Transformer paper year?

2017 (Attention Is All You Need).

Reinforcement Learning from Human Feedback.

Machine Intelligence History

Artificial Intelligence Timeline

From logic theorists and perceptrons to expert systems, statistical learning, deep neural networks, transformers, and generative AI, machines learned to perceive and create.

1950s -> Present Symbolic -> Neural Generative AI

AI Invention and Ideas

Alan Turing posed machine intelligence tests; Dartmouth workshops named AI; cycles of symbolic reasoning, knowledge engines, and neural networks led to today's large language models.

Big picture

Artificial intelligence evolution in one view

AI alternated between rule-based systems and data-driven learning until scale, GPUs, and big data made deep networks practical for vision, speech, and language.

Related timelines: computer timeline, processor timeline, internet timeline.

1950

Turing paper

1969

Perceptron limits

2012

Deep learning surge

Now

Generative AI

Early AI 1950s - 1960s

Turing, Dartmouth, and Symbolic AI

Timeline of artificial intelligence evolution from early theory to modern systems — AI evolution: how the field grew from theory into programmable intelligent systems.

Early artificial intelligence research — Turing, Dartmouth, and symbolic methods — The AI beginning: Dartmouth optimism, Lisp, and rule-based problem solvers.

1943: McCulloch-Pitts neuron — first mathematical model of a neural network.
1950: Turing asks Can machines think?
1951: Strachey's checkers program — among the first AI game programs.
1956: Dartmouth conference coins AI.
1966: ELIZA — first chatbot simulating conversation.
Logic Theorist proves theorems.

Researchers believed human reasoning could be encoded with symbols and rules, inspiring optimism about general problem solvers on early computers.

Technology Used

Symbolic rules: If-then systems.
Search algorithms: State spaces.
Lisp: AI-friendly language.

Features

Lab prototypes: Tiny problems only.
High hopes: AI summer begins.
Limited data: No big datasets.

Expert Systems 1970s - 1980s

Expert Systems and AI Winters

MYCIN diagnoses bacterial infections.
1980: XCON (R1) — first major commercial expert system success at DEC.
Japan fifth-generation computer project.
Expert systems commercialized then plateau.

Encoded human expertise in rule bases for narrow domains, but maintenance costs and brittleness led to funding pullbacks known as AI winters (see detailed winter timeline below).

Technology Used

Rule engines: Knowledge bases.
Inference: Forward/backward chaining.
Specialized Lisp machines: AI hardware.

Features

Narrow success: Medicine, configuration.
Expensive updates: Rules stale fast.
Winter cycles: Reduced investment.

ML Rise 1990s - 2005

Machine Learning and Statistical Methods

Support vector machines excel on data.
Spam filters use naive Bayes.
1989: ALVINN — neural network steers an autonomous van.
1997: IBM Deep Blue beats chess champion.

Cheaper disks and sensors made data-driven pattern recognition practical, shifting many tasks from hand-written rules to trained models.

Technology Used

SVMs: Kernel methods.
Ensembles: Random forests later.
Feature engineering: Human-crafted inputs.

Features

Real products: Search ranking, finance.
Competitions: Kaggle culture later.
Hybrid AI: Rules plus models.

Neural Return 2006 - 2017

Deep Learning Revolution

Deep learning algorithms — neural networks, CNNs, and GPU-accelerated training — Deep learning algorithms: layered networks trained on big data with parallel GPUs.

ImageNet 2012 breakthrough with CNNs.
GPUs accelerate matrix training.
2011: Watson wins Jeopardy! against human champions.
2015: AlphaGo defeats Go champion (Fan Hui 5-0 earlier).
2017: AlphaZero masters chess, Go, and shogi from scratch.

Deep convolutional and recurrent networks learned hierarchies from pixels and sequences, beating classical features on vision and speech benchmarks.

Technology Used

CNNs: Vision layers.
GPUs: Parallel training.
Frameworks: TensorFlow, PyTorch.

Features

Big data: Labeled datasets scale.
Assistants: Voice on phones.
Auto translation: Neural MT.

Transformers 2018 - 2022

Transformers and Large Language Models

Large language and transformer models — BERT, GPT, and scaled pretraining — Transformer LLMs: self-attention models pretrained on massive text corpora.

Retrieval-augmented generation — grounding language models with documents — RAG: retrieve trusted sources, then generate answers grounded in real data.

Transformer architecture (2017) scales.
BERT and GPT families emerge.
2021-2022: DALL-E and Stable Diffusion spark text-to-image generation.
ChatGPT popularizes conversational AI.

Self-attention models trained on web-scale text enabled few-shot learning and fluent generation, sparking global debate on jobs, truth, and safety.

Technology Used

Self-attention: Token relationships.
Pretrain + finetune: Transfer learning.
RLHF: Human preference tuning.

Features

Copilots: Code and writing help.
Multimodal: Text+image models.
Policy focus: Safety guidelines.

Governance 2023 - Present

Generative AI at Scale and Governance

2023: GPT-4 and Gemini advance multimodal LLMs.
Open models and on-device inference grow.
EU AI Act and global standards debated.
Agents combine tools, memory, and planning.

Enterprises deploy retrieval-augmented generation while regulators demand transparency, copyright clarity, and evaluations for bias and hallucination risk.

Technology Used

RAG: Grounding with documents.
Small models: Edge inference.
Agent frameworks: Tool use loops.

Features

Enterprise adoption: Support bots.
Watermarking debates: Content authenticity.
Human oversight: Review remains critical.

Key Historical Events

Landmark moments that shaped AI beyond the main era cards — foundations, commercial breakthroughs, and modern generative systems.

Event	Year	Why it matters
McCulloch-Pitts neuron	1943	First mathematical model of a neural network (predates Turing)
ELIZA	1966	First chatbot; demonstrated human-like conversation
XCON (R1)	1980	First successful commercial expert system (saved DEC millions)
ALVINN autonomous van	1989	Early neural network for self-driving
Watson wins Jeopardy!	2011	IBM AI beats human champions in open-domain Q&A
AlphaZero	2017	Mastered chess, Go, and shogi without human game data
DALL-E / Stable Diffusion	2021-2022	Text-to-image generation explosion
GPT-4 / Gemini	2023	Multimodal large language models at scale

AI Winters Explained

Periods when hype, funding, and research interest in AI fell sharply after ambitious promises outpaced results.

AI Winter 1 (1974-1980)

UK Lighthill Report criticized AI progress and cut government funding.
Minsky and Papert exposed perceptron limits (e.g., XOR problem).
Research grants and corporate enthusiasm dropped across the field.

AI Winter 2 (1987-1993)

Collapse of the specialized Lisp machine market.
Japan's Fifth Generation Computer Project failed to meet expectations.
Expert systems proved too expensive to maintain at scale.

AI “Firsts” at a Glance

First	Year	Achievement
First AI program	1951	Checkers-playing program (Strachey)
First industrial robot	1961	Unimate at GM plant
First autonomous car	1986	Mercedes-Benz van (~80 mph on empty roads)
First AI to beat human pro at Go	2015	AlphaGo Fan (5-0 vs Fan Hui)
First AI-generated art sold at auction	2018	“Edmond de Belamy” ($432,500)

Decade-by-Decade Visual Timeline

A quick scan of how AI milestones cluster across decades.

1940s McCulloch-Pitts neuron
1950s Turing test, Dartmouth AI
1960s ELIZA chatbot
1970s AI Winter 1 begins
1980s XCON expert system
1990s AI Winter 2, Deep Blue
2000s Statistical ML rise
2010s ImageNet, AlphaGo
2020s ChatGPT, GPT-4, EU AI Act

AI Pioneers and Contributors

Person	Contribution
Alan Turing	Turing Test; theoretical foundations of computation
John McCarthy	Coined “Artificial Intelligence”; Lisp
Marvin Minsky	Neural networks; Society of Mind
Frank Rosenblatt	Perceptron
Geoffrey Hinton	Backpropagation; deep learning revival
Yann LeCun	Convolutional neural networks
Yoshua Bengio	Deep learning sequence modeling
Fei-Fei Li	ImageNet dataset
Ilya Sutskever	Sequence-to-sequence learning; GPT development

Funding and Investment Waves

AI progress often tracks capital available for research, startups, and infrastructure — winters align with downturns.

Global artificial intelligence investment growth from expert systems boom to generative AI funding — AI investment waves: booms, winters, and the surge in venture and corporate AI spending.

Period	Global AI investment (approx.)
1980s	$1B+ (expert systems boom)
1990s	Decline (AI winter)
2010	~$2B
2015	~$12B
2020	~$68B
2023	$200B+

Hardware Evolution for AI

Breakthroughs in AI required matching gains in compute — from vacuum tubes to GPUs built for parallel math.

Era	Hardware	Compute power (approx.)
1950s	Vacuum tubes	~1,000 FLOPS
1960s	Transistors	~1M FLOPS
1970s	Microprocessors	~10M FLOPS
1990s	Early GPUs	~1B FLOPS
2012	GTX 580 (AlexNet era)	~1.5 TFLOPS
2024	NVIDIA H100 GPU	~1,979 TFLOPS

Ethics and Regulatory Timeline

Year	Event
2016	Asilomar AI Principles drafted
2018	GDPR includes “right to explanation”
2021	UNESCO AI ethics recommendation
2023	EU AI Act passes
2024	US Executive Order on AI Safety
2025	China deep synthesis regulations strengthened

Future Predictions (Speculative)

Forward-looking themes discussed by researchers and industry — not certainties, but directions the field is exploring.

2025-2027: AI agents with long-term memory and tool orchestration.
2028-2030: First controversial AGI capability claims.
2030-2035: AI-assisted scientific discovery (materials, drugs).
2040s: Human-AI brain interfaces moving toward mainstream medicine.