Even the AI companies are tired of talking about AGI

The cringe comes for us all, and for all our hot new turns of phrase. “Rizz” lost its luster when grandparents started asking about its meaning. Teachers who dressed up as “6-7” on Halloween drove a nail into the coffin of Gen Alpha’s rallying cry. And tech CEOs who once trumpeted the quest for “artificial general intelligence,” or AGI, are jumping ship for any other term they can find.

Until recently, AGI was the ultimate goal of the AI industry. The vaguely defined term was reportedly coined in 1997 by Mark Gubrud, a researcher who defined it as “AI systems that rival or surpass the human brain in complexity and speed.” The term still typically denotes AI that’s equal to or surpasses human intelligence. But now, several of the biggest companies are going for a rebrand — creating their own phrases or acronyms that (spoiler alert) still mean, essentially, the same thing.

CEOs have spent the past year downplaying the importance of “AGI” as a milestone. Dario Amodei, CEO of Amazon-backed Anthropic, has said publicly that he “dislike[s] the term AGI” and that he’s “always thought of it as a marketing term.” OpenAI CEO Sam Altman said in August that it’s “not a super useful term.” Jeff Dean, Google’s chief scientist and Gemini lead, has said he “tend[s] to steer away from AGI conversations.” Microsoft CEO Satya Nadella has said we’re getting “a little bit ahead of ourselves with all this AGI hype,” and that at the end of the day, “self-claiming some AGI milestone” is “just nonsensical benchmark hacking.” He also said on a recent earnings call that he doesn’t believe that “AGI as defined, at least by us in our contract, is ever going to be achieved anytime soon.”

In its place, they’re pushing a cornucopia of competing terminology. Meta has “personal superintelligence,” Microsoft has “humanist superintelligence,” Amazon has “useful general intelligence,” and Anthropic has “powerful AI.” It’s a sharp about-face for all of these companies, which previously bought into the AGI benchmark — and the fear of missing out that came from not chasing it — in recent years.

Part of the problem with “AGI” is that the more advanced AI gets, the more poorly defined the term seems — since the concept of AI that’s “equal to human intelligence” looks different to virtually everyone. “Lots of people have very different definitions of it, and the difficulty of the problem varies by factors of a trillion,” Dean said.

Yet some companies have billions of dollars riding on this nebulous phrase, a problem that’s clearest in the strange, ever-changing relationship between Microsoft and OpenAI.

In 2019, OpenAI and Microsoft famously signed a contract with an “AGI clause.” It gave Microsoft the right to use OpenAI’s tech until the latter achieved AGI. But the contract apparently didn’t fully define what that meant. When the deal was renewed in October, things got even more complicated. The terms shifted to say that “once AGI is declared by OpenAI, that declaration will now be verified by an independent expert panel” — meaning that now, it won’t just be OpenAI’s call to define what AGI means, it’ll be a group of industry experts — and Microsoft won’t lose all its rights to the tech once that happens, either. The simplest way to put this whole ordeal off? Just don’t say AGI.

Another problem is that AGI has developed some baggage. Tech companies have spent years detailing their own fears about how the technology could destroy everything. Books have been written (think: If Anyone Builds It, Everyone Dies). Hunger strikes have made headlines. For a while, it was still good publicity — saying your tech is so powerful that you’re worried about its influence on the Earth seems to draw big investor dollars. But the public, unsurprisingly, soured on that idea. So, with the complicated definitions, contract drama, and public fear around superpowerful AI, it’s a lot easier to market less-loaded terminology. That’s why every tech company seems to be making some new brand of “intelligence” its own.

One popular general-purpose replacement for AGI is “artificial superintelligence,” or ASI. ASI is AI that surpasses human intelligence in virtually every area — compared to AGI, which is now generally defined as AI that’s equal to human intelligence. But for some in the tech industry, even the idea of “superintelligence” has become amorphous and conflated with AGI. The multiple theoretical milestones don’t even have clearly distinguished timelines. Amodei says he expects “powerful AI” to come “as early as 2026.” Altman says he expects AGI to be developed in the “reasonably close-ish future.”

So companies have developed their own variants. Meta CEO Mark Zuckerberg said in January that the company needed “to build for [artificial] general intelligence,” but by July, he had pivoted to “personal superintelligence” in a manifesto. It was a power-to-the-people spin on AGI that “helps you achieve your goals, create what you want to see in the world, experience any adventure, be a better friend to those you care about, and grow to become the person you aspire to be.” Zuckerberg used the manifesto to combat public fears of AI taking jobs and throw shade at Meta’s competitors, calling the company’s vision “distinct from others in the industry who believe superintelligence should be directed centrally towards automating all valuable work, and then humanity will live on a dole of its output.”

Microsoft, however, has also rebranded its venture as chasing “Humanist Superintelligence (HSI),” which is essentially Zuckerberg’s manifesto in a different font. The company is defining HSI as “incredibly advanced AI capabilities that always work for, in service of, people and humanity more generally” and are “problem-oriented” instead of being “an unbounded and unlimited entity with high degrees of autonomy.” The rebrand came complete with a new website, topped with the term “Approachable Intelligence,” backed with a sepia-style background and a soft color palette, and awash with paintings and photos of nature.

For Amazon’s part, it has rebranded its AGI efforts as chasing “useful general intelligence,” or “AI that makes us smarter and gives us more agency.” Late last year, the company hired the founders of Adept, an agentic AI startup, and licensed its technology, in efforts to compete against others in the AGI race. Like the other companies’ branding efforts, though, Amazon is positioning its UGI efforts as useful, easily defined, and decidedly not all-powerful or scary: just “enabling practical AI that can actually do things for us and make our customers more productive, empowered, and fulfilled.”

With “powerful AI,” Anthropic has no interest in seeming down-to-earth. Amodei dubs it a “‘country of geniuses in a datacenter’” that is “smarter than a Nobel Prize winner across most relevant fields — biology, programming, math, engineering, writing, etc.” Powerful AI, he said, would be able to write compelling novels, prove unsolved theorems in mathematics, and write complex code. It would not just answer questions but complete complex, multistep tasks over hours, days, or weeks, similar to AI CEOs’ vision of a successful AI agent, and “absorb information and generate actions at roughly 10x–100x human speed.”

AGI and ASI were already a lot to reckon with. Now we’ve got PSI, HSI, UGI, and PI, too. Cheers to the new acronyms next year will bring.