Let people fail! Exploring the influence of explainable virtual and robotic agents in learning-by-doing tasks

TL;DR

Interestingly, participants autonomously performing the learning-by-doing task demonstrated superior knowledge acquisition than those assisted by explainable AI (XAI) and have significant implications for automated tutoring and human-AI collaboration.

Abstract

Collaborative decision-making with artificial intelligence (AI) agents presents opportunities and challenges. While human-AI performance often surpasses that of individuals, the impact of such technology on human behavior remains insufficiently understood, primarily when AI agents can provide justifiable explanations for their suggestions. This study compares the effects of classic vs. partner-aware explanations on human behavior and performance during a learning-by-doing task. Three participant groups were involved: one interacting with a computer, another with a humanoid robot, and a third one without assistance. Results indicated that partner-aware explanations influenced participants differently based on the type of artificial agents involved. With the computer, participants enhanced their task completion times. At the same time, those interacting with the humanoid robot were more inclined to follow its suggestions, although they did not reduce their timing. Interestingly, participants autonomously performing the learning-by-doing task demonstrated superior knowledge acquisition than those assisted by explainable AI (XAI). These findings raise profound questions and have significant implications for automated tutoring and human-AI collaboration.

Figures (12)

• Figure 1: A participant interacts with the humanoid robot iCub during the training phase (Robot group). On the screen is the nuclear power plant application running.
• Figure 2: Experimental design for the NPP experiment. Three macro-groups were considered: COM (with the computer), Self-taught, and Robot. Participants from the COM and Robot groups were divided into two experimental conditions (C-XAI and A-XAI) depending on the explanation strategy adopted by the agent.
• Figure 3: Example of DT where the leaf nodes 11 and 2 are the robot's suggestion and the predicted user's action, respectively. The classical XAI selects node 1 for the explanations since it is the most unused relevant node. The partner-aware XAI selects node 7 instead because it represents a perfect contrastive explanation for the fact 11 and foil 2.
• Figure 4: Distribution of the participant's knowledge about the functioning of nuclear power plants before the experiment. We classified them into three levels of knowledge (No, Some, A lot) by coding their open-ended questions to the pre-experiment questionnaire.
• Figure 5: The number of actions participants in the COM group performed during training. The * represents a statistically significant difference (p-value = .039, independent samples t-test).