How to define artificial general intelligence
T HE IDEA of machines outsmarting humans has long been the subject of science fiction. Rapid improvements in artificial-intelligence (AI) programs over the past decade have led some experts to conclude that science fiction could soon become fact. On March 19th Jensen Huang, the chief executive of Nvidia, the world’s biggest manufacturer of computer chips and its third most valuable publicly traded company, said he believed today’s models could advance to the point of so-called artificial general intelligence (AGI) within five years. What exactly is AGI—and how can we judge when it has arrived?
Mr Huang’s words should be taken with a pinch of salt: Nvidia’s profits have soared because of the growing demand for its high-tech chips, which are used to train AI models. Promoting AI is thus good for business. But Mr Huang did set out a clear definition of what he believes would constitute AGI: a program that can do 8% better than most people at certain tests, such as bar exams for lawyers or logic quizzes.
This proposal is the latest in a long line of definitions. In the 1950s Alan Turing, a British mathematician, said that talking to a model that had achieved AGI would be indistinguishable from talking to a human. Arguably the most advanced large language models already pass the Turing test. But in recent years tech leaders have moved the goalposts by suggesting a host of new definitions. Mustafa Suleyman, co-founder of DeepMind, an AI-research firm, and chief executive of a newly established AI division within Microsoft, believes that what he calls “artificial capable intelligence”—a “modern Turing test”—will have been reached when a model is given $100,000 and turns it into $1m without instruction. (Mr Suleyman is a board member of The Economist’s parent company.) Steve Wozniak, a co-founder of Apple, has a more prosaic vision of AGI: a machine that can enter an average home and make a cup of coffee.
