Explanatory machine learning for justified trust in human-AI collaboration: Experiments on file deletion...

Area of Science:

• Human-Computer Interaction
• Artificial Intelligence
• Cognitive Science

Background:

• Digital data accumulation leads to energy consumption and information overload.
• Effective digital data management is crucial for focus and productivity.
• Existing systems lack robust support for deleting irrelevant digital objects.

Purpose of the Study:

• To examine an explanatory AI system (Dare2Del) for deleting digital data.
• To optimize human-computer interactions by varying design features like explanations and verifiability.
• To investigate the link between digital file deletion and human memory forgetting.

Main Methods:

• Three experiments (N=61, 33, 73) tested Dare2Del with varied design features.
• Multilevel modeling analyzed the impact of explanations, familiarity, and verifiability.
• The study explored distributed cognition principles connecting digital and human memory.

Main Results:

• Explanations significantly increased acceptance of deletion suggestions across all experiments.
• Verifiability of explanations is crucial for building user trust in the AI system.
• No clear evidence was found that deleting digital files aids human forgetting.

Conclusions:

• Explanatory and verifiable AI systems can aid digital data management.
• AI design should prioritize user trust through transparent explanations.
• Further research is needed to understand the relationship between digital deletion and cognitive forgetting.