Source-linked AI summary
An Agent Model Abstraction for Human-AI Teaming Cognitive Coupling
Kolitha Kottagaha W. M, Jos A. C. Bokhorst, Ben Gaffinet, Christos Emmanouilidis
TL;DR
Human-AI teaming needs agents to align their perceptions, reasoning, and actions, yet existing work does not fully represent how their internal cognitive processes couple. The paper proposes a conceptual model of agent cognition and interaction; its supported conclusion is a framework for analysing and designing cognitive interoperability, bounded by the difficulty of measuring human internal processes.
Problem
Existing work does not fully specify how agents’ internal cognitive processes are organised and coupled during collaboration, making it difficult to trace goals, evaluations, and actions across agents.
Source: §2 · p. 2Method
Results
The conceptual model represents cognitive coupling through shared perception–action loops, goal-related exchanges, and direct agent communication.
Source: §4.2 · p. 4Takeaways & Limitations
The model provides a basis for analysing distributed cognition and designing cognitively interoperable human-AI systems.
Source: §1 · p. 1, §6 · p. 6Human internal cognitive states and processes are not fully measurable and can only be partially inferred from behaviour.
Source: §5.2 · p. 5
Abstract
from arXiv · showhide
Industrial environments increasingly rely on collaboration between humans and AI-enabled agents. Effective teamwork requires aligning how agents perceive situations, plan actions to pursue goals, and adapt to changing conditions, yet existing systems lack mechanisms for cross-agent cognitive processes coupling. This paper presents a conceptual cognitive agent model that formalises cognitive coupling through eight components: Input, Process, Output, State, Value, Memory, World Model, and Goal. The model abstracts how agents coordinate and co-regulate their cognitive cycles, providing a basis for analysing distributed cognition and designing cognitively interoperable human-AI systems.
1. INTRODUCTION
Human-AI teamwork in industrial settings depends on agents aligning their perceptions, reasoning, and actions toward shared objectives as conditions change. The paper proposes an eight-component cognitive agent model to represent how human and artificial agents coordinate and couple their cognitive processes.
1. INTRODUCTION
- Industrial collaboration increasingly treats humans and AI as team members, with perception, reasoning, and learning distributed across agents pursuing shared goals.Effective teaming requires alignment with common objectives while adapting to changing operational and environmental conditions.Source: §1 · p. 1
- Cognitive architectures model how agents perceive, reason, and learn, but the paper identifies coupling processes between human and artificial agents as a needed extension.The background includes ACT-R, LIDA, ICARUS, and SOAR.Source: §1 · p. 1
- The proposed model defines cognitive coupling as alignment between human and artificial agents’ cognitive characteristics, using eight components and a Shared Goal Space to represent common objectives.The model is intended as a foundation for analysing and designing cognitively interoperable human-AI teaming.Source: §1 · p. 1
2. BACKGROUND AND RELATED WORK
Prior work addresses team cognition, individual cognitive architectures, and system-level coordination, but these strands do not fully specify how agents’ internal processes couple during collaboration. The paper frames its model as a way to make those mechanisms explicit.
2. BACKGROUND AND RELATED WORK
- Team cognition research links shared mental models, communication, and mutual awareness with teams functioning as a coherent cognitive system.Joint and distributed cognition also treats cognitive functions as extending beyond individuals through interaction with agents and artefacts.Source: §2 · p. 2
- Cognitive architectures such as ACT-R, SOAR, ICARUS, and LIDA model how agents process inputs, retrieve knowledge, form intentions, and select actions.The Common Model of Cognition by Laird et al. (2017) is described as a high-level standard model of a cognitive agent.Source: §2 · p. 2
- Prior work identifies cognitive and decisional interoperability as relevant to human-AI collaboration, including compatible interpretations of goals, decisions, and context.Proposed dimensions of cognitive coupling include perception, situation awareness, memory, reasoning, planning, monitoring, and action.Source: §2 · p. 2
- The central gap is a representation combining agents’ internal cognitive organisation with how their cognitive cycles couple during collaborative work.The paper notes that treating team coordination, individual cognition, and system architecture separately makes it difficult to trace goals, evaluations, and actions across agents.Source: §2 · p. 2
3. DESIGN METHODOLOGY
The design methodology builds a functional abstraction of human and artificial agents from recurring cognitive functions, then represents their information flows and interactions in SysML. The model receives a conceptual consistency check.
3. DESIGN METHODOLOGY
- The design organises recurring cognitive functions from the cognitive-architecture literature into a generic cognitive unit applicable at a functional level to human and artificial agents.System-level structures extend these units to represent collaboration among agents.Source: §3 · p. 2
- SysML block definition and internal block diagrams represent system elements, information flow within an agent, and information flow among agents and shared components.The model’s representation is selected to express system structure, component relations, and information flow.Source: §3 · p. 2
- Conceptual validation checked component consistency with cognitive-architecture functions, coherence of the internal cognitive cycle, and whether system structures represent agent coupling.The validation was conceptual rather than an empirical performance evaluation.Source: §3 · pp. 2–3
4. MODEL DESCRIPTION
The model represents human and AI-enabled agents as cognitive units embedded in an environment and coordinated through shared goals. It specifies internal components and interaction pathways through which agents’ perceptions, intentions, and actions become coupled.
4. MODEL DESCRIPTION
- The Cognitive System combines agents, an external Environment, and a Shared Goal Space that maintains collective goals and performance information.Agents perceive, act, and receive feedback in the environment; the shared representation provides a common reference for collective objectives.Source: §4 · p. 3
4.1 Model Components
- Input selects and interprets environmental, communicative, and internal signals, while Process combines these with feedback and context to reason and select actions.Input also filters information by relevance; Process abstracts sub-processes such as situation interpretation, option evaluation, and action selection.Source: §4.1 · p. 3
- Output turns selected actions into environmental effects or communications, while Value evaluates outcomes against local and shared goals and feeds adjustments back.Value criteria can include efficiency, safety, workload, and alignment with shared goals.Source: §4.1 · p. 3
- Goal maintains an agent’s local goal hierarchy, deriving sub-goals from system goals and revising them in response to feedback or changing conditions.The SharedGoalSpace maintains system goals, constraints, and task states and supports distributing high-level goals into agent sub-goals.Source: §4.1 · pp. 3–4
- State represents an agent’s condition and may include information about others; Memory retains experience and task information, and the World Model represents the environment and other agents.These representations inform interpretation, coordination, and decisions, while the World Model is updated as new evidence or actions change the task context.Source: §4.1 · p. 4
4.2 Cognitive Coupling Between Agents
- Cognitive coupling links agents through shared perception–action loops, exchanges of goal information via the SharedGoalSpace, and direct communication.One agent’s actions can change what others perceive, while shared goal states constrain sub-goal formation and receive updates about progress.Source: §4.2 · p. 4
5. RELATION TO COGNITIVE ARCHITECTURES
The section relates the proposed agent model’s components to established cognitive functions and to the Common Model of Cognition. It highlights both functional correspondences and the model’s explicit representation of situational understanding across agents.
5.1 Cognitive Functions
- The model’s functional categories correspond to recurring functions in cognitive architectures, including perception, attention, memory, learning, reasoning, and action selection.The comparison also identifies the model’s extension toward collaboration between human and artificial agents.Source: §5.1 · p. 5
- Input filters sensory and communicative information, Process selects and integrates it for decisions, and Output executes actions and updates the environment and Shared Goal Space.These roles correspond to perception, attention, reasoning, and action selection.Source: §5.1 · p. 5
- Memory supplies long-term knowledge and episodic traces, World Model represents situational understanding, and State maintains transient conditions.Learning arises through Process, Output, and Value interactions that evaluate outcomes against goals and feed back into decisions, knowledge, and goals.Source: §5.1 · p. 5
5.2 The Common Model of Cognition
- The model accommodates both artificial and human agents, while human cognitive states and processes are only partially inferable from behaviour.Artificial agents can be designed to follow the architecture; humans are natural systems whose design cannot be altered.Source: §5.2 · p. 5
- The Common Model’s perception and motor blocks map to Input and Output, while the proposed model separates World Model from processing to represent task context, the environment, and other agents.The Common Model mainly handles this situational role within working memory.Source: §5.2 · pp. 5–6
6. CONCLUSION AND FUTURE WORK
The conclusion presents a conceptual model for how human and artificial agents coordinate cognition in shared tasks, using shared goals and interaction processes to represent their coupling. It remains non-operational, with implementation and validation challenges left for future work, while offering a conceptual basis for cognitive digital twins.
Conceptual model
- The model links eight internal components to the Environment and SharedGoalSpace, representing how agents use shared context and other agents’ decisions and actions in shared tasks.The SharedGoalSpace provides a common reference for collective objectives as tasks and situations change.Source: §6 · p. 6
- Perception–action relations, communication, and shared goal updates represent how agents influence one another and create coupling effects.The model abstracts coordination between human and artificial agents rather than specifying a particular implementation.Source: §6 · p. 6
Future work
- The representation is conceptual, not an operational system, and does not specify algorithms, behavioural policies, or domain-specific constraints.It also identifies human-cognition modelling and validation as challenges because individual differences matter and internal states are observable only through decisions and actions.Source: §6 · p. 6
- Future work will explore implementing aligned components and using them to analyse performance in simulated or real collaborative tasks.The modelled human’s working-memory capacity is one example of an interpersonal difference to account for.Source: §6 · p. 6
- The model offers a conceptual basis for cognitive digital twins that represent cognitive structures alongside agents’ physical or operational aspects.The proposed structure describes how inputs, memory, world representations, values, and goals are organised and coupled across agents.Source: §6 · p. 6