diff --git "a/4NAzT4oBgHgl3EQfD_rN/content/tmp_files/2301.00987v1.pdf.txt" "b/4NAzT4oBgHgl3EQfD_rN/content/tmp_files/2301.00987v1.pdf.txt" new file mode 100644--- /dev/null +++ "b/4NAzT4oBgHgl3EQfD_rN/content/tmp_files/2301.00987v1.pdf.txt" @@ -0,0 +1,1753 @@ +1 + +Book Series: +HUMAN-COMPUTER INTERACTION: FOUNDATIONS, METHODS, TECHNOLOGIES AND +APPLICATIONS +BOOK #5 - HUMAN COMPUTER INTERACTION: INTERACTING IN INTELLIGENT ENVIRONMENTS +Editors: Constantine Stephanidis & Gavriel Salvendy +Publisher: CRC Press + +Chapter 5 AI in HCI Design and User Experience +Wei Xu, Ph.D. +(Dec. 2022) +Zhejiang University, China + +5.1 Introduction +Recently, much progress has been made in artificial intelligence (AI) and machine-learning (ML); such +progress has also enabled human-computer interaction (HCI) and user experience (UX) professionals to deliver +solutions with better UX (Lu et al., 2022; Yang et al., 2018; Kuniavsky et al., 2017). The use of AI/ML capabilities +for improving HCI/UX work and delivering better UX in solutions is becoming a trend (Abbas et al., 2022; Wu et +al., 2019; Nikiforova et al., 2021) and creates many new opportunities for HCI/UX professionals (Holmquist, 2017; +Yang et al., 2020). Some even speculate “AI/ML is the new UX” (Yang et al., 2018). +Researchers proposed that AI can perform as an assistant, collaborator, researcher, or facilitator (Bertão & +Joo, 2021; Main & Grierson, 2020). AI technology will change the role of designers in the design process and +generate an opportunity for creative collaboration between AI and designers (McCormack et al., 2020). Also, +companies are moving fast to adopt AI for improving customer experience (CX). In 2018, the IBM Institute for +Business Value (IBV) surveyed 1,194 executives from seven industries worldwide who are responsible for the AI +initiatives of their companies (Schwartz et al., 2018). The results show that 74% said AI would fundamentally +change how they approach CX; 41% had an AI strategy considering the changes ahead. To some extent, AI and UX +designers have similar functions. They both gather data, analyze users’ behavior and interactions, and can predict +human behavior (Donahole, 2021). For example, Chatbots, Google Translate, and Alexa are good examples of AI +technology that uses big data to deliver enhanced UX. +Table 5.1 summarizes the benefits of AI technology in enhancing HCI/UX design (e.g., Yang et al., 2018; +Inkbot Design, 2021; Rogers, 2020; Schwartz et al., 2018; Baker, 2019; Donahole, 2021). Herein, AI refers to +technologies (e.g., AI, ML, big data) used to develop AI-based intelligent systems, applications, or services. + +Table 5.1 Benefits of AI for HCI and UX design + +2 + +Benefits of AI +Descriptions and Examples +Sensing users +intelligently and +supporting user +research effectively +• +Collect user personal knowledge (e.g., online shopping behavior, social +connections) +• +Recognize user’s activity (e.g., physical status and interaction with systems) +• +Infer the user’s internal status (e.g., intention, emotion, attitude) +• +Identify unique characteristics and behavioral patterns of collective users (e.g., +digital personas) +• +Sense context of user interaction (e.g., online historical shopping data) +Acting intelligently +based on insights +from sensing users +• +React with appropriate actions (e.g., inform, engage, assist, promote products) +• +React proactively or autonomously (e.g., empathy, influence, conflict +management) +• +Analyze and optimize customer journeys +Personalization +design + +• +Provider personalized online content and functionality based on user preferences +and interactions (e.g., data log, digital personas) +• +Promote marketing experience to another level by using users’ personal +information +• +Deliver advanced localization capabilities to handle language-related activities +• +Focus on satisfying the precise needs of users +Analyzing a large +amount of data +more quickly and +efficiently + +• +Reveal insights that help users rapidly make decisions by providing real-time +responses to user inquiries +• +Analyze large amounts of data to ascertain patterns and deliver meaningful +research results (e.g., quickly generate questionnaires, and provide relevant +responses for further inquiry) +• +Process vast datasets to gather information +• +Analyze massive sets of data to modify user experiences +• +Simulate intellectual cycles to empower decision-making +• +Automate mechanical errands at excellent paces +• +Gather and draw inferences from vast volumes of data in a time +Powering HCI/UX +activities for +efficiency + +• +Automate repetitive design activities (e.g., resize images, make color corrections, +crop images) +• +Help generate a creative idea in the early design stages +• +Leverage algorithms to create flow diagrams in UI design based on historical user +patterns +• +Develop wireframes based on the understanding of the context and the flow +• +Help generate multiple variants of a UI design solution for A/B testing +• +Automate design tasks, e.g., identify patterns in images and help designers stitch +them together + +3 + +• +Run dynamic A/B testing and analyze test results +• +Automate back-end processes (e.g., automate targeted marketing promotions) +• +Help designers quickly make design decisions (e.g., predictions based on +historical datasets, giving users the fewest potential choices +Providing more +natural and +effective +interaction +• +Enable new types of user interface technologies (e.g., voice input, face +recognition, gesture interaction, brain-computer interface) +• +Handle inaccurate input through reasoning (e.g., user intent detection, affective +interaction) +• +Build thinner UI with AI (e.g., using historical data) to anticipate a user’s action +or better prioritize the queries of users, provide a possible solution or pertinent +results +Better marketing +• +Help build a better connection among brands across target audiences and boost +their relationship. +• +Build customized eCommerce sites using their personal information, taking the +marketing experience to another level. +Working as a user +assistant +• +Help predict user behavior for improving UX (e.g., Siri, Alexa) +• +Integrate into end user-facing solutions for users to interact with directly (e.g., +chatbots, robots) +Working as a +capability + +• +Work as an app service tailored towards a particular action or use case (e.g., +search) +• +Provide ML-based speech-to-text services +• +Provide analytic capabilities (e.g., risk assessment, sentiment analysis, retroactive +analysis) +• +Provide text-related capabilities (e.g., natural language processing, text +recognition, speech-to-text conversion) +• +Provide visual capabilities (e.g., computer vision, augmented reality) + +Thus, it is apparent that AI is transforming how HCI and UX professionals work towards delivering optimal +UX in their solutions. The transformation impacts the HCI/UX activities such as user research, user interface (UI) +technologies and design, and user evaluation. +In this chapter, we review and discuss the transformation of AI technology in HCI/UX work and assess how +AI technology will change how we do the work. We first discuss how AI can be used to enhance the result of user +research and design evaluation. We then discuss how AI technology can be used to enhance HCI/UX design. +Finally, we discuss how AI-enabled capabilities can improve UX when users interact with computing systems, +applications, and services. + +4 + +5.2 AI in HCI/UX research and evaluation +5.2.1 Overview +The goal of HCI/UX research and evaluation is to systematically gather and analyze user data through +HCI/UX activities (e.g., user research, usability testing) to understand a problem space (e.g., user pain points, +usability issues) and guide the entire design process (Xu, 2005). Conventional approaches to user study rely on +methods such as surveys or user interviews; for UX evaluations, HCI/UX professionals manually conduct usability +testing of a proposed design to identify issues and then analyze data to generate recommendations for design +improvement (Xu, 2017). However, these methods are time-consuming. +A few years ago, researchers found that there is only little academic work at the intersection of UX and AI +(Chromik et al., 2020; Yang et al., 2018); they found even less research explicitly addressing AI/ML for user +research. However, the number of relevant publications has been increasing since 2015 and continues to do so as AI +is gaining popularity in many contexts. +The first AI-based approach uses big data-driven user research methods that gradually replace traditional +methods. With the development of the 5G, the Internet of Things, smart devices, etc., user data generation inevitably +grows explosively. Big data technology adoption in user research shows an upward trend. However, the way of +collecting user data is through some AI technology (e.g., sensing, face recognition) and user interactions (e.g., user +interaction log, online click streams, and social media); many of the big data-driven approaches are based on +traditional statistical techniques have not fully leveraged AI methods in their analysis for the collected user data +(Dan et al., 2020). +As an alternative approach, ML-based approaches are primarily used to support the analysis of already +collected user data (Chromik et al., 2020). For example, it analyzes textual user data. ML and natural language +processing (NLP) methods have been used to semi-automate the coding of interview transcripts (Marathe et al., +2018) and to extract UX-related problems from online review narratives through classification (Mendes et al., 2017). +Data-driven learning approaches have also been used to construct behavioral personas from user interaction log data. +The third approach combines the two methods discussed above, applying AI/ML technology to both data +collection and analysis stages. For example, Gartner (2018) provides recommendations for customer journey +analytics (CJA), leveraging existing analytics tools (when available) and incorporating ML as a service (i.e., +MLaaS) products to enhance analytical capabilities toward CJA. MLaaS incorporates ML-driven recommendations +to power customer journey orchestration solutions and enhances digital experiences, optimizing customer journeys +through Interaction Point Analysis across crucial interaction points. +Table 5.2 summarizes the main benefits of AI/ML across HCI/UX research and evaluation activities (e.g., +Chromik et al., 2020; Delrieu et al., 2018; Baker, 2019). + +Table 5.2 The benefits of AI/ML across HCI/UX research and evaluation activities +HCI / UX Activities +Benefits + +5 + +Data collection +• +Engaging surveys: simplify survey studies by leveraging the idea of +adaptive user interfaces (i.e., questionnaires might automatically be tailored +in real-time to the individual survey participant based on their previous +answers) +• +Remote tracking of user behavior over time +• +Applying conversational and voice user interfaces for more empathetic +survey studies +Evaluation of design +• +Data-Driven Design: Supporting design decisions by evaluating and +recommending UI options based on historical data of user behavior or user +preferences was considered another field of interest. +Analysis +• +Analysis of ordinal data (e.g., questionnaires) +• +Analysis of text (e.g., transcripts of user research/evaluation) +• +Analysis of voice/audio-based data (e.g., recordings, camera feed) +• +Analysis of log data (e.g., click behavior) +• +Analysis of time series data (e.g., mouse and eye movement) +• +Analysis of emotion and sentiment +• +Automated transcription: ML-based speech-to-text services for the post- +processing of contextual inquiries or interviews +• +Excel at quickly analyzing vast amounts of existing data to identify subtle +patterns in dispersed data silos and to inform UX insights +• +Detection of patterns within structured or unstructured data +• +User modeling +• +Augmented/predictive analytics: Insights automatically generated from ML +bring up new opportunities in conversion funnels +Generation of UX artifacts +• +User personas +• +Customer journeys: Leveraging ML functionality allows organizations to +contact or re-engage existing or potential customers at the optimal time in +the customer journey and through the optimal communication channel +• +Audience/customer segmentation: By quickly analyzing large datasets and +identifying patterns, these tools help analysts discover and validate new +customer cohorts or segments + +5.2.2 Digital personas +One practical approach to using the data from user research is to develop personas (Alan Cooper’s theory). +Persona refers to a group of users with similar behaviors and goals within the group and with differences in behavior +between the groups. Personas allow HCI/UX professionals to identify and understand the differences in how a +product is used by different groups of users based on their usages and behaviors so that the product can provide a + +6 + +tailored experience (e.g., functions, content) across personas. Generating personas can become challenging and time- +consuming in conventional user research, especially when HCI/UX researchers need to interview many users and plan +to build many personas. Also, a manual process of developing personas has been criticized for creating personas that +are not based on rigorous empirical data. The process often uses small samples, one-time data collection, and non- +algorithmic methods (Salminen et al., 2021). +The AI era generates a vast amount of user data, reflecting the user’s behavior, preferences, and demands, +and has high research value in building personas. Data-driven personas, called “digital personas,” have gained +popularity in HCI due to digital trends such as personified big data, online analytics, and the evolution of data +science algorithms. Specifically, three trends have significantly transformed the way how we build personas (Tan et +al., 2020): (1) availability of user data from online analytics and social media platforms; (2) democratization of data +science tools and algorithms that enable automated persona generation; and (3) Web technologies that remove the +limitations of static personas via interactive user interfaces. These three trends allow us to use algorithmic methods +to create accurate, representative, and refreshable personas from numerical data (Salminen et al., 2020, 2021). +The benefits of digital personas are apparent. AI/ML-based algorithms allow HCI/UX professionals to +develop personas much faster than conventional manual processes. An algorithm-based analysis of collected user data +also can help identify distinct personas and visualize how HCI/UX professionals reliably identified their attributes. +For example, Salminen et al. (2022) introduced Persona Analytics (PA), a system that tracks how users interact with +data-driven personas. PA captures users’ mouse and gazes behavior to measure users’ interaction with +algorithmically generated personas. The researchers also conducted a study with 144 participants, demonstrating +how PA could be deployed for remote user studies. +Salminen et al. (2021) summarize the distinct relative benefits of digital personas: +• +Enhanced objectivity. Digital personas tend to be replicable, and they use large sample sizes to +increase the user representativeness of the personas. The conventional manual process for developing +personas is associated with a high degree of subjectivity, which hinders the validity of the created +personas. The statistical robustness of digital personas boosts both the validity and credibility of the +developed personas. +• +Decreased cost. The manual process of creating personas is time-consuming and costly. Digital +personas mitigate this cost by relying on automation in persona creation, including data collection and +analysis, thus offering ways to “democratize” persona development for organizations of all kinds. +• +Updatability. Shifts in user demographics and behaviors are typical in many fast-moving industries, +such as Web-based businesses. Updating personas requires a high cost for the manual process, +resulting in outdated personas. Digital personas can capture the change in user behavior over time +based on their automated processes for systematic data collection and easy re-analysis using standard +algorithms. +• +Scalability. Manual data analysis is costly and requires specific expertise, making the personas built +using manual processes less compatible with large datasets. Large datasets are common with social + +7 + +media and Web analytics and are not a concern for digital personas, as data science and ML algorithms +have been developed to process large amounts of data + +More and more researchers are analyzing massive data and data characteristics to generate digital personas. +For example, Yu (2014) combined used tags to describe user characteristics. They extracted relevant information +tags through clustering in the big data environment to present the complete picture of users through digital personas. +Joni Salminen of Qatar Research Institute integrated data from Facebook, Twitter, YouTube, and generated real- +time personas based on user profiles and interactive behaviors, which provides users with competitive marketing +methods and strategies across different platforms (Salminen et al., 2017). From the perspective of cultural +differences, An et al. (2016) analyzed millions of content from the Middle East on YouTube to generate personas for +deeply analyzing cultural diversity’s impact on users’ social media use. +People have also attempted to build accurate personality traits and types for their target users as the +foundation for building personas (Salminen et al., 2020; Sun et al., 2018; Carducci et al., 2018). The automatic +assessment of personality dimensions relies on information gathered from social media platforms, such as a list of +friends and interests in music and movie endorsements. The work turned the collected data into signals as inputs. +Supervised ML approaches have been efficient and accurate in computing personality traits and types. Specifically, +Carducci et al. (2018) proposed a supervised ML approach to define personality traits by relying on what an +individual tweeted about publicly. The approach segments tweets in tokens and then learns word vector +representations as embeddings that are then used to feed a supervised learner classifier. This study demonstrates the +approach’s effectiveness by measuring the mean squared error of the learned model to compare it with an +international benchmark of Facebook status updates. Also, the study tested the transfer learning predictive power of +the proposed model with an in-house built benchmark created by twenty-four panelists who performed a state-of- +the-art psychological survey. The comparison shows that the proposed model received an excellent conversion while +analyzing the Twitter posts towards the personality traits extracted from the study. +Salminen et al. (2020) built social media-based personas with personality traits based on a deep ML +approach. They developed a deep learning classifier (a NN classifier) that predicts personality traits. The study used +three publicly available datasets and applied an automatic persona generation methodology to generate 15 personas +from the social media data of an online news platform. After developing the personas, the study aggregated each +persona’s YouTube comments and predicted the personality traits of each persona from the comments on that +persona. The results indicate an average performance increase of 4.84% in scores as compared with a baseline. + +However, there are challenges for digital personas (Salminen et al., 2021). The main challenges include (a) +data quality; (b) data availability; (c) method-specific weaknesses, such as the accuracy of the algorithm, people +behaving differently across different segmentations; (d) human and machine biases, such as the persistent need for +judgment calls (“manual labor”) that creates a potential source of bias and obstacles for completely automated +digital personas; (e) validation of digital personas methods; (f) lack of standardization; and (g) lack of consideration +for inclusivity. + + +8 + +5.2.3 Qualitative analysis +In user research and UX evaluation, HCI/UX professionals must conduct a qualitative analysis of the +collected data, such as interview transcripts in text or recorded video files (natural language), usability test video +recordings, and social media data. Qualitative analysis involves identifying themes, grouping data points based on +these themes, and establishing relations between these groups. This type of qualitative analysis is time-coming and +subjective to some extent. Many HCI/UX professionals have considerable training in analyzing human behavior +while users interact with computing systems. Still, AI applications for this kind of behavioral data have yet to +leverage their expertise fully. AI technology provides one approach to help the qualitative analysis for HCI/UX +professionals. Therefore, developing and leveraging AI-assisted capabilities for qualitative research is critical. +The application of AI in quantitative analysis has been increasingly popular over the last several years. For +example, researchers proposed using natural language processing (NLP) and ML to generate initial codes followed +by humans correcting the codes; other work utilized NLP to derive potential codes and models (Chen et al., 2018). +Although progress has been made in developing AI-assisted capabilities for qualitative analysis, there are still +challenges, and low accuracy has been considered the primary limitation of such automated approaches. Chen et al. +(2018) argued that we use AI/ML to support qualitative coding for identifying ambiguity. HCI/UX activities +generate petabytes of free-form text data, recording daily user experiences; NLI (natural language inference) allows +analysis of this large-scale data, but NLI used to be applied for analysis with organized datasets, and less is known +about how to design NLI for querying and analyzing text data (Mishra & Rzeszotarski, 2020). + +Specifically, Liu et al. (2020) deployed a semantic data analysis processing approach. The approach +introduced a specific implementation method using AI/ML semantic analysis technology to analyze language +materials in user research. The effectiveness of applying the AI semantic analysis techniques in user research was +tested and verified. The result shows that this application of AI technology has demonstrated the potential that AI +technology can replace some of the human manual analysis tasks for efficiency improvement. +User intent analysis is also essential in user research and UX evaluation. A method called Natural +Language Interaction (NLI) is to do user intent analysis (Setlur, 2020). For example, NLI was applied to a labeled +dataset that captures user intent distribution, co-occurrence, and flow patterns. Specifically, Setlur (2020) employed +deep ML techniques that approximate the heuristics and conversational cues for continuous learning in a chatbot +interface. These data-driven approaches help broaden the scope for visual analysis workflows across various chatbot +experiences. +Analyzing usability test videos is also challenging for HCI/UX professionals. Fan et al. (2022) have +explored how AI can help facilitate effective collaboration between UX evaluators and AI. Based on the previous +work in human and AI agent collaboration, they studied two primary factors: explanations and synchronization. +Explanations allow AI to inform UX professionals how it identifies UX issues from a usability test session; +synchronization refers to the two possible ways UX professionals and AI collaborate: synchronously and +asynchronously. By adopting a hybrid wizard-of-oz approach to simulating an AI solution with good performance, +they conducted a mixed-method study with 24 UX evaluators who were asked to identify UX issues from usability +test videos using the AI-assisted capability. The results show that AI with explanations, whether presented + +9 + +synchronously or asynchronously, provides better support for UX evaluators’ analysis; when without explanations, +synchronous AI better improved UX evaluators’ performance compared to asynchronous AI. This study also implies +that an AI-assisted UX evaluation can facilitate more effective human-AI collaboration. +Analyzing the structure of texts is an alternative way for qualitative analysis. Recently, personality +detection based on texts from online social networks has attracted more and more attention. Sun et al. (2018) +analyzed texts’ structure as an additional dimension for practical qualitative analysis. Previous models were based +on letters, words, or phrases, which is insufficient to get good results. Sun et al. (2018) present a preliminary +research result that shows the structure of texts can also be an essential feature in studying personality detection +from texts. More specifically, the study deployed a model called 2CLSTM. 2CLSTM is a bidirectional LSTM (Long +Short-Term Memory network) concatenated with CNN (Convolutional Neural Network), which can detect a user’s +personality based on text structures. They conducted evaluations across two datasets containing long and short texts. +The results have achieved better results, demonstrating the proposed model can efficiently learn useful text structure +features for qualitative analysis. +Chen et al. (2018) highlight two challenges for ML applications in qualitative coding. On the one hand, a +lack of understanding between disciplines may negatively impact trust, limiting the application of ML in qualitative +analysis because people using qualitative methods are generally not trained in ML techniques. In some cases, ML +experts’ limited understanding of social science values and methods can hamper effective collaborations. On the +other hand, there are fundamental differences between qualitative and quantitative methods. In quantitative analysis, +data points that appear very few times may be considered noise, but from a qualitative analysis perspective, the +quantity of instances does not always reflect significance. +As for future work supporting qualitative analysis in HCI/UX activities, we anticipate that more human- +centered and interpretable AI methods can potentially transform social science research (Chen et al., 2018). +Specifically, current AI models are not always interpretable, and the AI community needs to increase the +transparency and interpretability of AI technologies for qualitative analysis. Also, we need to explore ways to make +the usage of AI-based capabilities a meaningful task in HCI/UX practices, bridging the gap between AI and HCI/UX +communities. +5.2.4 UX evaluation +HCI/UX professionals conduct UX evaluations to achieve design improvements based on user feedback +and insights. However, traditional UX evaluation methods like questionnaires and usability testing are often +resource-intensive and not scalable. With the continuous evolution of many intelligent products (e.g., AI assistants, +autonomous driving, smart homes, intelligent robots), these UX evaluation methods will be difficult to meet new +scenarios (Lan et al., 2020). Emerging technologies such as AI and big data are currently influencing how HCI/UX +professionals conduct UX evaluations (Lan et al., 2020; Tan et al., 2020). +The methods of finding products with poor UX through big data-based analysis are gradually being adopted +by collecting the user’s stay time, login frequency, conversion rate, and other indicators (Tan et al., 2020). For +example, Yu (2018) designed a big data intelligent algorithm framework for UX evaluation of mobile media clients +through indicators such as the number of fans, page views, activity, stickiness, and emotional inclination, and finally + +10 + +implemented a cognitive algorithm framework. Li (2019) conducted an in-depth analysis of the user stickiness of +NetEase Cloud Music through big data analysis algorithms designed for popularity. Recent work has also attempted +to address other aspects of human behaviors on user interfaces, e.g., predicting human perception of UI interactivity +based on user behavior data such as mouse/keyboard logs, eye tracking, and usage log (Swearngin & Li, 2019). +Souza et al. (2022) establish a framework that employs eye and mouse tracking methods, keyboard input, self- +assessment questionnaires, and AI-based algorithms to evaluate UX and categorize users in terms of performance +profiles. +Furthermore, recent academic work explores the challenges of evaluating UX using multiple data sources +and proposes ML-based approaches (Asim et al., 2020). Connecting questionnaire results with log and time series +data about user behavior may be used as labeled data for input data to supervised ML. Such approaches may allow +for continuous monitor changes in users’ UX and inform HCI/UX professionals of the opportunity for improvement. +Chromik et al. (2020) propose that some equipment, such as electroencephalography (EEG) sensors, could be used +during real-time usability tests in lab contexts to record typical flows of interaction and users’ emotional responses. +These behavioral and emotional responses could be used as labels for an ML model (Chromik et al., 2020). +For usability testing, ML was used for selecting participants for usability tests (Gilbert et al., 2007) and +A/B tests (Kharitonov et al., 2017). In addition, automatic real-time evaluation of mobile-based experience via +emotional logging systems using video-captured facial expressions in lab contexts (Filho et al., 2015), using acoustic +data (Soleimani et al., 2017) and skin conductance signals (Liapis et al., 2015). +Studies show that AI technology may offer a more resource-effective approach (Chromik et al., 2020). For +example, Yang et al. (2020) proposed a methodology for measuring UX using AI-aided design (AIAD) technology +in mobile application design. AIAD focuses on the rational use of AI technology to measure and improve UX. The +researchers propose to obtain user behavior data from logs of mobile applications. They designed and used projected +pages of the application to train neural networks for specific tasks in terms of the click information of all users when +performing the tasks. The goal was to make the deep neural network model simulate the user’s experience in +operating a mobile application as much as possible. Thus, user behavior features could be aggregated and mapped in +the connection and hidden layers. Finally, the optimized design was executed on the application to verify the +efficiency of the proposed methodology. +We further provide two more examples illustrating how AI technologies can help facilitate UX evaluation +with three different approaches: visual search performance modeling, user emotion detection, and user interaction +behavior modeling. +The first example involves visual search performance modeling (Yuan et al., 2020). Modeling visual search +performance not only offers an opportunity to predict the usability of an application before actually testing it on real +users but also helps HCI/UX professionals better understand user behavior. The authors first analyzed a large-scale +dataset of visual search tasks on actual web pages. They then presented a deep neural network that learns to predict +the scannability of webpage content, i.e., how easy it is for a user to find a specific target. The model leveraged +heuristic-based features such as target size and unstructured features such as raw image pixels. The model then +analyzed the user behaviors to offer insights into how the salience map learned by the model aligns with human + +11 + +intuition and how the learned semantic representation of each target type relates to its visual search performance. +This approach allows HCI/UX professionals to model complex interactions involved in visual search tasks, which +traditional analytical methods cannot quickly achieve. +The second example is user emotion detection. Emotion is one aspect of UX that exploits an ML-based +automatic UX evaluation for understanding users’ emotions by analyzing the log data of the users’ interactions with +websites (Desolda et al., 2021). The evaluation results show the performance of each ML algorithm according to the +seven emotions. It is evident that emotions like sadness, anger, fear, disgust, and surprise were predicted with higher +accuracy; joy was instead predicted with medium accuracy, while contempt had lower accuracy in all cases. +Lastly, Bakaev et al. (2022) deployed neural networks-based approaches for predicting the visual +perception of UI. As testing and validation of graphical user interfaces (GUIs) increasingly rely on computer vision, +convolutional neural networks (CNN) models that predict UX start to achieve decent accuracy. However, CNN +models require vast amounts of human-labeled training data, which are costly or unavailable for HCI/UX activities. +This study compares the prediction quality of CNN and artificial neural networks (ANN) models to predict visual +perception in terms of aesthetics, complexity, and orderliness scales for about 2700 web UIs assessed by 137 users. +The results suggest that the ANN architecture produces a smaller mean squared error (MSE) for the training dataset +size (N) available in our study but that CNN should become superior with N > 2912. +While using AI technology in UX evaluation is promising, we still face challenges (Li et al., 2020). For +instance, deep ML methods are often data-hungry, while interaction data is relatively scarce compared to classic ML +problems such as computer vision or natural language processing. Deep ML models are not easy to analyze. While +better modeling accuracy is of great benefit, the interpretability of a model is crucial for HCI/UX professionals to +gain more insights about UX. Further collaborative work is needed between the AI and HCI/UX communities. +5.3 AI in HCI/UX design +5.3.1 AI for UI design +Designing an excellent GUI requires much innovation and creativity, but the process is time-consuming +and error-prone (Lu et al., 2022). Recently, AI-driven design (e.g., algorithmically powered tools) has become +popular. AI-driven design vows to move UX design to another degree of digital experience in support of +wireframing automation, visual design analysis, and UI pattern-driven design (Baker, 2019). Many researchers have +worked on building design support tools to improve the efficiency of UI work. Also, many commercial design +prototyping tools are developed. These tools have greatly helped designers create UI prototypes supporting UX +work. +Many initial AI-based capabilities were released to support UI design. For example, Adobe’s Creative +Cloud software can realize the intelligent analysis function of multimedia files such as images and videos. They can +provide intelligent material recommendations according to the designer’s design needs. Autodesk’s Dream Catcher +can quickly generate thousands of design proposals for designers to choose. Google released a new AI-based +Alphard (Alpha Graphic design) with the output of high-quality graphic design solutions. Microsoft and Airbnb +experimented with converting paper sketches directly into GUI code, bypassing much of the digital wireframing +phase (Wilkins, 2018). Some of the tools are even beyond GUI. They utilize pre-trained AI algorithms that have the + +12 + +potential to support new forms of interaction by processing eye, face, body, and hand movements captured through +webcams, speech commands captured through the browser’s audio channel, and text through web elements (Li et al., +2020). +There are many (or potential) benefits to leveraging AI technology in UX design. Figure 5.1 illustrates how +an AI-based approach could help remove some redundant work from human designers and developers from a design +process perspective. The research presents some inspiring illustrations of ML-based UX design tools. For example, +Gajjar et al. (2021) proposed Akin, a UI wireframe generator that uses a fine-tuned SAGAN model to generate UI +wireframes for smartphone UI design. The researchers annotated and classified 500 UI screens from RICO into five +commonly used mobile design patterns. The SAGAN model was trained with the dataset to generate UI wireframes +for a given UI design pattern. An evaluation of Akin conducted with 15 UX designers shows that the designers rated +the quality of wireframes developed by Akin as approximately equal to designer-made wireframes. Also, the +designers could not distinguish UI wireframes generated by Akin from designer-made wireframes 50% of the time. + + +Figure 5.2 Illustration of an AI-based approach for UI work (Babich, 2020) + +AI-based design capabilities also support design creativity, such as Simon’s optimization-based design +(Yang, 2017), which can all be linked to today’s interest in employing ML to assist human creativity. TensorFlow.js +is an open-source AI platform for developing, training, and using models in a browser or anywhere Javascript can +run (Li et al., 2020). At Google, TensorFlow.js has been leveraged as a platform for AI + HCI collaborative +research. TensorFlow.js is a browser-based ML framework to enable new forms of HCI/UX design innovation. +TensorFlow.js provides a rich set of features accessible to researchers with different levels of ML experience. The +library allows design experts to build models from scratch but also makes it easy to integrate pre-trained models. +The following list further summarizes some benefits (or potential benefits) of using AI-based capabilities +supporting UI design (Baker, 2019; Lu et al., 2022; Vetrov, 2022; Chen et al., 2018; Abbas et al., 2022): +• +Quickly make various design varieties per the user’s response. +• +Support design creativity +• +Make wireframing and prototyping work more efficiently and less monotonously +• +Transform UI sketches directly into a prototype +• +Potential to immediately change a whiteboard sketch over into a functional prototype + +Designer +Front-endDeveloper +品 +O +Sketchinterface +Layoutinterface +Designinterface +implementintertace +lmplementfeatures +Redundantwork +Redundantwork +Somestepsinthedesignworkflowareredundant,requiringtime +designerscouldusetofocusmoreoncreativetasks.Imagecredit +Tony Beltramelli.13 + +• +Quickly design alternative exploration +• +Support design customization to support personalized UX +• +Do design guideline violation check +• +Empower design decision-making +• +Help prepare UI assets and content +• +Translate digital UI mockups into UI specifications + +A representative approach of AI-based UI design is called Generative UI Design. Generative UI design is +based on AI generative technology. Generally speaking, with the development of AI generative technology, people +can realize collaborative creation with AI in music, painting, writing, design, dance, etc. (Li et al., 2020). AI +generative technology can quickly generate new samples that meet the specifications based on specific data sets so +that novices can quickly start creation or reduce the repetitive work of designers. For example, analyzing big data on +clothing design through modeling and visualization techniques to expand the ideas of clothing designers and gain +insights (Glauser et al., 2019). +With the millions of websites and mobile apps available, many UX problems an HCI/UX designer +encounters may have already been considered and solved by someone else. Specifically, in the UI design area, +generative models (e.g., Variational Autoencoders) were trained on a large set of UI design examples that can +suggest design alternatives for HCI/UX designers (Li et al., 2020). Systems based on these AI methods often +leverage human support or” Wizard of Oz” techniques to collect the data from large design samples and eventually +generate design solutions informed by the collected data (Vaccaro et al., 2018). +Researchers have promoted a hybrid intelligence approach for effective generative UI design. Specifically, +rather than using humans solely for data collection to train an AI system, the hybrid intelligence approach +incorporates human users, often crowd workers, as an essential and permanent component in an interactive system +for complex design tasks (Lasecki, 2019; Li et al., 2020). Such a hybrid intelligence approach provides rich +opportunities to combine human and machine intelligence to collaborate on a task and improve each other +dynamically and interactively. A system powered by the hybrid intelligence approach needs to synthesize responses +from multiple designers to achieve acceptable performance or availability for the system. +Another approach for effective generative UI design is to foster a creative, generative ML approach +(Kayacik et al., 2019). Research shows that such an approach is more robust when multiple designers with different +points of view actively contribute to them. Currently, many UXers do not have the ML education needed in the +industry. This lack of education is hampering ML research teams’ capacity to have a broad impact on their projects. +To address the issue, the Google People and AI Research (PAIR) group developed a novel program method in which +UXers are embedded into an ML research group for three months to provide a human-centered perspective on the +creation of ML models. The first full-time cohort of UXers was embedded in a team of ML research scientists +focused on deep generative models to assist in music composition (Kayacik et al., 2019). At the end of three months, +the UXers had new ML knowledge, and ML research scientists had a greater understanding of user-centered + +14 + +practices. The PAIR program results show that UX research and design involvement in creating ML models help +ML research scientists more effectively identify human needs that ML models will fulfill. +However, the generative UI design method is limited, so the design realization is still limited to the +traditional UI visualization level (Xu & Ge, 2018). It attaches great importance to the novelty of the appearance but +lacks attention to UX design. It is not mature enough to deliver optimal UX to HCI/UX designers, especially with +the business processes and structure built. Researchers also found that the innovative algorithm’s information +overload and uncertain output in human-AI collaboration are the key challenges (He et al., 2019). Also, AI +algorithms will produce inaccurate judgments but lack explanations, reducing the user’s perception of them (Glauser +et al., 2018). Morris et al. (2022) proposed two design spaces for consideration when developing future generative +AI models: how HCI can impact generative models (i.e., interfaces for models) and how generative models can +impact HCI (i.e., models as an HCI prototyping material). +5.3.2 AI as a new design material +As AI technology advances, HCI/UX professionals regularly integrate AI capabilities into new apps, +devices, and systems (Dove et al., 2017). AI becomes available as a resource to use by non-experts like HCI/UX +professionals. First and foremost, intelligence is becoming a new design material (Holmquist, 2017). The options of +a designer are, to a large extent, defined by the materials they have to work with. For instance, a product designer +would need to be aware of the physical characteristics of materials such as plastic, wood, and metal, as well as how +these fit together mechanically, to design an aesthetically and functionally pleasing experience. As AI becomes a +more vital part of everyday products, HCI/UX designers will have to figure out how to work with intelligence as a +new material. +With AI as a new design material, the primary role for HCI/UX designers is currently transitioning to +augment end users with extended capabilities (e.g., new ideas, emotion design), besides routine UI design work. +HCI/UX designers are becoming creators of the interface between humans and technology by leveraging algorithms. +AI as a design material means that HCI/UX designers should view AI as a capability as an application service +tailored to a specific functionality to support a particular experience (e.g., a “search” function). For instance, for an +application with conversational UI. The traditional approach is that a customer asks a question, and a human agent +responds for help. If we add AI to that conversation function by training models of the language, having those +models process that language and algorithmically build the best response and return that response with an AI-based +virtual support agent. Thus, this new material of invisible, personalized, conversational design is algorithms. +HCI/UX designers can take an active role in bridging algorithms and the UI to bring significant experience to end +users with technology. +AI as a new design material is essentially an algorithm as a new material. However, algorithms have many +limitations, impacting the outcome of using these “design materials.” Pavliscak (2016) lists several limitations of +algorithms: +• +Algorithms are not neutral or objective. Algorithms have a point of view with potentially biased outputs. +Humans create algorithms, so their point of view gets embedded in the system +• +Algorithms don’t understand you as a complex individual. Algorithms generalize, simplify, and filter out + +15 + +things they consider to be irrelevant. +• +In many cases, algorithms use other people’s data to fill in missing bits and pieces. The result is that +algorithms don’t reflect complicated humans. +• +Algorithms are opaque. It’s not always clear how or why they work the way they do. People who +write them don’t fully understand how they work. + +There has been continuing study on how to approach UX design practice while working with AI as a design +material (Dove et al., 2017; Yang et al., 2018; Amershi et al., 2019). Researchers argue that while data tell us about +people and organizations, algorithms create guidelines, and ML shapes the experience (Pavliscak, 2016). +Algorithms are considered a set of guidelines on how to perform a task. When you send a text message or do an +Internet search, HCI/UX triggers a nested set of interdependent algorithms. To best make use of algorithms for UX, +Holmquist (2017) proposed the following design guidelines when algorithms are used for design: +• +Reveal the effects of algorithms: users don’t understand how algorithms work, and experience designers +need to make algorithms’ results more apparent +• +Participatory design for algorithms: let users participate in their data creation for a personalized experience +and choose a level of trust using different personal preferences +• +Designing for transparency: let users understand how AI affects their interaction with applications +• +Designing for opacity: it is no longer possible to explain exactly why or how an AI does what it does +• +Designing for unpredictability: no matter how well-trained a neural network is, it is still drawing its +conclusions from given data. +• +Designing for learning: the learning must be built into the interaction and completely unobtrusive, so it +does not feel like the user doubles as the AI’s training wheels. +• +Designing for evolution: AI systems will continue to evolve. It will be necessary to communicate this to +users so that they know what to expect and can benefit while avoiding unpleasant surprises +• +Designing for shared control: how AI systems can be designed to allow the sharing of power with users + +While AI technology has brought in values for HCI/UX design, we still face challenges to fully leverage +this new type of design materials. Dove et al. (2017) conducted a survey that shows some significant challenges: (1) +There are challenges with using AI capabilities from a human-centered perspective. Current HCI/UX design +education cannot prepare future design graduates to incorporate AI into their work. (2) While ML pushes the +boundaries of design, the balance of collaboration with engineers and developers is currently such that design-led +innovation is still rare. (3) UX/UI prototyping with AI/ML is difficult. HCI/UX designers used to create prototypes +in the form of sketches, plans, and physical models made of paper, cardboard, or foam (Hallgrimsson, 2012). + +5.3.3 AI as a design collaborator +Design ideation is a source of innovation in the early stages of a development process. Beyond the AI +capability to support UI design as discussed above, we also need to rethink the current role of HCI/UX designers. UI + +16 + +design should be a creative process involving multiple iterations of different prototyping fidelities to create a UI +design. With the help of AI, we need to consider how to enable AI to perform repetitive tasks for the designer while +allowing the designer to take command of the creative process. This approach would greatly benefit designers in co- +creating design solutions with AI (Liao et al., 2020). Such a collaborative creation with AI may further promote +optimal experience in solutions (Oh et al., 2018). +Researchers have been promoting this approach. For instance, Liao et al. (2020) proposed a framework of AI- +augmented design support for the early stages where AI’s role in creativity is related to creating representation, +triggering empathy, and promoting engagement. Similarly, McCormack et al. (2020) characterized AI as a creative +agent system that provokes, challenges, and enhances human creativity. Verganti et al. (2020) further claimed that +AI reinforces design principles such as human-centered design, leading to potentially more creative solutions. AI +will enable the designer’s work, boost their creativity, and help experts create the best quality design products in a +minimum time (Inkbot Design, 2021). +Main & Grierson (2020) proposed that AI can perform as an assistant, collaborator, researcher, or facilitator +but might also play the role of future co-creator. Furthermore, McCormack et al. (2020) consider AI a system that +allows creative collaboration with designers. Li et al. (2020) argued that rather than using humans solely for data +collection to train an AI system, hybrid intelligence incorporates human users as an essential and permanent +component in an interactive system for complex design tasks. +De Peuter et al. (2021) challenged the current approach and argued that AI for supporting designers needs to +be rethought. It should aim to cooperate, not automate, by supporting and leveraging the creativity and problem- +solving of designers. How to infer designers’ goals and help develop a creative design needs to be figured out. They +believe there is an urgent need to develop AI methods to cooperate with designers, working as assistants +communicating with a designer about the design goal while supporting them in working toward that goal. In such a +collaborative process, HCI/UX designers should remain the primary actor in the design process. As active +participants, the designers explore and try things out to refine their goals. Further, the collaborative process can +leverage the designer’s creative abilities and expertise to build innovative designs. +Specifically, De Peuter et al. (2021) proposed a general-purpose approach for cooperative assistants in design +problems. Collaborative design assistance has been offered for specific design problems, but the proposed method is +to support a wide range of interactions. It uses a generative user model to infer a designer’s goal from their behavior +and plan how to assist the designer best. De Peuter et al. (2021) demonstrated the approach in a trip planning +example (see Figure 5.2). As illustrated in Figure 5.2, AI should appreciate the explorative character of designers’ +thinking. Within this design process, designers generate solutions not only to solve a problem but also to learn about +it, including its objectives and constraints. They can mentally plan over a design space based on a utility function +(shown as contours in Figure 5.2). The utility function evolves as the design progresses. The AI should collaborate +in this creative process, for example, by proposing high-quality solutions and complementing the designer’s +problem-solving. To do so, it needs to know the designer’s utility function. The study suggested creating AI +assistants (shown in blue) that can infer this utility from observations and then use it to assist a designer. + + +17 + + +Figure 5.2 An illustration of the designer-AI collaborative design activities +on a trip planning example (De Peuter et al., 2021) + +Also, Chen et al. (2019) proposed an integrated approach for enhancing design ideation by applying AI and +data mining techniques. This approach consists of two models, a semantic ideation network and a visual +combination model, which inspire semantically and visually based on computational creativity theory. The semantic +ideation network provokes new ideas by mining knowledge across multiple domains. A generative adversarial +networks model is proposed for generating UI objects for the visual combination model. An implementation of these +two models was developed and tested, indicating that the approach can create a variety of cross-domain concept +associations and advance the ideation process quickly and easily. +Liao et al. (2020) also proposed a framework of AI-augmented design support that involves the human +ideation components and design tools related to AI in the early design stages. The framework describes the explicit +roles of AI in design ideation as representation creation, empathy trigger, and engagement. The framework suggests +approaches to assist cognitive patterns in the design process. An empirical study was conducted to investigate the +cognitive patterns of design representations and design rationales. The study involved 30 designers with concurrent +think-aloud protocols and behavior analysis. The study identified the opportunities for AI to support human +creativity, and AI could provide inspiration, inform design scope, and request design actions. +5.3.4 Challenges and future work +Despite attempts to integrate HCI and AI, these HCI/UX designers experience challenges in incorporating +AI into common UX design paradigms (Policarpo et al., 2021). We summarize the overall challenges of applying AI +in HCI/UX design. +First, the AI-based approach challenges the typical activity of UX/UI prototyping. It is often difficult to +convince leadership to commit to more innovative designs. The AI-based method requires an unwieldy amount of +data to create a functional prototype. This approach could conflict with UX mantras like “fail fast, fail often” (Dove +et al. 2017). Consequently, it isn’t easy in research to experiment with many different design solutions in searching +for the best. In practice, designers could not demonstrate or validate their designs’ value through a working +prototype as they traditionally did. Recent research also founds a lack of research integrating UX and AI (Abbas et +al., 2022). One example of the obstacle is UX designers’ struggle when collaborating with data scientists. Another +obstacle is the lack of the tools and abilities needed to sketch or prototype when using AI as a design material. + +Designerdesigns +planning +planning +designspace +Alassists +AI +planning18 + +Second, many AI-based research projects’ impact remained within the academic research community and +haven’t succeeded in making practical influences on industry practices (Jiang et al., 2022). For instance, there is a +lack of research on ML algorithms and UX, especially in envisioning how ML might improve UX (Bertão & Joo, +2021). Bridging this gap requires research that identifies practitioners’ specific needs and provides translational +resources to benefit from the latest technological advances and academic research findings. +Third, Abbas et al. (2022) argue that I/ML has remained underutilized to assist designers and has yet to be +fully integrated into design patterns, education, and prototype tools. Therefore, tools are still in the early stages and +cannot cover all conceivable questions. Also, tools were not designed with the participants of UX designers (Abbas +et al. (2022). +Finally, the target users of many AI-based prototyping tools are mostly software developers rather than +HCI/UX designers (Sun et al., 2020). Many HCI/UX designers lack AI knowledge, so it is still challenging to use +these tools for prototyping. Further research is needed on prototyping tools for HCI/UX designers to help quickly +build prototypes, discover design problems early, and reduce product development risk. +As we look forward, with the emergence of new AI-based design paradigms, HCI/UX design activities +require the support of new tools. The development of these new tools should fully consider the knowledge +background and way of thinking of HCI/UX designers and the collaboration between these designers and +AI/software engineers in a real design environment (Sun et al., 2020). Undoubtedly, AI technology can’t replace +creative HCI/UX designers since these human professionals have unique capabilities to set the foundation for UX +design. Still, AI definitely can support these designers in UX design as a new design material and collaborator for +co-creative HCI/UX design. +5.4 AI for enhancing UX +5.4.1 Intelligent UI +AI technology is also transforming traditional UI into intelligent user interfaces (IUI). Traditional UI +techniques (e.g., mice, keyboards, and touch screens) require the user to provide inputs explicitly. AI-based +approaches are now robust to inherent ambiguity and noise in real-world data to analyze and reason about natural +human behavior (e.g., speech, motion, gaze patterns, or bio-physical responses). AI-based techniques can also learn +high-level concepts such as user preference, user intention, and usage context to adapt the UI and proactively present +information (Gebhardt et al., 2019). As a paradigm shift, AI technology holds great promise in shifting how we +interact with machines from an explicit input model to a more implicit interaction paradigm in which the machine +observes and interprets our actions (Li, Kumar, et al., 2020). +The idea of introducing intelligence to HCI and UI sprouted decades ago in the form of intelligent +computer-assisted instructions, which later gained a wider following and application as IUIs (Maybury, 1998). IUI +aims to improve the efficiency, effectiveness, and naturalness of human interaction with machines by representing, +reasoning and acting on models of the user, domain, task, discourse, and media. Different disciplines support the +field, including AI, software engineering, HCI, human factors engineering, psychology, etc. +Different from traditional UI, IUI should be able to adapt its behavior to other users, devices, and situations +(Gonçalves et al., 2019). A non-IUI considers an “average user” in design, i.e., the UI is not designed for all types of + +19 + +users but for an “average” of all potential users (Ehlert, 2003). Typically, we have one context of use in non-IUI; but +the context of use can change over time in an IUI. IUIs use adaptation techniques to be “intelligent/adaptive” with +the ability to adapt to the user, communicate with the user and solve problems for the user” (Ehlert, 2003). Its +difference from traditional UI is that they represent and reason concerning the user, task, domain, media, and +situation (Jaquero, 2009). +The goal of developing IUI is to make full use of advanced AI technology to provide natural and effective +human-machine dialogue. For example, new technologies (e.g., language recognition, facial recognition, gesture +input, gaze tracking) offer a natural UI for systems; multi-channel interaction through multiple modalities captures +user intent, behavior, and contextual scenarios to improve further the naturalness, accuracy, and effectiveness of +interaction. At the same time, the effective user-centered design method is used to optimize the design of IUI. +In addition, there are several reasons why we need to research IUI. First, IUI helps promote the +development of new AI technologies. Throughout modern technology development, GUI and mouse have promoted +the popularization of personal computer technology; multi-touch screen technology has improved mobile phone and +mobile user experience. Therefore, IUI research will find suitable application scenarios for developing AI +technology. +IUI also helps further exert human capabilities and enhance human intelligence. For example, brain- +computer interface research helps develop human potential and enhance the abilities of disabled people with +disabilities through rehabilitation therapy. IUI actively understand user status (e.g., physiology, psychology, +intention), so it will better understand users and predict their needs and behaviors, adaptively support users’ +activities, and ultimately make users more comfortable and interact with machines efficiently and securely. +Lastly, IUI aims to provide benefits to users such as adaptivity, context sensitivity, and task assistance +(Gonçalves et al., 2019). IUI research will provide more natural and efficient intelligent systems, bringing economic +benefits and considerable returns on investment to users, developers, and manufacturers. Effective IUI can improve +users’ work efficiency and make their work and life more convenient. The productivity of HCI can be significantly +enhanced not only by contact (mouse pointing device, joystick, touchpad, keyboard, etc.) methods but also by +contactless (speech and gesture commands, head and body movements, facial expressions, user’s look direction, +etc.) ones (Karpov & Yusupov, 2018). +Historically, interaction paradigms have guided UI development in HCI work, e.g., the WIMP (window, +icon, menu, pointing) paradigm. However, WIMP’s narrow sensing channels and unbalanced input/output +bandwidth restrict human-machine interaction (Fan et al., 2018). Table 5.3 summarizes the new HCI characteristics +that AI technology has brought in, emerging human factors issues, and critical issues for future HCI/UX work (Xu, +2019, 2020). + +Table 5.3 New characteristics and human factors issues of AI technology, critical issues for HCI/UX work +Transformative +Characteristics of +AI technology) +Emerging Human Factors Issues in AI +Technology +Critical Issues for HCI/UX Work + +20 + +From “one-way” to +“man-machine +collaboration-based +two-way” UI + +• +AI systems no longer passively accept +user input and produce expected output +according to fixed rules +• +AI-based agents can actively perceive to +capture and understand the user’s +physiological, cognitive, emotional, +intentional, and other states and actively +initiate interaction and push services to +users +• Human-machine teaming/collaboration- +based interaction models and paradigms +• Cognitive models of the user’s states +(e.g., situation awareness, physiology, +cognition, emotion, and intention) +From “usability” to +“explainable AI” +UI +• +AI “black box” effects can lead to +inexplicable and incomprehensible +system outputs +• +AI “black box” effect raises AI trust +issues +• +Innovative UI technologies (such as +visualization) and design +• +“Human-centered” explainable and +understandable AI (Ehsan et al., 2021) +• +Accelerated transformation of +psychological explanation theories +From “simple +attributes” to +“contextualized” +UI +• +AI system input includes “contextualized” +data (e.g., the context of usage, user +behavior), besides traditional information +(e.g., simple objects such as target +location and colors) +• +Modeling and intelligent deduction of +“situational” features (e.g., user +characteristics, digital personas) based +on data such as operating context and +user behavior +• +Personalized functionality suitable for +user needs and usage scenarios +From “precise +input” to “fuzzy +reasoning,” UI +• User input is not just a single precise form +(e.g., keyboard, mouse), but may also be +multimodal, ambiguous interactions (e.g., +user intent) +• Ambiguous interaction issues in operating +scenarios (e.g., random interaction signals +and ambient noise) +• Methods and models for inferring user +interaction intentions under uncertainty +• Naturalness and effectiveness of HCI in +an ambiguous state +From “interactive” +to “collaborative” +UI + +• UI supporting both human-machine +interaction and human-machine teamwork +• UI supporting effective human-machine +collaboration +• +Alternative design paradigms and +models for human-machine collaborative +UI +• +UI design standards for intelligent HCI +• +Interaction design effectively supports +human-machine collaboration (e.g., +human-machine control hand-over in an +emergency) + + +As Table 5.3 lists, these transformative characteristics of AI technology lead to the need for innovative UI +capabilities and interaction paradigms (e.g., two-way, collaborative UI). This will ultimately prompt the +development of more natural and effective IUI and will require HCI/UX professionals to develop more effective +approaches to explore the design of innovative UI design. +From a methodology perspective, research shows that there is a lack of effective methods for designing +IUI, and HCI/UX professionals have had difficulty performing the typical HCI activities of conceptualization, rapid +prototyping, and testing (Yang et al., 2020; Holmquist, 2017; Dove et al., 2017). The HCI/UX community has + +21 + +realized the need to enhance existing methods (Stephanidis, Salvendy, et al., 2019; Xu, 2018; Xu & Ge, 2020). To +this end, Xu, Dainoff, et al. (2021) assessed existing methods of HCI, human factors, and other related fields. As a +result, they proposed alternative approaches that can support the effective design of IUI better. These alternative +methods can help HCI/UX professionals overcome the limitations of conventional HCI methods when designing +IUI. +From a process perspective, research shows HCI/UX professionals have challenges integrating HCI/UX +processes into the process of developing IUI systems. For instance, many HCI/UX professionals joined AI projects +only after the requirements were defined (Yang, Steinfeld et al., 2020). Consequently, the design recommendations +from HCI/UX professionals could be quickly declined (Yang, 2018). AI professionals often claim that many +problems that HCI could not solve in the past have been solved through IUI technology (e.g., voice UI), and they +can design the interaction by themselves. Still, studies have shown that the outcomes may not be acceptable from a +UX perspective (e.g., Budiu & Laubheimer, 2018). Some HCI/UX professionals find collaborating effectively with +AI professionals challenging due to a lack of a shared process and a common language (Girardin & Lathia, 2017). +Also, studies have shown that HCI/UX professionals are not prepared to provide effective design support for AI +systems (Yang, 2018). +For future work, we offer several strategic recommendations. Firstly, HCI/UX professionals need to +integrate HCI/UX methods into the development process of IUI to maximize interdisciplinary collaboration. For +instance, to understand the similarities and differences in practices between HCI/UX professionals and other +professionals, Girardin & Lathia (2017) summarize a series of touch points and principles. Within the HCI +community, researchers have indicated how the HCI/UX process should be integrated into the process of developing +IUI systems (Lau et al., 2018). Specifically, Cerejo (2021) proposed a “pair design” process that puts two people +(one HCI/UX professional and one AI professional) working together as a pair across the development stages of IUI +systems. + +Secondly, HCI/UX professionals must update their skillsets and knowledge in AI. While AI professionals +should understand HCI/UX approaches, HCI/UX professionals also need to have a basic understanding of AI +technology and apply the knowledge to facilitate the process integration and collaboration so that HCI professionals +can fully understand the design implications posed by the unique characteristics of AI technology and be able to +overcome weaknesses in the ability to influence IUI systems as reported (Yang, 2018). +Thirdly, future work needs to adapt AI technology to human capability. Human-limited cognitive resources +become a bottleneck of HCI design in the pervasive computing environment. For instance, in an implicit interaction +scenario initiated by intelligent ambient systems, intelligent systems may cause competition between human +cognitive resources in different modalities, and users will face a high cognitive workload. Thus, HCI design must +consider the “bandwidth” of human cognitive processing and resource allocation while developing innovative +approaches to reduce user cognitive workload through appropriate interaction technology, adapting AI technology to +human capabilities. +Fourthly, we need to develop new interaction paradigms that better fit IUI. IUI requires effective UI +paradigms. In the realization of IUI, hardware technology is no longer an obstacle, but the user’s interaction ability + +22 + +has not improved. Designing effective multimodal integration of sight, hearing, touch, gestures, and other parallel +interaction paradigms is an essential part of HCI research in the age of intelligence. Historically, interface paradigms +and models have guided the development of human-computer interaction (e.g., WIMP). However, the limited +perception channels and unbalanced input/output bandwidth of WIMP restrict the further evolution of the UI in the +AI age. Existing studies have proposed the concepts of Post-WIMP and Non-WIMP, but the effectiveness remains +to be further verified. HCI/UX community should support defining paradigms, metaphors, and empirical validation +to solve unique problems in IUI. It requires HCI/UX professionals to explore innovative ideas that can effectively +facilitate interaction in IUI. +Finally, we need to develop HCI design standards that specifically support the development of IUI. +Existing HCI design standards are primarily grown for non-IUI, and there is a lack of design standards and +guidelines explicitly supporting IUI design. IUI design standards need to consider the unique characteristics of AI +technology fully. There are initial design guidelines available, such as the “Google AI + People Guidebook” +(Google PAIR, 2019) and Microsoft’s 18 Design Guidelines (Amershi et al., 2019). The HCI/UX community must +play a key role in developing these design standards. +5.4.2 AI assistants +An intelligent assistant (IA) is an AI/ML-based computer system capable of intelligently assisting people. +IAs have gained in popularity over recent years; it ranges from helping people develop skills and exercise properly +to rehabilitate physically (Islas-Cota et al., 2022], among other application domains. IAs are being deployed across +domains, such as health, education, online social services, driving, domestic environment, enterprise/industry, +fitness, and learning. To perform their users’ daily tasks or services, IAs can send a message, make a phone call, +search for specific information, set a reminder or calendar, and provide personalized recommendations. IAs are +intelligent agents that employ AI techniques to provide a human-like interface (e.g., voice, vision) (Hu et al., 2019). +They are also expected to perform more complex tasks, such as making purchases and accessing or managing smart +IoT devices (Han & Yang, 2018). Natural language processing and AI technologies enable IAs to self-learn users’ +schedule and taste through daily interactions and collecting awareness data (e.g., location and context) from the +Internet of Things (IoT), and then autonomously perform tasks based on user preferences and habits (Santos et al., +2016). +The objectives of IAs are to increase efficiency in an activity, better cope with an illness, resolve a +problem, support everyday situations, refine skills, and attain a healthy life. Ultimately, IAs aim to enhance UX in +their daily work and life. One good example of IA is voice assistants, which have been rising recently, such as +Amazon Alexa, Google Assistant, and Siri from Apple (Zwakman et al., 2021). Voice assistants help facilitate +human–computer dialogue naturally and intuitively, like conversations between humans. +Islas-Cota et al. (2022) presented a systematic review aiming to classify recent advances in IAs in terms of +IAs’ objectives, application domains, and workings. They identified what AI/ML techniques are used to enable the +AI assistants. As a result, the study proposes a taxonomy of IAs, as illustrated in Figure 5.3. + +23 + + + +Figure 5.3 Taxonomy of IAs (Islas-Cota et al., 2022) +Research into AI-based digital assistants has a long history, dating back to Joseph Weizenbaum’s well- +known ELIZA in 1966 (Maedche et al., 2019). In parallel, global technology companies such as Microsoft, IBM, +Google, and Amazon have been working with AI-based digital assistants to provide significant opportunities. The +rise of IAs has opened a broad research area for HCI/UX professionals. It is a technology with an explicit interface +to users and could therefore provide a fruitful avenue for HCI/UX research (Enhancing UX/AI assistant 5). Much +research has been done, but the work primarily focused on improving the technology. Their indirect objective is to +enhance the usability of the IAs, and not on the usability aspect per se. Budiu & Laubheimer’s (2018) usability study +found that both voice-only and screen-based intelligent assistants worked well for only minimal, simple queries with +relatively simple, short answers. Users had difficulty with anything else. In addition, as part of experience issues, +the privacy and security aspect of the IAs (e.g., voice assistants) still exists, as many IAs are prone to various attacks +that might steal user information (Zwakm et al., 2021). There are several ways in which IAs could be used that can +create new ethical and legal issues (Almeida et al., 2020). +Besides the usability design and interaction issues of IAs, the collaborative relationship between humans +and AI-based IA systems is another important topic for HCI/UX professionals. Traditionally, AI-based applications +are considered a tool in support of humans. We should stop thinking of AI as a developing phenomenon independent +of humans, and it is necessary to move on to the consideration of hybrid intelligence. Hybrid intelligence can be +further understood from three aspects, considering hybrid intelligence as the sum of human and machine efforts in +achieving a goal; the amplifier of human intelligence at the physiological level; and a partnership between humans +and machines (Shichkina et al., 2017). + +IntelligentAssistant +Objective +Targetuser +Enabler +Software Interface +Increaseefficiencyinanactivity +.Workers +:Artificialintelligence +.Mobileapp +Bettercopewithanillness +.Patients +.Machine Learning +Desktopapp +.Resolveaproblem +.Students +.Hybrid +Webapp +:Supporteverydaysituations +.Drivers +Embeddedsystem +.Refine skills +.Healthprofessionals +.Attaina healthy life +Generalusers +Device +Learn +·Computerexperts +Internet/Online servicesusers +. +Mobiledevice +Input +.Leamer/Trainee users +.Fitnessusers +. +Robot +Seniorcitizens +PersonalComputer +. +.Text +Wearable +Smartspeaker +:Audio +. +Gesture +:Posture +Triggering stimulus +:Image +Video +Domain +Sensorsignal +.Userrequest +:Userstate +Health +Capability +.Arequest +:Education +·Environmentstate +andn +.Social +Onlineservices andloT +:Recognition/Detection +.Driving +Tracking/Monitoring +Triggering actor +.Text +·Domesticenvironment +.Notification/Recommendation +Audio +Computerscience +:Natural LanguageProcessing +.Video +.Art +.Dataprocessing +.User +Images +·Enterprise/lndustry +Learning +.IA +.Haptic +.Fitness +Personalization +User/A +.Deviceactivation24 + +Shichkina et al. (2017) further argued that IAs should not be just the creation of hybrid intelligence but a +co-evolutionary hybrid intelligence (CHI). CHI is a symbiosis of artificial and natural intelligence, mutually +developing, teaching, and complementing each other in co-evolution. Human-machine intelligence co-evolution is +the fundamental building of more robust intelligent systems (Krinkin et al., 2021). Based on the concept of CHI, the +goal of IAs is the mutual development of human and artificial intelligence as a single indivisible organism. +From the perspective of human-machine intelligence complementarity, the most significant potential for +IAs is a mutually beneficial collaboration (Maedche et al., 2019). Both humans and machines have relative +strengths. While machines are ideal for conducting repeatable, highly structured tasks, collecting, storing, processing +vast amounts of data, and predicting the future in stable environments, humans can handle abstract problems and +deal with fragmented information much more efficiently. +Functional and task allocation between humans and machines are a classical activity for HCI/UX +professionals. There is an intensive discourse on how humans interact with AI-based technologies and how the +performance of a particular task should be divided between these two entities. It is crucial to involve humans to an +appropriate level in task performance, depending on the task characteristics and the context (Maedche et al., 2019). +A significant challenge for future research is to investigate how to distribute the tasks between these two entities at +an appropriate level to achieve optimal performance. +Autonomy is another topic that little research has been done to examine the issue from the perspective of +autonomy as intelligent agents (Hu et al., 2019). In the past, many topics about autonomy focused on human +autonomy, such as job autonomy, human autonomy, and community autonomy. IA autonomy refers to the fact that +IA, as an intelligent agent, can independently complete tasks in some scenarios. Autonomy is a double-edged sword +factor for IAs, as it can increase benefits (e.g., exerting specific risky tasks) (Robert & You, 2018). It may allow a +machine to over-control without human authority, which may put humans in a risk situation in some domains (Xu, +2019, 2020). +The over-emphasis on human autonomy is because machines were not smart enough in the past. They can +be regarded as relatively automated rather than autonomous. Recently, AI-enabled IAs to self-learn users’ +preferences through daily interactions and personal data, which ensures intelligent agents’ autonomy (Hu et al., +2019). For instance, IAs will set the alarm clock to wake up according to the user’s habits, reflecting the IA +scheduling autonomy. Still, the complexity of IA executing both instructions is hidden, which will result in the user +losing control over the specific execution process. Hu et al. (2019) raised several research questions for future work +of AIs, such as whether decision-making autonomy will have a positive influence on perceived competence, whether +perceived competence will have a positive effect on the intention of a user to IA continuous usage, and whether +perceived uncertainty will harm the purpose to IA continuous usage. +We summarize the future research directions for improving the UX of IAs (Islas-Cota et al., 2022; +Maedche et al., 2019; Zwakman et al., 2021). +• +Understanding human users’ needs of IA. Ensure that the design of IAs is in line with the user +population’s and society’s goals and values. Research is needed to create a rich understanding of the +needs and usage of the potential users of such assistants + +25 + +• +Interaction technology: Need to improve the technology empowering these IAs to provide better +capabilities, such as voice recognition, the ability to understand multiple languages, providing human- +like speech output, adding emotions to these devices, and likewise +• +User privacy: Ensure that the users can trust them in their daily usage because many IAs collect +potentially sensitive data from users’ activities, such as visited locations +• +Collaboration: Need to explore further how to design effective collaboration between humans and IAs. +Technically speaking, the agent-based paradigm of IAs supports collaborative problem-solving either +with other agents or with agent-human teams. Future work needs to exploit further the agent paradigm +where multiple agents (namely, IAs) can coordinate, collaborate, and negotiate among themselves to +provide users with a multi-domain ubiquitous assistance +• +Evaluation: Need to evaluate the overall system performance from a collaboration perspective. A +common practice is comparing with utilized benchmark datasets to evaluate their IAs. We need to find +practical evaluation approaches to assess the efficiency and effectiveness of IAs in a collaborative way +• +Functional and task allocations: Need to investigate from a conceptual perspective the interplays +between humans and machines when using IA, investigating design variants of IA for different task +types. Collaboration between humans and IA may depend on the task type to achieve optimal +experience. +• +Context-aware assistance: Exploit unsupervised ML techniques to discover users’ context and +behavioral patterns. IAs can provide users with personalized and contextual assistance by establishing +a context. Features such as users’ activities, interactions, status, and intent detection, can establish a +context that enables IAs to determine how and when assistance should be provided. +• +Emotionally aware assistance: Need to explore more elaborate emotional models. Emotions are +critical to humans in decision-making and communication, among other everyday activities. +• +Virtual, augmented, and mixed reality: Need to leverage these technologies to help improve user +experience. Currently, there is a lack of IAs taking advantage of virtual, augmented, and mixed reality. +For instance, IA can use virtual reality devices to assist patients with physical rehabilitation and train +surgeons. Further HCI work is needed to enhance user experience while interacting with IA. +• +Human characteristics: Need to assess human characteristics in the design of IA, such as user +expertise with the technology, personality, culture, social norm, delivering personalized assistance + +5.4.3 Recommender systems +Recommender Systems (RS) are software tools that support human decision-making, especially when +choices are made over large product or service catalogs (Ricci et al., 2021). Recommender systems are integral to +many of today’s websites and online services. After 30 years, personalized recommendations are ubiquitous, fueled +by advances in AI technology. To a large extent, making recommendations is an HCI/UX topic, which aims to +determine how a computerized system can effectively support users in information search or decision-making +contexts for an optimal experience. + +26 + +The academic and industrial communities have proposed many recommender software and algorithms +(Elahi et al., 2021). Most of these algorithms can gather various data types and exploit them to generate +recommendations. These data types can describe either the item content (e.g., category, brand, and tags) or the user +preferences (e.g., ratings, likes, and clicks). A recommendation list for a specific user is then made by filtering the +items representing similar features to the rest of the items that the user liked/rated high. However, users may be +exposed to risks, such as bad user experience and decision difficulty. If the set of recommendations is unfortunate +(e.g., poor decisions, the pre-selection of items, or the decision bias ), this might lead to a poor experience. +The goal of a recommender system is to predict user interests and infer their mental processes. For +example, personalized recommender systems are one of the most widely used fields of big data technology. It is +implemented by mining user attributes through user behavior data and realized through the inference of basic +information (e.g., user’s age, gender, residence, and educational background based on the user’s online browsing +behavior) (Wang et al., 2013). Based on the results of the inferred data (e.g., digital personas, user profile, behavior, +preference), systems can intelligently send recommendations to target users with a personalized experience. As a +success case, Amazon’s recommendation engine provides it with a conversion rate of up to 60% and a sales +contribution rate of 30% (Li et al., 2015). +In general, there are two traditional recommender systems (Elahi et al., 2021) +• +Collaborative filtering: The method predicts users’ preferences (i.e., ratings) by learning the +preferences that a group of users provided and suggests to users the items with the highest predicted +priorities. It is used in almost all application domains and relies on big data of ratings acquired from a +typically extensive network of users (Desrosiers & Karypis, 2011). The underlying assumption is that +users with similar preferences will also have similar preferences in the future. +• +Content-based: Content-based methods adopt content-based filtering (CBF) algorithms to build user +profiles by associating user preferences with the item content (Deldjoo & Atani, 2016). Content-based +approaches recommend items that share characteristics with items that the user has previously liked +(e.g., items with a similar description or genre) (Calero Valdez et al., 2016; Zangerle et al., 2022). +Typical fields of application are recommending movies, music, or related products in e-commerce. +Despite these traditional methods’ effectiveness, as we enter the age of big data and AI, more advanced +techniques have been developed to build intelligent systems for quicker, more accurate, and personalized +recommendations tailored to each user’s needs and preferences (Elahi et al., 2021). +There are several types of AI-enabled recommender systems that have been explored in academia and the +industry: +• +Data-driven recommendations: This method enables leveraging ML technologies to contextualize the +big data to enhance the precision of suggestions, which facilitates the use of content (Beheshti et al., +2020). The approach moves from traditional statistical modeling to advanced AI-based models, which +will improve mining patterns between items and user descriptors to build better suggestions. + +27 + +• +Knowledge-driven recommendations: This method empowers simulating the expertise of the domain +experts (e.g., crowdsourcing methods) and adopting techniques such as reinforcement ML to enhance +the system’s capability for making relevant and accurate recommendations (Beheshti et al., 2018). +• +Conversational recommendations: This method provides more sophisticated interaction paradigms for +preference elicitation, item presentation, or user feedback through conversational interactions between +users and recommender systems (Lei et al., 2020). +• +Intelligent ranking-based recommendations: It can be trained by the domain experts’ knowledge and +experience to understand the context, extract related features, and determine the causal connections +among various features over time. The goal is to change from statistical modeling to novel forms of +modeling, such as deep learning, to improve potential similarities among descriptors and build a more +accurate ranking (Chen et al., 2020). +• +Intelligent personalization-based recommendations: It can support analytics around users’ cognitive +activities to provide intelligent and time-aware recommendations. The method tailors product and +content recommendations to users’ profiles and habits by analyzing users’ behavior, preferences, and +history. This process requires automatic data processing to identify meaningful features, select suitable +algorithms, and use them for training a proper personalization model (Herath & Jayarathne, 2018). +• +Cognition-aware recommendations: It aims to recognize the users’ personalities and emotions and +analyze their characteristics and affinities over time. The system needs to interpret social information +(at a group level or on a one-to-one basis) and provide context-aware recommendations (Beheshti, +Yakhchi et al., 2020). +Many AI models have been adapted for use in these AI-enabled recommender systems. For instance, deep +neural networks for collaborative filtering to model the user-item interactions, including deep factorization machines +or (variational) autoencoders (Zangerle & Bauer, 2022). Convolutional Neural Networks (CNN) are primarily used +for learning features from (multimedia) sources for learning the data from audio signals (Van den Oord et al., 2013) +or modeling latent features from user reviews and items (Zheng et al., 2017). Recurrent Neural Networks (RNN) are +used to model sequences for sequential recommendations (Quadrant et al., 2017). Reinforcement learning models +incorporate user contexts while continuously updating and optimizing the recommendation model based on user +feedback (Zheng et al., 2018). +Early research on recommender systems focuses on algorithms and their evaluation to improve +recommendation accuracy (Calero Valdez et al., 2016). After a few decades, the field of recommender systems has +been driving toward consensus; that is, accuracy only partially constitutes the UX of a recommender system. As a +result, there is an evolution from research on algorithms to research on UX with recommender systems (Konstan & +Terveen, 2021). +Human-centered recommender systems are an approach that focuses on understanding the characteristics of +recommender systems and users as well as the relationships between them. The goal is to design recommender +systems' algorithms and interactions to fulfill better users' goals (Konstan & Terveen, 2021). Different from +traditional technology-centered recommender systems, a different set of questions need to be answered from the + +28 + +human-centered recommender systems. For instance, what does it mean for a recommendation to be good? How +many products are too many to recommend? Should I show the best recommendations or save some for later? When +should the recommendations be diverse concerning each other or the user’s history? What type of recommendations +leads to better UX? Konstan & Terveen (2021) presented HCI research work focusing on UX and interactive +visualization techniques to support the transparency of results. In addition, there is also a need for frameworks to +combine human-centered recommender systems research with the best ML algorithms to achieve scalable, efficient +human-centered recommender systems (Konstan & Terveen, 2021). +Furthermore, to enhance user experience, Ekstrand et al. (2014) conducted research to understand how +users perceive their performance, including dimensions of accuracy, diversity, novelty, personalization, and +satisfaction. The work shows that these factors should be included in an analysis of algorithm performance while +building a structural equation model. Willemsen et al. (2016) studied user choice overload in the context of +recommender systems. The results show that the diversity of items recommended affected the effort required to +make a choice; diversity led to higher satisfaction choices but not always the highest-scoring choices for users. +Research also shows that a recommender system built to optimize user engagement (rather than predictive accuracy) +leads to recommendations that increase subsequent user engagement compared to predictive accuracy +recommenders (Zhao et al., 2018). +How to effectively measure the UX of recommender systems is essential, so HCI/UX professionals can +identify the pain point to close the gap in design. The performance of recommender systems is typically evaluated +using offline and online experiments (Zangerle & Bauer, 2022). When assessing the effectiveness of a recommender +system, people largely adopt offline rather than live user studies methods. Conversely, real users are requested to +evaluate the recommendations in online studies. Offline studies are more popular than user studies, which are more +complex and time-consuming. However, measuring the UX of recommender systems is often challenging. Konstan +& Riedl (2012) argue that evaluating the UX requires a broader set of measures. Regular algorithmic work can be +done by using existing datasets; measuring UX requires developing additional capabilities that include both +algorithms and UI. +More specifically, Knijnenbur et al. (2012) propose a user-centric approach for evaluating recommender +systems. The framework links objective system aspects to accurate user behavior through a series of perceptual and +evaluative constructs. It also incorporates the influence of personal and situational characteristics on the UX. The +framework was validated using a method called structural equation modeling. The results show that subjective +system aspects and experience variables are invaluable in explaining why and how the UX of recommender systems +comes about; the perceptions of recommendation quality and variety are essential mediators in predicting the effects +of objective system aspects on the three components of UX: process (e.g., perceived effort, difficulty), system (e.g., +perceived system effectiveness) and outcome (e.g., choice satisfaction). Also, the study finds that these subjective +aspects strongly correlate to user behaviors (e.g., reduced browsing for higher system effectiveness). +Based on current literature, the following list summarizes the suggestions for future HCI/UX work in +designing recommender systems (Calero Valdez et al., 2016; Konstan & Riedl, 2012; Konstan & Terveen, 2021; +Jannach et al., 2021): + +29 + +• +Better understand how users make decisions: For example, we need to know how recommender +systems can adapt to different needs (e.g., new users vs. experienced users) and how they can balance +short-term with longer-term value. +• +Putting the user in control: Users are often more satisfied when given control over how the +recommender system functions. We need to design for the sweet spot so that the recommender system +can balance serving users effectively while the users have the desired control. +• +Developing adaptive recommender systems: Previous research shows that user satisfaction does not +always correlate with high recommendation accuracy and the user’s knowledge level and interests. +There is a need to adapt recommender systems and their user interfaces to these other personal and +situational characteristics. +• +Supporting affective design: Emotions play a crucial role in human decision-making. Future work +needs to explore novel sensing technologies for capturing user behavioral data (e.g., physiological +data, facial expressions, speech) so that recommender systems can detect emotions and adapt +recommendations based on emotional responses. +• +Conducting ongoing research: For instance, the research on real applications that allow incorporates +diverse contexts, including multi-interaction modalities (e.g., voice/audio vs. text vs. visual interaction) +and decision nature (e.g., health/habit, low- vs. high-stakes) +• +Developing rigorous methods for evaluating UX: We need to continue to adopt rigorous methods for +assessing the UX and user satisfaction, such as the structural equation modeling +• +Designing for high-risk domains: Spending money on an undesired product is the most significant risk +for a user of e-commerce sites. Risk-aware algorithms or predictions of risks need to be further +investigated. For instance, how to effectively visualize or communicate the uncertainty and risk of a +recommendation to users, which is crucial for the systems in high-risk domains (e.g., medicine). +• +Developing insightful “beyond-accuracy” measures: Many current methods rely on data-centric +“offline” experiments that do not involve the human in the loop. We should focus much more on how +systems affect both organizations and entire experience journeys, the diversity of the +recommendations, or the novelty of the identified items. +• +Developing integrated solutions for better UX: A fundamental challenge to the field of recommender +systems is the integration of content-based approaches (e.g., product information, user profiles), +collaborative approaches (e.g., explicit and implicit ratings, tagging, and user preference), and +contextual approaches (e.g., business rules, location, user task and mood, UI ) into comprehensive +recommender systems. + +5.5 Conclusions +AI technology is transforming how HCI and UX professionals work towards delivering optimal UX in their +solutions, including all aspects of user research, UI technologies and design, and UX evaluation. AI-based solutions +have raised user and academic awareness of technical innovation. As a result, AI is becoming increasingly popular + +30 + +in improving the quality of UX. This chapter summarizes how AI technology can help HCI/UX professionals in +HCI/UX research and evaluation, HCI/UX design, and enhancing UX by leveraging AI-based capabilities. It also +highlights the benefits of deploying AI technology for HCI/UX activities, the challenges that HCI/UX professionals +face, and future HCI/UX work. +To push the boundaries of what AI might be and might do, we need to continue to identify major unknown +topics as a basis for future research endeavors (Yang, 2018). We must bridge the gap between HCI/UX +professionals’ work practices and AI-enabled capabilities. Professionals across disciplines need to seize this +opportunity to create something entirely new. It’s essential to treat the adoption of AI as a significant strategic and +cultural shift for improving UX, not simply the installment of new technology (Schwartz et al., 2018). We need to +collaborate on developing new methods, tools, and processes to help HCI/UX and AI professionals better innovate +with AI (Bertão & Joo, 2021). +References +Amershi, S., Weld, D., Vorvoreanu, M., Fourney, A., Nushi, B., Collisson, P., ... & Horvitz, E. (2019, May). Guidelines for +human-AI interaction. In Proceedings of the 2019 chi conference on human factors in computing systems (pp. 1-13). +Almeida, V., Furtado, E. S., & Furtado, V. (2020). Personal digital assistants: the need for governance. IEEE Internet +Computing, 24(6), 59-64. +Asim, M.N., Ghani, M.U., Ibrahim, M.A., Mahmood,W., Dengel, A., Ahmed, S.: Benchmarking performance of machine and +deep learning-based methodologies for Urdu text document classification. Neural Comput. Appl. 33(11), 5437–5469 (2020) +Abbas, A. M., Ghauth, K. I., & Ting, C. Y. (2022). User Experience Design Using Machine Learning: A Systematic +Review. IEEE Access. +An, J., Cho, H., Kwak, H., Hassen, M. Z., & Jansen, B. J. (2016, August). Towards automatic persona generation using social +media. In 2016 IEEE 4th International Conference on Future Internet of Things and Cloud Workshops (FiCloudW) (pp. 206- +211). IEEE. +Babich, N. (2020). Wireframing Automation and Artificial Intelligence for UX Design. XD Ideas +https://xd.adobe.com/ideas/process/wireframing/wireframing-automation-ai-ux-design/ +Bakaev, M., Heil, S., Chirkov, L., & Gaedke, M. (2022). Benchmarking Neural Networks-Based Approaches for Predicting +Visual Perception of User Interfaces. In International Conference on Human-Computer Interaction (pp. 217-231). Springer, +Cham. +Beheshti, A., Benatallah, B., Sheng, Q. Z., & Schiliro, F. (2020, January). Intelligent knowledge lakes: The age of artificial +intelligence and big data. In International conference on web information systems engineering (pp. 24-34). Springer, Singapore. +Beheshti, A., Benatallah, B., Nouri, R., & Tabebordbar, A. (2018). CoreKG: a knowledge lake service. Proceedings of the VLDB +Endowment, 11(12), 1942-1945. +Beheshti, A., Yakhchi, S., Mousaeirad, S., Ghafari, S. M., Goluguri, S. R., & Edrisi, M. A. (2020). Towards cognitive +recommender systems. Algorithms, 13(8), 176. +Bertão, R. A., & Joo, J. (2021). Artificial intelligence in UX/UI design: a survey on current adoption and [future] practices. +Brdnik, S., Heričko, T., & Šumak, B. (2022). Intelligent User Interfaces and Their Evaluation: A Systematic Mapping +Study. Sensors, 22(15), 5830. +Budiu, R. & Laubheimer, P. (2018). Intelligent Assistants Have Poor Usability: A User Study of Alexa, Google Assistant, and +Siri. https://www.nngroup.com + + +31 + +Cerejo, J. (2021) The design process of human-centered AI - part 1. Bootcamp. https://bootcamp.uxdesign.cc/human-centered-ai- +design-process-part-1-8cf7e3ce00 +Carducci, G., Rizzo, G., Monti, D., Palumbo, E., & Morisio, M. (2018). Twitpersonality: Computing personality traits from +tweets using word embeddings and supervised learning. Information, 9(5), 127. +Chen, N. C., Drouhard, M., Kocielnik, R., Suh, J., & Aragon, C. R. (2018). Using machine learning to support qualitative coding +in social science: Shifting the focus to ambiguity. ACM Transactions on Interactive Intelligent Systems (TiiS), 8(2), 1-20. +Chen, C., Su, T., Meng, G., Xing, Z., & Liu, Y. (2018, May). From UI design image to GUI skeleton: a neural machine translator +to bootstrap mobile GUI implementation. In Proceedings of the 40th International Conference on Software Engineering (pp. 665- +676). +Chen, J., Lian, D., & Zheng, K. (2020). Collaborative filtering with ranking-based priors on unknown ratings. IEEE Intelligent +Systems, 35(5), 38-49. +Chen, L., Wang, P., Dong, H., Shi, F., Han, J., Guo, Y., ... & Wu, C. (2019). An artificial intelligence based data-driven approach +for design ideation. Journal of Visual Communication and Image Representation, 61, 10-22. +Chromik, M., Lachner, F., & Butz, A. (2020, October). ML for UX?-An Inventory and Predictions on the Use of Machine +Learning Techniques for UX Research. In Proceedings of the 11th Nordic Conference on Human-Computer Interaction: Shaping +Experiences, Shaping Society (pp. 1-11). +Calero Valdez, A., Ziefle, M., & Verbert, K. (2016, September). HCI for recommender systems: the past, the present and the +future. In Proceedings of the 10th ACM Conference on Recommender Systems (pp. 123-126). +Colusso, L., Bennett, C. L., Hsieh, G., & Munson, S. A. (2017, June). Translational resources: Reducing the gap between +academic research and HCI practice. In Proceedings of the 2017 Conference on Designing Interactive Systems (pp. 957-968). +Deldjoo, Y., & Atani, R. E. (2016). A low-cost infrared-optical head tracking solution for virtual 3d audio environment using the +nintendo wii-remote. Entertainment Computing, 12, 9-27. +De Peuter, S., Oulasvirta, A., & Kaski, S. (2021). Toward AI Assistants That Let Designers Design. arXiv preprint +arXiv:2107.13074. +Deb, K. (2014). Multi-objective optimization. In Search methodologies (pp. 403-449). Springer, Boston, MA. +Desolda, G., Esposito, A., Lanzilotti, R., & Costabile, M. F. (2021, August). Detecting Emotions Through Machine Learning for +Automatic UX Evaluation. In IFIP Conference on Human-Computer Interaction (pp. 270-279). Springer, Cham. +Desrosiers, C., & Karypis, G. (2011). A comprehensive survey of neighborhood-based recommendation methods. Recommender +systems handbook, 107-144. +Donahole, S. (2021). How Artificial Intelligence Is Impacting UX Design; UXmatters. +https://www.uxmatters.com/authors/archives/2021/04/stephanie_donahole.php April 19, 2021 +Dove, G., Halskov, K., Forlizzi, J., & Zimmerman, J. (2017, May). UX design innovation: Challenges for working with machine +learning as a design material. In Proceedings of the 2017 chi conference on human factors in computing systems (pp. 278-288). +Delrieu, B. & Barot, S. (2018). Using Machine Learning to Analyze and Optimize Customer Journeys. Gartner +Delrieu, B. & Skowron, J. (2018). Optimize Digital Customer Experiences With Machine-Learning Enhanced Analytics. Gartner. +Ehlert, P. (2003). Intelligent user interfaces: introduction and survey. Delft University of Technology. +Elahi, M., Beheshti, A., & Goluguri, S. R. (2021). Recommender systems: Challenges and opportunities in the age of big data +and artificial intelligence. Data Science and Its Applications, 15-39. +Ekstrand, M. D., Harper, F. M., Willemsen, M. C., & Konstan, J. A. (2014, October). User perception of differences in +recommender algorithms. In Proceedings of the 8th ACM Conference on Recommender systems (pp. 161-168). +Fan, M., Yang, X., Yu, T., Liao, Q. V., & Zhao, J. (2022). Human-AI Collaboration for UX Evaluation: Effects of Explanation +and Synchronization. Proceedings of the ACM on Human-Computer Interaction, 6(CSCW1), 1-32. + +32 + +Fan, J. J., Tian, F., Du, Y et al. (2018). Thoughts on human-computer interaction in the age of artificial intelligence (in Chinese). +Sci Sin Inform, 48, 361–375, doi: 10.1360/N112017-00221 + +Filho, J. F., Valle, T., & Prata, W. (2015, August). Automated usability tests for mobile devices through live emotions logging. +In Proceedings of the 17th International Conference on Human-Computer Interaction with Mobile Devices and Services +Adjunct (pp. 636-643). +Gajjar, N., Sermuga Pandian, V. P., Suleri, S., & Jarke, M. (2021, April). Akin: Generating UI wireframes from UI design +patterns using deep learning. In 26th International Conference on Intelligent User Interfaces-Companion (pp. 40-42). +Gebhardt, C., Hecox, B., van Opheusden, B., Wigdor, D., Hillis, J., Hilliges, O., & Benko, H. (2019, October). Learning +cooperative personalized policies from gaze data. In Proceedings of the 32nd Annual ACM Symposium on User Interface +Software and Technology (pp. 197-208). +Girardin, F. & Lathia, N. (2017). When user experience designers partner with data scientists. In the AAAI Spring Symposium +Series Technical Report: Designing the User Experience of Machine Learning Systems. The AAAI Press, Palo Alto, California. + +Gilbert, J. E., Williams, A., & Seals, C. D. (2007). Clustering for usability participant selection. Journal of Usability +Studies, 3(1), 40-52. +Glauser, O., Wu, S., Panozzo, D., Hilliges, O., & Sorkine-Hornung, O. (2019). Interactive hand pose estimation using a stretch- +sensing soft glove. ACM Transactions on Graphics (ToG), 38(4), 1-15. +Gonçalves, T. G., & da Rocha, A. R. C. (2019, October). Development process for intelligent user interfaces: an initial approach. +In Proceedings of the XVIII Brazilian Symposium on Software Quality (pp. 210-215). +Google PAIR (2019). People + AI Guidebook: Designing human-centered AI products. pair.withgoogle.com/ + +Hallgrimsson, B. (2012). Prototyping and modelmaking for product design. Hachette UK. +Han, S., & Yang, H. (2018). Understanding adoption of intelligent personal assistants: A parasocial relationship +perspective. Industrial Management & Data Systems, 118(3), 618-636. +He, Z., Spurr, A., Zhang, X., & Hilliges, O. (2019). Photo-realistic monocular gaze redirection using generative adversarial +networks. In Proceedings of the IEEE/CVF International Conference on Computer Vision (pp. 6932-6941). +Herath, D., & Jayarathne, L. (2018). Intelligent Recommendations for e-Learning Personalization Based on Learner's Learning +Activities and Performances. International Journal of Computer Science and Software Engineering, 7(6), 130-137. +Holmquist, L. E. (2017). Intelligence on tap: artificial intelligence as a new design material. interactions, 24(4), 28-33. +Hu, Q., Pan, Z., & Liu, J. (2019). The Duality of Autonomy on Continuous Usage of Intelligent Personal Assistants (IPAs): From +Agency Perspective. +InkbotDesign (2021). 10 Ways How AI is Shaping UI/UX Design. https://inkbotdesign.com/how-ai-is-shaping-ui-ux-design/ +June 10, 2021 +Islas-Cota, E., Gutierrez-Garcia, J. O., Acosta, C. O., & Rodríguez, L. F. (2022). A systematic review of intelligent +assistants. Future Generation Computer Systems, 128, 45-62. +Jannach, D., Pu, P., Ricci, F., & Zanker, M. (2021). Recommender systems: Past, present, future. Ai Magazine, 42(3), 3-6. +Jansen, B. J., Salminen, J. O., & Jung, S. G. (2020). Data-driven personas for enhanced user understanding: Combining empathy +with rationality for better insights to analytics. Data and Information Management, 4(1), 1-17. +Jiang, Y., Lu, Y., Nichols, J., Stuerzlinger, W., Yu, C., Lutteroth, C., ... & Li, T. J. J. (2022, April). Computational Approaches +for Understanding, Generating, and Adapting User Interfaces. In CHI Conference on Human Factors in Computing Systems +Extended Abstracts (pp. 1-6). +Jaquero, V. L., Montero, F., & Molina, J. P. (2009). Intelligent user interfaces: Past, present and future. In Engineering the User +Interface (pp. 1-12). Springer, London. + +33 + +Kayacik, C., Chen, S., Noerly, S., Holbrook, J., Roberts, A., & Eck, D. (2019, May). Identifying the intersections: User +experience+ research scientist collaboration in a generative machine learning interface. In Extended Abstracts of the 2019 CHI +Conference on Human Factors in Computing Systems (pp. 1-8). +Karpov, A. A., & Yusupov, R. M. (2018). Multimodal interfaces of human–computer interaction. Herald of the Russian Academy +of Sciences, 88(1), 67-74. +Kharitonov, E., Drutsa, A., & Serdyukov, P. (2017, February). Learning sensitive combinations of a/b test metrics. +In Proceedings of the Tenth ACM International Conference on Web Search and Data Mining (pp. 651-659). +Konstan, J., & Terveen, L. (2021). Human-centered recommender systems: Origins, advances, challenges, and opportunities. AI +Magazine, 42(3), 31-42. +Krinkin, K., Shichkina, Y., & Ignatyev, A. (2021, September). Co-evolutionary hybrid intelligence. In 2021 5th Scientific School +Dynamics of Complex Networks and their Applications (DCNA) (pp. 112-115). IEEE. +Kuniavsky, M., Churchill, E., & Steenson, M. W. (2017). The 2017 AAASI spring symposium series technical reports: +Designing the user experience of machine learning systems. (Technical Report SS-17-04). +Lau, N., Fridman, L., Borghetti, B. J., & Lee, J. D. (2018). Machine learning and human factors: Status, applications, and future +directions. In Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 62(1), 135-138. Sage CA: Los +Angeles, CA: SAGE Publications. + +Liu, Z., Liu, Y., Gao, X. (2020). Application of Artificial Intelligence Semantic Analysis Techniques in User Research. +Packaging Engineering, 41(18), 53-59. +Lei, W., He, X., de Rijke, M., & Chua, T. S. (2020, July). Conversational recommendation: Formulation, methods, and +evaluation. In Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information +Retrieval (pp. 2425-2428). +Liao, J., Hansen, P., & Chai, C. (2020). A framework of artificial intelligence augmented design support. Human-Computer +Interaction, 35(5–6), 511–544. +Lan, Y., Liu, S. (2020). Overview of Product User Experience Research under Artificial Intelligence Technology. Packaging +Engineering, 41(240), 22-28. +Li, L. (2019). Research on User Viscosity Based on Big Data Intelligence Algorithm. New Media Research, 5(4), 4-6. +Li, Y., Kumar, R., Lasecki, W. S., & Hilliges, O. (2020, April). Artificial intelligence for HCI: a modern approach. In Extended +Abstracts of the 2020 CHI conference on human factors in computing systems (pp. 1-8). +Li, C., Lan, M., Zou, B. et al. (2015). Big Data and Recommendation System. Big Data Research, 3, 23-35. +Li, N. (2020). TensorFlow.js: an AI platform for HCI. CHI’20,, April 25–30, 2020, Honolulu, HI, USA +Liao, J., Hansen, P., & Chai, C. (2020). A framework of artificial intelligence augmented design support. Human–Computer +Interaction, 35(5-6), 511-544. +Liapis, A., Katsanos, C., Sotiropoulos, D., Xenos, M., & Karousos, N. (2015, September). Subjective assessment of stress in +HCI: a study of the valence-arousal scale using skin conductance. In Proceedings of the 11th Biannual Conference on Italian +SIGCHI Chapter (pp. 174-177). +Lu, Y., Zhang, C., Zhang, I., & Li, T. J. J. (2022, April). Bridging the Gap Between UX Practitioners’ Work Practices and AI- +Enabled Design Support Tools. In CHI Conference on Human Factors in Computing Systems Extended Abstracts (pp. 1-7). +Main, A., & Grierson, M. (2020). Guru, Partner, or Pencil Sharpener? Understanding Designers’ Attitudes Towards Intelligent +Creativity Support Tools. Computer Science, arXiv abs/(Jul 2020) +Maedche, A., Legner, C., Benlian, A., Berger, B., Gimpel, H., Hess, T., ... & Söllner, M. (2019). AI-based digital +assistants. Business & Information Systems Engineering, 61(4), 535-544. +Marathe, M., & Toyama, K. (2018, April). Semi-automated coding for qualitative research: A user-centered inquiry and initial +prototypes. In Proceedings of the 2018 CHI conference on human factors in computing systems (pp. 1-12). + +34 + +Maybury, M. (1998, December). Intelligent user interfaces: an introduction. In Proceedings of the 4th international conference +on Intelligent user interfaces (pp. 3-4). +McCormack, J., Hutchings, P., Gifford, T., Yee-King, M., Llano, M. T., & D’inverno, M. (2020). Design considerations for real- +time collaboration with creative artificial intelligence. Organised Sound, 25(1), 41-52. +Mendes, M. S., & Furtado, E. S. (2017, November). UUX-Posts: a tool for extracting and classifying postings related to the use +of a system. In Proceedings of the 8th Latin American Conference on Human-Computer Interaction (pp. 1-8). +Mishra, S. & Jeffrey Rzeszotarski (2020). Towards Natural Language Interactions for Qualitative Text Analysis. CHI’20, April +25–30, 2020, Honolulu, HI, USA +Majumder, N., Poria, S., Gelbukh, A., & Cambria, E. (2017). Deep learning-based document modeling for personality detection +from text. IEEE Intelligent Systems, 32(2), 74-79. +Morris, M. R., Cai, C. J., Holbrook, J. S., Kulkarni, C., & Terry, M. (2022). The Design Space of Generative Models. +Nikiforova, O., Zabiniako, V. M., Kornienko, J., Gasparovica-Asite, M., & Silina, A. (2021). Mapping of Source and Target Data +for Application to Machine Learning Driven Discovery of IS Usability Problems. Appl. Comput. Syst., 26(1), 22-30. +Oh, C., Song, J., Choi, J., Kim, S., Lee, S., & Suh, B. (2018). I lead, you help but only with enough details: Understanding user +experience of co-creation with artificial intelligence. In Proceedings of the CHI 2018 conference on human factors in computing +systems (pp. 1–13). +Pazzani, M. J., & Billsus, D. (2007). Content-based recommendation systems. In The adaptive web (pp. 325-341). Springer, +Berlin, Heidelberg. +Pavliscak, P. (2016). Algorithms as the New Material of Design. UXMatters +https://www.uxmatters.com/authors/archives/2014/06/pamela_pavliscak.php +Policarpo, L. M., da Silveira, D. E., da Rosa Righi, R., Stoffel, R. A., da Costa, C. A., Barbosa, J. L. V., ... & Arcot, T. (2021). +Machine learning through the lens of e-commerce initiatives: An up-to-date systematic literature review. Computer Science +Review, 41, 100414. +Quadrana, M., Karatzoglou, A., Hidasi, B., & Cremonesi, P. (2017, August). Personalizing session-based recommendations with +hierarchical recurrent neural networks. In proceedings of the Eleventh ACM Conference on Recommender Systems (pp. 130-137). +Ricci, F., Massimo, D., & De Angeli, A. (2021, July). Challenges for Recommender Systems Evaluation. In CHItaly 2021: 14th +Biannual Conference of the Italian SIGCHI Chapter (pp. 1-5). +Robert Jr, L. P., & You, S. (2018). Are you satisfied yet? Shared leadership, individual trust, autonomy, and satisfaction in virtual +teams. Journal of the association for information science and technology, 69(4), 503-513. +Rogers, M. (2020). Artificial Intelligence Is Shifting Our Perception of Great. UXmatters. User Experiences. +https://www.uxmatters.com/authors/archives/2020/01/margaret_rogers.php January 6, 2020 +Salminen, J., Guan, K., Jung, S. G., & Jansen, B. J. (2021). A survey of 15 years of data-driven persona +development. International Journal of Human–Computer Interaction, 37(18), 1685-1708. +Salminen, J., Rao, R. G., Jung, S. G., Chowdhury, S. A., & Jansen, B. J. (2020, July). Enriching social media personas with +personality traits: A deep learning approach using the big five classes. In International Conference on Human-Computer +Interaction (pp. 101-120). Springer, Cham. +Salminen, J., Jung, S. G., & Jansen, B. (2022, March). Developing Persona Analytics Towards Persona Science. In 27th +International Conference on Intelligent User Interfaces (pp. 323-344). + +Santos, J., Rodrigues, J. J., Silva, B. M., Casal, J., Saleem, K., & Denisov, V. (2016). An IoT-based mobile gateway for +intelligent personal assistants on mobile health environments. Journal of Network and Computer Applications, 71, 194-204. +Schwartz, R., Mooney, K., & Baird, C. H. (2018). The AI-enhanced customer experience: A sea change for CX strategy, design +and development. IBM Institute for Business Value. March 2018. https://www935. ibm. com/services/us/gbs/thoughtl +eadership/custexperience/ai-cx. + +35 + +Setlur, V. (2020). Data-driven Intent Models for Visual Analysis Tools and Chatbot Platforms. CHI’20,, April 25–30, 2020, +Honolulu, HI, USA +Sermuga Pandian, V. P., Suleri, S., Beecks, C., & Jarke, M. (2020, December). MetaMorph: AI Assistance to Transform Lo-Fi +Sketches to Higher Fidelities. In 32nd Australian Conference on Human-Computer Interaction (pp. 403-412). +Shichkina, Y., & Krinkin, K. (2022, September). Principles of building personalized intelligent human assistants. In 2022 Fourth +International Conference Neurotechnologies and Neurointerfaces (CNN) (pp. 148-151). IEEE. +Soleimani, S., & Law, E. L. C. (2017, June). What can self-reports and acoustic data analyses on emotions tell us? In +Proceedings of the 2017 Conference on Designing Interactive Systems (pp. 489-501). +Souza, K. E. S. D., Aviz, I. L. D., Mello, H. D. D., Figueiredo, K., Vellasco, M. M. B. R., Costa, F. A. R., & Seruffo, M. C. D. R. +(2022). An Evaluation Framework for User Experience Using Eye Tracking, Mouse Tracking, Keyboard Input, and Artificial +Intelligence: A Case Study. International Journal of Human–Computer Interaction, 38(7), 646-660. +Stephanidis, C., Salvendy, G., Antona, M., Chen, J. Y., Dong, J., Duffy, V. G., ... & Zhou, J. (2019). Seven HCI grand +challenges. International Journal of Human–Computer Interaction, 35(14), 1229-1269. + +Sun, X., Liu, B., Cao, J., Luo, J., & Shen, X. (2018, May). Who am I? Personality detection based on deep learning for texts. +In 2018 IEEE international conference on communications (ICC) (pp. 1-6). IEEE. +SUN Ling-yun1,2, ZHANG Yu-yang1,2, ZHOU Zhi-bin1,2, ZHOU Zi-hong (2020). Current Situation and Development Trend of +Intelligent Product Design under the Background of Human-centered AI. PACKAGING ENGINEERING, 41(2), 1-6. +Swearngin, A., & Li, Y. (2019, May). Modeling mobile interface tappability using crowdsourcing and deep learning. +In Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems (pp. 1-11). +TAN Hao, YOU Zuo, PENG Sheng-lan. Big Data-Driven User Experience Design[J]. Packaging Engineering, 2020, +41(2), 7-12. +Vaccaro, K., Agarwalla, T., Shivakumar, S., & Kumar, R. (2018, April). Designing the future of personal fashion. In Proceedings +of the 2018 CHI Conference on Human Factors in Computing Systems (pp. 1-11). +Van den Oord, A., Dieleman, S., & Schrauwen, B. (2013). Deep content-based music recommendation. Advances in neural +information processing systems, 26. +Verganti, R., Vendraminelli, L., & Iansiti, M. (2020). Innovation and Design in the Age of Artificial Intelligence. Journal of +Product Innovation Management, 37(3), 212–227. +Vetrov, Y. (2022). Algorithm-Driven Design: How Artificial Intelligence is Changing Design. https://www.algorithms.design/ +Wu, C. J., Brooks, D., Chen, K., Chen, D., Choudhury, S., Dukhan, M., ... & Zhang, P. (2019, February). Machine learning at +Facebook: Understanding inference at the edge. In 2019 IEEE international symposium on high performance computer +architecture (HPCA) (pp. 331-344). IEEE. +Wang, W., Zhao, D., Luo, H., & Wang, X. (2013, July). Mining user interests in web logs of an online news service based on +memory model. In 2013 IEEE Eighth International Conference on Networking, Architecture and Storage (pp. 151-155). IEEE. +Willemsen, M. C., Graus, M. P., & Knijnenburg, B. P. (2016). Understanding the role of latent feature diversification on choice +difficulty and satisfaction. User Modeling and User-Adapted Interaction, 26(4), 347-389. +Wilkins, B. (2018). Sketching Interfaces. Airbnb.Design. [Online]. Available: https://airbnb.design/sketching-interfaces +Xu, W. (2005). Recent trend of research and applications on human-computer interaction. Journal of Ergonomics, 11(4), 37-40. +Xu, W. (2017). User-centered design (II): New challenges and opportunities. Journal of Ergonomics, 23(1), 82-86. +Xu, W. (2018). User-centered design (III): Methods for user experience and innovative design in the intelligent era. Chinese +Journal of Applied Psychology, 25(1), 3-17. +Xu, W. (2019). User-centered design (IV): Human-centered artificial intelligence. Chinese Journal of Applied Psychology, 25(4), +291-305. + +36 + +Xu, W. (2020). User-centered design (V): From automation to the autonomy and autonomous vehicles in the intelligence era. +Chinese Journal of Applied Psychology, 26(2), 108-129. +Xu, W. & Ge, L. (2018). New trends in human factors. Advances in Psychological Science, 26(9), 1521-1534. +Xu, W. & Ge, L. (2020). Engineering psychology in the era of artificial intelligence. Advances in Psychological Science, 28(9), +1409-1425. +Xu, W., Dainoff, J.M., Ge, L, Gao, Z. (2021). From human-computer interaction to human-AI Interaction: new challenges and +opportunities for enabling human-centered AI arXiv preprint arXiv:2105.05424 5 +Xu, X., Liu, J., Yan, Y., Yang, J. (2020). Development and Theoretical Trend of Intelligent Design Methods. Packaging +Engineering, 41(4), 10-19. +Yang, Q., Banovic, N., & Zimmerman, J. (2018, April). Mapping machine learning advances from hci research to reveal starting +places for design innovation. In Proceedings of the 2018 CHI conference on human factors in computing systems (pp. 1-11). +Yang, Q., Steinfeld, A., Rosé, C., & Zimmerman, J. (2020, April). Re-examining whether, why, and how human-AI interaction is +uniquely difficult to design. In Proceedings of the 2020 chi conference on human factors in computing systems (pp. 1-13). +Yang, Q., Scuito, A., Zimmerman, J., Forlizzi, J., & Steinfeld, A. (2018, June). Investigating how experienced UX designers +effectively work with machine learning. In Proceedings of the 2018 designing interactive systems conference (pp. 585-596). +Yang, Q. (2018). Machine learning as a UX design material: How can we imagine beyond automation, recommenders, and +reminders? Conference: 2018 AAAI Spring Symposium Series: User Experience of Artificial Intelligence, March 2018, At: Palo +Alto, CA. + +Yang, B., Wei, L., & Pu, Z. (2020). Measuring and improving user experience through artificial intelligence-aided +design. Frontiers in Psychology, 11, 595374. +Yu, M. (2014). Data modeling of user portrayal in product development: From concrete to abstract'. Design Act Research, 6, 60- +64. +Yu, G. (2018). Media User Experience’s Effect Evaluation Based on the Paradigm of Big Data Intelligence Algorithm. Journal of +Education and Media Studies, 16(5), 10-12 +Yuan, A., & Li, Y. (2020, April). Modeling human visual search performance on realistic webpages using analytical and deep +learning methods. In Proceedings of the 2020 CHI conference on human factors in computing systems (pp. 1-12). +Zangerle, E., & Bauer, C. (2022). Evaluating Recommender Systems: Survey and Framework. ACM Computing Surveys (CSUR). +Zhang, X., Brown, H. F., & Shankar, A. (2016, May). Data-driven personas: Constructing archetypal users with clickstreams and +user telemetry. In Proceedings of the 2016 CHI conference on human factors in computing systems (pp. 5350-5359). +Zheng, L., Noroozi, V., & Yu, P. S. (2017, February). Joint deep modeling of users and items using reviews for recommendation. +In Proceedings of the tenth ACM international conference on web search and data mining (pp. 425-434). +Zheng, G., Zhang, F., Zheng, Z., Xiang, Y., Yuan, N. J., Xie, X., & Li, Z. (2018, April). DRN: A deep reinforcement learning +framework for news recommendation. In Proceedings of the 2018 world wide web conference (pp. 167-176). +Zhao, Q., Harper, F. M., Adomavicius, G., & Konstan, J. A. (2018, April). Explicit or implicit feedback? Engagement or +satisfaction? A field experiment on machine-learning-based recommender systems. In Proceedings of the 33rd Annual ACM +Symposium on Applied Computing (pp. 1331-1340). +Zwakman, D. S., Pal, D., & Arpnikanondt, C. (2021). Usability evaluation of artificial intelligence-based voice assistants: the +case of Amazon Alexa. SN Computer Science, 2(1), 1-16. +