The Delusional Hedge Algorithm as a Model of Human Learning From Diverse Opinions.

Area of Science:

• Cognitive Science
• Machine Learning
• Decision Making

Background:

• Traditional cognitive learning models often require direct experience with events and outcomes.
• Everyday learning frequently involves inferring information from others' opinions without direct experience or ground truth.
• Existing algorithms for learning from diverse sources, like the hedge algorithm, lack mechanisms for semi-supervised learning.

Purpose of the Study:

• To investigate how humans learn to trust opinions without direct experience or ground truth.
• To extend the hedge algorithm to accommodate both supervised and unsupervised learning experiences.
• To compare human judgment alignment with the standard hedge, a novel "delusional hedge" algorithm, and a heuristic baseline.

Main Methods:

• Introduction of the "delusional hedge" algorithm, a semi-supervised variant of the hedge algorithm.
• Conducting two experiments to evaluate human judgments against algorithmic predictions.
• Utilizing both labeled (supervised) and unlabeled (unsupervised) data in the learning process.

Main Results:

• Human judgments aligned significantly with the predictions of the delusional hedge algorithm.
• Evidence suggests humans effectively integrate both labeled and unlabeled information when learning from opinions.
• Findings indicate that human learners evaluate not only the accuracy but also the consistency of information sources.

Conclusions:

• Human learning from diverse opinions is effectively modeled by the delusional hedge algorithm.
• People assess the reliability of information sources by considering their accuracy and consistency with other trusted sources.
• This research advances understanding of human learning from conflicting information and informs the development of more sophisticated learning algorithms.