Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cognitive LLMs: Towards Integrating Cognitive Architectures and Large Language Models for Manufacturing Decision-making

Siyu Wu, Alessandro Oltramari, Jonathan Francis, C. Lee Giles, Frank E. Ritter

TL;DR

This work addresses the gap between human-like cognitive reasoning and the broad, often noisy inferences of large language models in production settings. It introduces LLM-ACTR, a neuro-symbolic framework that grounds LLMs with intermediate representations from the ACT-R cognitive architecture, enabling human-aligned and explainable manufacturing decisions. By extracting ACT-R decision traces, embedding them, and finetuning an open LLM with LoRa, the approach improves grounded decision-making and task performance over LLM-only baselines. The results on a Design for Manufacturing task show promise for scalable, cognitively grounded AI that can operate reliably in real-world production environments, with clear pathways for future enhancements and data-driven improvements.

Abstract

Resolving the dichotomy between the human-like yet constrained reasoning processes of Cognitive Architectures and the broad but often noisy inference behavior of Large Language Models (LLMs) remains a challenging but exciting pursuit, for enabling reliable machine reasoning capabilities in production systems. Because Cognitive Architectures are famously developed for the purpose of modeling the internal mechanisms of human cognitive decision-making at a computational level, new investigations consider the goal of informing LLMs with the knowledge necessary for replicating such processes, e.g., guided perception, memory, goal-setting, and action. Previous approaches that use LLMs for grounded decision-making struggle with complex reasoning tasks that require slower, deliberate cognition over fast and intuitive inference -- reporting issues related to the lack of sufficient grounding, as in hallucination. To resolve these challenges, we introduce LLM-ACTR, a novel neuro-symbolic architecture that provides human-aligned and versatile decision-making by integrating the ACT-R Cognitive Architecture with LLMs. Our framework extracts and embeds knowledge of ACT-R's internal decision-making process as latent neural representations, injects this information into trainable LLM adapter layers, and fine-tunes the LLMs for downstream prediction. Our experiments on novel Design for Manufacturing tasks show both improved task performance as well as improved grounded decision-making capability of our approach, compared to LLM-only baselines that leverage chain-of-thought reasoning strategies.

Cognitive LLMs: Towards Integrating Cognitive Architectures and Large Language Models for Manufacturing Decision-making

TL;DR

This work addresses the gap between human-like cognitive reasoning and the broad, often noisy inferences of large language models in production settings. It introduces LLM-ACTR, a neuro-symbolic framework that grounds LLMs with intermediate representations from the ACT-R cognitive architecture, enabling human-aligned and explainable manufacturing decisions. By extracting ACT-R decision traces, embedding them, and finetuning an open LLM with LoRa, the approach improves grounded decision-making and task performance over LLM-only baselines. The results on a Design for Manufacturing task show promise for scalable, cognitively grounded AI that can operate reliably in real-world production environments, with clear pathways for future enhancements and data-driven improvements.

Abstract

Resolving the dichotomy between the human-like yet constrained reasoning processes of Cognitive Architectures and the broad but often noisy inference behavior of Large Language Models (LLMs) remains a challenging but exciting pursuit, for enabling reliable machine reasoning capabilities in production systems. Because Cognitive Architectures are famously developed for the purpose of modeling the internal mechanisms of human cognitive decision-making at a computational level, new investigations consider the goal of informing LLMs with the knowledge necessary for replicating such processes, e.g., guided perception, memory, goal-setting, and action. Previous approaches that use LLMs for grounded decision-making struggle with complex reasoning tasks that require slower, deliberate cognition over fast and intuitive inference -- reporting issues related to the lack of sufficient grounding, as in hallucination. To resolve these challenges, we introduce LLM-ACTR, a novel neuro-symbolic architecture that provides human-aligned and versatile decision-making by integrating the ACT-R Cognitive Architecture with LLMs. Our framework extracts and embeds knowledge of ACT-R's internal decision-making process as latent neural representations, injects this information into trainable LLM adapter layers, and fine-tunes the LLMs for downstream prediction. Our experiments on novel Design for Manufacturing tasks show both improved task performance as well as improved grounded decision-making capability of our approach, compared to LLM-only baselines that leverage chain-of-thought reasoning strategies.
Paper Structure (42 sections, 7 equations, 8 figures, 2 tables)

This paper contains 42 sections, 7 equations, 8 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Cognitive Decision Making
  4. Integration of Cognitive Architectures and LLMs
  5. Problem Definition: Design for Manufacturing
  6. Cognitive LLMs: Hybrid Architectures for Human-Aligned Decision Making
  7. Cognitive Architectures based on ACT-R
  8. Human-aligned Cognitive Models
  9. Reinforcement Mechanisms in Production Systems
  10. VSM-ACTR model evaluation
  11. The LLM-ACTR Framework
  12. Selecting Salient Decision Information
  13. Learning an Embedding Space of Decision Traces
  14. Injecting Decision Information into LLMs
  15. Experiments
  16. ...and 27 more sections

Figures (8)

  • Figure 1: Decision augmentation using a neural-symbolic cognitive architecture approach. (1) Tasks are modeled with cognitive architecture. (2) Cognitive model used to run stochastic simulation of task at scale. (3) Synthetic data are distilled from simulation and combined with prompt requests. (4) A fine tuning pipeline is used to calibrate open source LLM to perform decision augmentation for task in exercise.
  • Figure 2: A Value Stream Map of manufacturing process.
  • Figure 3: (a) Obtaining decision representations from VSM-ACT-R. (b) LLM feature extraction for behavior prediction.
  • Figure 4: ACTR embedding mapping
  • Figure 5: Finetuning pipeline.
  • ...and 3 more figures