Artificial Intelligence Timeline

How to use the timeline

The timeline above reads from left to right, from the past to today. Here is how to get the most out of it:

• Zoom in on a period to spread out the points when they crowd together, as they do in recent years.
• Filter by domain (language, vision, games, biology…) to follow one story at a time.
• Filter by organization to see which teams, academic or corporate, produced which models.
• Click a point: a detail card opens with the available information on the model, such as its date, the organization behind it and its domain.

Milestones added by hand by our editorial team stand apart from the models in the dataset. They act as landmarks so you never lose your bearings.

The major eras of AI

Symbolic beginnings

In 1950, mathematician Alan Turing published a paper proposing an “imitation game,” now known as the Turing test: can a machine hold a conversation well enough that you cannot tell it from a human? In 1956, the Dartmouth workshop brought together researchers who gave the field its name and its agenda.

Early approaches were symbolic: people wrote rules and logical reasoning by hand for the computer to apply. That worked well on narrow, well-defined problems, far less so in the messy real world.

The AI winters

More than once, promises outran what the technology could do. Funding shrank and interest faded: these were the AI winters. Between them, expert systems, which encode specialists’ knowledge as rules, enjoyed real success in business before hitting their limits: they were expensive to maintain and hard to adapt.

Machine learning

Gradually, another idea took over: instead of writing the rules, let the machine learn them from examples. That is machine learning. In 1997, IBM’s Deep Blue computer beat world chess champion Garry Kasparov. Its strength came mostly from raw computing power and searching through moves, but the moment stuck: a machine could outperform a human at a task seen as intellectual.

Deep learning

Deep learning uses artificial neural networks with many layers, trained on large amounts of data with ever more powerful processors. It transformed image and speech recognition. In 2016, DeepMind’s AlphaGo beat Lee Sedol, one of the world’s best Go players, in a game long thought out of reach for machines. In 2017, the paper “Attention Is All You Need” introduced the Transformer, the architecture behind most of today’s large language models.

Generative AI

Models learned to create text, images, sound and code. In November 2022, OpenAI’s launch of ChatGPT put this technology in the hands of the general public and set off a wave of chat assistants.

The rise of agents

The latest chapter: AI agents, systems that no longer just answer but chain actions together to reach a goal (searching, comparing, filling in a form, writing and then testing code). This is the heart of this site, and a period still taking shape: capabilities are growing, and the limits are coming into focus too.

What “notable model” means

The points on the timeline come from Epoch AI’s “Data on AI Models” database, which tracks AI models back to the field’s origins. Among them, Epoch AI flags certain models as notable. According to its documentation, these include models that advanced the state of the art, are highly cited in research, have historical significance or are widely used.

This is an expert judgment, not an absolute truth. It mostly reflects what has been published and documented: models kept secret or poorly described are less likely to appear. To see how the site uses this data, read our methodology.

Go further

If you are new to the topic, the Start track covers the basics step by step, and the AI glossary explains the terms you will meet on the timeline. To see how the number of models, computing power or investment have evolved, head to AI in numbers.