1950
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.
Timeline of AI history
From thought experiment to operated AI system.
Twelve selected milestones trace the path from a research question to learned models, generative systems and tool-using agents. This is not a complete chronicle: it links to primary sources and keeps events separate from our interpretation.
Scroll, drag or use the arrow keys to explore.Follow the milestones from 1950 to today.
The foundations
1950–19971956
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.
1997
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.
Deep learning
2012–20162012
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.
2016
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.
Transformers
2017–20202017
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.
2020
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.
Generative AI
2022–20232022
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.
2023
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.
Tool-using systems
2024–20262024
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.
2025
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.
2026
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.
Which process deserves a robust evaluation?
We start with the process boundary, baseline, evaluation cases, human approvals and documented data flows — not an arbitrary agent demo.
Book a conversation