Toward an Intelligent Agent Framework for Enterprise Integration

AAAI 1991

TL;DR

This paper introduces an Intelligent Agent (IA) framework designed to integrate people and computer systems in large, geographically dispersed manufacturing enterprises. By modeling enterprise tasks as cognitive activities and assigning them to specialized agents, the framework aims to automate processes, improve communication, and create a more responsive and efficient operational model.

What Problem Does This Paper Address?

Integrating complex, large-scale manufacturing enterprises is a significant challenge. These enterprises involve numerous people, diverse computer systems, and intricate business processes that often operate in silos. Traditional approaches struggle to bridge the gap between human intuition and automated systems, leading to inefficiencies, communication breakdowns, and a lack of real-time responsiveness. The authors aim to address this by creating a unified framework that augments human workers with intelligent assistants (IAs) capable of understanding and executing specific tasks, thereby creating a more cohesive and intelligent enterprise.

Methods

The core of the proposed framework is an Intelligent Agent (IA) system built upon three synergistic technologies:

  1. Intelligent Agents (IAs): These are active, autonomous processes designed to model and perform specific tasks. They are built by first modeling activities in cognitive terms (what to look for, what to do, who to tell) and then "agentifying" these models.
  2. MKS Knowledge Service: This acts as a central repository for a comprehensive, object-oriented model of the enterprise. It captures personnel, facilities, equipment, inventory, processes, and information flow. IAs interact with this model to access information, register for events, and understand the enterprise's state.
  3. Distributed System Infrastructure: While not detailed, this underpins the communication and coordination between agents.

Key methodological aspects include:

  • Activity Modeling: Complex job functions are broken down into elementary tasks, each modeled in cognitive terms. This makes the models understandable to human users.
  • Agentification: Passive activity models are transformed into active IAs through instantiation, process creation, and activation.
  • Personal Assistants (PAs): A special class of IA that facilitates human-IA interaction by translating between human modalities (email, graphical editors) and IA network protocols. PAs maintain personal models of their users to manage information flow and requests.
  • IA Network: A dedicated communication bus for IA-to-IA messaging, using a specific IA protocol that supports requests, information sharing, inquiries, and bidding.

Datasets

Not discussed in the paper. The paper focuses on the framework and its implementation in a semiconductor fabrication facility at Stanford's Center for Integrated Systems (CIS).

Key Insights and Ideas

  • Cognitive Modeling for Tasks: Breaking down complex jobs into elementary, cognitively modeled tasks makes the system understandable and maintainable by the very people who work in the enterprise. This is crucial for adoption.
  • Augmenting, Not Replacing, Humans: The framework emphasizes augmenting human workers with IAs, creating a partnership rather than aiming for full automation, which is seen as unrealistic in the near future.
  • Personal Assistants as Bridges: PAs are vital for seamless human-IA interaction, acting as intelligent intermediaries that manage communication and information flow based on individual user preferences and context.
  • IA Network for Coherent Communication: A specialized IA protocol and network enable concise, natural, and comprehensive communication among agents, abstracting away low-level network details.
  • Model-Based Approach: The entire framework is driven by an enterprise model (MKS), which serves as a shared understanding and a dynamic representation of the enterprise's state.

Key Results

The paper presents preliminary experimental results suggesting the model-based, man-machine approach is a viable path for applying Distributed AI (DAI) to real-world enterprises. The authors are developing a prototype implementation for the semiconductor fabrication facility at Stanford's CIS. The paper describes a scenario where:

  • A "shadowing" wafer-lot IA tracks lots through processing steps, dynamically assigning equipment.
  • Operator IAs manage equipment queues and ensure machines run efficiently.
  • Monitor IAs track equipment health, detect malfunctions, and initiate shutdown procedures or alert maintenance personnel.
  • The system demonstrates how IAs can coordinate tasks, manage resources, and handle exceptions, including escalating issues to human PAs.

Limitations and Future Work

Limitations: * The current prototype restricts IAs to run within a single workstation, though they can access remote MKS models. * The paper acknowledges that while preliminary results are promising, extensive real-world testing and refinement are still needed.

Future Work: * Developing fully distributed IA systems where agents can live in various geographically dispersed workstation environments. * Further development of the distributed infrastructure using low-level agents called Proxies. * Collaborating with DAI researchers to establish common protocols for integrating agents with sophisticated planning and reasoning capabilities. * Combining multi-media CSCW tools with Personal Assistants to support all information-related enterprise activities in an integrated way. * Exploring how this interdisciplinary approach can lead to a major impact on mainstream information technology.

Summary updated

BibTeX
@inproceedings{1991toward,
  title = {{Toward an Intelligent Agent Framework for Enterprise Integration}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {1991},
}
Figure