On 30 November 2022, OpenAI released ChatGPT, a chatbot built on a large language model, free for anyone to use on the web. It became one of the fastest-growing consumer applications in history; by one widely reported estimate, it reached a hundred million users within about two months. For many people, it was their first encounter with modern AI, and it set off a global race. This final level surveys the period since, which is still unfolding as you read. Because the field moves fast, some details here will date quickly.
The race
Technology companies moved quickly. In 2023, OpenAI released GPT-4, which could also interpret images. Anthropic released its Claude models, Google its Gemini models, and Meta released its Llama models with their internal parameters, the “weights,” available for others to download and modify. Chinese companies built their own systems. In January 2025, the Chinese company DeepSeek released a model that performed strongly on reasoning tasks and whose developers reported it had been trained at a fraction of the cost of rivals, briefly causing a sharp fall in the stock of the chipmaker Nvidia and intensifying debate about competition and costs.
Investment poured into AI, especially into data centers full of specialized chips. Models gained new abilities: reasoning models, introduced from 2024, were trained to work through problems step by step before answering, improving their performance in mathematics and programming. By 2025, companies were deploying AI agents, systems that can use tools, browse the web, write and run code, and carry out multistep tasks with less human supervision.
Uses
AI tools spread into writing, programming, customer service, education, medicine, scientific research and creative work. Programmers used AI assistants to write and check code. Scientists used models to analyze data and design experiments. Students and teachers wrestled with how to use chatbots honestly. Supporters argued that AI could greatly increase productivity and help solve problems in health and science; skeptics questioned how reliable and transformative it would prove.
Concerns
The rapid spread of AI brought a range of concerns, described below.
- Accuracy. Models still produced confident errors.
- Work. Writers, artists, translators and others worried about their jobs. In 2023, the Writers Guild of America strike in Hollywood won limits on studios' use of AI.
- Copyright. Authors, artists and news organizations, including the New York Times in a lawsuit filed in December 2023, challenged the use of their work to train models without permission.
- Misinformation. Generated text, images and voices made deception cheaper.
- Energy. Data centers' growing demand for electricity and water raised environmental questions.
- Power. A few companies controlled the most advanced systems.
Debates about risk
Some prominent researchers warned of larger dangers. In May 2023, Geoffrey Hinton left Google so that he could speak freely about AI risks, including the possibility that systems might one day become more capable than humans and difficult to control. That month, hundreds of researchers and industry leaders signed a one-sentence statement that mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war. Other researchers, such as Yann LeCun, argued that such fears were exaggerated and distracted from present harms like bias and misinformation. The debate remains open.
Rules
Governments began to respond. In November 2023, Britain hosted an international AI Safety Summit at Bletchley Park, where Turing had worked. The European Union's AI Act, the first broad law regulating AI, entered into force on 1 August 2024, with obligations phased in over the following years, classifying AI uses by risk and banning some, such as certain forms of social scoring. Other countries adopted their own approaches, and national policies sometimes shifted with changes in government. How to encourage innovation while preventing harm, and whether rules should be national or international, remain central questions.
Looking back
The history of AI shows recurring patterns: bold predictions that came too early, winters when hopes collapsed, quiet progress that was ignored, and sudden breakthroughs when data, hardware and ideas came together. It also shows that AI is built by people, from the hidden chess player in the Mechanical Turk to the online workers who labeled ImageNet and the reviewers who rate chatbot answers. Its benefits and harms depend on human choices.
Worked example: reading an AI headline
Suppose you read: “New AI Outperforms Doctors at Diagnosing Disease.” Apply lessons from this course.
- What was actually shown? Was it tested on real patients, or on a benchmark dataset in a lab, like ImageNet?
- Which doctors, which disease? A narrow task, like reading one kind of scan, is very different from general diagnosis, as the history of expert systems shows.
- Who was in the data? As Gender Shades showed, a system can work well for some groups and badly for others.
- Who made the claim? A company selling the product has incentives, as the Mechanical Turk's exhibitor did.
- What happens when it fails? MYCIN was accurate but never used, partly because no one had settled who was responsible for errors.
The story of AI is still being written, and you are part of it as a user, a voter and a citizen. Understanding its history is one of the best ways to judge its present: to recognize real progress, see through hype, and insist that these powerful tools serve the dignity and wellbeing of every person, not just the few who build them.