The History of AI, Part 1: When Machines Learned to See and Play (2012–2017)

The Starting Gun: Deep Learning Becomes Real

The AI revolution didn't start with ChatGPT. It started quietly – in research labs and at conferences that hardly anyone outside the tech bubble knew about. But between 2012 and 2017, the foundations were laid on which everything is built today.

2012: AlexNet and the ImageNet Moment

In September 2012, a neural network called AlexNet won the ImageNet competition – and not by a narrow margin, but with such a dramatic lead that it shook the entire computer vision community. The error rate dropped from 26% to 16%.

What was new? AlexNet used GPUs to train deep neural networks. What previously took weeks now took days. Deep learning was suddenly no longer theory, but practice.

Why This Mattered

• Proved that deep neural networks work
• Established GPUs as training hardware
• Triggered billions in AI research investment

2014–2015: GANs and the Creative Machine

Ian Goodfellow introduced Generative Adversarial Networks (GANs) in 2014 – two neural networks playing against each other. One generates images, the other evaluates them. The result: machines that appeared creative for the first time.

The first GAN images were blurry and eerie. But the concept was groundbreaking – and laid the foundation for everything that later came with DALL-E, Midjourney, and Stable Diffusion.

2016: AlphaGo Defeats the World Champion

In March 2016, Google's AlphaGo defeated Go world champion Lee Sedol. This wasn't an ordinary computer victory over a human. Go was considered too complex for brute-force computation – it has more possible positions than atoms in the universe.

AlphaGo used a combination of deep learning and reinforcement learning. In Game 2, the AI made a move (Move 37) that no human player would ever have made – and won with it. It was the moment when it became clear: AI can't just calculate, it can simulate intuition.

"After humanity spent thousands of years refining the game of Go, the machine comes along and says: actually, you've been playing it wrong." – Fan Hui, European Go champion

2017: Attention Is All You Need

In June 2017, a Google team published the paper "Attention Is All You Need" – introducing the Transformer architecture. No other research paper has changed the world as much since then.

What Makes Transformers Special?

Before (RNNs/LSTMs)	Transformer
Sequential processing	Parallel processing
Slow training	Fast training on GPUs
Forgets in long texts	Attention across the entire text
Limited scaling	Scales with more data & compute

Transformers are the architecture behind GPT, BERT, Claude, Gemini, LLaMA and virtually every modern language model. Without this paper, there would be no ChatGPT.

What We Learn from This Era

The years 2012–2017 were the foundational research phase. Few outside of research suspected what was brewing. But three patterns emerged:

1. Hardware drives progress – GPUs made deep learning possible in the first place
2. Architecture innovations change everything – AlexNet, GANs, Transformers
3. Scaling works – more data + more compute = better results

This insight – that you can simply "build bigger" – became the guiding idea of the years to come.

---

Continue with Part 2: The Language Revolution – When Machines Learned to Read and Write (2018–2020)