Model Expert System MES

TL;DR

This paper introduces MES, a model expert system designed to unify numerical calculations with logical inference for oil log interpretation. By introducing the concept of "TASK" and defining Task Transformation Rules, MES provides a flexible framework for integrating diverse problem-solving approaches, paving the way for more comprehensive and intelligent expert systems.

What Problem Does This Paper Address?

Traditional expert systems in fields like oil log interpretation often focus heavily on logical or plausible inference. However, the real-world process of oil log interpretation involves a crucial interplay between numerical calculations and logical reasoning. Experts first perform qualitative analysis to determine parameters for quantitative calculations, then use these results for further inference, and so on. This iterative and integrated approach is challenging to capture within a purely rule-based inference system. The authors aim to create a uniform framework that can effectively handle both numerical computation and logical inference without sacrificing the strengths of existing rule-based systems.

Methods

The core of MES is its novel approach to knowledge representation and task solving.

  • Knowledge Representation:
    • ASSERTION: Defined as (?X, OBJECT, VALUE, SUPPORT, PLAUSIBILITY), where SUPPORT and PLAUSIBILITY quantify evidence for or against a VALUE.
  • TASK: Defined as T=(TN, ?TV), representing a problem to be solved with a binary outcome (TRUE/FALSE). Each OBJECT has a corresponding TASK.
  • Uncertainty Representation: Leverages a simplified form of Dempster-Shafer's Evidence Theory, using SUPPORT (belief) and PLAUSIBILITY (plausibility) to express uncertainty.
  • Task Solving:
    • Task Transformation Rules: The central mechanism for solving TASKS. These rules are divided into:
  • Task Decomposition Rules: Break down a TASK into sub-TASKS and define how to evaluate the parent TASK.
  • Task Derivation Rules: Define the order in which TASKS are solved, focusing on sequence rather than strict logical relations.
  • Inference Rules: * Production Rules: Used to bind values to OBJECTS based on conditions.
  • Procedure Rules: Call subroutines (numerical calculations) when conditions are met.
  • Integration: MES integrates numerical calculations (via Procedure Rules and subroutines) and logical inference (via Production Rules) within a unified TASK-solving framework. The "side effect" of solving a TASK is the production of new ASSERTIONS.

Datasets

Not discussed in the paper. The paper focuses on the design of the MES system and uses the WELIES (comprehensive oil log interpretation expert system) as an example to illustrate its functionality.

Key Insights and Ideas

  • Unified Framework: The most significant insight is the creation of a unified framework that seamlessly integrates numerical computation with logical inference. This addresses a key limitation of many existing expert systems.
  • The "TASK" Concept: Introducing "TASK" as a fundamental unit of problem-solving allows MES to manage complex processes that involve both calculation and reasoning.
  • Task Transformation Rules: These rules provide a flexible and powerful mechanism for defining how complex problems are decomposed and solved, allowing for both top-down (decomposition) and sequential (derivation) approaches.
  • Handling Uncertainty: The use of Dempster-Shafer's Evidence Theory for SUPPORT and PLAUSIBILITY provides a robust way to represent and propagate uncertainty, which is critical in domains like oil log interpretation.

Key Results

The paper primarily presents a conceptual design and methodology for the MES system. It describes the architecture and the mechanisms for knowledge representation and task solving. The system WELIES, built upon MES, is mentioned as an implementation example, but specific quantitative results or performance metrics for WELIES are not detailed in this paper. The key result is the successful design of a model expert system that can integrate numerical calculation and logical inference.

Limitations and Future Work

  • Limitations:
    • The paper focuses on the conceptual design of MES and uses WELIES as an example, but does not provide detailed performance evaluations or benchmarks for the system.
  • The explanation of uncertainty representation is simplified, and the full implications of the chosen simplification are not deeply explored.
  • The paper mentions a "fixed threshold" for rule invocation but does not specify its value or how it is determined.
  • Future Work:
    • The authors plan to establish other oil expert systems based on the MES framework.
  • Further development and application of the MES system to other domains are implied.

Summary updated

BibTeX
@inproceedings{1985model,
  title = {{Model Expert System MES}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {1985},
}
Figure