Social technologies have profoundly transformed how people communicate, form relationships, and experience social connection. This chapter synthesizes two decades of research on social media and messaging platforms to extract lessons for new forms of social technology, such as empathic artificial intelligence. We propose that prior attempts to identify universal effects of social technology on connection are largely uninformative due to substantial heterogeneity across studies. We highlight and briefly review three key sources of heterogeneity: (1) the type and features of the technology, (2) how it is used, and (3) who is using it. We recommend that future research explicitly consider these sources of heterogeneity to better understand how technological advances will affect social life – and how they can be better designed to promote connection.

Efforts to develop “empathic AI” often assume a universal definition for empathy – a singular, objective understanding of “human” experiences. This framing treats achieving empathic AI as an optimization challenge, where researchers should find the “correct” way to align AI with human empathy. In this chapter, we challenge this assumption by emphasizing the empathy’s multifaceted nature and examining how variation across its forms and the social groups we are embedded within complicates efforts to design a universally applicable empathic AI. We argue that codifying empathy into artificial systems is difficult, given that empathy is an inherently context-dependent phenomenon. Drawing on recent evidence of cultural biases in large language models, we highlight broader ethical, epistemic, and potentially existential concerns in designing machines that “understand” or “feel” others. We conclude by advocating for a more contextualized and culture-aware approach to empathic AI, moving beyond the reductionist notion of “humans” as a monolith.

Most definitions of empathy stress its interpersonal nature: Empathy requires a human sender and a human receiver, where all parties involved have the theoretical capacity to accurately understand each other’s minds. Because empathy is inherently interpersonal, some scholars have questioned whether empathy between humans and artificial intelligence as it currently exists (e.g., large language models) is “real.” However, the requirement for two minds, each with the theoretical capacity to fully understand the thoughts and emotions of the other, is also not clearly met in other domains where empathy is discussed, such as empathy between humans and non-human animals or empathy with nature. This chapter critically examines these assumptions about what is needed for empathy to occur, offering several possible sets of empathy requirements that researchers might engage with, and outlining potential challenges researchers might face if they adopt any particular set of requirements in their work. Ultimately, we pose the question of whether two minds and/or the ability for minds to understand each other are truly necessary for empathy, or whether all that is required for empathy to occur is an entity’s perception of giving or receiving it.

Who – or what – people engage for supportive conversations is rapidly evolving as technology becomes more sophisticated and users adapt their expectations and perceptions of communication. Empathy is central to understanding and responding to emotions along with communicating sensitive supportive messages. Questions surrounding the role of empathy when communicating with artificial intelligence, however, have yielded mixed and interesting conclusions. This chapter discusses perspectives on empathy, supportive communication, and the influence of machine agents in the process of communicating sensitive, empathic messages. We consider whether machine agents can simulate empathy and convey sensitive support or merely produce reactive content that might simply appear empathic. This chapter offers insights into the relationship between empathy and supportive interactions with artificial intelligence.

Recent advances in large language models (LLMs) have enabled artificial systems to generate responses widely perceived as empathic, raising fundamental questions about the nature and value of empathy. This chapter examines why, despite evidence that AI can equal or even surpass humans in tasks of cognitive empathy, human empathy remains uniquely valued by recipients. We review empirical findings comparing human and AI empathic responses, highlighting the critical role of perceived source, effort, motivation, and trust in shaping recipients’ experiences. We argue that the value of empathy depends on the type of empathy sought – cognitive, affective, or motivational – and that AI may be well suited for some empathic functions but not others. Finally, we consider the broader societal implications of artificial empathy, including cultural variation, trust, skill erosion, and the future of human relationships, outlining key open questions for research on empathy in an AI-mediated world.

The notion of “AI empathy” has been criticized on the grounds that AI lacks the capacity to authentically share in the experience of the empathy recipient, or to be selective in its choice to empathize. In this chapter, we challenge the argument that human empathy is inherently or universally superior to AI empathy and suggest that a more fruitful way to frame this debate is in terms of the possible trade-offs between what human and AI sources of empathy can offer in particular situations. Emphasizing the empathy recipient’s perspective, we explore possible dimensions in which each empathizer source may excel, shaping the recipient’s possible preference for one source over the other. Drawing upon different ethical frameworks, we then discuss the possible normative consequences of whether one ought to accept or avoid AI empathy, given these trade-offs.

A wealth of studies suggest that children anthropomorphize artificial intelligence (e.g., voice assistants, humanoid robots) to a greater degree than do adults, suggesting that children may recognize at least some forms of artificial intelligence as meaningful "social" partners. Adults, however, seem more divided on this premise. For example, adults sometimes admit to feeling "more heard" by AI-generated empathic responses than human-generated empathic responses – but only when they believe the empathy came from a human. Would children similarly reject "artificial empathy"? In this chapter, I review developmental evidence that suggests otherwise: Today’s children (and future children) may very well accept artificial empathy as on par with empathy from humans. This may even come to affect how future generations value empathy in human-human interaction.

Psychological theories of empathy have traditionally focused on the empathizer’s ability to perspective take and share another’s emotions. But people today live in a day and age where AI models, like Large Language Models (LLMs) devoid of emotions and other empathic capabilities, can still produce text that makes people feel heard and comforted. Surely what people perceive in these artificially-generated responses still amounts to some form of “empathy”? To better understand this, we motivate a focus on received empathy. How empathy is received is impacted by many factors, such as recipients’ concern about maintaining self-esteem, that are nevertheless not considered by traditional psychological theories of empathy. Most importantly, the existence of "empathy" expressed by these artificial agents – what we call "LLMpathy" – allows us to shift the discussion away from givers of empathy toward empathy as perceived by the recipient. We consider the implications of receiving LLMpathy from these agents.

Artificial Intelligence systems increasingly match or surpass human performance across a wide range of cognitive tasks. Recent advances in large language models have yielded conversational abilities often indistinguishable from human dialogue. In this chapter, we argue that whether artificial agents should be considered empathetic depends on how empathy itself is defined and how it relates to subjective experience. We show that different philosophical schools produce distinct, and often incompatible, answers to the question of whether such a system should be considered genuinely empathetic. On one hand, empathy is a functional trait that can be fully characterized by observable behavior; on the other hand, empathy is inseparable from subjective experience and conscious feeling. Drawing on the philosophical notion of the “zombie” we introduce a thought experiment involving a hypothetical chatbot that exhibits perfect empathic behavior across all conceivable benchmarks while lacking any subjective experience. This framing allows us to disentangle functional performance from conceptual attribution and to examine whether empathy judgments depend on observable behavior alone or on assumptions about inner experience. As artificial agents become increasingly integrated into emotionally contexts, understanding how and why humans attribute empathy to machines becomes a pressing scientific, ethical, and societal question.

If artificial agents can behave empathically, could this contribute to the development of more ethical AI systems? This question becomes especially important and timely with the speed of development and increasing use-cases in AI, has been raising concerns about their ethical use. In this chapter, I tried to give a nuanced exploration of different components of computational empathy, and the opportunities and challenges they present on different stages of the moral decision-making process. Rather than offering a simple “yes” or “no” answer, it suggests that the integration of empathy into AI systems depends on which components are employed and when. The chapter reviews existing research on empathy’s connection to moral and ethical behavior in humans, followed by a critical examination of how these insights could be applied to artificial agents.

Chapter 2 lays the technological foundations for the analysis developed in the rest of the book. In contrast to Chapter 1, which is largely AI-free, this chapter is deliberately entrepreneurship-light, focusing instead on the technological and socio-economic features of artificial intelligence. After a brief introductory overview of AI, the chapter provides a structured and accessible representation of deep learning technologies, explaining how they operate, what their core capabilities are, and where their main limitations lie. The chapter presents AI as a digital ‘statistical machine’ and examines the inferential architectures underlying contemporary deep learning systems, with particular attention to data-driven learning effects and feedback loops. Emphasis is placed on the socio-economic implications of these technologies for the functioning of markets, including issues related to data, infrastructures, and computational resources. Building on this foundation, the chapter then introduces the functional capabilities enabled by AI – predictive, generative, and agentic – and discusses how they are deployed within the economic system, illustrating the structure of AI supply chains. By isolating the technological dimension from entrepreneurial dynamics, the chapter provides a clear analytical baseline for understanding how AI reshapes economic environments before turning explicitly to entrepreneurship in subsequent chapters.

The conclusions argue that artificial intelligence, despite its disruptive nature, does not undermine the essential role of innovative entrepreneurship within capitalism. While AI reshapes entrepreneurial processes and competitive dynamics, the persistence of entrepreneurial space depends primarily on institutional arrangements and policy choices. Preserving a dynamic and socially valuable form of capitalism in the AI era ultimately hinges on human agency rather than on technology itself.

This brief introduction outlines the premises and fundamental questions that the book asks and then explains its structure.

This chapter argues that processes of racialisation have been radically transformed by large-scale data-processing techniques that use algorithmic sorting, identification, and classification. These techniques employ what we call “proxy logics”: using implicit forms of feedback and complex statistical transformations on data to sort individuals into contingent groupings or clusters. In these instances, race is ascribed not through modes of identification grounded in the body (via phenotype or genotype) or through cultural ascriptions (self-identification and processes of belonging), but by latent associations between data features (i.e. proxy indicators). To cluster individuals into groups, such systems don’t discern our “identity.” Rather, they draw inferences and approximations, creating clusters that group people together not on “common characteristics, or meeting specified criteria” but on “a spatialised proximity or distance” (Amoore, 2021, p. 6). We contend that these new techniques for constituting race require a new politics of identity: a “politics of proximity.”

In Chapter 8, the ease of weaponizing technology is discussed, using the example of machine learning and general AI. The examination applies the lens of technological determinism to the growth of public–private efforts to develop AI that can harm and kill humans. Examples are provided of promising litigation in the US that would allow for attaching civil and criminal liability to human actors responsible for the conduct of AI.

Algorithmic governmentality at the border has gained attention in recent years on debates around migration. Within these debates, algorithmic risks and limitations have triggered contestations and mobilised campaigns given the harmful consequences they generate at the racialised border, such as automated bias and discrimination, privacy or surveillance. Yet this chapter invites us to critically interrogate how algorithmic governmentality at the border invisibilises, fosters and recrafts these racialised mechanisms. By theoretically building on Foucault’s repressive hypothesis about power relations, we analyse two case studies that show how algorithmic governmentality has transformed the ways in which power is exercised and resistance is enacted at the border. This chapter concludes that in order to truly engage with a critical perspective on algorithmic governmentality, we should start from the consideration that the border still enacts racialised mechanisms.

This chapter examines the foundations and limits of computation and their implications for artificial intelligence. It begins with the Entscheidungsproblem and Alan Turing’s formulation of the universal Turing machine, introducing the idea that computation itself can be studied independently of any physical machine. The chapter explains the halting problem and related undecidability results, showing that some well-defined questions cannot be answered by any algorithm, even in principle. It then turns to computational complexity, distinguishing between problems that are solvable in theory and those that are solvable efficiently, and introduces the P versus NP problem as a central open question. The chapter explores how computational hardness can function not only as a limitation but also as a resource, underpinning modern cryptography through one-way functions and key-exchange protocols. Finally, the discussion turns to artificial intelligence, contrasting early rule-based approaches with modern machine learning and neural networks. Recent advances in deep learning and large language models are presented as the culmination of decades of theoretical insight, highlighting both their remarkable capabilities and the enduring structural limits imposed by computation itself.

The epilogue reflects on the unifying themes that emerge across the scientific ideas explored in this book and considers their implications for the future of knowledge. A central lesson is that modern science has revealed fundamental limits – to prediction, to certainty, and even to what can be known in principle. From Gödel’s incompleteness theorems and Turing’s limits of computation to quantum uncertainty and chaotic dynamics, uncertainty appears not as a failure of understanding but as a structural feature of reality. A second recurring theme is that extraordinary complexity can arise from simple underlying rules, whether in computation, physics, biology, or planetary processes. Looking forward, the epilogue considers the growing role of artificial intelligence in scientific discovery, suggesting that future insights may increasingly be generated by systems whose internal reasoning is difficult for humans to interpret, even when their results can be verified. The book concludes by arguing that making sense of modern science requires embracing both humility and curiosity, as humanity navigates a future shaped by powerful ideas, computational tools, and enduring uncertainty.