The history of artificial intelligence includes different ideas about representing knowledge, learning from data, searching for solutions and evaluating performance. The selected events below are a reading guide, not an exhaustive chronology or a claim that a single invention created modern AI.
Selected research milestones
-
1950
Alan Turing and the imitation game
Turing published Computing Machinery and Intelligence, a foundational contribution to debate about machine intelligence. [22]
-
1956
The Dartmouth research project
John McCarthy helped organize the summer project associated with establishing artificial intelligence as a named field. The 1955 proposal and 1956 meeting are distinct events. [23]
-
1960s onward
Expert systems and represented knowledge
Systems such as DENDRAL explored the use of specialist knowledge. This strand of AI helps explain why the field is broader than today’s generative models. [27]
-
1997
Deep Blue and chess
IBM’s Deep Blue defeated world chess champion Garry Kasparov in a six-game match, a milestone in task-specific computer performance. [24]
-
2012
Deep neural networks and ImageNet
Krizhevsky, Sutskever and Hinton’s image-classification research became a prominent milestone in deep learning. [25]
-
2016
AlphaGo and Go
AlphaGo defeated Lee Sedol 4-1, combining learning and search in a challenging game domain. [26]
-
2017
The Transformer
Vaswani and colleagues introduced the architecture in Attention Is All You Need, a major contribution to sequence modeling. [6]
-
2020
Retrieval-augmented generation
Lewis and colleagues studied combining a generative model with retrieved information for knowledge-intensive language tasks. [7]
-
2025-2026
Adoption and infrastructure measurement
Surveys and statistical reports documented changing use, investment and resource demand. Their figures refer to different populations and periods. [13]
How to interpret progress in AI
A benchmark result and a general capability are different claims
A chess match supplies clear rules and a score. A language task may have several acceptable answers. An agent’s task can involve tools, changing information and uncertain stopping points. Comparing these settings requires more than arranging their most impressive results in chronological order. The test conditions, available resources and definition of success are part of the evidence.
Methods, systems and institutions evolve together
A research paper introduces or tests a method. A product combines methods with engineering choices. An institution supplies funding, infrastructure or standards. Keeping those roles separate makes the history more informative: a familiar product name should not displace the underlying research, and a research result should not be inflated into a claim about every application.
The original Transformer and retrieval-augmented-generation papers illustrate this distinction. They identify particular contributions that later systems can build on. Readers can use the source notes to examine those contributions directly, without assuming that every current system follows the same architecture or implementation. [6] [7]
Publication history of this reference
September 30, 2026 · Initial edition
Created the “What is AI?” article, Resources guide and this History page using the supplied wiki templates. Added an infobox, contents navigation, original explanatory examples, three data charts, a glossary and dated source notes.
Statistics and source descriptions belong to this edition. No earlier revisions, editors or review events are claimed.
This is a static reference package. Future editors can add a dated entry describing substantive corrections or updates. A publication record should distinguish a factual correction from a new observation, a changed definition or a design-only revision.
Return to the main AI article or consult the Resources guide for a reading path.