CEOs keep crying AGI. What does that even mean?

Nvidia CEO Jensen Huang (middle right) and OpenAI President Greg Brockman (right) discuss AI with President Donald Trump after a recent meeting. Tech CEOs have started saying AI has reached a critical — and highly debated — threshold. AGI is typically taken to mean an AI that can match or exceed human intelligence. Engineers told Business Insider the term is overblown. "Welcome to the AGI era," OpenAI president Greg Brockman declared on a call with reporters in September, touting the company's latest AI model. "For me, personally, I do think we're there." Recently, tech CEOs and people on X alike are throwing around a buzzword that was once a distant, largely theoretical goal for artificial intelligence. AGI, or artificial general intelligence, is a nebulous term often taken to mean an AI model that can perform a wide variety of tasks as well as or better than a human. Aside from Brockman, influential people who have said we've reached that point include Elon Musk, Sam Altman, and Jensen Huang. However, others in the industry told Business Insider that the term is, at best, too poorly defined to serve as a meaningful benchmark. At worst, they called claims that AGI is here dishonest marketing practices. "It's all just marketing talk, and it's complete nonsense," said Peter Voss, who was among the first software engineers to use the term. "If you had AGI right now, we would certainly know it." What is AGI? And why are tech leaders saying we've reached it? In 2002, Ben Goertzel was having trouble deciding on a book title for a project on a speculative, versatile type of AI called a "thinking machine," which is capable of, say, beating a human at chess while also learning to drive a car. After reaching out to friends, he settled on the term: "Artificial General Intelligence." Voss, who contributed a chapter for the book, said such a machine represented a "dream that has really not been realized in AI." The leading chatbots today, he said, have difficulty branching out beyond certain limited capabilities and learning new skills with the same plasticity as humans. Humans, he said, can learn to work at a call center given one or two days of training. An AGI system should be able to do the same, but current AI systems struggle with such tasks. "Right now, to try and do even something as simple as customer support, you need billions of dollars of engineering," he said. "You need hundreds of thousands of example conversations, and the results are still very poor." Still, that hasn't stopped tech CEOs and venture capitalists from crying AGI. Nvidia CEO Jensen Huang next to OpenAI president Greg Brockman. Both have declared "AGI" this year. Following the release of OpenAI's Astra model in September — the one Brockman was gushing about — Nvidia CEO Jensen Huang posted a triumphant message on X declaring that "AGI has arrived." "From ChatGPT to o1 to Astra in 4 years," Huang said. "Congratulations @OpenAI team." Elon Musk, ever the speculator, posted on X this week in response to a Claude-generated short film. "I really felt the AGI profoundly this time," he said. Factory CEO Matan Grinberg recently told the Sources podcast that "I think AGI is already here," and that "we are living in a post-AGI world right now." "We're kind of past that event horizon," he said. 'A marketing term' Other CEOs, like Anthropic's Dario Amodei, have been more reticent to declare the dawning of the AGI era. "AGI has never been a well-defined term, for me," he said last year. "I've always thought of it as a marketing term." Frontier AI companies like Anthropic and OpenAI are under intense financial pressure to showcase ever-more-powerful technology as they prepare for what will likely be blockbuster, trillion-dollar-plus IPOs. By tapping into futuristic nomenclature that evokes images of fully autonomous, free-thinking robots, as seen in "Star Wars" or "Her," Emily Bender, a professor at the University of Washington, said the companies propagate an "illusion" that chatbots are more advanced than they actually are. "We have expectations based on those fictional ideas," she said. "It refers to an imagined technology rather than any actually existing technology." "AGI, like AI, is a marketing term," she added. Alan Chan, an AI research fellow at GovAI, an AI policy think tank, told Business Insider that AGI isn't the best term to gauge progress because of how vague it can be in practice. Chan said that some definitions of AGI conflate AI's ability to do new things with "the ability to actually do things in the world." "We might have a system that's really good at coding, but it's just really bad at learning to do new things," Chan said. "So, it's really good at coding, but when we deploy it in a new job, it just takes so much time and effort to train it to actually be good at that job. "I think when people talk about AGI, they mix those two things together," he added. Are we getting AGI anytime soon? If you buy into Voss's definition of AGI, you might have to wait a while for an AI system that can learn to drive a car in 24 hours. Voss said that the popularity of LLMs with investors and the general public makes it hard to get competing systems off the ground. "They sucked all of the oxygen out of the air for any other approaches," he said. Voss isn't the only engineer who has said that the ChatGPT-era strategy of scaling LLMs with more data and chips won't lead to AGI anytime soon. AI "godfather" Yann LeCun, who was previously Meta's chief AI scientist before he left to launch his own AI lab. "You cannot just assume that more data and more compute means smarter AI," Yann LeCun, an "AI godfather" and Meta's former chief AI scientist, said last year. OpenAI cofounder Ilya Sutskever said in 2025 that the tides of the AI industry will have to shift back to the research phase. "These models somehow just generalize dramatically worse than people," he said. "It's super obvious. That seems like a very fundamental thing." Others, like Bender, the researcher at the University of Washington, said that AGI, as a concept, is impossible for humans to create. "A general-purpose thinking machine is not within the realm of possibility," she said. "That would require an approach to engineering that just doesn't work." Read the original article on Business Insider

Read it at Business Insider ↗

← Front page