The History of Artificial Intelligence
iAs of: September 2026
Historical facts and dates in this section are long-lived. The table of current top models reflects the state as of September 2026 — new model versions may have shifted since.
Wissen
The Birth: Dartmouth Conference 1956
Every great idea has a beginning. For Artificial Intelligence, that moment was the summer of 1956 in Dartmouth, New Hampshire (USA). A group of young scientists -- including John McCarthy, Marvin Minsky, Allen Newell, and Herbert Simon -- met for a workshop with a bold goal: they wanted to find out whether machines can think.
John McCarthy coined the term "Artificial Intelligence" during this event. The researchers were optimistic: they believed that within a generation, machines would be as intelligent as humans. This prediction was wrong, but the vision set a revolution in motion.
iDid You Know?
The original proposal for the Dartmouth Conference put it this way: "Every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it." This conviction drives AI research to this day.
Early AI Systems: ELIZA and Expert Systems
In the 1960s and 1970s, the first AI programs emerged:
ELIZA (1966) was one of the first chat programs, developed at MIT by Joseph Weizenbaum. ELIZA simulated a psychotherapist and could hold simple conversations -- not because it understood what was being said, but by recognizing keywords and giving pre-made responses. Nevertheless, many people were convinced they were talking to a "thinking" machine.
Expert systems (1970s-1980s) were programs that captured the knowledge of human experts in rules. For example, MYCIN could diagnose blood infections by working through hundreds of "if-then" rules. The problem: every single rule had to be programmed by hand -- a tedious and inflexible process.
The AI Winters: When the Excitement Freezes
Twice in history, AI research has experienced a dramatic collapse -- the so-called "AI winters":
First AI winter (mid-1970s): The high expectations of the 1960s were not met. Computers were too slow, the problems too complex. Funding was cut, research projects were shut down.
Second AI winter (late 1980s to 1990s): Expert systems proved to be too rigid and too expensive to maintain. The business world lost interest, and funding collapsed again.
!Lesson from History
The AI winters show: exaggerated expectations lead to disappointment. Even today, some experts warn of a possible new AI winter if the current promises of the technology are not fulfilled.
The Deep Learning Revolution: AlexNet 2012
The turning point came in 2012. Geoffrey Hinton's team won the ImageNet competition (a major image recognition competition) with a deep learning model called AlexNet. The gap to the second-place entry was so large that the entire research community took notice.
What was different? Instead of programming rules by hand, AlexNet learned on its own from millions of images to recognize objects. The combination of neural networks with many layers ("deep learning"), powerful GPUs, and large datasets suddenly worked spectacularly well.
After 2012, things moved fast:
- 2014: Generative Adversarial Networks (GANs) create artificial images
- 2016: AlphaGo defeats the world's best Go player
- 2017: Google publishes the Transformer architecture -- the foundation of all modern language models
- 2018: Google BERT revolutionizes text processing
- 2020: OpenAI releases GPT-3 with 175 billion parameters
ChatGPT: The Moment AI Went Mainstream
On November 30, 2022, OpenAI released ChatGPT -- and the world changed overnight. ChatGPT reached 100 million users in just two months, making it the fastest-growing app in history.
Why was ChatGPT so special? Not because the technology was entirely new, but because for the first time anyone could talk to an AI system as easily as chatting with a friend over a messenger. The barrier to entry disappeared.
iFor Comparison
Instagram took 2.5 years to reach 100 million users. TikTok took 9 months. ChatGPT did it in 2 months. This explosive growth shows how strong the demand for accessible AI was.
Current State 2026: The Model Explosion
Since ChatGPT, development has accelerated further. As of September 2026, several high-performance AI models compete for the top position:
| Model | Company | Year Released | Special Feature |
|---|---|---|---|
| Claude Fable 5 | Anthropic | June 2026 | First Mythos-class model, 1M context, 95 % SWE-bench |
| GPT-5.6 | OpenAI | July 2026 | Multimodal, large ecosystem |
| Claude Opus 5 | Anthropic | July 2026 | 1M token context, strong coding (~96 % SWE-bench) |
| Gemini 3.1 Pro | Google DeepMind | February 2026 | 1M token context, Google integration |
| Llama 4 | Meta | 2025/2026 | Open source, 10M token context |
2023–2026: What Has Changed
The impact of the AI revolution is already measurable — in development teams, companies, and entire industries:
Development accelerates dramatically: GitHub reported in 2023 that Copilot users completed programming tasks 55% faster (GitHub/Microsoft study, 2022) and that in supported languages, 46% of code was suggested by AI. The term "Vibe Coding" — coined by Andrej Karpathy in February 2025 — describes a new programming style where developers set the direction and AI writes the code.
Companies restructure: Swedish fintech company Klarna reported in 2024 that their AI chatbot was handling the work of 700 customer service agents. The company reduced its workforce by about 25% — not through mass layoffs, but primarily by not filling vacated positions.
Knowledge platforms transform: Stack Overflow, for years the go-to resource for programming questions, saw traffic decline by approximately 50% since 2023. The platform has repositioned itself, launching OverflowAI with AI-powered features.
!Numbers in Context
Many of these statistics come from the companies themselves and have not been independently verified. The actual impact of AI on productivity and the job market is still being studied by researchers. Healthy skepticism toward impressive numbers is warranted.
We live in a time when AI is developing faster than any other technology before. What was science fiction yesterday is everyday reality today.
What was special about ChatGPT when it was released in 2022?
Verstehen
AI in the Context of the Great Revolutions
Human history has been shaped by great technological revolutions -- and each one arrived faster than the last. AI is the newest and potentially the most profound of these revolutions.
Gutenberg's printing press democratized knowledge. Books were no longer just for elites -- anyone could read and learn.
Parallele zu KI
The steam engine and factory replaced muscle power. For the first time, energy could be generated independently of humans and animals.
Parallele zu KI
Electricity changed everything: light, communication, production. The world became connected.
Parallele zu KI
Personal computers and the internet gave everyone access to computing power and global knowledge.
Parallele zu KI
AI democratizes intelligence itself. For the first time, anyone can access cognitive capabilities that were previously reserved for experts.
Parallele zu KI
Do you notice a pattern? Each revolution democratized something -- made it accessible to everyone:
- Printing press: Knowledge for all
- Industrial Revolution: Energy for all
- Electricity: Communication for all
- Computers & Internet: Computing power for all
- AI: Intelligence for all
iThe Speed Is Increasing
From the printing press to the Industrial Revolution, about 320 years passed. From the Industrial Revolution to electricity, 110 years. From computers to AI, only about 50 years. The time between revolutions keeps getting shorter -- and the impact keeps getting larger.
And there is another important parallel: Each of these technologies was initially underestimated. The printing press was seen as a threat to the Church. Electricity was dismissed as a dangerous toy. The internet was still laughed at as a fad in 1995. And AI? AI is still misunderstood by many today as a "better search engine."
Think of the AI winters like the ice age in nature: life (research) does not disappear completely, but it is pushed back and slowed down. And just like after an ice age, a new spring follows.
Why did the first AI winter occur in the 1970s?
What pattern do the great technological revolutions show?
Where Is This Heading?
We are currently in the steepest growth phase that technology history has ever seen. The development of AI models follows an exponential curve -- it is not slowing down, but getting faster and faster.
What does this mean in practice? AI performance currently doubles roughly every 6-8 months. GPT-3 (2020) had 175 billion parameters. GPT-4 (2023) was orders of magnitude more capable. And the current models of 2025/2026 surpass everything that was imaginable just two years ago.
*Nobody Fully Understands It -- and That Is Okay
Here is a secret that even many experts admit: Nobody fully understands why large language models work so well. That sounds worrying, but it is historically normal. When humanity began using electricity, the underlying physics were not yet fully understood. Yet people built power plants and light bulbs. It is similar with AI: we can use it and benefit from it, even while science is still working to understand all the details.
Anwenden
What does this mean for you as a learner? Three things:
- You do not need to be an expert to use AI. Just as you use electricity without being an electrician, you can use AI without being a data scientist.
- Now is the best time to start. Those who learn early how to work with AI have a decisive advantage.
- The journey never stops. AI is developing so fast that lifelong learning is not optional -- it is necessary.
Why is it okay to use AI even though nobody fully understands it?
Reflect
The history of AI shows that progress does not move in a straight line -- there are phases of great excitement and phases of disappointment. Yet with each cycle the technology becomes more capable, and AI joins the ranks of the great revolutions that each democratized something fundamental. In the next section, we will look at which companies and models shape the market today.