A New Framework for Designing Explainable AI for Augmented Reality Applications

Artificial intelligence (AI) and machine learning tools are commonly used today to power technical applications, but the underpinnings of many of these tools are difficult to decipher. This is because most of them are based on “black box” models, i.e. models that learn how to analyze data and make predictions, but the processes behind these predictions are not shared with human users. yeah.

Researchers at Meta Reality Labs recently created XAIR, a framework that helps developers understand the processes behind AI predictions. This framework is Proceedings of the 2023 CHI Conference on Human Factors in Computing Systemsis specifically designed to create explainable AI (XAI) systems that can be applied to various augmented reality (AR) settings.

“As black-box models are increasingly adopted in everyday life, there is growing concern about humans misusing AI and losing control,” said two of the researchers who conducted the study. One Xuhai Xu and Anna Yu told Tech Xplore. “This created a need to make algorithms easier to understand, which led to the proliferation of XAI. Existing research shows that XAI can help resolve end-user confusion and build trust. , industry professionals are looking to use XAI to improve user experience. ”

AR technology allows users to view a modified version of their surrounding environment that integrates digital elements, sound, and/or visual enhancements. These “digitally enhanced” versions of reality can be seen through head-mounted displays, goggles, other wearable gear, or even simply through smartphone screens.

Recently, some researchers are looking at using AI to power AR applications. For example, it is more responsive to changes in the user’s environment, and it is able to analyze and predict certain objects. Xu and Yu and their colleagues set out to create a framework that could make the results of his AI tools for AR applications easier to understand and increase user confidence.

“Since context-aware everyday AR requires an AI model, XAI will also be essential as end users will interact with all kinds of AI results,” said Xu and Yu. “XAI can help in many ways, including making intelligent AR behavior interpretable, resolving confusion and surprise with unexpected AI results, promoting privacy awareness, and building trust. Given the nature of this, we aim to answer research questions such as: What is the right way to create effective XAI experiences for AR in everyday scenarios?”

Meta’s team created the XAIR framework in hopes of facilitating the design of XAI for AR applications. Their framework basically addresses his three open questions: when, what and how? The answers to these questions can be used to provide a better explanation for AI predictions in AR scenarios. In addition to helping developers create AI that can answer these three questions, XAIR has a set of important guidelines for researchers and developers working on his XAI for AR applications. provides an overview of

“Based on an extensive literature review, we identified five key factors,” explained Xu and Yu. “These elements are the ‘when, what, how’ aspects, including two AR-specific elements: user state and contextual information, and three non-AR-specific elements: system goals, user goals, and user profiles. determine the design of

Essentially, to use the Teams framework, developers must first address these five elements to provide accurate contextual information about users, user states, system-wide goals, and potential user goals and profiles. must be specified. Once you do this, you can adapt and refine your XAI system for AR applications by simply referencing the XAIR framework.

“AI is getting more and more powerful, and we can expect AI to help us automatically identify at least a subset of these five factors in the near future,” said Xu and Yu. “So it makes the framework an automation or self-automating tool that allows a designer to improve his design of her XAI in AR.”

As part of their research, Xu, Yu and others summarized over 100 research findings rooted in various disciplines, identified key aspects to consider when developing XAI for AR applications, and identified when and what , answered how to ask. The researchers then conducted a large-scale survey of over 500 of his users and held a workshop with 12 experts in the field. The questionnaire responses and views shared by the experts during the workshop provided valuable insights to guide the development of XAIR.

“XAIR is the first framework for XAI design in AR scenarios and also includes guidelines to support the designer’s design thinking process,” said Xu and Yu. “The results of a design workshop with 10 designers showed that XAIR can provide designers with meaningful and insightful creativity support.We also implemented a real-time AR system based on one design. , tested with 12 end-users.”

To assess the framework’s value, researchers used it to create a real-world XAI system and tested it in real-time in a series of AR scenarios. They found that users perceived this system as both transparent and trustworthy, suggesting that their framework effectively guided its development. In the future, the XAIR framework can be used to create various AI systems to power AR applications. This helps users perceive their predictions as more credible because they can explain their predictions.

“Next research will explore things like automating the design framework, creating long-term personalized XAI experiences in AR, and allowing users to provide feedback to further improve the system.” added Xu and Yu. . “The XAIR framework creates a foundation and guidelines for exploring XAI interactions in future AR systems. The recent explosion of generative AI is also very exciting, and this trend will help shape the future of XAI in AR. I’m interested in exploring how it impacts design.”