In human-machine trust, humans rely on a simple averaging strategy

Area of Science:

• Cognitive Science
• Human-Computer Interaction
• Artificial Intelligence

Background:

• Artificial intelligence (AI) is increasingly integrated into daily life, necessitating research into effective human-AI collaboration.
• Understanding how humans integrate machine recommendations, especially when they conflict with personal judgment, is crucial for trust in human-machine teaming.

Purpose of the Study:

• To investigate trust dynamics in human-machine teaming using a perceptual judgment task.
• To analyze how the 'advice distance' (discrepancy between human and machine judgment) influences trust and behavioral adjustments.

Main Methods:

• Participants performed a perceptual estimation task and received a recommendation from a machine agent.
• Trust was measured by the degree participants shifted their second response towards the machine's recommendation.
• The study analyzed how participants' judgments changed in response to varying distances between their initial estimate and the AI's advice.

Main Results:

• While some participants either increased or decreased trust based on advice distance, the most common behavior was not extreme.
• A simple averaging model best explained participants' trust behavior, indicating a nuanced integration of AI recommendations.
• Human trust in machine agents did not follow simple distrust or over-reliance patterns when recommendations differed significantly.

Conclusions:

• Human trust in AI is not binary; it involves complex integration strategies, often leaning towards averaging judgments.
• Findings suggest that simple models of trust, like averaging, are more accurate in predicting human-AI collaboration behavior.
• Implications for developing more effective human-machine teaming systems and refining theories of trust in AI are discussed.