Recently large language models (LLMs), such as ChatGPT, have received unrivalled attention not only in AI research but also in the greater public. The massive coverage in the mainstream media makes it utterly hard for an end user to get a decent overview of systems and an understanding of what these systems are actually capable of, plus how they are best and safely used. In the field of LLMs, where news and updates about the newest features are published on almost a daily basis, users, or prospective users, are overwhelmed by new possibilities on how to retrieve information. What is the best approach to tackle the challenging situation where users are inundated with various options of different LLMs, and which performs best for various use cases? This chapter presents the outcomes of a project initiated by AI bachelor students that aim to better understand different LLMs from an end-user perspective. We have thoroughly tested the capabilities of these systems in various group sessions and present our results in this chapter. As well as try to offer some guidance on how to learn about these systems by testing them in a structured way, we reflect on to which extent different LLMs met our expectations and compare them to determine which LLM to use in different scenarios. We believe this chapter can serve as a decent introduction to LLMs for researchers of all disciplines, as well as practitioners interested in exploring the potential of various systems.

Decision-making often involves a choice between options whose outcomes are probabilistic and thus cannot be predicted with certainty. A common methodology for understanding people’s risky choices is to develop formal models of their responses to monetary gambles, usually collected in lab experiments. The quest for veridical models of decisions under risk has been one of the most actively debated areas in the behavioural sciences, in particular in psychology and economics. This chapter outlines and compares different routes that have been taken to capture the psychological underpinnings of people’s risky choices: neo-Bernoullian models (e.g., cumulative prospect theory, security-potential/aspiration theory, transfer of attention exchange model), models with context-dependent valuation (proportional difference model, decision-by-sampling theory), heuristics and sequential sampling models (e.g., decision field theory, drift–diffusion model). These modelling frameworks propose diverse psychological constructs, including distorted representation of outcomes, asymmetric representation of gains and losses, asymmetric weighting of outcomes and probabilities, various notions of how decision weights for risky options are derived from probability information, and imbalanced and limited attention. Despite the richness of the theoretical arena, relationships between the constructs are poorly understood. This chapter argues that this disconnect between theories can be overcome by greater efforts towards theory integration, which maps constructs from different models onto each other. We present two examples illustrating the theory integration approach and its possible merits, such as indicating possible mechanistic explanations for patterns in probability weighting.

This chapter critically reviews the success of existing algorithms in explaining and predicting human behaviour. While traditional statistical methods have limitations in this area, algorithmic approaches have gained popularity. The chapter covers a range of algorithms, including decision trees, neural networks and clustering algorithms, evaluating their strengths and limitations in various applications. It also considers ethical concerns, such as bias and privacy violations and the need for transparency and explainability. The chapter emphasises the importance of interdisciplinary collaboration between computer science, statistics and behavioural science, and the need for ongoing development and refinement of these methods. By evaluating the effectiveness of algorithmic approaches to human behaviour, this chapter is a valuable resource for researchers and practitioners in the field.

Context-dependent behaviour is a central feature of human and algorithmic decision-making. Unlike context-free models, which assume stable preferences or uniform mechanisms across environments, context-dependent models account for how behaviour shifts depending on external, social, temporal or emotional surroundings. This chapter reviews conceptual frameworks and computational techniques used to capture context-dependence in behavioural data science. We explore psychological theories such as framing and cue-dependence, formal models including multi-level regression, contextual bandits and conditional generative models and the role of context in real-world domains such as health, finance and human–machine interaction. We provide a taxonomy of modelling approaches, examine challenges in data acquisition and generalisability and present a worked example using environmental nudges in energy consumption. The chapter concludes with future directions for context-sensitive modelling, highlighting its importance for ethical, fair and adaptive behavioural data systems.

Anthropomorphic learning in behavioural data science refers to a hybrid modelling paradigm that integrates human-centric principles from decision theory with data-driven methodologies of machine learning to simulate, predict and optimise human-like behaviours in artificial systems. This chapter introduces and explores this hybrid approach, examining its theoretical underpinnings, methodological framework and practical applications. Anthropomorphic learning seeks to bridge the normative rigour of decision theory – grounded in models of preference, choice and rationality – with the empirical flexibility of machine learning, particularly in settings marked by uncertainty, complexity and interactivity. The chapter distinguishes anthropomorphic learning from conventional hybrid modelling techniques and critiques its promise through a detailed analysis of recent implementations in user modelling, digital choice architectures and autonomous agent systems. It also discusses its limitations and ethical implications, especially concerning transparency, replicability and human interpretability. By embedding agency, intentionality and bounded rationality into algorithmic structures, anthropomorphic learning represents a compelling frontier for advancing both human-aligned AI and behavioural data science.

Humans are emotional beings that reason, not reasoning beings that occasionally experience emotions. In fact, emotions are essential to good decision-making. In the world of big data and machine learning, the human potential for emotions can influence the way machine learning systems are developed and deployed and also influence the way they are perceived and used. The two manifestations of emotion that can influence decision-making are bias and noise, both of which have emotional undertones. In this chapter, I consider the impact of emotion, bias and noise and suggest a way to acknowledge and measure their impact in the realm of machine learning and big data.

This chapter explores how traders’ performance may be influenced by the rationality levels of their peers in the market. Using a Behavioural Data Science approach, the study integrates experimental methods, machine learning and large-scale digital trace analysis to examine this relationship. Specifically, we analysed data from a cryptoasset exchange over a five-week period in late 2017 and early 2018, covering over 700,000 transactions across 17 trading pairs. We complemented this behavioural trace data with an online guessing game involving 2,622 active traders, of whom 273 participated. By combining survey results and trading histories, we applied clustering algorithms to identify seven distinct trader profiles, including ‘jokers’, ‘focal point traders’ and those operating at different levels of strategic reasoning (first, second and third order), as well as ‘professional’ and ‘Nash equilibrium’ traders. The findings suggest that traders engaging in higher-order reasoning generally achieve better financial outcomes, yet even experienced professionals are not immune to behavioural biases. The chapter highlights how Behavioural Data Science methods – linking experimental insight with real-world data and computational tools – can illuminate the cognitive patterns underlying economic decision-making in digital markets.

This chapter argues that two of the common methods used in behavioural and social sciences to reduce the chances that models overfit the available data, namely heavy reliance on benchmark models and rigorous parameter estimation techniques, can slow the advancement of these sciences. An examination of classical decision research highlights how applying these methods shaped the field but have also led to limited success. As an alternative, the chapter proposes a prediction-oriented approach to the development of behavioural models. Evaluating and comparing models based on their predictive power inherently guards against overfitting and also facilitates accumulation of knowledge. The chapter reviews research employing the prediction-oriented approach in behavioural decision research and demonstrates that, in contrast to a common misconception, the focus on predictions can also facilitate better understanding of the underlying processes.

Extended reality (XR), encompassing virtual reality (VR) and augmented reality (AR), has become a crucial tool in Behavioural Data Science. This chapter explores the applications of XR, VR and AR in this field, with a focus on analysing human behaviour and decision-making in immersive environments. The chapter begins with an overview of XR, VR and AR technologies and their potential in Behavioural Data Science. It discusses the advantages of using immersive environments for studying human behaviour, such as the ability to control and manipulate variables, measure behaviour in real time and simulate complex scenarios. It reviews various applications of XR, VR and AR in Behavioural Data Science. The chapter covers how immersive environments aid in studying decision-making, social interaction, learning, training and cognitive processes, with specific examples like using VR for consumer behaviour studies and AR for employee training. The challenges and opportunities of applying XR, VR and AR in Behavioural Data Science are also discussed. This includes the need for advanced data collection and analysis tools, ethical considerations around data privacy and security and potential new applications in fields like healthcare, education and entertainment. The chapter emphasises the significance of XR, VR and AR in understanding human behaviour and decision-making in immersive environments. It calls for ongoing research to further explore the potential applications of these technologies in Behavioural Data Science and to develop new tools and methods for analysing data from immersive environments.

Network and collective choice models are foundational tools in Behavioural Data Science, offering deep insight into how individual decisions scale into system-level outcomes. These models underpin everything from public health planning and traffic optimisation to social media influence and climate action coordination. Yet, as they become more integrated into decision-making architectures – especially under the regulatory framing of the European Union’s AI Act – questions of bias, equity, explainability and accountability become unavoidable. This chapter argues for a responsible approach to network and collective choice modelling, grounded in legal foresight, social ethics and behavioural realism. Beginning with an overview of the theoretical foundations and methodological advantages of these models, it then unpacks critical concerns: selection and confirmation bias, representational fairness, algorithmic opacity and privacy loss. Special attention is paid to the risks of amplification of systemic inequality and marginalisation through flawed modelling assumptions. The chapter draws on real-world applications to show how stakeholder co-design, model interpretability and participatory governance can mitigate harm. By weaving legal obligations under the AI Act with behavioural science principles, this chapter offers a pathway to designing socially beneficial, transparent and context-aware models of collective decision-making in digital systems.

Over the last few decades, psychologists have increasingly found that the mind stores and uses the statistics of its environment. However, less work has analysed whether the environmental statistics have changed and what that would imply for the mind. In this chapter, we consider human memory as the solution to the computational problem of predicting what events will happen next given a history of past events. Prior work examining two years of data (1986–1987) found that the environmental statistics of events occurring in the world are reflected in human memory of events, such as practice and retention effects. We analyse the last century of event statistics by assuming that words in the headlines of The New York Times are each an event. While presenting our methods, we do so in the form of a case study – we discuss general practices for Behavioural Data Science projects, standard issues that arise and how to resolve different issues as they arise for the presented analyses. After replicating prior work analysing event statistics in this manner during 1986–1987, we extend the methodology to the last century (1919–2019). Our analyses suggest that the events are occurring in denser bursts, meaning that, if a new event occurs in the last few years, this event reoccurs more often in the short-term and less often in the long-term (as compared to events that first occurred in the early twentieth century). This suggests that human memory faces different environmental demands than it has in the past and may be adapting to the dynamics of event statistics.

The A-IQ project evaluates conversational artificial intelligence (AI) performance through a behavioural perspective, measuring observable external behaviours rather than internal processes considering algorithms developed pre-generative AI. This approach is valuable for applied research targeting end-user perception and where it is not necessary to consider internal processes. The Interdisciplinary Artificial Intelligence Model was developed, consisting of seven domains to solve problems, from which the Interdisciplinary Artificial Intelligence Quotient Scale (iAIQs Scale) was created. The iAIQs Scale contains 62 questions that are evaluated by human testers through a multi-level system of response categories. The A-IQ tests were conducted on Google Now (2018 release) and Google Assistant (2021 release), Siri (Apple), Cortana (Microsoft) and Alexa (Amazon). The results revealed that Siri had the best overall performance due to high scores in the working memory domain, while Cortana scored highest in explicit knowledge. However, no conversational AI scored in the critical or creative thinking domains. Testing, conducted in 2021 (pre-ChatGPT release), showed improvements in Google Assistant’s performance, followed by Alexa, while Siri showed minimal improvements. A prototype was developed to automate the testing process and facilitate continuous monitoring. Limitations were identified regarding reproducibility and objectivity, but the A-IQ project contributes to the evolving field of human–machine interaction, focusing on communication. This chapter focuses specifically on conversational agents developed prior to the release of ChatGPT, with generative AI systems examined in Chapter 15.

This chapter reviews the latest methods and datasets on cognitive networks as models of human cognition and behaviour. Despite some limitations, cognitive networks have significant potential for various case studies. This chapter identifies three primary applications for cognitive networks: (i) interpretation of cognitive processing across visual, auditory and semantic tasks and among different populations; (ii) predicting cognitive development, decline and performance in both healthy and clinical populations; and (iii) reconstructing semantic framing in texts and media. The chapter emphasises that cognitive networks can provide relevant tools for modelling human behaviour and the field can advance through careful statistical modelling, collaboration with other interpretable frameworks and the availability of rich datasets across tasks and contexts. The review highlights that cognitive networks offer new insights into complex human cognitive processes that were previously difficult to model, making them promising for further research and development. The chapter provides valuable insights for researchers and practitioners interested in cognitive networks, emphasising their potential for a wide range of applications in cognitive and behavioural sciences. The findings suggest that cognitive networks can be an essential tool for advancing our understanding of human cognition and behaviour.

The rise of social media as big social data offers opportunities for behavioural science and Behavioural Data Science researchers. The availability of large volumes of data generated by users of social media opens up the possibility of being able to study human behaviour in detail, at scale and in real-time. Stimulated in part by such opportunities, machine learning and natural language processing (NLP) techniques have advanced rapidly. Applications of social media analytics now range from attitude extraction, sentiment analysis and behavioural monitoring, to rumour verification, event prediction, detection and tracking. However, the challenges have also grown and opinions on the value of social media analytics remain divided, with some seeing great potential while others express concerns about, for example, the absence of transparency in machine learning and NLP models and problems of bias in – and ethics of – harvesting social media data. Some are also concerned about the lack of methodological rigour and even question whether social media analytics is capable of delivering meaningful and robust insights into human behaviour. If they are to address these issues, then behavioural science researchers will need to master the skills and knowledge that will be needed to choose and apply machine learning and NLP techniques in ways that can guarantee robust and scientifically valid findings.

The use of Bayesian modelling has been growing in the cognitive sciences, as it provides a flexible framework for modelling complex and heterogeneous data, as well as incorporating prior knowledge and uncertainty into the modelling process. In this chapter, we focus on predictive Bayesian modelling, which allows for the construction of models that can make accurate and meaningful predictions about future events, behaviours or outcomes. We begin by providing a brief overview of Bayesian statistics and the key concepts necessary for understanding predictive Bayesian modelling. We then discuss the benefits and limitations of using predictive Bayesian modelling in cognitive science research. Next, we explore several examples of how predictive Bayesian modelling has been applied to different areas of cognitive science, including perception, memory, decision-making and language. We discuss how these models have contributed to our understanding of cognitive processes and how they can be used to make predictions about human behaviour. Finally, we outline some of the challenges and future directions for predictive Bayesian modelling in cognitive science research. We discuss the importance of evaluating model fit and model comparison, the need for more accurate and informative prior distributions and the potential for combining predictive Bayesian modelling with other methods, such as machine learning.

This chapter reconsiders the prevailing logic of risk categorisation in artificial intelligence regulation, focusing on how the European Union?s Artificial Intelligence Act operates in practice as a four-tiered, domain- and use-case-based framework. While the Act introduces a structured and ostensibly proportionate approach to governance, it rests on an assumption that risk is a function of application domain rather than the behavioural, contextual and technical dynamics that shape real-world harm. Drawing on the interdisciplinary field of behavioural data science – which integrates behavioural science, cognitive psychology and empirical data analytics – we argue for a scenario-specific model of risk assessment. This model accounts for how artificial intelligence systems interact with human cognitive biases, demographic vulnerabilities and shifting deployment conditions. By reconceptualising risk as an emergent property of human–machine co-production, the chapter introduces a semi-quantitative scoring framework grounded in behavioural indicators. This framework enables regulators and developers to assess risk more adaptively and responsively across general-purpose and domain-specific artificial intelligence applications. In doing so, the chapter proposes a shift from assumption-driven to evidence-based governance, positioning behavioural data science as a critical lens for advancing socially robust artificial intelligence regulation.

Behavioural Data Science has become a crucial tool in finance, helping researchers and practitioners understand the behaviours of individuals and markets. This chapter investigates how unstructured customer feedback data, collected from the online review platform Trustpilot, can be used to model consumer behaviour in financial services. Through a large-scale text analysis of customer reviews, we examine (i) the differences in how customers perceive traditional financial institutions compared to fintech firms; (ii) the predictive power of context-dependent sentiment in forecasting customer satisfaction; and (iii) the methodological advantages of using real-time, unstructured behavioural data for improving customer experience analytics. Our findings indicate that traditional financial service providers and fintech companies often elicit orthogonal sentiment patterns from customers – even when offering similar services – highlighting the importance of brand identity and user expectation in behavioural outcomes. We also demonstrate that models trained on Trustpilot-derived textual data outperform conventional natural language processing approaches in predicting customer satisfaction. By embedding sentiment analysis within a broader Behavioural Data Science framework, this chapter illustrates how financial institutions can more accurately interpret and respond to consumer feedback, contributing to more adaptive, customer-centric service design in both traditional and emerging financial ecosystems.

Behavioural Data Science represents the convergence of behavioural theory, computational modelling and empirical analysis to understand, predict and shape human, algorithmic and systems behaviour. As the field matures, its relevance hinges not only on its conceptual elegance but on its demonstrable value in addressing real-world challenges. This chapter serves as an introduction to Part V, which presents a broad spectrum of applied domains – ranging from healthcare and education to financial services, digital marketing, cybersecurity, public policy and environmental sustainability. Each application domain embodies the core tenets of Behavioural Data Science: interdisciplinarity, responsiveness to contextual variability and methodological agility. The chapter explores how behavioural theory is operationalised through data science techniques and how complex behavioural systems are navigated and intervened upon using algorithmic strategies. It critically assesses the translational challenges encountered when models developed in controlled settings are deployed in dynamic, heterogeneous environments. Ethical considerations – ranging from data governance and fairness to manipulation and informed consent – are shown to be not peripheral but integral to application design and implementation. Drawing on illustrative cases and cross-cutting insights from the chapters that follow, this chapter argues that applications of Behavioural Data Science must balance predictive power with interpretability, impact with equity and automation with accountability. In doing so, the chapter positions Part V not merely as a repository of case studies but as a reflection of how Behavioural Data Science performs under real-world constraints – and how its promises are tested, refined or reimagined in practice.