The Use of Qualitative and Quantitative Simulations

Reid G. SimmonsAAAI 1983

Introduction

In the early development of artificial intelligence, a central challenge was reconciling high-level, symbolic reasoning with precise, numerical computation. In his 1983 paper, Reid Simmons introduced "imagining," a hybrid technique that integrates qualitative and quantitative simulations to solve complex verification tasks. By leveraging the strengths of both approaches, the system can determine if a proposed sequence of events logically and geometrically results in a specific observed state.

Problem Statement: The Verification Challenge

The research addresses the "test" phase of the "generate and test" problem-solving paradigm. Specifically, given an initial state, a final goal state, and a proposed sequence of events, how can an automated system verify the validity of that sequence?

In domains such as geology, this is particularly difficult. A purely quantitative model requires exact parameters (e.g., the precise angle of a tectonic tilt) which are often unknown. Conversely, a purely qualitative model is too abstract; stating that a formation "tilted clockwise" lacks the geometric detail necessary to match a precise geological diagram. The core challenge lies in "parameter inference"—calculating the initial values required to reach a known final state.

A complex geological cross-section representing a goal state to be verified.

The "Imagining" Framework

Simmons proposes a three-step process to bridge this gap:

  1. Qualitative Simulation: The system symbolically represents changes for each event, generating "change equations" that relate parameters (such as layer height or angle) before and after an event.
  2. Parameter Inference: By taking measurements from the goal diagram (the final state) and working backward through the symbolic change equations, the system infers the original numeric values for each process.
  3. Quantitative Simulation: Using these inferred numeric values, the system runs an algorithmic simulation to construct a sequence of diagrams. Verification is achieved if the final generated diagram matches the target goal diagram.

This method distinguishes between algorithmic descriptions (quantitative "how-to" steps) and assertional descriptions (qualitative "what happens" statements).

Application in Geology

The system was tested using geologic interpretation, where the primary data consists of cross-sectional diagrams and sequences of geologic processes such as deposition, uplift, intrusion, tilting, faulting, and erosion.

Legend for geological cross-sections showing symbols for Shale, Sandstone, and Mafic-Igneous rock.

Key insights from this application include:

  • The Skeleton and the Flesh: Qualitative simulation provides the structural "skeleton" of changes over time, while quantitative simulation provides the spatial "flesh."
  • Error Detection and Model Debugging: Mismatches between the simulation and the goal often reveal "bugs" in the qualitative model. For example, if a faulting process is described only by horizontal movement but the diagram shows vertical displacement, the resulting inaccuracy highlights exactly where the qualitative description is incomplete.
  • Measurement Correction: The system recognizes that final state measurements cannot be used directly as initial parameters. They must be "corrected" by reasoning about the cumulative effects of subsequent geological events.

Limitations and Future Directions

While effective, the "imagining" technique faces challenges in qualitative shape description, which remains difficult to formulate symbolically. Additionally, some parameters (like the magnitude of an uplift event followed by total erosion) are unmeasurable from the final state. In such cases, the system uses arbitrary values that do not affect the final geometric match.

Simmons suggests that this framework could extend beyond geology to fields like economics or chemistry, where testing theories of how a current state was reached is of paramount importance.

Summary updated

BibTeX
@inproceedings{simmons1983use,
  author = {Reid G. Simmons},
  title = {{The Use of Qualitative and Quantitative Simulations}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {1983},
}
Figure