Summer 1997. At a nanotechnology conference in Palo Alto, a physicist named Mark Gubrud He presents a paper discussing autonomous weapons and advanced artificial intelligence. He uses a term, AGI, that no one has ever heard before: "artificial general intelligence." No one pays attention. The term ends up buried in the proceedings of a niche conference, one of those halfway between scientists and obsessive nerds.
Twenty-eight years later, that same acronym dominates the headlines, fuels tens of billions of dollars in deals between OpenAI and Microsoft, and justifies colossal investments by Meta and Google. The term AGI has become the Holy Grail of technology. But who actually invented it? The answer is stranger than you might imagine. And it involves a man who now lives in a cabin in Colorado, with a "useless" doctorate and no money.
A forgotten conference and a definition, AGI, which was prophetic
In 1997, Mark Gubrud He was obsessed with nanotechnology and its dangers. Sitting (literally) in the basement of theUniversity of Maryland He read everything he could about the risks of emerging technologies. In particular, he was concerned about autonomous weapons. So he wrote a paper for the Fifth Foresight Conference on Molecular NanotechnologyThe title was: “Nanotechnology and International Security”Inside, a phrase that would change the world.
The AGI definition that Gubrud proposed in 1997 was already complete:
“artificial intelligence systems that match or surpass the human brain in complexity and speed, capable of acquiring, manipulating, and reasoning with general knowledge, usable in any phase of industrial or military operations where human intelligence would be required.”
Twenty-eight years later, that definition still holds true. It's the foundation of everything.
The paper circulated little. Its impact was minimal. Gubrud continued to work on arms control, without ever really returning to the term. The AGI had just been born—in fact, it had just been christened—but no one knew it.
2002: Shane Legg, Ben Goertzel and the “reinvention”
Five years later, Shane Legg was working on a book with Ben Goertzel and other researchers. They wanted to give artificial intelligence a name that wasn't specific, but general. Like human intelligence. They tried "real AI" or "synthetic intelligence." Nothing worked. Then Legg had an idea: AGI, "artificial general intelligence." It sounded good. They adopted it. The book came out in 2007, but already in 2002 the term began to circulate in forums and conferences, as if it had only appeared at that moment.
That's when Gubrud showed up. In the mid-2000s, he wrote to Goertzel and Legg to point out that he had used the term as early as 1997. They checked. It was true. Legg himself recalls it:
“Someone comes out of nowhere and says, 'I invented the AGI definition in '97,' and we say, 'Who the hell are you?' Then we checked, and indeed there was a paper. So I didn't invent the term, I reinvented it.”
Why Gubrud disappeared and why nobody knows him today
Gubrud admits that the lack of follow-up left him out. "I accept credit for the first citation and give them credit for all the work I didn't do," he explains. His focus was elsewhere: autonomous weapons, the arms race, military risks. The AGI definition he proposed was a means, not an end. A way to warn of danger. And then, he simply didn't do anything about it.
Meanwhile, Legg co-founded DeepMind (later acquired by Google) and became Chief AGI Scientist. Goertzel founded SingularityNET and became an icon in the industry. Gubrud? He remained in the shadows. Today he's 66, caring for his mother, and has no steady job. "It's a paradox that doesn't escape me," he says.
“This term is literally worth trillions of dollars. And I'm a 66-year-old with a useless doctorate, no name, no money, no job.”
AGI's definition today: billion-dollar deals and the race to the future
In 2025, AGI is everywhere. OpenAI just announced which will have a fully autonomous AI researcher by 2028. Microsoft has signed agreements with OpenAI that hinge exactly on achieving AGI. Meta e Google They invest hundreds of billions to get there before everyone else. Demis Hassabis of DeepMindNobel laureate says that “we are in the final years of pre-AGI civilization.” American politicians warn: if China gets there first, we’ve lost.
The definition of AGI that everyone uses today is essentially Gubrud's: systems capable of equaling or surpassing human intelligence in any cognitive task, with general learning, reasoning, and adaptation capabilitiesOf course, there are variations. Google DeepMind proposed five levels of performance to classify the AGI. OpenAI He talks about “autonomous systems that outperform humans in most economically relevant jobs.” But the substance remains the same as in 1997.
Recap: When is it coming? The forecasts converge: according to leading experts, the AGI will arrive between 2027 and 2030. Shane Legg, the one who reinvented the term in 2002, has always said he believed in a 50% probability by 2028. Geoffrey Hinton, the “godfather of AI,” has lowered his estimate from 30-50 years to just 5-10 years. Ben Goertzel He mentions 2027 as a plausible date.
Gubrud's warnings remain relevant
The most interesting thing? Gubrud's warnings in 1997 are still valid. He, remember, used the AGI definition in a paper on military risks. He skipped over the entire "visible" future and went straight to the worst-case scenario (a bit like Snowden, but decades earlier).
He talked about autonomous weapons, systems that make life-or-death decisions without human supervision, an arms race that no one can stop. "My goal was to warn of that danger," he explains. Some thought he was crazy, but today the debates onAI that could feign stupidity to dominate or on risk of extinction for humanity they exactly reflect his fears from 1997.
Gubrud has written other papers on the topic. He has called for a ban on autonomous killer robots. He has spoken at conferences on AI safety. But his career has been fragmented, his voice marginal. Meanwhile, the acronym he coined has become the center of the 21st-century technological revolution.
A name for the future, a life in the shadows
Mark Gubrud's story is one Silicon Valley isn't keen on telling. To hear them say, every garage in Palo Alto and Cupertino has created a billionaire, but the rule is exactly the opposite. The rule is like Gubrud: an outsider who anticipates everyone, proposes a definition of perfect AGI, warns of the risks, and then disappears. No billion-dollar startup, no cover story on... Wired, no TED Talk. Just a 1997 paper that very few have read, but which gave a name to the future.
Because the future is created by those without a name, even if those who are best at shaping it pass into history.
