Artificial intelligence (AI)

The term was coined by John McCarthy in the 1955 proposal for the 1956 Dartmouth Summer Research Project, which conjectured that every aspect of learning or intelligence could in principle be described precisely enough for a machine to simulate it. Alan Turing had already framed the question operationally in 1950 with the imitation game. From the start the field split into two traditions: symbolic AI (logic, search, knowledge representation, planning), and connectionist AI based on networks of simple units trained from data. Symbolic methods dominated until the 1980s, peaking with rule-based expert systems; their brittleness and maintenance cost, together with over-promising, contributed to two funding contractions usually called the AI winters (mid-1970s and late 1980s).

Russell and Norvig define the field around rational agents: systems that perceive an environment and act to maximise an expected performance measure. This framing covers search algorithms, constraint solvers, probabilistic reasoning such as Bayesian networks, planning, robotics and machine learning under one roof. Since the 2010s, machine learning and in particular deep learning has become the dominant technique, which is why everyday usage now treats AI and ML as near-synonyms, and more recently AI and large language models. The two are not interchangeable: a route planner using A* search or a tax-rules engine is AI in the textbook sense but involves no learning.

Legal definitions matter more than academic ones for compliance. Article 3(1) of the EU AI Act (Regulation (EU) 2024/1689) defines an AI system as a machine-based system designed to operate with varying levels of autonomy, that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers from the input it receives how to generate outputs such as predictions, content, recommendations or decisions that can influence physical or virtual environments. The wording is aligned with the OECD definition revised in 2023. The key criterion is the capability to infer; Recital 12 and the Commission's 2025 guidelines on the definition exclude systems based solely on rules defined by natural persons to execute operations automatically. ISO/IEC 22989:2022 provides the corresponding vocabulary used by the ISO/IEC 42001 management-system standard.

Two recurring misconceptions: first, the distinction between narrow AI (competence on a defined task) and artificial general intelligence has no agreed test, and capability claims should be evaluated per task with benchmarks rather than by the label. Second, the so-called AI effect: once a technique works reliably (optical character recognition, spam filtering, route finding) people stop calling it AI, so the term tends to refer to whatever is currently new. For procurement and risk work the useful question is always what the system infers, from which data, and what decision it feeds.