The History of Artificial Intelligence: From Dartmouth to GPT-4

The Birth of AI (1940s-1950s)

Early Foundations

The theoretical groundwork for AI was laid in the 1940s and 1950s:

• 1943: Warren McCulloch and Walter Pitts publish “A Logical Calculus of Ideas Immanent in Nervous Activity,” proposing the first mathematical model of a neural network
• 1950: Alan Turing publishes “Computing Machinery and Intelligence,” introducing the Turing Test
• 1956: The Dartmouth Conference, organized by John McCarthy, Marvin Minsky, Nathaniel Rochester, and Claude Shannon, coins the term “Artificial Intelligence”

The Golden Age (1956-1974)

Early AI research made rapid progress:

• 1958: Frank Rosenblatt creates the Perceptron, an early neural network
• 1964: Joseph Weizenbaum develops ELIZA, a natural language processing program
• 1966: Shakey the Robot, the first general-purpose mobile robot
• 1969: Minsky and Papert publish “Perceptrons,” highlighting limitations of single-layer networks

First AI Winter (1974-1980)

Funding and interest in AI declined due to:

• Failure to deliver on early promises
• Limited computational power
• The Lighthill Report (UK) criticizing AI research progress
• Collapse of DARPA funding

Expert Systems Era (1980-1987)

AI experienced a resurgence with expert systems — rule-based programs that encoded human expertise:

• 1980: XCON, a production rule system, saves DEC $40M annually
• 1981: Japan launches the Fifth Generation Computer project
• 1986: Backpropagation algorithm popularized by Rumelhart, Hinton, and Williams

Second AI Winter (1987-1993)

Another downturn followed:

• Collapse of the Lisp machine market
• Expert systems proved brittle and hard to maintain
• Government funding dried up

The Rise of Machine Learning (1993-2012)

Key Milestones

• 1997: IBM’s Deep Blue defeats world chess champion Garry Kasparov
• 2006: Geoffrey Hinton coins “deep learning” and introduces greedy layer-wise pre-training
• 2009: ImageNet dataset released, enabling computer vision breakthroughs
• 2011: IBM Watson wins Jeopardy!

The Deep Learning Revolution (2012-Present)

2012: The AlexNet Moment

Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton win ImageNet with a deep convolutional neural network, dramatically outperforming traditional computer vision approaches.

2014-2017: Building Blocks

• 2014: Generative Adversarial Networks (GANs) by Ian Goodfellow
• 2014: DeepMind’s AlphaGo defeats European Go champion Fan Hui
• 2017: Google researchers publish “Attention Is All You Need,” introducing the transformer architecture

2018-2021: Scaling Up

• 2018: GPT-1 (117M parameters) by OpenAI
• 2019: GPT-2 (1.5B parameters), initially withheld due to safety concerns
• 2020: GPT-3 (175B parameters), demonstrating few-shot learning
• 2021: DALL-E for text-to-image generation

2022-2023: Mainstream Breakthrough

• 2022: ChatGPT launched, reaching 100M users in 2 months
• 2023: GPT-4 with multimodal capabilities
• 2023: Open-source models (LLaMA, Mistral) match proprietary quality

2024-2026: Current Era

• 2024: GPT-4o with real-time voice, image, and text
• 2024: Claude 3 Opus with 200K context window
• 2025: AI agents, multimodal models, on-device AI
• 2026: Continued rapid advancement in capability and accessibility

Key Lessons from AI History

1. Cycles of Hype and Disappointment: AI has experienced multiple boom-bust cycles. Current enthusiasm is unprecedented but warrants measured expectations.

2. Hardware Enables Progress: Each wave of AI advancement has been enabled by more powerful hardware — from early computers to GPUs to specialized AI chips.

3. Data is Critical: The deep learning revolution was fueled by the availability of large datasets (ImageNet, Common Crawl, the internet itself).

4. Open Research Drives Innovation: The transformer paper, PyTorch, and open-source models have accelerated progress enormously.

5. Commercial Applications Sustain Funding: Real-world applications (search, advertising, cloud services) provide the economic foundation for AI research.

What’s Next?

While predicting AI’s future is notoriously difficult, current trends suggest progress toward more capable, multimodal, and agentic systems. Understanding AI’s history helps contextualize both the remarkable achievements and the persistent challenges that remain.