COMEX: A Commodities Support System

J. Stansfield

Abstract

In the late 1970s, commodities analysts faced a significant data bottleneck due to the high volume of agricultural reports containing both quantitative data and descriptive text. COMEX-0 was developed as an intelligent assistant to address this challenge using frame-based knowledge representation. By modeling human-centric concepts like trading and transport through structured templates, the system enabled automated reasoning and qualitative simulation. This paper explores the architecture of COMEX-0, its inferencing mechanisms, and its ability to handle incomplete information in the commodities market.

Introduction

The commodities market of the 1970s was characterized by an influx of data from diverse sources, including weather reports, crop status updates, and market movement summaries. Analysts required tools that could go beyond simple data storage to "understand" the implications of these reports. COMEX (Commodities Expert) was designed to bridge this gap by representing complex events in a format suitable for computer-aided reasoning and qualitative simulation.

Frame-Based Knowledge Representation

At the core of COMEX-0 is the concept of "frames," a knowledge representation technique popularized by Marvin Minsky. Frames serve as structured templates for specific concepts or events. For instance, a MOVING event is defined by its relationship to other events and the specific data points (slots) it must contain.

Figure 1: A frame definition for the 'MOVING' event, illustrating slots and inheritance from the 'CHANGING' category.

As shown in Figure 1, the MOVING frame inherits properties from a more general CHANGING event. It includes slots for the object being moved (SUBJECT), its origin (OLD-VALUE), and its destination (NEW-VALUE). This hierarchical structure allows the system to automatically infer that a MOVING event also possesses attributes like "time," inherited from higher-level parent categories.

Inferencing and Event Aggregation

COMEX-0 employs three primary inferencing methods to process information:

  1. Definition Expansion: Complex events are decomposed into simpler sub-components. For example, an EXPORTING event is modeled as a combination of TRADING (an exchange of goods) and TRANSPORTING (causing an object to move).
  2. Constraints and Procedural Attachment: Frames are not merely static data structures. They utilize "if-added" theorems—procedural code triggered when data is entered. These procedures can validate input or automatically update related values, such as recalculating supply when production figures change.
  3. Matching: This mechanism allows the system to identify when two different frame structures describe the same real-world entity, merging their information to create a more complete picture.

Qualitative Reasoning

A significant innovation in COMEX-0 is its focus on qualitative reasoning. Unlike traditional systems that rely solely on precise mathematical models, COMEX was designed to handle descriptive and incomplete information. This approach allows the system to reason about abstract concepts like "market pressure" or environmental factors such as "rain falling on soil," mimicking the way human analysts interpret narrative reports without requiring perfect mathematical equations for every variable.

Results and Future Work

COMEX-0 served as a successful proof-of-concept, capable of representing approximately 20 distinct event types alongside price and weather data. The system demonstrated that formal dialogues could successfully instantiate complex frame structures, allowing the computer to "understand" the narrative flow of commodities reports.

However, the system faced challenges regarding the management of constraints. As the network of interconnected frames expanded, the computational complexity of managing cascading changes (ripples) across the system increased. Future work aimed to extend COMEX to include qualitative simulations of biological processes, such as wheat growth, to provide a more holistic view of how environmental variables impact market outcomes.

Conclusion

COMEX-0 represented a pioneering effort in applying AI to the commodities sector. By leveraging frame-based representation and qualitative reasoning, it provided a framework for managing the complexities of agricultural data, paving the way for more sophisticated intelligent assistants in market analysis.

Summary updated

BibTeX
@inproceedings{stansfield1977comex,
  author = {J. Stansfield},
  title = {{COMEX: A Commodities Support System}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {1977},
}
Figure