Yueqing Xuan , Edward Small , Kacper Sokol , Danula Hettiachchi , Mark Sanderson
{"title":"Comprehension is a double-edged sword: Over-interpreting unspecified information in intelligible machine learning explanations","authors":"Yueqing Xuan , Edward Small , Kacper Sokol , Danula Hettiachchi , Mark Sanderson","doi":"10.1016/j.ijhcs.2024.103376","DOIUrl":null,"url":null,"abstract":"<div><div>Automated decision-making systems are becoming increasingly ubiquitous, which creates an immediate need for their interpretability and explainability. However, it remains unclear whether users know what insights an explanation offers and, more importantly, what information it lacks. To answer this question we conducted an online study with 200 participants, which allowed us to assess explainees’ ability to realise <em>explicated information</em> – i.e., factual insights conveyed by an explanation – and <em>unspecified information</em> – i.e, insights that are not communicated by an explanation – across four representative explanation types: model architecture, decision surface visualisation, counterfactual explainability and feature importance. Our findings uncover that highly comprehensible explanations, e.g., feature importance and decision surface visualisation, are exceptionally susceptible to misinterpretation since users tend to infer spurious information that is outside of the scope of these explanations. Additionally, while the users gauge their confidence accurately with respect to the information explicated by these explanations, they tend to be overconfident when misinterpreting the explanations. Our work demonstrates that human comprehension can be a double-edged sword since highly accessible explanations may convince users of their truthfulness while possibly leading to various misinterpretations at the same time. Machine learning explanations should therefore carefully navigate the complex relation between their full scope and limitations to maximise understanding and curb misinterpretation.</div></div>","PeriodicalId":54955,"journal":{"name":"International Journal of Human-Computer Studies","volume":"193 ","pages":"Article 103376"},"PeriodicalIF":5.3000,"publicationDate":"2024-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"International Journal of Human-Computer Studies","FirstCategoryId":"94","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S1071581924001599","RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, CYBERNETICS","Score":null,"Total":0}
引用次数: 0
Abstract
Automated decision-making systems are becoming increasingly ubiquitous, which creates an immediate need for their interpretability and explainability. However, it remains unclear whether users know what insights an explanation offers and, more importantly, what information it lacks. To answer this question we conducted an online study with 200 participants, which allowed us to assess explainees’ ability to realise explicated information – i.e., factual insights conveyed by an explanation – and unspecified information – i.e, insights that are not communicated by an explanation – across four representative explanation types: model architecture, decision surface visualisation, counterfactual explainability and feature importance. Our findings uncover that highly comprehensible explanations, e.g., feature importance and decision surface visualisation, are exceptionally susceptible to misinterpretation since users tend to infer spurious information that is outside of the scope of these explanations. Additionally, while the users gauge their confidence accurately with respect to the information explicated by these explanations, they tend to be overconfident when misinterpreting the explanations. Our work demonstrates that human comprehension can be a double-edged sword since highly accessible explanations may convince users of their truthfulness while possibly leading to various misinterpretations at the same time. Machine learning explanations should therefore carefully navigate the complex relation between their full scope and limitations to maximise understanding and curb misinterpretation.
期刊介绍:
The International Journal of Human-Computer Studies publishes original research over the whole spectrum of work relevant to the theory and practice of innovative interactive systems. The journal is inherently interdisciplinary, covering research in computing, artificial intelligence, psychology, linguistics, communication, design, engineering, and social organization, which is relevant to the design, analysis, evaluation and application of innovative interactive systems. Papers at the boundaries of these disciplines are especially welcome, as it is our view that interdisciplinary approaches are needed for producing theoretical insights in this complex area and for effective deployment of innovative technologies in concrete user communities.
Research areas relevant to the journal include, but are not limited to:
• Innovative interaction techniques
• Multimodal interaction
• Speech interaction
• Graphic interaction
• Natural language interaction
• Interaction in mobile and embedded systems
• Interface design and evaluation methodologies
• Design and evaluation of innovative interactive systems
• User interface prototyping and management systems
• Ubiquitous computing
• Wearable computers
• Pervasive computing
• Affective computing
• Empirical studies of user behaviour
• Empirical studies of programming and software engineering
• Computer supported cooperative work
• Computer mediated communication
• Virtual reality
• Mixed and augmented Reality
• Intelligent user interfaces
• Presence
...