Artificial intelligence

Official definitions describe an AI system by what it does with its input. The OECD Council adopted this definition: “An AI system is a machine-based system 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 EU AI Act uses nearly the same wording and explains that inference covers producing such outputs and deriving models or algorithms from inputs or data. It also says the definition does not cover systems that only execute rules written entirely by people.

Arvind Narayanan and Sayash Kapoor call AI “an umbrella term for a set of loosely related technologies” and point out that ChatGPT “has little in common with, say, software that banks use to evaluate loan applicants” (AI Snake Oil, 2024). This glossary separates four kinds of AI by what they do. Predictive AI estimates an unknown value or future outcome from data, for example a credit score or a price forecast. Computer vision interprets images and video, for example by reading the amount written on a photographed check. Generative AI produces new text, images or code, for example a chatbot’s answer or a draft trading script. Agentic AI pursues a goal over several steps by choosing and taking actions with tools, for example an AI agent that pulls prices, runs a backtest and then places an order. Machine learning is not a fifth kind: it is the method most of these systems are built on, in which software learns its behavior from examples instead of following rules a person wrote. The kinds also combine; an AI agent usually uses a large language model, which is generative AI, to choose its next step.

The term comes from A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence by J. McCarthy, M. L. Minsky, N. Rochester and C.E. Shannon, dated 31 August 1955. It proposed a two-month study by ten researchers at Dartmouth College in the summer of 1956, “on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it.” Nature’s obituary of McCarthy (26 October 2011) credits him with coining the term and reports Stanford’s view that the proposal was its first use in publication. McCarthy later defined AI as “the science and engineering of making intelligent machines, especially intelligent computer programs” (What is artificial intelligence?, revised 12 November 2007). The OECD Council adopted the definition quoted above on 8 November 2023, in part to make clear that it covers generative AI systems, which produce content (OECD Recommendation on Artificial Intelligence; current text on OECD.AI). The EU AI Act’s version is in Article 3(1), and its explanation of inference is in Recital 12.