Goal relevance as a quantitative model of human task relevance

Area of Science:

• Cognitive Science
• Information Theory
• Artificial Intelligence

Background:

• The concept of relevance is widely used but lacks a precise, quantitative definition, particularly for sensory observations related to goals.
• Existing frameworks struggle to numerically assess the informational value of data concerning goal attainment.

Purpose of the Study:

• To propose a theoretical definition for the information value of data observations with respect to a goal, termed "goal relevance."
• To establish a quantitative measure for the previously elusive concept of relevance.

Main Methods:

• Defined goal relevance using the Kullback-Leibler divergence between an agent's subjective belief distributions before and after observing new data.
• Utilized simulated environments to test theoretical predictions against human participant responses.

Main Results:

• Theoretical predictions derived from the goal relevance definition aligned with human participant judgments in 83.5% of trials.
• The proposed goal relevance measure outperformed several machine learning models in predicting relevance.

Conclusions:

• The introduced definition of goal relevance provides a general, quantitative, and explicit method for measuring the relevance of observations to specific goals.
• This framework offers a novel approach to understanding and quantifying information value in goal-directed behavior.