History of data, AI, and statistics The ideas behind the tools.
Technology can look inevitable after its dead ends are removed. These archives put the papers, people, disputes, and failed turns back into the record.
Every timeline links to original research and archival records. Attribution disputes and later interpretations are labeled.
Choose an archive- Published
- Last updated
- Research and curation
- Let's Data Science
- Standards
- Research methodCorrections
Open archives
Choose a field.
Start with the larger fields of artificial intelligence and machine learning, then follow focused archives on language models, retrieval and context engineering, generative and multimodal systems, or agents that plan and act. The collection also traces databases, SQL, statistics, and probability. Each archive is built record by record.
- Broad field archive Open archive
History of Artificial Intelligence
How intelligence became a machine problem.
Trace the programs, laboratories, arguments, failures, and public demonstrations that repeatedly changed what artificial intelligence meant.
- Records
- 52
- Sources
- 94
- Eras
- 6
- Span
- 1943–2025
Before AI1943–1955Search & symbols1956–1969Knowledge & robots1970–1987Uncertainty & agents1988–2009Games & scale2010–2017Models & governance2018–2025Three points in the record- Artificial intelligence receives a name
- Deep Blue defeats the world chess champion
- ChatGPT puts an instruction-following model into dialogue
- Technical lineage archive Open archive
History of Machine Learning
How machines learned.
Follow the papers, people, datasets, and arguments that connected formal learning theory, statistical methods, neural networks, and systems built at scale.
- Records
- 45
- Sources
- 57
- Eras
- 6
- Span
- 1943–2024
Foundations1943–1969Hidden structure1970–1989Generalization1995–2001Deep revival2006–2012Learning systems2013–2017Foundation systems2018–2024Three points in the record- The perceptron learns a boundary
- Backpropagation trains hidden features
- Attention replaces recurrence
- Language model archive Open archive
History of Large Language Models
How prediction became a general interface.
Follow the probability models, corpora, neural architectures, training methods, evaluations, and system boundaries that shaped modern language models.
- Records
- 45
- Sources
- 52
- Eras
- 6
- Span
- 1913–2026
Probability & language1913–1987Neural sequences1995–2016Transformer pretraining2017–2019Scale & instruction2020–2022Open & multimodal2023–2024Reasoning systems2025–Jul 2026Three points in the record- Markov tests dependence on the letters of a poem
- The Transformer routes context with attention alone
- DeepSeek-R1 publishes an open-weight reasoning pipeline
- Evidence and memory archive Open archive
History of RAG, AI Memory, and Context Engineering
How models learned to find, carry, and remember the right evidence.
Follow the retrieval methods, memory architectures, benchmarks, grounded generators, long-context studies, and agent systems behind modern context engineering.
- Records
- 72
- Sources
- 96
- Eras
- 6
- Span
- 1945–2026
Finding records1945–1957Ranking evidence1960–1999Learned retrieval2003–2019RAG takes shape2020–2021RAG systems2022–2024Context & memory2025–July 2026Three points in the record- Rare terms receive more weight through inverse document frequency
- Retrieval-Augmented Generation gives a name to the architecture
- Context engineering is popularized as a broader systems term
- Media model archive Open archive
History of Generative and Multimodal AI
How machines learned to make what they could once only recognize.
Follow the samplers, representations, datasets, generators, and multimodal systems that connected language with images, audio, video, and interactive worlds.
- Records
- 50
- Sources
- 122
- Eras
- 6
- Span
- 1953–2026
Sampling & latents1953–2013Neural media2014–2016Scale & diffusion2017–2021Text to mediaDec 2021–2022Native multimodality2023–2024Audiovisual systems2025–Jul 2026Three points in the record- A generator learns by trying to fool a discriminator
- Stable Diffusion makes capable weights downloadable
- Muse Image adds tools and self-refinement to generation
- Autonomous systems archive Open archive
History of AI Agents
How software learned to pursue goals and act.
Trace the feedback loops, planning systems, robots, agent societies, learning environments, tool interfaces, and control layers behind software that acts.
- Records
- 45
- Sources
- 77
- Eras
- 6
- Span
- 1948–2026
Feedback & plans1948–1971Agent architectures1975–1994Agent societies1995–2009Learning agents2013–2017Language & tools2019–2023Protocols & control2024–July 2026Three points in the record- Shakey links perception, planning, and physical action
- ReAct interleaves reasoning traces with environment actions
- Agent safety becomes a problem of controlling privileged insiders
- Data systems archive Open archive
History of Databases and SQL
How data became queryable.
Follow the machines, models, languages, recovery protocols, distributed systems, and open formats that changed how people store, query, and share data.
- Records
- 58
- Sources
- 105
- Eras
- 6
- Span
- 1890–2024
Machine records1890–1959Navigation1960–1969Relations & SQL1970–1985Standards & scale1986–1999Web scale2000–2011Cloud convergence2012–2024Three points in the record- Codd separates logical data from storage paths
- MapReduce packages distributed batch processing
- DuckDB stabilizes embedded analytical SQL
- Uncertainty and evidence archive Open archive
History of Statistics and Probability
How uncertainty became evidence.
Follow the games, population records, experiments, arguments, algorithms, and software that changed how people measure variation and reason from incomplete information.
- Records
- 66
- Sources
- 111
- Eras
- 6
- Span
- 1494–2021
Calculated chance1494–1812Population evidence1829–1898Designed inference1900–1937Decision & simulation1945–1964Models & resampling1970–1986Open computation1990–2021Three points in the record- Bayes and Price reason backward from observations
- Tests are designed around errors and power
- Efron lets a sample stand in for repeated experiments
Research method
Built as a record, not a recap.
A milestone earns its place when it changed a field and the claim can be traced to reliable evidence. The archive keeps the original contribution separate from later interpretation.
- Primary evidencePapers, books, reports, archives, and official records.
- Named contributorsPeople, roles, affiliations, and profile links.
- Claims with contextDisputed firsts, limits, and later retellings are marked.