This handbook provides valuable information on using artificial intelligence in education (AIED). It was designed to address both technical and human-centred, societal aspects of the subject from various perspectives. The cover image symbolically reflects the book's approach, depicting four children: two working together at a laptop, one working individually at another laptop, and one using paper. A screen is positioned in the background, unnoticed by the children, who display a calm, focused demeanour. Although technology is present, the image centres on the children themselves, underscoring an important message. The book is not about idealizing artificial intelligence (technology) but instead explores how this technology can support student-centred learning, which is also a current approach within the European Higher Education Area.

The idea for the handbook originated in late 2019, before the recent surge of interest in AI. Daniel Mathers from Edward Elgar Publishing approached Benedict du Boulay to see if he would be interested in editing the book. Du Boulay first encountered AIED during his PhD studies in the Department of AI at Edinburgh in the 1970s. Antonija Mitrovic joined the project in January 2020, having first learned about Intelligent Tutoring Systems in 1990 while working on her MSc project at the University of Calgary. The last editor, Kalina Yacef, was introduced to the world of intelligent tutoring systems in the late 1990s. The book was published in 2023.

Although AI may seem like a recent development, it has a long history, including in education. One of the earliest AI-based educational systems was Carbonell's Scholar (1970), which could ask and answer questions about South American geography. This system used a symbolic AI knowledge base that could be swapped out for other topics, making it relatively versatile—though not as universally adaptable as models like ChatGPT. Going back even further, Pressey (1926) designed a special-purpose machine that presented multiple-choice questions to students, requiring correct answers before moving on to the next question. Building on Pressey's work, Skinner (1958) later developed his theory of programmed learning. Chapter 2 provides more details about the history of AIED.

The second part of the book focuses on theories underpinning AIED. For instance, it discusses instructional design principles, such as how cognitive conflict can stimulate students' interest. Chapter 4 explains the role of metacognition and pedagogical theories in AIED, while the following chapter explores emotions, motivation, affect, meta-affect, and affective pedagogy. This section references Dweck's (2002) growth mindset theory, which suggests that students who believe intelligence is fixed tend to avoid academic challenges, while those who see it as malleable are more resilient. Dweck proposes that communicating to students that intelligence is flexible and that mistakes are essential to learning can help them sustain their efforts.

The message is clear: when designing and implementing AIED systems, we must consider the psychological characteristics of our students and help them understand how their minds work. One practical example of technology's role is the visualization of emotions, which can enhance self-reflection— a key step toward emotional regulation. Technology can also improve classroom dynamics; for example, Washington et al. (2017) used Google Glass to develop an emotion recognition system for children with autism.

The third part of the book is dedicated to the architecture and design of AIED systems. It introduces domain model paradigms such as rules, constraints, Bayesian networks, and machine learning. Another chapter explores how to design open-ended learning environments. Various examples are presented, such as PhET (https://phet.colorado.edu/), which provides interactive math and science simulations grounded in extensive validation and research. In this intuitive, game-like environment, students learn through exploration and discovery, receiving visual feedback on their progress.

Another example is Betty's Brain (https://lab.vanderbilt.edu/live/2024/06/20/bettys-brain/), where a virtual character named Betty interacts with students as a “good” learner who possesses self-regulatory skills. By incorporating feedback focused on these self-regulatory skills, the system has shown that students are better able to perform in future learning tasks.

A subsequent chapter presents six instructional approaches supported by AIED systems: learning through problem-solving, learning from examples, exploratory learning, collaborative learning, game-based learning, and learning by teaching. One example is the Tactical Language and Culture Training System (https://www.29palms.marines.mil/training/magtftcsims/tlts/#about), which uses AI-driven Non-Player Characters. Studies have shown that students using this system, which focuses on Arabic language and Iraqi cultural knowledge, gained a better understanding of the Arabic language.

Overall, this section presents approaches that enhance student interaction, collaboration, and problem-solving skills. Another chapter examines the role of natural language processing and how students can benefit from learning through self-explanations and social constructivism—where students learn through social interactions. Animated pedagogical agents are used to support and stimulate these approaches.

The fourth part focuses on the use of learning analytics, illustrating how data-driven models can automatically define a set of skills for assessment. It presents two main approaches: explicit and implicit models. Explicit models use data-driven feature detectors to define a discrete set of skills, evaluating them as correct or incorrect. This approach is more interpretable and transparent for students and can be enhanced by expert input, where even a small group of experts can significantly improve the accuracy of data-driven skill detectors. Implicit models, based on neural networks, are fully automated and can uncover new patterns that might be missed by humans. This approach is particularly useful when raw predictive performance is prioritized. Chapter 14 is dedicated to the role of AI in Human-AI co-orchestration, extending beyond learner-technology interaction to consider interactions across the broader learning environment.

The following chapter features co-author Dragan Gašević, a prominent figure in learning analytics, and focuses on how learning analytics can support teachers. Teachers can use descriptive learning analytics for an overview of students' activities, while predictive learning analytics helps indicate whether students' past behaviour can predict changes in future behaviour or learning outcomes. One case study in this chapter introduces ZoomSense (https://gitlab.com/action-lab-aus/zoomsense/zoomsense), a tool that provides real-time support for monitoring class activities when students use Zoom and Google Documents, as well as post-class reflection that enables teachers to review student progress and interactions in each breakout room session.

Chapter 16 specifically addresses predictive modelling of student success, discussing critical issues with this approach, such as biases based on sample demographics (e.g., discrimination against students from minority groups or low socioeconomic backgrounds). Challenges also arise when new students enter or when new circumstances develop. Learning analytics can also foster interaction in learning communities like MOOCs. The authors of Chapter 17 demonstrate how mentor support during COVID positively impacted students' success.

The fifth part of the book explores various AIED systems in practice. Chapter 18 presents a systematic review of intelligent systems for psychomotor learning—a relatively uncommon topic, as a comprehensive meta-review identified only twelve relevant papers. The tools used in this field include VR goggles, simulators, phantom devices, robotic tutors, and various sensors. Feedback and results from these tools are provided in multiple forms, such as visual, verbal, text-based, haptic, or vibrotactile responses. These systems can be applied in diverse fields, from martial arts and surgical training to pilot instruction and rehabilitation. The chapter's authors recommend future research that integrates Bloom's taxonomy's three domains—psychomotor, cognitive, and affective. They also emphasize the benefits of collaborative approaches in this field, noting that future applications could learn from gamification techniques, as some sensors in psychomotor AIED systems originated in the gaming industry.

Chapter 19 provides an overview of various AI techniques to support collaborative learning, including group formation, formative and summative feedback, adaptive scripting, and enhanced group and teacher awareness. For example, one framework, Quantitative Ethnography (Shaffer, 2017), offers a conceptual framework for mapping data to constructs using codes—the contextual meanings of actions. Codes can be defined from an etic perspective, based on relevant theories, or from an emic perspective, representing the actors' viewpoints. Emic coding helps balance informativeness with interpretability.

Chapter 20 reviews digital AI-enhanced games, explaining how AI can be used to support game operations, interact with players (e.g., through NPCs based on AI), or apply AI techniques such as data mining, learning analytics, and machine learning. The following chapter focuses on the use of AI for assessment, a popular topic. AI-based assessment can provide personalized learning instructions and formative assessments using diverse data sources, including performance metrics, behaviour, language, and biometric data. Ideally, multimodal data, combining inputs from various sources, is used. Assessment can be either visible or stealth, with the latter often employed in game environments. One example is the game Use Your Brainz, a modified version of the commercial game Plants vs. Zombies 2.

In VanLehn's chapter, a simple overview introduces options for designing comprehensive evaluations in AIED, summarizing factors essential to successful AI-supported assessment. The next chapter addresses the commercialization of AI in school-based environments, focusing on K-12 education in the U.S. The authors discuss three key areas: personalized instruction, automated assessment, and data-driven support for teachers and administrators. For example, Khan Academy's diagnostic systems are used to create personalized learning pathways, and automated essay scoring systems and reading tutors are also commonly implemented.

The final chapters of this section cover the political economy, commercialization, discrimination, and ethics of AI in education. The authors examine how political factors, such as the situation in China, impact AI technology, and discuss algorithmic discrimination. The chapter also highlights the public education sector's reliance on private-sector technology infrastructure, a topic also addressed by the European Digital Education Hub, which examines interoperability and the challenge of open-vendor solutions. The concluding chapter focuses specifically on ethical issues in AIED, mapping various AI biases and their implications for education.

The final part of the book, titled The Future, is a single chapter divided into eleven sub-sections, each authored by a different contributor. Each section addresses a specific challenge or opportunity anticipated in the next twenty years. The opening section examines UNESCO's SDG 4, with a focus on equity and inclusivity in AI-supported education. The following section tackles how to engage learners in an era of information overload. Inclusion is further explored in the third section, where the author discusses how pedagogical agents can foster inclusivity in STEM disciplines. The fourth section explores the concept of intelligent textbooks, while the fifth returns to STEM, discussing open-ended learning environments. The sixth section considers the ubiquity of AIED, followed by an essay examining the cultural, ontological, and learner modelling challenges associated with AIED. The eighth section addresses the pathway from crowdsourcing to personalized learning. The ninth section reflects on the challenges facing developed countries in implementing effective AIED, once again raising the issue of the AI gap. The author of the tenth section speculates on the future of assessment, while the final section discusses intelligent mentoring systems.

Overall, I highly recommend the book as a comprehensive resource on AIED. It provides valuable insights and cross-references across chapters, with a well-structured approach that guides readers from historical context and theoretical frameworks to technological foundations, practical applications, and forward-looking reflections. Readers can also explore individual chapters for in-depth knowledge on specific topics. Each chapter is thoroughly supported by a wide range of sources, making the book a useful guide to further resources in EdTech, pedagogy, and psychology.

One area the book does not cover is a detailed analysis of generative-pre-trained transformers (GPTs) and their current educational applications, as this is a relatively recent development. However, the book provides a strong foundation for educators and educational researchers, emphasizing that AIED extends well beyond the use of GPTs alone.