What Is Artificial Intelligence?

Artificial intelligence (AI) is a branch of computer science concerned with the design and study of computational systems that exhibit behaviors associated with human intelligence, including learning, reasoning, perception, language understanding, and decision-making. The field traces its formal origins to the 1956 Dartmouth Conference, where researchers first articulated the goal of building machines capable of simulating any aspect of human cognition. Over the following seven decades, AI has expanded from rule-based symbolic programs into a broad discipline encompassing machine learning, probabilistic reasoning, and large-scale neural network models.

AI draws on mathematics, logic, statistics, linguistics, neuroscience, and engineering. Its methods range from formal theorem provers and constraint satisfaction algorithms to deep neural networks trained on billions of examples. This breadth makes AI both a foundational discipline and an enabling technology for dozens of applied fields.

Machine Learning and Neural Networks

Machine learning is the sub-discipline of AI in which systems improve their performance on a task by exposure to data rather than through explicit programming. Artificial neural networks are the dominant architecture for machine learning today: layered arrangements of weighted computational units, loosely modeled on biological synapses and neurons, that learn internal representations by adjusting weights through backpropagation. Feedforward networks, including deep convolutional networks, process inputs in a single forward pass and have achieved benchmark performance on image classification, speech recognition, and language generation tasks. Generative adversarial networks (GANs), introduced by Goodfellow et al. in 2014, pit two networks against each other in a minimax game, enabling synthesis of highly realistic images, audio, and video.

As reviewed in the IEEE Xplore survey of deep learning algorithms and architectures, the decade from 2012 to 2022 saw repeated breakthroughs in visual recognition, natural language processing, and protein structure prediction, driven largely by scale: larger datasets, more parameters, and greater computational resources.

Reasoning, Knowledge, and Expert Systems

Before the machine-learning era, much of AI research focused on symbolic approaches: encoding domain knowledge explicitly in rules, ontologies, and logical frameworks. Expert systems, which emerged in the 1970s and 1980s, encoded human specialist knowledge as if-then production rules and used inference engines to answer queries in domains such as medical diagnosis and chemical analysis. While early systems such as MYCIN and DENDRAL demonstrated the viability of knowledge-driven AI, they required enormous manual effort to maintain and struggled to generalize beyond their encoded domains.

The Semantic Web initiative extends this tradition into distributed computing, representing knowledge in machine-readable formats that allow automated agents to query and integrate information across disparate data sources. Minimax search, used in game-playing systems, and Bayesian prediction methods, which reason under measurement uncertainty by maintaining probability distributions over possible states, represent other strands of the reasoning tradition that remain relevant in planning and decision support.

Autonomous and Adaptive Agents

Software agents are programs that perceive their environment, pursue goals, and act with some degree of autonomy. When embodied in physical systems, they become the control cores of autonomous vehicles, industrial robots, and unmanned aerial systems. Reinforcement learning, a branch of machine learning in which agents learn by receiving rewards and penalties from their environment, has produced notable results in robot locomotion, board game mastery, and real-time strategy, areas reviewed in research on computational intelligence published by PMC/NIH. Neurocontrollers, which use neural networks as feedback controllers, bridge AI and control engineering by learning stable policies for systems too complex to model analytically.

Pervasive computing environments, in which sensors and networked devices generate continuous streams of data, create both a challenge and an opportunity for AI agents, as decisions must be made quickly with noisy, incomplete inputs. This has made data mining and anomaly detection core competencies for deployed intelligent systems.

McKinsey's analysis of what AI is and how it is used provides a practitioner-oriented survey of current deployment patterns across industries.

Applications

Artificial intelligence has applications in a wide range of fields, including:

• Autonomous vehicles and advanced driver assistance systems
• Medical imaging, clinical decision support, and drug discovery
• Natural language processing, machine translation, and conversational interfaces
• Data mining and fraud detection in financial systems
• Robot learning and adaptive control in industrial automation
• Scientific research acceleration in genomics, materials science, and climate modeling