This chapter introduces the field of artificial intelligence (AI) through a review of its core themes, history, major research areas, and current trends. It gives a brief history of AI with an account of some of its relatively recent major accomplishments. These include knowledge-based expert systems, chess players, theorem provers, natural language processing, and a new killer application. There are almost a dozen distinct subfields of AI research, each with its own specialized journals, conferences, workshops, and so on. The chapter provides a concise account of the research interests in each of these subfields. Flourishing recent trends include soft computing, AI for data mining, agent-based AI, cognitive computing (including developmental robotics and artificial general intelligence), and the application of AI in cognitive science. Research into artificial intelligence is thriving as never before, and promises continuing contributions, both practical to engineering and theoretical to science.

Much work in artificial intelligence (AI) has built on concepts and theories developed by philosophers and logicians. This chapter introduces this foundational work, surveying different conceptions of AI, the philosophical dream of mechanizing human reasoning, the conceptual roots of AI, and the major theories of mind that have underpinned different strands of AI research. It discusses an area that serves as an exemplar of AI that is bound up with philosophy. The three principal philosophical criticisms of strong AI that helped to change the tide in the AI community and point to new research directions are the critique of Hubert Dreyfus, Block's critique of machine functionalism via the China brain thought experiments and Searle's Chinese Room thought experiment. Neural networks are capable of a certain type of learning; they can be trained to compute, or approximate, a target function.

Most contemporary philosophers would accept that the intelligence is provided by the material brains, and thus would be disinclined to challenge the possibility of artificially intelligent devices on the ground of their materiality. The questions and problems about artificial intelligence (AI) that remain can be divided into those that are largely independent of particular approaches to AI, and those that are prompted by more specific ideas about artificially realizable cognitive architectures. The chapter shows that questions about representations and their use play an important role in many challenges to AI. Connectionist devices are characterized by their rule for finding a set of connection weights that will yield patterns on their output units that are appropriate to each pattern on their input units. Dynamical systems theory (DST) views cognition as depending on a continuous interaction of a cognitive agent with its surroundings.

This chapter concerns the symbolic artificial intelligence (AI). Both technological and psychological AI employ the full range of AI methodologies, Good Old-Fashioned AI (GOFAI) included. The GOFAI methodology employs programmed instructions operating on formal symbolic representations. It is well suited to the binary, serial nature of the von Neumann digital computer. More accurately, evolutionary programming was initially developed within GOFAI, and is often used in a GOFAI context, although it can also be used for evolving connectionist networks. The major strengths of GOFAI are its abilities to model hierarchy and sequential order, to allow for precision in problem solving, and to represent specific propositional contents. GOFAI planning is even important in video games and Hollywood animation, for example, to prevent the virtual reality characters from bumping into one another. Invisibility is only one reason why GOFAI's successes go largely unsung.

This chapter reviews briefly the history of connectionist models, identifying major ideas and major areas of applications. It gives a brief review of major paradigms of learning in neural networks. The chapter addresses the issue of symbolic processing in connectionist models. It looks in particular at the following types of learning: supervised learning, unsupervised learning, and reinforcement learning. The connectionist revolution has spurred vigorous theoretical debates about the nature of cognition and the various approaches toward understanding. The chapter discusses the hybrid connectionist models, which incorporate both connectionist and symbolic processing methods. In contrast to connectionist implementationism, hybrid connectionist models may be considered a synthesis of connectionist models and traditional symbolic models. Hybrid models, such as CLARION, have been used to address a wide variety of issues in cognitive science and artificial intelligence, including human learning, reasoning, problem solving, creativity, motivational dynamics, metacognitive processes, and above all, human consciousness.

This chapter traces some of the history of the individual intellectual threads of situated activity, embodiment, and dynamics that underlie the situated, embodied, dynamical (SED) approach. It particularly focuses on the years 1985- 1995. The first intellectual thread making up the SED approach is situated activity. Situated activity stresses three ideas such as concrete action, situatedness, and interactionism that have been traditionally neglected in artificial intelligence (AI) and cognitive science. A second intellectual thread in the SED approach is embodiment. There are three distinct ideas such as physical embodiment, biological embodiment, and conceptual embodiment that have been advanced by advocates of embodied cognitive science. The final intellectual thread constituting the SED approach is dynamics, within which one must distinguish at least three ideas: dynamical systems theory (DST), dynamical framework, and dynamical hypothesis. The chapter articulates an integrated theoretical framework that combines the insights from situatedness, embodiment, and dynamics.

This chapter focuses on machine learning as a general way of thinking about the world, and provides a high-level characterization of the major goals of machine learning. Structural inference is the basis of many, and arguably most, machine learning frameworks and methods, including many well-known ones such as various forms of regression, neural-network learning algorithms such as back propagation, and causal learning algorithms using Bayesian networks. Machine learning algorithms must balance three factors: complexity of the learned model, which provides increased accuracy in representing the input dataset; generalizability of the learned model to new data, which enables the use of the model in novel contexts; and computational tractability of learning and using the model, which is a necessary precondition for the algorithms to have practical value. The practice of machine learning inevitably involves some human element to specify and control the algorithm, test various assumptions, and interpret the algorithm output.

This chapter focuses on visual perception, which is the dominant sense in humans and has been used from the first days of building artificial machines. It highlights the state of the art in computer vision methods that have been found to operate well and that led to the development of capabilities. The chapter summarizes the work structured into four key topics: object recognition and categorization, tracking and visual servoing, understanding human behavior, and contextual scene understanding. Scene geometry is an important intermediate representation in the interpretation process of an image. Object recognition can be seen as the challenge to determine the where and what of objects in a scene. Surveillance systems often work in two phases: a learning phase and a run-time phase. The chapter concludes with a critical assessment of what computer vision has achieved and what challenges remain.

Reasoning and decision making are fundamental parts of the knowledge representation and reasoning (KR&R) artificial intelligence (AI) approach. Research on the two topics of reasoning and decision making is often done in isolation, with different methods and different theoretical understandings for the two topics. The chapter distinguishes research along representation lines, taking particular aim at logic-based and probability-based representations. Research on representation languages that permit tractable query answering yielded specialized languages with large bodies of applications. The chapter describes work along these research paths, focusing on logical reasoning, probabilistic reasoning, and commonsense reasoning. Decision making as a research area spans the disciplines of economics, psychology, computer science, and virtually all the engineering disciplines. The chapter looks at approaches to autonomous decision making developed over the past fifty years. Research on commonsense reasoning is divided into three main streams: logical theory, large commonsense knowledge bases, and ad hoc commonsense reasoning techniques.

Language and communication are considered as relevant to artificial intelligence. Linguists are not the only scientists wishing to test theories of language functioning: so do psychologists and neurophysiologists. This chapter briefly looks at samples of important and prescient early work, and shows two contrasting, slightly later, approaches to the extraction of content, evaluation, representation, and the role of knowledge. It considers a range of systems embodying natural language processing (NLP)/computational linguistics (CL) aspects since the early seventies, and divides them by their relationships to linguistic systems and in relation to concepts normally taken as central to AI, namely logic, knowledge, and semantics. Broadly, statistical methods imply the use of only numerical, quantitatively based, methods for NLP/CL, rather than methods based on representations, whether those are assigned by humans or by computers. The chapter discusses the role of annotations to texts and the interpretability of core AI representations.

This chapter introduces and surveys the emerging agent-centered artificial intelligence (AI), and highlights the importance of developing theories of action, learning, and negotiation in multi-agent scenarios such as the Internet. It examines in detail the main functionalities software systems would display in a social, agent-centered AI or, in other words, the principles of behavior of the new AI. Approaches to multi-agent behavior differ mainly in regards to the degree of control that the designer should have over individual agents and over the social environment, that is, over the interaction mechanisms. Research on machine learning has been mostly independent of agent research and only recently has it received attention in connection with agents and multi-agent systems. Agent-based applications have enjoyed considerable success in manufacturing, process control, telecommunications systems, air traffic control, traffic and transportation management, information filtering and gathering, electronic commerce, business process management, entertainment, and medical care.

This chapter first focuses on artificial emotions, and then moves on to machine consciousness, reflecting the fact that emotions and consciousness have been treated independently and by different communities in artificial intelligence (AI). It reviews the philosophical perspectives of two pioneers in AI and philosophy of mind, Alan Turing and Hilary Putnam, respectively. The chapter discusses the philosophical implications of AI research on emotions and consciousness. Much research on the role of emotions in artificial agents has been motivated by an analysis of possible functional roles of emotions in natural systems. Work on emotions in AI can be roughly divided into two strands (with a small overlap): communicative aspects and architectural aspects. Emotion research has become an active interdisciplinary subfield in AI, and machine consciousness is on the verge of establishing a research community that pursues the design of conscious machines.

This chapter starts with a brief sketch of the history of robotics and then gives some background on traditional approaches. The central goal of classical industrial robotics is to move the end of an arm to a predetermined point in space. Control of classical industrial robots is often based on solutions to equations describing the inverse-kinematics problem. These usually rely on precise knowledge of the robot's mechanics and its environment. The chapter focuses on the classical approach to intelligent mobile robotics. An industrial robot's working environment is often carefully designed so that intricate sensory feedback is unnecessary; the robot performs its repetitive tasks in an accurate, efficient, but essentially unintelligent way. The chapter concentrates on two important and influential areas: evolutionary robotics and insect-inspired approaches to visual navigation. It outlines an important area of robotics that emerged at about the same time as behavior-based and biologically inspired approaches.

Artificial life research is mainly a scientific activity, but it also raises and illuminates certain philosophical questions. This chapter explains what artificial life is and how it is connected with artificial intelligence (AI). It briefly describes some of its representative scientific achievements. Some soft artificial life models focus on self-organization and study how structure can emerge from unstructured ensembles of initial conditions. The chapter discusses some associated philosophical issues involving emergence, creative evolution, the nature of life, the connection between life and mind, and the social and ethical implications of creating life from scratch. The science and engineering of artificial life impinges on a number of broad philosophical issues, including how life emerges from non-life, whether the evolution of life has a directional arrow, what life is, whether software systems could ever be literally alive, and what the social and ethical implications of creating artificial life are.

This chapter surveys some of the ethical challenges that may arise as one can create artificial intelligences (AI) of various kinds and degrees. Some challenges of machine ethics are much like many other challenges involved in designing machines. There is nearly universal agreement among modern AI professionals that artificial intelligence falls short of human capabilities in some critical sense, even though AI algorithms have beaten humans in many specific domains such as chess. In creating a superhuman chess player, the human programmers necessarily sacrificed their ability to predict Deep Blue's local, specific game behavior. A different set of ethical issues arises when one can contemplate the possibility that some future AI systems might be candidates for having moral status. One also has moral reasons to treat them in certain ways, and to refrain from treating them in certain other ways. Superintelligence may be achievable by increasing processing speed.