Introduction: The Information Explosion
In the early 1980s, organizations faced a burgeoning "paperwork and information explosion." While relational databases were becoming standard, they functioned primarily as digital filing cabinets—efficient at storage but lacking the intelligence to help users navigate complex information landscapes. The emerging field of Information Resources Management (IRM) sought to treat information as a corporate asset, yet existing tools were limited to metadata catalogs that identified data locations without assisting in its application.
Charles H. Kellogg of the System Development Corporation proposed a vision for "Intelligent Information Resources Assistants" (IIRA). These systems were designed to bridge the gap between raw data and human expertise, acting as an intelligent interface that could function as a librarian, researcher, and secretary.
System Architecture and Methodology
The core of this vision was realized through Deductively Augmented Data Management (DADM). This approach integrated logic-based AI with high-performance relational databases. The prototype system, KM-1 (Knowledge Manager-1), was built upon three primary components:
- Deductive Processor (Reasoning Engine): Implemented on a Xerox 1100 Lisp machine, this component planned strategies to decompose user requests into executable search plans.
- Searching Engine: A Britton-Lee IDM-600 relational database management system handled large-scale data retrieval.
- Knowledge Base: A repository of logic statements representing application-specific expertise.
When a user submitted a query, the deductive engine broke it down into a "search/compute" plan, queried the databases via Ethernet, applied logical rules to the results, and presented a final answer.
Key Insights: The "Amalgam" Approach and Explanatory AI
The most significant contribution of the KM-1 project was the "Amalgam" approach: the realization that knowledge bases and databases are complementary. By combining them, the system could provide the "why" and "how" of AI alongside the "what" and "scale" of traditional data management.
Furthermore, KM-1 introduced "Explanation as a Feature." Unlike black-box systems, KM-1 provided a "proof tree" to justify its answers. For instance, if a manager asked for a personnel transfer recommendation, the system would not just name a candidate but would show the underlying logic—such as availability due to project overstaffing or specific qualifications.
Results and Evaluation
The KM-1 prototype successfully demonstrated the feasibility of a Manager’s Assistant. It handled complex, multi-layered queries, such as identifying employees who could be reassigned between specific projects based on overlapping skill sets and availability.
In one test case, the system coordinated a plan involving 13 constrained relations. Ten of these were distributed to external databases, while three were computed locally in Lisp. This distributed processing proved that a deductive engine could effectively guide database searches, significantly reducing the cognitive load on human managers.
Limitations and Future Directions
Despite its success, the author noted that IRM was in its infancy. Challenges remained in knowledge acquisition—the manual process of encoding expertise into logic—and the need for qualitative reasoning regarding the "value" and "credibility" of information sources. Scaling these systems to handle the disparate resources of global corporations or federal agencies remained a significant engineering hurdle for the future.
Summary updated