Evolution · 1 min read · Updated 2026-10-05
The evolution of generative AI
1950s to 2000s: rules, then learning from data
Early AI followed rules that people wrote by hand. In the 1990s, programs started learning from data instead. Spam filters and movie recommendations came from this time.
2010s: deep learning
In 2012, a deep learning system won a big image contest by a wide margin. Soon computers got much better at recognising pictures and speech.
2017 to 2020: the transformer
In 2017, researchers introduced the transformer, a better way to read text. GPT-3 followed in 2020 and surprised people by how well it could write.
2022 until now: chatbots for everyone
ChatGPT arrived in November 2022 and millions tried it. Since then, AI has learned to handle pictures and sound, think through problems step by step, and use tools.
Key takeaways
- Rules first, then learning from data, then deep learning, then transformers.
- Bigger models with more data gave big jumps.
- Since 2022, AI chat has gone mainstream.
Quick questions
When did generative AI start?
Ideas are old, but the big change came after 2017 and with ChatGPT in 2022.
What is a transformer?
The design behind most chatbots. It helps the computer see how the words in a sentence relate to each other.
What is multimodal?
AI that can work with more than words, such as pictures, sound and video.