Using Knowledge Acquisition and Representation Tools to Support Scientific Communities

Brian R. Gaines, Mildred L. G. ShawAAAI 1994

Published in the mid-1990s, at the dawn of the widespread Internet, this research explores how Artificial Intelligence (AI) tools—originally conceived for building standalone "expert systems"—can be repurposed to manage the collective "intellectual life" of global scientific communities. As research becomes increasingly distributed, the challenge shifts from simple communication to the complex task of intellectual management.

The Challenge of Intellectual Management

By 1994, the Internet was already facilitating "invisible colleges"—geographically dispersed communities collaborating via email and FTP. However, while the infrastructure for communication was maturing, the tools for managing large-scale, international research projects remained primitive. Projects like the Intelligent Manufacturing Systems (IMS) program faced three primary hurdles:

  1. Coherence: Aligning dozens of sub-projects across different nations with a single, unified mission.
  2. Formalization: Translating vague, high-level policy goals (e.g., "manufacturing should be congenial to society") into operational technical requirements.
  3. Knowledge Bottlenecks: Capturing and managing the "commonsense" knowledge that experts often assume but rarely document.

Gaines and Shaw argue that AI should not merely aim to replace human experts but should instead serve as a facilitator for the collective intelligence of these research communities.

Methodology: Mapping the Scientific Mind

The authors applied a suite of Knowledge Acquisition (KA) and Representation (KR) techniques to the "GNOSIS" project, a major IMS test case involving over 100 participants across 14 countries. Their approach integrated several distinct methods:

  • Text Analysis: Using empirical induction on word co-occurrences in research papers to automatically generate concept maps. This visualized the relationships between technical terms in the community's discourse.
  • Repertory Grid Analysis: Based on Kelly’s Personal Construct Psychology, this technique elicited how experts distinguish between concepts. By comparing different "soft machine" projects, researchers identified the underlying dimensions of their work.
  • Laddering: A structured interview technique used to move between abstract goals and concrete technical definitions by asking "how" and "why."
  • Formal Logic (CLASSIC): Visual maps were translated into a formal KR language. This allowed an inference engine to detect "constraint violations," effectively using logic to find contradictions in project goals.

Results: Visualizing Research Landscapes

A primary success of the study was the reconstruction of the "mission logic" for the GNOSIS project. The tools allowed researchers to see how high-level socio-economic goals, such as environmental sustainability, linked directly to specific technical work packages (TWs).

Figure 2: The GNOSIS Mission Concept Map illustrates the hierarchical linkage between high-level socio-economic goals and specific technical work packages.

By applying Principal Component Analysis (PCA) to the repertory grids, the authors identified that the "Soft Machinery" research was primarily split along two axes: Human vs. Machine Intelligence, and System-led vs. User-led reconfiguration. This mapping provided a clear overview of the project's intellectual diversity and helped identify areas of overlap or neglect.

Figure 3: Principal Component Analysis (PCA) of repertory grids identifies the primary axes of research, such as Human vs. Machine Intelligence and System-led vs. User-led reconfiguration.

Key Insights: AI as a Reflective Mirror

The most significant outcome was not a computer "thinking" for the scientists, but the scientists seeing their own logic mapped out. This "mirror" effect revealed that certain work packages were poorly integrated into the core mission, explaining their slower progress.

Furthermore, the authors suggest that KR systems should function like the scientific method itself, following Karl Popper’s philosophy of refutation. Rather than trying to "prove" a research goal is correct, the system should be used to refute inconsistent conjectures. If a new technical requirement contradicts the mission statement, the system flags it as an "incoherence," prompting human intervention.

Conclusion and Future Directions

While the tools proved powerful, the authors noted several limitations. Individual projects cannot build a complete world model from scratch; future systems must interface with massive, shared ontologies (like the CYC project) to handle basic human concepts. Additionally, there is a need for tighter integration of KR structures into the standard document formats and "pragmatic" tools (like email) that scientists use daily. Ultimately, the paper suggests that practical scientific reasoning may require less rigid formal logic and more flexible coordination tools to support the evolving needs of global research communities.

Summary updated

BibTeX
@inproceedings{gaines1994using,
  author = {Brian R. Gaines and Mildred L. G. Shaw},
  title = {{Using Knowledge Acquisition and Representation Tools to Support Scientific Communities}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {1994},
}
Figure