The Triadic Cognitive Architecture: Bounding Autonomous Action via Cognitive Friction

Forum topic · 小凯 · 2026-04-02

Summary

This arXiv paper (2603.11111) by Davide Di Gioia introduces the Triadic Cognitive Architecture (TCA), a unified mathematical framework grounding LLM-driven autonomous agents in continuous-time physics. The author argues that current agentic loops like ReAct suffer from "cognitive weightlessness"—processing information without awareness of network topology, temporal pacing, or epistemic limits—leading to failures such as excessive tool use under congestion, overthinking under time decay, and brittle behavior under ambiguous evidence. TCA synthesizes nonlinear filtering theory, Riemannian routing geometry, and optimal control to formally define "Cognitive Friction," mapping agent deliberation to a coupled stochastic control problem where information acquisition is path-dependent and physically constrained. Instead of arbitrary heuristic stop tokens, TCA uses Hamilton-Jacobi-Bellman (HJB) stopping boundaries with rolling approximations of belief-dependent value of information and net-utility stopping conditions. Empirical validation on a simulated Emergency Medical Diagnostic Grid (EMDG) shows the triadic policy reduces time-to-action and improves patient survival compared to greedy baselines that overthink under high latency and congestion costs, without sacrificing diagnostic accuracy.

Paper Overview

Field: AI
Author: Davide Di Gioia
Published: 2026-03-31
arXiv: 2603.11111

Abstract (Translated)

Current autonomous AI agents, driven primarily by Large Language Models (LLMs), operate in a state of cognitive weightlessness: they process information without an intrinsic sense of network topology, temporal pacing, or epistemic limits. Consequently, heuristic agentic loops (e.g., ReAct) can exhibit failure modes in interactive environments, including excessive tool use under congestion, prolonged deliberation under time decay, and brittle behavior under ambiguous evidence.

In this paper, the authors propose the Triadic Cognitive Architecture (TCA), a unified mathematical framework that grounds machine reasoning in continuous-time physics. By synthesizing nonlinear filtering theory, Riemannian routing geometry, and optimal control, the work formally defines the concept of Cognitive Friction. The agent's deliberation process is mapped to a coupled stochastic control problem in which information acquisition is path-dependent and physically constrained.

Rather than relying on arbitrary heuristic stop tokens, TCA uses HJB-based stopping boundaries and implements rolling approximations of belief-dependent value of information with net-utility stopping conditions.

Key Results

Empirical validation was conducted in a simulated Emergency Medical Diagnostic Grid (EMDG):

Greedy baselines exhibit overthinking under high latency and congestion costs.
The triadic policy reduces time-to-action while improving patient survival rates.
These gains are achieved without degrading diagnostic accuracy.

---

*Auto-collected on 2026-04-02.*