention). The definition of 'AI Agent' provided in this extension relaxes such requirements and instead focuses on classifying these agents based solely on their use of AI technologies to allow compatibility with the evolving definitions.

[coolharsh55]

Separately from the above, I also propose to discuss Agentic AI concepts as following:

• ai:AgenticAI
  • parent: ai:AISystem
  • definition: Agentic AI is an AI system that utilises AI Agents to perform tasks for specific goals without human intervention.
  • Usage note: 'Agentic AI' is a broad term that is evolving through use and advancement of technologies. Agentic AI systems involve AI agents, commonly utilising LLMs and Generative AI technologies, that may utilise additional AI techniques such as reasoning, NLP, and ML techniques to achieve capabilities such as computer vision, information retrieval, and generation of media in order to complete specific goals (functions).

Note: in above, LLMs and GenAI are missing from the current AI taxonomy.

[bact]

* `ai:AgenticAI`
  
  * parent: `ai:AISystem`
  * definition: Agentic AI is an AI system that utilises AI Agents to perform tasks for specific goals without human intervention.
  * Usage note: 'Agentic AI' is a broad term that is evolving through use and advancement of technologies. Agentic AI systems involve AI agents, commonly utilising LLMs and Generative AI technologies, that may utilise additional AI techniques such as reasoning, NLP, and ML techniques to achieve capabilities such as computer vision, information retrieval, and generation of media in order to complete specific goals (functions).

Does it possible to define this by using dpv:AutomationLevel and probably a combination of some properties like:

• ai:AISystem tech:hasSensor tech:ComputerVision
• ai:AISystem tech:hasActuator tech:ButtonPusher

[coolharsh55]

Does it possible to define this by using dpv:AutomationLevel and probably a combination of some properties like:

* `ai:AISystem` `tech:hasSensor` `tech:ComputerVision`

* `ai:AISystem` `tech:hasActuator` `tech:ButtonPusher`

Hi @bact yes (e.g. OWL-style). Though if we do that then it implies that Agentic AI always has that (specific) automation level, and that specific properties...

[coolharsh55]

(using dpvbot:) This was discussed in Meeting 2025-10-15
concepts accepted ; next step to add them to spreadsheet, produce RDF, produce documentation