History › Our time › World Wide Web and the internet age
Artificial intelligence boom
deep learning from 2012; ChatGPT 2022· Invention or discovery· Global
Well documentedSolid records, physical evidence or scientific consensus.
A surge in computers that learn from data, from 2012 breakthroughs in image recognition to chatbots such as ChatGPT from 2022.
See it on the History map →What happened
AI research began in the 1950s but repeatedly stalled. Around 2012, 'deep learning', layered artificial neural networks trained on huge datasets with graphics chips (GPUs), began beating older methods at recognizing images and speech. In 2016 DeepMind's AlphaGo beat Go champion Lee Sedol in Seoul. A 2017 Google design, the 'transformer', led to large language models. OpenAI's ChatGPT (2022) reached about 100 million users in two months.
Why it matters
It is changing how software, search, writing and research are done, driving a boom in chips, data centers and power demand, and raising fears about jobs and safety.
How we know
Published research papers, benchmark test results, company disclosures and adoption surveys; investment and data-center electricity use are tracked by industry and government agencies.
Still debated
How far current methods can go, whether AI will add or destroy jobs on net, and how to keep powerful systems safe are open questions.
Connections (2)
▸Artificial intelligence boom is often compared to the Industrial RevolutionMostly agreed
The connection
Economists and tech leaders often compare AI to the Industrial Revolution or to electricity: a general-purpose technology that could raise productivity across many industries at once. Steam replaced muscle; AI may automate some mental work. The comparison carries a warning too: early industrialization brought decades of disruption before ordinary wages rose.
When and how
A common comparison since the 2010s, made by economists such as Erik Brynjolfsson and by leaders at Google, Microsoft and OpenAI.
How sure are we?
It is an analogy, not a causal link. Whether AI's impact matches industrialization will only show in productivity data over the coming decades.
Debates and open questions
Skeptics note productivity gains so far are modest, and computers took about 15 years to show up in the statistics. AI's real scale is unknown.
▸World Wide Web and the internet age supplied the training data behind the Artificial intelligence boomWell documented
The connection
Modern AI learns from examples, and the Web created an enormous supply: billions of pages, photos and videos. ImageNet, built from millions of labeled web images, enabled the 2012 deep-learning breakthrough. Large language models such as ChatGPT were trained largely on text gathered from the Web.
When and how
ImageNet, released in 2009 by Fei-Fei Li's team at Princeton and Stanford; the Common Crawl archive of web pages, from 2008, fed many language models.
How we know
AI developers' published papers describe their training data, including Common Crawl and Wikipedia.
Debates and open questions
Authors, artists and publishers have sued over whether training on web content without permission is legal; early court rulings have been mixed.