From thought experiment to operated AI system.

Turing formulates the Imitation Game

Alan Turing examines whether machines can think and proposes an observable conversation experiment. The paper still shapes debate about machine intelligence.

Artificial intelligence becomes a research term

The Dartmouth proposal uses the phrase “artificial intelligence” for a planned research programme. It marks an institutional starting point for the field.

Deep Blue wins the rematch against Kasparov

IBM's chess system wins a match against reigning world champion Garry Kasparov. It relied on specialised search, evaluation functions and substantial compute — not today's language-model methods.

AlexNet accelerates the deep-learning shift

The published neural network achieves a decisive lead in the ImageNet competition. GPU training, large datasets and deep networks become central components of computer vision.

AlphaGo defeats Lee Sedol

DeepMind's system wins the closely watched Go match 4–1. The underlying work combines neural networks with search and reinforcement learning.

4–1in the match against Lee Sedol

Transformers rely on attention

“Attention Is All You Need” describes a sequence architecture without recurrent layers. The idea becomes an important foundation for later large language models.

GPT-3 demonstrates few-shot behaviour at scale

OpenAI documents a 175-billion-parameter autoregressive language model and studies tasks performed from examples in the prompt rather than task-specific training.

175bnparameters reported in the GPT-3 paper

ChatGPT brings dialogue models into broad use

OpenAI releases ChatGPT as a research preview. Its accessible dialogue interface lets a broad audience experience both the abilities and the limits of generative language models.

GPT-4 documents multimodal input

The technical report describes a model that can process image and text inputs and produce text. It publishes extensive evaluations, but not full architecture or training details.

MCP standardises tool and data access

Anthropic publishes the Model Context Protocol as an open standard for connecting AI applications to data sources and tools. A protocol alone does not solve permissions or quality assurance.

Agents become an organisational and operational problem

Microsoft's Work Trend Index describes a move from assistants towards human-agent teams. This is a vendor framework, not proof of maturity: production still requires evaluation, limited permissions, logging and clear human accountability.

Microsoft packages agent governance and Cowork

Microsoft describes Agent 365 as a control and governance layer and makes Copilot Cowork generally available to Microsoft 365 Copilot customers. Licensing, data flows, model providers and regional availability must be checked for each use case.