An fMRI and effective connectivity study investigating miss errors during advice utilization from human and machine...

Area of Science:

• Neuroscience
• Human-Computer Interaction
• Cognitive Science

Background:

• Societal reliance on automation necessitates understanding human advice utilization.
• Differentiating neural responses to human versus machine advice is crucial for effective human-AI collaboration.

Purpose of the Study:

• To investigate the neural mechanisms underlying advice utilization from human and machine agents.
• To examine how manipulated reliability and agent type (human vs. machine) affect advice seeking and neural processing.

Main Methods:

• Functional magnetic resonance imaging (fMRI) and multivariate Granger causality analysis were employed.
• Participants performed an X-ray luggage-screening task, evaluating advice from human and machine 'experts' with varying reliability.

Main Results:

• The machine-agent group exhibited reduced advice utilization compared to the human-agent group.
• Missed advice errors significantly degraded advice utilization, impacting attention and expectations.
• Neural correlates included salience and mentalizing networks, with specific roles for the lingual and fusiform gyri in decision and feedback phases, respectively.

Conclusions:

• Advice utilization is sensitive to agent type and reliability, with errors diminishing trust and engagement.
• Neural networks supporting salience detection, self-processing, and attentional modulation of sensory information are critical for processing and utilizing advice.
• Findings highlight the importance of perceived reliability and error management in designing effective human-AI advisory systems.