Dancing With Algorithms: Interaction Creates Greater Preference and Trust in Machine-Learned Behavior.

Area of Science:

• Robotics and Artificial Intelligence
• Human-Computer Interaction

Background:

• Machine learning (ML) often lacks user interaction, hindering trust.
• Interactive machine learning (IML) integrates user feedback for improved algorithm recognition and collaboration.
• Developing trustworthy control for unmanned vehicles is crucial for their application.

Purpose of the Study:

• To evaluate interactive machine learning (IML) for developing trustworthy control of unmanned vehicle area search behaviors.
• To assess the impact of user interaction levels on the perception and acceptance of ML-driven behaviors.
• To investigate user trust, preference, and recognition of ML behaviors developed through IML.

Main Methods:

• Participants evaluated and selected behaviors under low and high IML interaction conditions.
• Behavior evolution was observed over time using ML.
• User trust, preference, and discriminability of behaviors were quantitatively measured.

Main Results:

• Behaviors developed with IML were significantly more trusted, preferred, and recognizable than those from non-interactive methods.
• IML-developed behaviors achieved performance comparable to pure ML models.
• User involvement in IML enhanced the perception and acceptance of the control algorithms.

Conclusions:

• IML demonstrates significant potential for developing user-satisfactory control behaviors for unmanned vehicles.
• This study is the first to extend IML to vehicle behavior modeling with a multifaceted evaluation of user perception, trust, and implementation.
• IML offers a promising avenue for generating trustworthy and high-performing autonomous systems, though further research is needed.