Artificial Intelligence - A Very Short Introduction

Margaret Boden, a founding figure in cognitive science, demystifies artificial intelligence by framing it as the project of making computers do the sorts of things minds can do. She maps the field's five major approaches—symbolic (GOFAI), neural networks, evolutionary computing, cellular automata, and dynamical systems—and explains the unifying idea of the 'virtual machine': the information-processing system that matters, not the hardware. Across chapters on general intelligence, language, creativity, emotion, neural networks, robotics, artificial life, the philosophy of mind, and the Singularity, Boden shows that AI has produced spectacular narrow successes while artificial general intelligence remains a distant Holy Grail. She argues that AI's greatest lesson is how astonishingly rich human minds actually are, and she replaces hype with careful, sceptical judgement about both AI's promise and its real present-day dangers. Anyone wanting a rigorous yet accessible orientation to the conceptual foundations and stakes of AI will find it here.

What it argues

Artificial Intelligence - A Very Short Introduction

Key ideas it contributes

• AI Method Selection — The choice among the major families of AI techniques—symbolic/GOFAI, neural networks, evolutionary programming, cellular automata, and dynamical systems—and their combination, tailored to the task at hand.
• Virtual Machine Sophistication — The informational power and structural richness of the virtual machine (the information-processing system) implemented, independent of the underlying hardware, including its layered and hybrid organization.
• Computational Resources — Available hardware power, memory, data availability, and growth following Moore's Law, which enable but do not by themselves guarantee advanced virtual machines.
• Tractability Strategies — Techniques for making problems solvable in practice—heuristics, planning, mathematical simplification, and efficient search ordering—that direct or shrink the search space.
• Knowledge Representation Quality — The adequacy of how a problem and its domain knowledge are encoded—rules, frames, semantic nets, logic, word-vectors, or distributed networks—for enabling the system to reason and act effectively.
• Biological Inspiration — The degree to which AI designs draw on real brains, organisms, evolution, and self-organization (e.g., situated robotics, GasNets, reaction-diffusion, neuromorphic computing).
• Learning Capacity — The system's ability to improve through supervised, unsupervised, reinforcement, or deep learning, including discovering multilevel representations from data.
• Relevance and Common-Sense Handling — The capacity to judge what is relevant, apply everyday knowledge, and avoid the frame problem—the central bottleneck distinguishing narrow systems from genuinely general intelligence.
• Architectural Hybridization — The integration of symbolic and connectionist (and other) processing within a single system to combine complementary strengths, including whole-mind architectures.
• Narrow-Task Performance — Measurable success of AI on specialized tasks such as game playing, image recognition, translation, planning, and expert advice, often matching or exceeding humans.
• Progress Toward General Intelligence — The degree to which systems approach integrated, broad, common-sense, human-level intelligence spanning reasoning, perception, language, creativity, and emotion.
• Societal Impact and Risk — The real-world consequences of AI deployment, including productivity gains alongside unemployment, privacy erosion, autonomous weapons, and deceptive emotional companions.