SYSTEM / DEC.2024
AGI. Who cares.
In the vocabulary of artificial intelligence (AI), AGI, or Artificial General Intelligence, holds a special place. The term fascinates as much as it irritates, with a simple idea behind it...
Loading page…
SYSTEM / DEC.2024
In the vocabulary of artificial intelligence (AI), AGI, or Artificial General Intelligence, holds a special place. The term fascinates as much as it irritates, with a simple idea behind it...
In the vocabulary of artificial intelligence, AGI, short for Artificial General Intelligence, occupies a strange place. The word attracts as much as it irritates. It generally refers to a machine capable of learning a complex task without specific training, observing its environment, then adjusting its behavior. The short version: human intelligence simulated by a machine.
DeepMind and OpenAI publicly state that they are pursuing this direction. Yann LeCun, whom I had the good fortune to hear at a conference at Station F, considers the idea of imminent AGI unrealistic, even misleading. His argument can easily be expressed in one sentence. Discussing a technology that is still largely hypothetical today makes the conversation more confusing than useful.
The reference points are shifting nonetheless. What has long been presented as a technological pinnacle, sometimes with a Terminator-style threat, could end up looking like a milestone. Sam Altman, CEO of OpenAI, takes this idea even further. That milestone could pass almost unnoticed.
During a recent interview at the New York Times DealBook Summit, reported by The Verge, Sam Altman played down the importance of AGI. He describes its arrival as a stage whose impact could be less visible than many imagine.
My intuition is that we will reach AGI sooner than most people think, but that it will matter much less, he said.
Altman explains that the arrival of AGI, often imagined as a historic turning point, could extend the acceleration already underway. In his view, the biggest change would come afterwards, with what he calls superintelligence: AI that would far exceed human capabilities.
He adds that the major security and ethical problems associated with AI would be concentrated more in the progression towards this superintelligence. They would become more serious as systems grow more capable. Between these two moments, Altman imagines a long phase of progress, with economic and social effects gradually taking hold.
Altman’s position encourages a less theatrical view of AGI. Treating it as a final destination seems questionable to me. The history of technology suggests instead that each tool becomes a foundation for the next stage, especially when it increases our ability to create.
The telephone illustrates this well. In the Middle Ages, a machine capable of transmitting voices over a distance would probably have been seen as the pinnacle of technology. Today, the telephone is an ordinary object, integrated into everyday life. AGI could follow the same path. A very advanced tool at first, then one tool among others in a continuous succession of technological advances.
The idea of a singularity, with a culminating point at which machines would definitively surpass humans, remains heavily influenced by science fiction. These systems can become extremely capable. They can accelerate research and coding, as well as writing or analysis. Yet presenting them as the end of technological history amounts to overestimating the machine while underestimating what humans do with their tools.
AGI deserves to be assessed on what it will actually make possible, far removed from the myth that is adored or feared. If it arrives, its value will lie in its ability to accelerate the resolution of complex problems, including global challenges. Humans retain a central role in areas where creativity, judgment and intention matter.
AGI can be an important milestone. It remains a milestone.