What Is AGI? The Goal Nobody Can Define

Artificial general intelligence is the most talked-about milestone in tech — and the one nobody can actually pin down. Executives predict it, venture capital floods toward it, and critics warn it'll upend civilization once it lands. But ask five researchers what AGI actually means and you'll get five different answers. That definitional chaos isn't incidental. It's starting to look like the point.

What Does 'Artificial General Intelligence' Actually Mean?

AGI is generally understood as an AI system capable of understanding, learning, and applying knowledge across a wide range of tasks at something approaching human-level performance — not a narrow tool trained for one job, but a genuinely flexible reasoner. The concept has roots in AI research going back to the 1950s, though the specific term 'artificial general intelligence' was popularized in the early 2000s by researchers including Ben Goertzel, Shane Legg, and Peter Voss, who wanted to draw a clear line between narrow AI systems and the original grand ambition of the field.

That line has been blurring ever since. Artificial general intelligence has become a phrase that prominent figures use to mean wildly different things — sometimes within the same press release. 'There's a bunch of different definitions,' said Malo Bourgon, CEO of the Machine Intelligence Research Institute. 'When we start to talk about, is this system AGI? Is that system AGI? What precisely qualifies as AGI by what definition? I think that's kind of difficult to do.'

The Definitional Arms Race — Who's Already Claiming Victory?

Some people think AGI is already here. Shaw Walters, founder of Eliza Labs, told reporters during ETHDenver that current leading models already meet his personal definition of AGI. 'I think that we're at the inflection point where we have AGI,' he said. 'I completely believe that this is general intelligence. It's nothing like us. It learns in a completely different way, but it is intelligent.' Recent large language models — Gemini, ChatGPT, Grok, Claude — can write essays, generate code, create images, and answer complex questions. That's enough for some to call the race won.

Ben Goertzel disagrees. Hard. Goertzel, CEO of SingularityNET and one of the people who popularized the term in the first place, said the media discourse has gotten muddy in ways that serve a specific constituency. 'The term has become rather confused now in the media,' he said. 'Tech CEOs find it convenient to say, Hey, we've launched AGI already, and people sensationalize things.' His argument is precise: today's models aren't reasoning their way to general capability — they're interpolating across an enormous compressed representation of the internet. 'They get there not by learning to do all of it,' Goertzel said. 'They get there by having the whole internet crammed into their knowledge base.' A model trained only on music from before 1900, he argued, would never invent hip hop or grindcore. That's the gap between pattern matching at scale and genuine generalization.

Tech CEOs find it convenient to say, 'Hey, we've launched AGI already,' and people sensationalize things.

Does Elon Musk Think Grok 5 Could Be AGI?

What probability did xAI assign to Grok 5 reaching AGI?

10%. That's the figure Elon Musk put on it. Speaking at the Baron Investment Conference, Musk said Grok 5 would be 'the first time where I thought, well, we have a non-zero chance of achieving artificial general intelligence.' His broader timeline is aggressive even by the standards of an industry not known for modesty. In a December interview with XPRIZE Foundation executive chairman Peter Diamandis, Musk said he expected AGI by 2026 — and added that by 2030, AI would exceed the combined intelligence of every human alive.

The competitive edge Musk keeps citing is live data from X — the social platform's real-time stream, which he argues gives xAI's models a freshness advantage over rivals training on static datasets. Whether that actually moves the needle on general intelligence or just improves certain benchmarks is, predictably, contested. Bourgon's framework cuts to the real issue: autonomy. Most serious definitions of AGI require not just capability but independent agency — systems that can 'accomplish tasks in a wide variety of environments with a large amount of autonomy,' as he put it, rather than functioning as elaborate chat interfaces.

Why the AGI Timeline Debate Looks Different From Beijing

American AI labs are consumed by the existential dimensions of AGI. China's tech community mostly isn't. Kyle Chan, a researcher at the Brookings Institution studying global AI policy, described a striking divergence when he spoke to reporters about the state of the field. 'AGI is not such a big thing in China, especially from the policymakers, the broader AI community, the broader tech industry,' he said. 'Most people are focused on trying to make money on this thing, and especially on the physical side.'

That physical side — robotics, autonomous systems, drones — is where Chan says Chinese companies see their clearest competitive advantage over US rivals. The hardware supply chains are there. The manufacturing infrastructure is there. Why chase an undefined philosophical milestone when you can build things that roll, fly, and move? Chan acknowledged some Chinese AI founders do discuss AGI and even artificial superintelligence, but characterized it as a peripheral concern rather than an organizing mission. The contrast says something about how different the commercial incentives look from either side of that divide.

Goertzel sees the boundary between pre-AGI and AGI systems as inherently fuzzy — not a bright line but a gradient, like biology's uncertain territory around viruses and retroviruses. 'There doesn't have to be a completely crisp boundary between AGI and pre-AGI,' he said. We can still tell a dog is alive and a rock is not, even if some edge cases resist clean categorization. That framing matters because it deflates the dramatic 'AGI day' narrative that fundraising announcements tend to require.

Does the Label Even Matter?

Bourgon's honest answer is: not as much as people think. The Machine Intelligence Research Institute's work focuses less on whether any given system clears the AGI bar and more on what those systems can actually do — and what they do next. 'What are the effects and the capabilities of these systems?' he said. 'That's more the frame of mind we want to be in now.'

There's a cynical read here that deserves airtime. When a term is undefined, it can be claimed by anyone. Every lab that announces 'AGI-level performance' on some benchmark is doing something real: raising its valuation, locking in talent, and making competitors look behind. The vagueness of AGI is commercially useful in ways that a precise, agreed-upon definition would immediately destroy. Goertzel helped build the vocabulary. He also seems to be the one most troubled by what's been done with it. That's not irony — it's physics. Once a word enters the hype cycle, the people who coined it are the last ones in control of it.