The motivations for acting and planning with probabilistic models are about handling uncertainty in a quantitative way, with optimal or near-optimal decisions. The future is never entirely and precisely predictable. Uncertainty can be due to exogenous events in the environment, from nature and other actors, to noisy sensing and information gathering actions, to possible failures and outcomes of imprecise or intrinsically nondeterministic actions. Models are necessarily incomplete. Knowledge about open environments is partial. Part of what may happen can be only be modeled with uncertainty. Even in closed predictable environments, complete deterministic models may be too complex to develop. The three chapters in Part III tackle acting, planning, and learning in a probabilistic setting.

This chapter is about planning techniques for solving MDP problems. It presents algorithms that seeks optimal or near-optimal solution policies for a domain. Most of the chapter is focused on indefinite-horizon goal reachability domains that have positive costs and a safe solution; they may have dead ends, but those are avoidable. The chapter presents dynamic programming algorithms, heuristics search methods and their heuristics, linear programming methods, and online and Monte Carlo tree search techniques.

Chapter 5 investigates “seamlessly” networked self-tracking tools as symbols of idealized professional mobility and looks to the Quantified Self (QS) as a forum that responds to and registers these business challenges and ambitions. Technologists tend to fetishize frictionless digital mobility. Conversations with digital professionals who participate in forums such as QS, however, indicate that the attractiveness of well-networked devices resonates less with the emerging realities of wearable technology or consumer “needs and wants” (a concern thematized in Chapter 3) than the ideals of lasting and sustainable tech sector careers that are otherwise punctuated by instability and breakdown. These are the additional entrepreneurial desires that motivate the making of self-tracking technology and become embedded in its design. QS also acts as a practical source of mutual aid that facilitates the desired connectivity and agility of working bodies. This chapter thus investigates QS as an interface that reconciles the technological fantasy and its repetitious recital with the difficulties tech executives face in their personal lives and professional work.

Chapter 3 details how business executives have interacted with the Quantified Self (QS) as a site that materializes a particular consumer “segment” and consumer “demand” in ways that accord with the binary and voyeuristic principles of consumer-centric design. QS offers visibility into ways technologists produce the distance they seek to see between themselves and their customers. However, the manner in which they interact with the forum also testifies to the involved role digital professionals frequently play in formulating consumer desire.

The participation of tech executives in collectives such as the Quantified Self (QS) belies their desire to occupy the position of a professional participant observer who is simply looking in on an emerging social scene as though from afar. Chapter 4 looks at the way technologists have leveraged QS to cultivate a professional identity of a digital devotee. In particular, it analyzes how the popular staging of QS as a space for private explorations of self-tracking makes it possible for technologists to recoup their business-driven engagements with the forum as hallmarks of personal – as well as of more general – passion for self-quantification, a display of which has become increasingly necessary for success in the tech sector. Innovation is often enough framed as a product of masculinized heroics and individual acts of daring. Examining QS as an instrument of professional development refocuses attention on the feminized modes of free and affective labor that continue to move the tech industry forward. As these chapters explore the forum both as a mechanism and as a mirror of these professional imperatives, they highlight the knottier role desire plays in the digital economy.

In the conclusion, I consider how the Quantified Self (QS) has evolved since I completed my research in 2017. The composition and social function of this collective have been partially reshaped by its original organizers who have continued to focus group activities on citizen science and academic research. Groups such as QS have also become affected by the COVID-19 pandemic, which has altered the nature and function of in-person and post-work socializing in the commercial sphere more broadly. Nevertheless, the industry practices, challenges, and promises refracted through the QS interface in this book remain germane as they speak to some of the central dynamics that continue to impact the self-tracking market, if now in a different guise.

In probabilistic models, an action can have several possible outcomes that are not equally likely; their distribution can be estimated relying on statistics of past observations. The purpose is to act optimally with respect to an optimization criterion of the estimated likelihood of action effects and their cost. The usual formal probabilistic models are Markov decision processes (MDPs). An MDP is a nondeterministic state-transition system with a probability distribution and a cost distribution. The probability distribution defines how likely it is to get to a state 𝑠′ when an action 𝑎 is performed in a state 𝑠. The chapter presents MDPs in flat then structured state-space representations. Section 8.3 covers modeling issues of a probabilistic domain with MDPs and variants such as the stochastic shortest path model (SSP) or the constrained MDP (C-MDP) model. Section 8.4 focuses on acting with MDPs. Partially observable MDPs and other extended models are discussed in Section 8.5.

This chapter introduces stochastic gradient MCMC (SG-MCMC) algorithms, designed to scale Bayesian inference to large datasets. Beginning with the unadjusted Langevin algorithm (ULA), it extends to more sophisticated methods such as stochastic gradient Langevin dynamics (SGLD). The chapter emphasises controlling the stochasticity in gradient estimators and explores the role of control variates in reducing variance. Convergence properties of SG-MCMC methods are analysed, with experiments demonstrating their performance in logistic regression and Bayesian neural networks. It concludes by outlining a general framework for SG-MCMC and offering practical guidance for efficient, scalable Bayesian learning.

This chapter is about a refinement acting engine (RAE) used on a hierarchical task-oriented representation. It relies on an expressive, general-purpose language that offers rich programming control structures for online decision-making. A collection of refinement methods describes alternative ways to handle tasks and react to events. A method can be any complex algorithm, decomposing a task into subtasks and primitive actions. Subtasks are refined recursively. Nondeterministic actions trigger sensory-motor procedures that query and change the world nondeterministically. We assume that the methods are manually specified and that RAE chooses the appropriate method for the task and context at hand heuristically.

The recent developments of large language models (LLMs) and their extension in multimodal foundation models have introduced new perspectives in AI. An LLM is basically a very large neural net trained as a statistical predictor of the likely continuation of a sequence of words. LLMs have excellent competencies over a broad set of NLP tasks. Additionally, LLMs demonstrate the emergence of deliberation capabilities for reasoning, common sense, problem solving, code writing, and planning. These abilities have not been designed for in LLMs. They are unexpected and remain to a large extent poorly understood. Although error-prone and imperfect, they open up promising perspectives for acting, planning, and learning, which are presented in this chapter.

This chapter is about domain-independent classical-planning algorithms, which until recently were the most widely studied class of AI planning algorithms. The chapter classifies and describes a variety of forward search, backward search, and plan-space planning algorithms, as well as heuristics for guiding the algorithms.

This chapter sets the foundation for the next two chapters. It introduces the reader to robotics platforms for the development of acting, planning, and learning functions. The study of motion is based on classical mechanics for the modeling of forces and their effects on mouvements. Robotics builds on this knowledge to master computational motion, navigation, and manipulation over different types of devices and environments. Robotic devices are informally introduced in the following section. Motion problems and the metric representations with continuous state variables needed for geometric, kinematic, and dynamic operational models are then presented. Section 20.3 introduces localization and navigation problems, followed by a section on manipulation problems and their representations.

This chapter is about representing HTN planning domains and solving HTN planning problems. Several of the formal definitions require the same "classical planning" restrictions as in Part I, but most practical HTN implementations loosen or drop several of these restrictions. We first discuss ways to represent and solve planning problems in which there is a totally ordered sequence of tasks to accomplish. We then generalize to allow partially ordered tasks and describe ways to combine classical planning and HTN planning. Finally, we briefly discuss heuristic functions, expressivity, and computational complexity.