Vocabulary Note: Shared Understanding, Delegated Cognition, and Agent Coordination
Concepts extracted from three papers for thinking about Stera’s cognition-net collaborator
---
1. Shared Mental Model (SMM)
Source: The role of shared mental models in human-AI teams (Frontiers)
Definition: A structured, overlapping cognitive representation held by both human and AI team members that encodes task objectives, team roles, situational dynamics, and each party’s capabilities and limitations, enabling implicit coordination without exhaustive communication.
Use for Stera: SMM is the closest existing construct to the shared cognition Stera envisions—but current SMM work treats the model as a static alignment artifact, whereas Stera’s collaborator builds shared understanding dynamically through interaction, making the model a process rather than a prerequisite.
---
2. Team Situation Awareness (TSA)
Source: The role of shared mental models in human-AI teams (Frontiers)
Definition: The degree to which every team member—human or AI—possesses an accurate, current understanding of the unfolding situation, including what has happened, what is happening now, and what is likely to happen next, in a form that supports interdependent action.
Use for Stera: TSA highlights a gap Stera directly addresses: most AI agents maintain their own context, but Stera’s collaborator maintains joint situation awareness—it tracks not just the task state but the human’s understanding of the task state, enabling it to notice when the two diverge and repair the gap.
---
3. Shared Task Representation
Source: Coordinated LLM Multi-Agent Systems (2411.12345)
Definition: A structured description of a task—its goal, subtask decomposition, constraints, and success criteria—that is explicitly codified and shared across agents in a multi-agent system, allowing each agent to reason about its own role in relation to the whole.
Use for Stera: In multi-agent systems, shared task representation is a coordination artifact; in Stera, it becomes a collaborator’s working memory—the representation the cognition-net holds with the human, continually updated as understanding evolves, not a static plan handed down at the start.
---
4. Role Negotiation
Source: Coordinated LLM Multi-Agent Systems (2411.12345)
Definition: The process by which agents in a multi-agent system dynamically determine who does what—assigning subtasks, resolving overlaps, and reallocating responsibilities—based on current capabilities, availability, and task demands, rather than relying on a fixed pre-assignment.
Use for Stera: Role negotiation captures the dynamic allocation of cognitive work that Stera enables: the collaborator does not assume a fixed division of labor (“I do X, you do Y”) but negotiates in real time who holds which piece of the problem, based on where understanding currently resides.
---
5. Delegated Cognition
Source: Agentic AI Systems Applied to tasks in Financial Services (2412.10954) — introduced implicitly through agent delegation patterns; term crystallized through cross-paper synthesis
Definition: The act of transferring a bounded cognitive task or subtask—reasoning, analysis, decision preparation—to an agent, with the delegator retaining oversight and the agent returning structured outputs that the delegator can incorporate without redoing the work.
Use for Stera: This is the operational primitive of a cognition-net: Stera’s collaborator is not a tool you query but a cognitive resource you delegate to—and, crucially, the collaborator can also delegate back, distributing cognition across the human-machine boundary bidirectionally. Most current agent delegation is one-way.
---
6. Agent-to-Agent Communication Protocol
Source: Coordinated LLM Multi-Agent Systems (2411.12345)
Definition: A structured format and sequence for messages between agents—often natural language wrapped in a coordination envelope—that ensures each agent receives the information it needs, when it needs it, in a form it can act on, without overwhelming the system with redundant or misaligned communication.
Use for Stera: Stera replaces explicit agent-to-agent protocols with a shared cognitive space: agents do not send messages to each other; they write into and read from a common understanding surface. This collapses the communication overhead that limits current multi-agent systems and is the architectural insight that makes a cognition-net scalable.
---
7. Task Decomposition and Allocation
Source: Agentic AI Systems Applied to tasks in Financial Services (2412.10954)
Definition: The two-step process of breaking a complex task into independent or sequential subtasks and then assigning each to the most capable available agent—or to a human—based on a capability match, with the allocation logic often driven by LLM reasoning about agent profiles.
Use for Stera: Current decomposition-and-allocation assumes tasks can be cleanly split. Stera’s approach acknowledges that cognitive tasks often cannot be—they require overlapping understanding. The collaborator decomposes not just the task but the understanding required, holding pieces that no single agent fully possesses but that the net collectively maintains.
---
8. Human-in-the-Loop Integration
Source: Agentic AI Systems Applied to tasks in Financial Services (2412.10954)
Definition: An architectural pattern where the agent system pauses at designated decision points and presents its intermediate reasoning or proposed action to a human for approval, modification, or override, ensuring that consequential actions remain under human authority.
Use for Stera: Human-in-the-loop is a checkpoint; Stera’s collaborator is a continuous presence in the cognitive loop. It does not surface for approval—it shares the reasoning space throughout, so the human is never outside the loop needing to be brought in. This shifts the human’s role from gatekeeper to co-reasoner.
---
9. Mutual Predictability
Source: The role of shared mental models in human-AI teams (Frontiers)
Definition: The property of a team where each member can anticipate, with useful accuracy, what another member will do next—not by explicit coordination but because their shared mental model makes behavior legible and expectations aligned.
Use for Stera: Mutual predictability is a goal of SMM-based teams; Stera’s collaborator achieves it by making its reasoning visible within the shared net. The human does not predict the AI’s behavior from a model—they see the reasoning unfold in the common space, making prediction unnecessary because understanding is shared in real time.
---
10. Cognitive Load Distribution
Source: The role of shared mental models in human-AI teams (Frontiers) — embedded in SMM theory; surfaced here as a named concept for Stera analysis
Definition: The deliberate allocation of cognitive demands—tracking context, holding intermediate results, monitoring for anomalies, switching between subtasks—across human and machine team members such that neither is overloaded and the team’s total cognitive capacity is effectively utilized.
Use for Stera: This is the human-centered justification for the cognition-net. Stera’s collaborator takes on not just computational tasks but cognitive tasks—remembering what was discussed, noticing contradictions, holding unresolved threads—that currently burden the human, freeing human cognition for what it does best. Current agent systems offload execution; Stera offloads understanding maintenance.
---
11. Coordination Overhead
Source: Coordinated LLM Multi-Agent Systems (2411.12345)
Definition: The computational and communicative cost—in tokens, latency, and coherence loss—incurred by maintaining alignment among multiple agents, which grows with the number of agents and the complexity of their interactions, and which can erode the net benefit of multi-agent architectures.
Use for Stera: Coordination overhead is the scaling bottleneck Stera’s architecture is designed to escape. By replacing pairwise or hub-spoke communication with a shared cognition surface, the collaborator makes coordination cost constant with respect to the number of participants—each reads and writes the same surface—which is a structural advantage over current multi-agent paradigms.
---
12. Adaptive Autonomy
Source: Agentic AI Systems Applied to tasks in Financial Services (2412.10954)
Definition: The capacity of an agent system to vary the degree of autonomy it exercises—from fully autonomous execution to consultative recommendation to complete deference—depending on task criticality, confidence, and domain risk, often governed by predefined thresholds or learned policies.
Use for Stera: Adaptive autonomy is about how much the agent does on its own; Stera’s collaborator adds how much understanding is shared while doing it. Autonomy in Stera is not a slider between “agent acts” and “human approves” but a quality of the shared cognitive field—even when the collaborator acts independently, the reasoning remains visible in the net, so autonomy never becomes opacity.
---
These twelve concepts form a working vocabulary for analyzing Stera’s cognition-net collaborator against the current AI agent landscape. Each highlights a dimension where existing systems operate and where Stera’s approach introduces a structural shift—not by improving coordination protocols or refining human-in-the-loop checkpoints, but by replacing the communication-and-control architecture with a shared cognitive surface that makes joint understanding the foundation, not the aspiration.
Comments
No comments yet — be the first.