How we got here: more than 80 years of AI
A fast-talking podcast sent me back to 1943. A short journey through the ideas behind modern AI, and the generations of research that made ChatGPT possible.
Marc Andreessen seems to think at the speed of light. Unfortunately, he talks almost as quickly. So while listening to his conversation with Patrick O'Shaughnessy on Invest Like the Best, I slowed the podcast down to 0.8 speed. At one point, while discussing DeepSeek and OpenAI, he traced the origins of modern artificial intelligence back to 1943. I knew parts of this history, but had never quite connected them. And I suspect I'm not alone.
The roots reach back decades
In 1943, Warren McCulloch and Walter Pitts published a mathematical model of artificial neurons. Their work helped establish the foundations of neural networks, the idea that complex computational behaviour could emerge from interconnected, relatively simple units. Over the following decades, researchers developed these ideas further, through periods of excitement, disappointment and renewed progress.
Google's breakthrough in 2017
Fast-forward to 2017. A team of researchers, primarily at Google, published Attention Is All You Need, introducing the Transformer architecture. It enabled neural networks to model relationships between words while allowing much more of the training to happen in parallel. The Transformer would become the foundation of modern large language models. Google continued developing the technology, but it was another organisation that turned it into a global consumer phenomenon.
From OpenAI to Anthropic and beyond
OpenAI, founded in 2015, built upon the Transformer architecture and developed its GPT models. In November 2022, it launched ChatGPT, making decades of research accessible to almost anyone with a browser. Anthropic, established in 2021 by former OpenAI researchers, subsequently developed Claude. Other companies, including Meta, Mistral and DeepSeek, followed their own paths, drawing upon much of the same scientific heritage.
Nobody started from scratch
This was Andreessen's observation that caught my attention. OpenAI didn't invent the Transformer. Google didn't invent neural networks. And the researchers in 1943 were themselves building upon earlier work in mathematics, logic and neuroscience.
Much of this intellectual development took place at universities and research institutions, often supported by public funding. Today's AI companies have made substantial contributions of their own, but their achievements rest upon generations of accumulated knowledge.
From scientific discovery to societal value
There's an interesting distinction here between discovering something, developing it into a working technology and bringing it into widespread use. These are different capabilities, and they don't necessarily reside within the same organisation.
What appeared to arrive almost overnight with ChatGPT in November 2022 was, in reality, more than eighty years in the making. That's what I took away from a few minutes of Andreessen's rapid-fire conversation. A useful bit of history that helped me better understand how we got here.
