Measuring trust in artificial intelligence with the N2pc component

Area of Science:

• Cognitive Science
• Neuroscience
• Human-Computer Interaction

Background:

• Efficient attention allocation is crucial for human-AI collaboration, where users must monitor AI performance to prevent errors.
• Over-reliance or excessive monitoring of AI can lead to performance degradation and critical failures.
• Trust in AI is a key factor influencing attentional effort offloading but is challenging to measure directly.

Purpose of the Study:

• To introduce and validate an electroencephalography (EEG)-based approach for directly tracking attentional resource sharing between humans and AI.
• To investigate how AI competency influences human attention deployment and trust calibration during collaborative tasks.
• To establish neurophysiological markers as implicit measures of trust in AI systems.

Main Methods:

• Participants engaged in a visual search task, collaborating with an AI of varying competency levels.
• Electroencephalography (EEG) was utilized to record brain activity.
• The N2pc component, a neural marker of selective visual attention, was measured to quantify attention deployment.

Main Results:

• N2pc amplitude was significantly modulated by the AI's competency.
• Smaller N2pc amplitudes correlated with increased attentional offloading and trust when interacting with a high-competency AI compared to a low-competency AI.
• These findings suggest that neural markers can implicitly reflect trust calibration in human-AI collaboration.

Conclusions:

• The N2pc component serves as a valid and non-disruptive neurophysiological marker for quantifying attention allocation in human-AI collaborative search tasks.
• This EEG-based approach offers a promising method for implicitly measuring trust in AI, advancing our understanding of trust calibration.
• The study extends the application of the N2pc from visual attention research to the critical domain of trust in automation.