Why does the history matter?

You don’t need to be a historian to understand AI, but knowing where it comes from helps you understand where we are. AI didn’t appear out of nowhere with ChatGPT — it’s been evolving for 70 years, with spectacular successes and resounding failures.

1950: Alan Turing and the key question

It all starts with a simple question: “Can a machine think?”

Alan Turing, a British mathematician considered the father of computer science, proposed in 1950 what we now know as the Turing Test: if a person talks to a machine and can’t tell whether the responses come from a human or a machine, then the machine can be considered “intelligent”.

Turing never lived to see a working AI — he died in 1954. But he planted the seed.

1956-1970: The initial euphoria

In 1956, a group of scientists gathered at Dartmouth University and coined the term “Artificial Intelligence”. They were convinced that within a generation they’d have machines as intelligent as humans.

Those early years were bursting with optimism. Programs were created that could play chess, solve logic problems, and hold basic conversations. ELIZA (1966), a program that simulated being a psychotherapist, surprised many people — even though it simply rephrased what you said as a question.

1970-1980: The first winter

Optimism crashed against reality. Computers in the 70s were too slow and had too little memory to do anything truly intelligent. Governments cut funding because the promises weren’t being kept.

This is known as the first AI winter — a period of disillusionment where almost nobody wanted to invest in AI.

1980-1990: Expert systems and the second attempt

In the 80s, AI revived with expert systems: programs with rules written by hand by human experts. For example, a medical expert system had thousands of rules like “if the patient has fever AND cough AND difficulty breathing, then consider pneumonia”.

They worked in very specific domains, but had a fundamental problem: every rule had to be written manually. As the domain grew more complex, maintaining the rules became unsustainable.

1990-2000: The second winter

Expert systems showed their limitations. Funding was cut again. AI fell into disrepute once more. Second winter.

Meanwhile, quietly, some researchers were working on a different idea: instead of programming rules by hand, what if we let the machine learn from examples? This was called Machine Learning.

2012: The Deep Learning revolution

The turning point came in 2012, when a Deep Learning system won the ImageNet image recognition competition by a huge margin over all previous methods.

The key was the combination of three things that hadn’t existed together before:

  • Massive data — The internet had generated enormous amounts of text, images, and videos
  • Powerful GPUs — Graphics cards designed for video games turned out to be perfect for training neural networks
  • Improved algorithms — Neural networks with many layers (“deep”) that could learn complex patterns

From 2012 onwards, AI hasn’t stopped growing.

2017: The Transformer architecture

A paper from Google titled “Attention Is All You Need” introduced the Transformer architecture — the foundation of all modern language models. Without this paper, GPT, Claude, and Gemini wouldn’t exist.

Transformers solved a key problem: how to process long text sequences efficiently and in parallel. Previously, models read text word by word. Transformers can “see” the entire text at once and decide which parts to pay attention to.

2022: ChatGPT and the explosion

In November 2022, OpenAI launched ChatGPT — and changed everything. It wasn’t the first chatbot, nor the first LLM, nor even the most powerful model at the time. But it was the first that anyone could try for free in their browser.

In 5 days it reached 1 million users. In 2 months, 100 million. AI went from being a lab topic to being a conversation topic at the pub.

2023-2026: The current race

Since ChatGPT, the pace has been dizzying:

  • Anthropic launched Claude, with emphasis on safety and helpfulness
  • Google responded with Gemini
  • Meta bet on open source with LLaMA
  • Mistral emerged as the European alternative
  • DeepSeek and other Chinese models proved that innovation is global

Today, models don’t just generate text — they understand images, audio, and video. And with the emergence of frameworks like OpenClaw, anyone can have their own AI agent up and running.

The pattern of winters and summers

If you look at the history, you see a clear pattern: euphoria → exaggerated promises → disappointment → winter → new technology → euphoria.

Are we in an AI summer? Probably. Will another winter come? Perhaps, but with one difference: this time AI is already integrated into products that millions of people use every day. It’s hard to “uninstall” something that has already changed how we work.

Key concepts today

  • Turing Test: If you can’t tell a machine apart from a human in a conversation, the machine is “intelligent”
  • AI Winter: A period of disillusionment where investment in AI is cut (there have been two)
  • Deep Learning: Machine learning with deep neural networks (many layers)
  • Transformer: The architecture that made modern LLMs possible (2017)
  • Machine Learning: Getting machines to learn from examples instead of programming them rule by rule

Next lesson: Types of AI — how we classify intelligent machines.