AIDED HUMAN PROCESSING OF INCONCLUSIVE EVIDENCE IN DIAGNOSTIC SYSTEMS: A SUMMARY OF EXPERIMENTAL EVALUATION.

Abstract

The report describes three experimental evaluations of a procedure for aiding men in combining probabilistic (inconclusive) information. The procedure is called 'Semi PIP' (Probabilistic Information Processing). In the semi PIP procedures, a computer relieves man of part of the task of combining probabilistic evidence. The experiments involved a simulated military threat-diagnosis task in which probabilistic data are used as a basis for deciding among alternative hypothesized threats. The Semi PIP was compared with an unaided procedure called POP (Posterior Probability), and both procedures compared with a mathematically ideal combination of evidence (Bayes' theorem). It was found from the three experiments that: (1) overall, the difference between Semi PIP and POP performance was very slight; (2) POP was best when a small total amount of evidence accumulated very rapidly; and, (3) Semi PIP was slightly, though consistently, superior when the total amount of evidence to be evaluated was large, when the total diagnostic impact in a set of evidence was large, or when both these conditions prevailed. Specific implications of the results for diagnostic system design and for research on basic human inference processes are summarized. A glossary of key terms from Bayesian decision theory is included in the report. (Author)

Document Details

Document Type
Technical Report
Publication Date
May 01, 1969
Accession Number
AD0691239

Entities

People

  • David A. Schum
  • Jack F. Southard
  • Louise F. Wombolt

Organizations

  • Ohio State University

Tags

Communities of Interest

  • Materials and Manufacturing Processes

DTIC Thesaurus Topics

  • Computers
  • Cooperation
  • Decision Theory
  • Information Processing
  • Mathematical Analysis
  • Mathematics
  • Probability
  • Test And Evaluation
  • Theorems

Readers

  • Artificial Intelligence
  • Human-Computer Interaction (HCI).
  • Systems Analysis and Design

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference
  • AI & ML - Neural Networks