Mental Model Edit Distance as a Measure of Difficulty for Adaptation to Novelty

Abstract

The amount of change in an agents mental model required to accommodate novel experiences is proposed as a measure of difficulty for a task. Adaptation to novelty is viewed as learning to change and augment existing skills to confront unfamiliar situations. Agents have skills to perform tasks in an environment. Skills are viewed as algorithms to accomplish a range of tasks. If new skills are needed for adapting to a novel situation, these skill programs will need modification or augmentation. The degree of modification is considered as an edit distance in representation space using algorithmic information theory. The representation edit distance is related to simple, interpretable, and intuitive models in a variety of environments. Simple models almost as accurate as the best complex machine learning models often exist in practical domains as an example of the Rashomon effect. In this paper, we propose that the amount of editing of an effective representation (the Representation Edit Distance or RED) used in a set of skill programs in an agents mental model is a measure of difficulty for adaptation to novelty. A further approximation, the Mental Model Edit Distance (MMED) is a practical, semantic, and intuitive approximation to RED. We present some notional examples of how to use RED for predicting difficulty and how MMED can be used for the representations found in the DARPA SAIL-ON program.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Mar 01, 2023
Accession Number
AD1211378

Entities

People

  • Joshua Alspector

Organizations

  • Institute for Defense Analyses

Tags

Communities of Interest

  • Autonomy
  • Cyber
  • Weapons Technologies

DTIC Thesaurus Topics

  • Algorithms
  • Artificial Intelligence
  • Artificial Intelligence Software
  • Computer Languages
  • Computer Programming
  • Computer Programs
  • Computers
  • Environment
  • Infectious Diseases
  • Information Processing
  • Information Science
  • Information Systems
  • Information Theory
  • Machine Learning
  • Neural Networks
  • Ontologies
  • Supervised Machine Learning

Fields of Study

  • Computer science

Readers

  • Artificial Intelligence
  • Computer Vision.
  • Systems Analysis and Design

Technology Areas

  • AI & ML
  • Space