Mental Models of Mere Mortals with Explanations of Reinforcement Learning

XGate: Explainable Reinforcement Learning for Transparent and Trustworthy API Traffic Management in IoT Sensor Networks

The rapid proliferation of Internet of Things (IoT) devices and their associated application programming interfaces (APIs) has significantly increased the complexity of sensor network traffic management, necessitating more sophisticated and transparent control mechanisms. In this paper, we introduce XGate, a novel explainable reinforcement learning framework designed specifically for API traffic management in sensor networks. XGate addresses the critical challenge of balancing optimal routing decisions with the interpretability demands of network administrators operating large-scale IoT deployments. Our approach integrates transformer-based attention mechanisms with counterfactual reasoning to provide human-comprehensible explanations for each traffic management decision across distributed sensor data streams. Through extensive experimentation on three large-scale sensor API traffic datasets, we demonstrate that XGate achieves 23.7% lower latency and 18.5% higher throughput compared to state-of-the-art black-box reinforcement learning approaches. More importantly, our user studies with sensor network administrators (n=42) reveal that XGate's explanation capabilities improve operator trust by 67% and reduce intervention time by 41% during anomalous sensor traffic events. The theoretical analysis further establishes probabilistic guarantees on explanation fidelity while maintaining computational efficiency suitable for real-time sensor data management. XGate represents a significant advancement toward trustworthy AI systems for critical IoT infrastructure, providing transparent decision making without sacrificing performance in dynamic sensor network environments.

Towards reconciling usability and usefulness of policy explanations for sequential decision-making systems

Safefy-critical domains often employ autonomous agents which follow a sequential decision-making setup, whereby the agent follows a policy to dictate the appropriate action at each step. AI-practitioners often employ reinforcement learning algorithms to allow an agent to find the best policy. However, sequential systems often lack clear and immediate signs of wrong actions, with consequences visible only in hindsight, making it difficult to humans to understand system failure. In reinforcement learning, this is referred to as the credit assignment problem. To effectively collaborate with an autonomous system, particularly in a safety-critical setting, explanations should enable a user to better understand the policy of the agent and predict system behavior so that users are cognizant of potential failures and these failures can be diagnosed and mitigated. However, humans are diverse and have innate biases or preferences which may enhance or impair the utility of a policy explanation of a sequential agent. Therefore, in this paper, we designed and conducted human-subjects experiment to identify the factors which influence the perceived usability with the objective usefulness of policy explanations for reinforcement learning agents in a sequential setting. Our study had two factors: the modality of policy explanation shown to the user (Tree, Text, Modified Text, and Programs) and the "first impression" of the agent, i.e., whether the user saw the agent succeed or fail in the introductory calibration video. Our findings characterize a preference-performance tradeoff wherein participants perceived language-based policy explanations to be significantly more useable; however, participants were better able to objectively predict the agent's behavior when provided an explanation in the form of a decision tree. Our results demonstrate that user-specific factors, such as computer science experience (p < 0.05), and situational factors, such as watching agent crash (p < 0.05), can significantly impact the perception and usefulness of the explanation. This research provides key insights to alleviate prevalent issues regarding innapropriate compliance and reliance, which are exponentially more detrimental in safety-critical settings, providing a path forward for XAI developers for future work on policy-explanations.

Essential properties and explanation effectiveness of explainable artificial intelligence in healthcare: A systematic review

Background

Significant advancements in the field of information technology have influenced the creation of trustworthy explainable artificial intelligence (XAI) in healthcare. Despite improved performance of XAI, XAI techniques have not yet been integrated into real-time patient care.

Objective

The aim of this systematic review is to understand the trends and gaps in research on XAI through an assessment of the essential properties of XAI and an evaluation of explanation effectiveness in the healthcare field.

Methods

A search of PubMed and Embase databases for relevant peer-reviewed articles on development of an XAI model using clinical data and evaluating explanation effectiveness published between January 1, 2011, and April 30, 2022, was conducted. All retrieved papers were screened independently by the two authors. Relevant papers were also reviewed for identification of the essential properties of XAI (e.g., stakeholders and objectives of XAI, quality of personalized explanations) and the measures of explanation effectiveness (e.g., mental model, user satisfaction, trust assessment, task performance, and correctability).

Results

Six out of 882 articles met the criteria for eligibility. Artificial Intelligence (AI) users were the most frequently described stakeholders. XAI served various purposes, including evaluation, justification, improvement, and learning from AI. Evaluation of the quality of personalized explanations was based on fidelity, explanatory power, interpretability, and plausibility. User satisfaction was the most frequently used measure of explanation effectiveness, followed by trust assessment, correctability, and task performance. The methods of assessing these measures also varied.

Conclusion

XAI research should address the lack of a comprehensive and agreed-upon framework for explaining XAI and standardized approaches for evaluating the effectiveness of the explanation that XAI provides to diverse AI stakeholders.

Measures for explainable AI: Explanation goodness, user satisfaction, mental models, curiosity, trust, and human-AI performance

If a user is presented an AI system that portends to explain how it works, how do we know whether the explanation works and the user has achieved a pragmatic understanding of the AI? This question entails some key concepts of measurement such as explanation goodness and trust. We present methods for enabling developers and researchers to: (1) Assess the a priori goodness of explanations, (2) Assess users' satisfaction with explanations, (3) Reveal user's mental model of an AI system, (4) Assess user's curiosity or need for explanations, (5) Assess whether the user's trust and reliance on the AI are appropriate, and finally, (6) Assess how the human-XAI work system performs. The methods we present derive from our integration of extensive research literatures and our own psychometric evaluations. We point to the previous research that led to the measurement scales which we aggregated and tailored specifically for the XAI context. Scales are presented in sufficient detail to enable their use by XAI researchers. For Mental Model assessment and Work System Performance, XAI researchers have choices. We point to a number of methods, expressed in terms of methods' strengths and weaknesses, and pertinent measurement issues.

Finding AI’s Faults with AAR/AI: An Empirical Study

Would you allow an AI agent to make decisions on your behalf? If the answer is “not always,” the next question becomes “in what circumstances”? Answering this question requires human users to be able to assess an AI agent—and not just with overall pass/fail assessments or statistics. Here users need to be able to localize an agent’s bugs so that they can determine when they are willing to rely on the agent and when they are not. After-Action Review for AI (AAR/AI), a new AI assessment process for integration with Explainable AI systems, aims to support human users in this endeavor, and in this article we empirically investigate AAR/AI’s effectiveness with domain-knowledgeable users. Our results show that AAR/AI participants not only located significantly more bugs than non-AAR/AI participants did (i.e., showed greater recall) but also located them more precisely (i.e., with greater precision). In fact, AAR/AI participants outperformed non-AAR/AI participants on every bug and were, on average, almost six times as likely as non-AAR/AI participants to find any particular bug. Finally, evidence suggests that incorporating labeling into the AAR/AI process may encourage domain-knowledgeable users to abstract above individual instances of bugs; we hypothesize that doing so may have contributed further to AAR/AI participants’ effectiveness.

A survey on artificial intelligence assurance

Artificial Intelligence (AI) algorithms are increasingly providing decision making and operational support across multiple domains. AI includes a wide (and growing) library of algorithms that could be applied for different problems. One important notion for the adoption of AI algorithms into operational decision processes is the concept of assurance. The literature on assurance, unfortunately, conceals its outcomes within a tangled landscape of conflicting approaches, driven by contradicting motivations, assumptions, and intuitions. Accordingly, albeit a rising and novel area, this manuscript provides a systematic review of research works that are relevant to AI assurance, between years 1985 and 2021, and aims to provide a structured alternative to the landscape. A new AI assurance definition is adopted and presented, and assurance methods are contrasted and tabulated. Additionally, a ten-metric scoring system is developed and introduced to evaluate and compare existing methods. Lastly, in this manuscript, we provide foundational insights, discussions, future directions, a roadmap, and applicable recommendations for the development and deployment of AI assurance.