This is the transcript of a general-audience talk I gave in July 2024. In preparing it for publication I have corrected a few facts and phrasings, but left the perspective and conclusions as they stood at the time.
Hello everyone. Today let’s talk about the history of artificial intelligence. From the moment the idea was first proposed to where we are now, it has passed through many important stages and upheavals.
The concept of artificial intelligence can be traced back to the 1950s. The 1956 Dartmouth workshop is regarded as the formal starting point of AI research. Organized by John McCarthy, Marvin Minsky, Claude Shannon, and Nathaniel Rochester, it coined the term “artificial intelligence” and imagined machines that could carry out tasks normally requiring human intelligence.
In the 1960s, AI research produced its first tentative results. Work in this period centered on problem solving and symbol manipulation, continuing the line of the earliest AI programs from the late 1950s — such as the Logic Theorist (1956) and the General Problem Solver (1957). These programs showed the potential of computers to solve mathematical problems and carry out logical reasoning.
Next came the 1970s, when the arrival of expert systems marked an important advance. Expert systems are computer programs that simulate the decision-making of a human expert; DENDRAL and MYCIN are two representatives. DENDRAL was used for chemical analysis, MYCIN for medical diagnosis. Although these systems performed well within their narrow domains, their generality and flexibility were limited.
Entering the 1980s, AI went through an important shift as machine learning began to emerge. Researchers of this period realized that symbol manipulation and hand-written expert-system rules were not enough — machines needed to be able to learn from data. The revival of neural networks became one of the defining markers of the era. The idea of neural networks actually reaches back much further — to the McCulloch–Pitts model of 1943 and the perceptron proposed by Rosenblatt in 1958 — but it was not until the 1980s, as computing power grew and learning algorithms such as backpropagation were popularized again around 1986, that neural networks drew widespread attention once more.
In the 1990s, with the rapid advance of computer hardware, AI entered a new phase. The spread of the internet and the explosive growth in the amount of available data gave machine learning and data-driven AI models a rich supply of resources. During this period, statistical learning methods and new techniques such as support vector machines made notable progress.
In the 21st century, and especially after 2010, the rise of deep learning transformed the face of AI completely. Deep learning is a branch of neural networks that uses multi-layer network structures to automatically extract features and recognize patterns in data. In 2012, Geoffrey Hinton’s team achieved a breakthrough result in the ImageNet competition with a deep convolutional neural network — an event widely seen as the opening of the deep learning era.
On November 30, 2022, the release of ChatGPT was an extraordinarily important milestone in the history of AI. In a sense, the “Tower of Babel” that legend says collapsed halfway because people could no longer understand one another was, on that day, raised again in another form — for the first time, humanity had a machine that could wield language freely. ChatGPT not only demonstrated the power of large language models (LLMs); it set off boundless speculation about the future of AI. The LLM is an intelligence revolution, one that will bring enormous and far-reaching change to human society — the coming decades, perhaps even the coming centuries, will be profoundly shaped by it.
There is also something intriguing about artificial general intelligence (AGI): the core ideas that drive these systems are often simpler than people imagine. The Transformer architecture behind today’s large models supports astonishing capability out of just a handful of basic building blocks. It brings to mind a line from Westworld — “You think the human code is complex, but it’s surprisingly simple.” That leaves us full of anticipation for where AI is headed, and makes us reconsider what intelligence really is.
All in all, the history of artificial intelligence is a story full of invention and upheaval. From its first tentative theories to today’s widespread applications, AI has kept pushing past its own limits and reshaping our world. Thank you all for listening.