How far has
intelligence come?
From early rules to systems that converse and act.
A history of intelligence, still unfolding.
Moments that changed the directionMoments that changed the direction
Point to a year to preview · Click for detailsTap a year for details and sources
The illustration is unavailable. The history remains readable.
When learning meets the game
AlphaGo defeated Lee Sedol 4–1.
A breakthrough in a specialized task is not general intelligence.
Explore details & sources
What happened
AlphaGo defeated Lee Sedol 4–1.
What changed
Neural networks, reinforcement learning and search helped navigate a vast game space.
What remains unresolved Editorial interpretation
A breakthrough in a specialized task is not general intelligence.
Selected milestones, not a complete history. Different approaches coexist; progress has never been a straight line.
Original visual metaphors, not historical artifacts, actual game positions or experimental data.
All milestones & sources
1950A question becomes a test
What happened
Turing published Computing Machinery and Intelligence, proposing the imitation game.
What changed
From an abstract debate to observable behavior.
What remains unresolved Editorial interpretation
Human-like conversation does not establish consciousness.
1956A field takes shape
What happened
Following a 1955 proposal, the Dartmouth summer research project brought researchers together around artificial intelligence in 1956.
What changed
Machine intelligence became a shared research agenda.
What remains unresolved Editorial interpretation
Naming an ambition did not solve it.
1986Learning internal representations
What happened
Rumelhart, Hinton and Williams published influential work on learning representations through backpropagation.
What changed
Networks could adjust internal features using training errors.
What remains unresolved Editorial interpretation
Learning from examples did not establish open-ended general intelligence.
1997A specialist beats the champion
What happened
IBM Deep Blue defeated Garry Kasparov in a six-game chess match.
What changed
A machine surpassed the reigning world champion in a tightly defined domain.
What remains unresolved Editorial interpretation
Chess expertise did not generalize to everyday tasks.
2012Learning to recognize
What happened
AlexNet demonstrated the strength of deep convolutional networks in the ImageNet challenge.
What changed
Learned visual features advanced image classification.
What remains unresolved Editorial interpretation
Classifying images is not a complete understanding of the world.
2016When learning meets the game
What happened
AlphaGo defeated Lee Sedol 4–1.
What changed
Neural networks, reinforcement learning and search helped navigate a vast game space.
What remains unresolved Editorial interpretation
A breakthrough in a specialized task is not general intelligence.
2017Attention changes the architecture
What happened
The Transformer introduced an attention-based architecture for sequence modeling.
What changed
It provided a new foundation for later large-scale language models.
What remains unresolved Editorial interpretation
Fluent language is not sufficient evidence of general intelligence.
2022Conversation becomes an interface
What happened
OpenAI released ChatGPT as a public research preview on November 30, 2022.
What changed
People could explore a language model directly through conversation.
What remains unresolved Editorial interpretation
Helpful-sounding answers can still be wrong.
2024More modalities. Deeper reasoning.
What happened
Releases such as GPT-4o and o1-preview separately explored multimodal interaction and additional computation for reasoning.
What changed
Text, audio and image interfaces developed alongside reasoning-oriented models.
What remains unresolved Editorial interpretation
Higher test scores do not guarantee reliability on unfamiliar tasks.
2025From answers to actions
What happened
Operator launched as a research preview of an agent that could perform tasks through a browser.
What changed
A model could operate software, not only produce a reply.
What remains unresolved Editorial interpretation
Tool use does not establish dependable autonomy.
The next chapter
is still open.
An outlook, not an arrival timetable.
Can knowledge transfer to unfamiliar tasks?
Can systems keep learning and adapting?
Can they act reliably, under human oversight?
Broader capability is not the same as reliable general intelligence.
AGI: artificial general intelligence. Its definition and evaluation remain under discussion.
Make intelligence understandable. Give learners more agency.
A student-focused Agent is being prepared, exploring support for learning under human oversight. It is not yet available.