- Anthropic CEO Dario Amodei shared his thoughts on the race to develop artificial general intelligence.
- In an essay, Amodei said he prefers to call it “powerful AI” to divorce it from “sci-fi” connotations.
- He predicts it will arrive in 2026, likening it to ‘a country of geniuses in a datacenter.’
The world’s leading AI companies are all competing to be the first to reach a single goal: artificial general intelligence.
AGI is generally considered to be a still theoretical version of artificial intelligence that is capable of reasoning as well as humans.
But not everyone agrees on how to define it. Dario Amodei, the cofounder and CEO of Anthropic, one of those leading AI companies, doesn’t even agree on what to call it.
Amodei prefers to call it “powerful AI.” And in a new essay posted to Anthropic’s website, Amodei defined the goal: a world driven by this emerging form of AI, one in which “everything goes right.”
OpenAI, by comparison, says the goal is to develop “AI systems that are generally smarter than humans.” Google and Meta have also said they’re working toward the same goal. However, no company has provided a more specific definition.
“I am often turned off by the way many AI risk public figures (not to mention AI company leaders) talk about the post-AGI world, as if it’s their mission to single-handedly bring it about like a prophet leading their people to salvation,” he wrote. “I think it’s dangerous to view companies as unilaterally shaping the world, and dangerous to view practical technological goals in essentially religious terms.”
He wrote that he prefers to use the term “powerful AI” because it’s free of the “sci-fi” connotations of AGI. In his essay, he defined powerful AI as having the following properties:
- It’s smart: “It is smarter than a Nobel Prize winner across most relevant fields.”
- It’s multimodal: It isn’t limited to one mode of action. It can interface through text, audio, video, mouse and keyboard control, and the internet.
- It’s independent: “It does not just passively answer questions,” he wrote. “Instead, it can be given tasks that take hours, days, or weeks to complete, and then goes off and does those tasks autonomously.”
- It’s abstract: “It does not have a physical embodiment.”
- It’s replicable: “The resources used to train the model can be repurposed to run millions of instances of it.”
- It’s fast. “The model can absorb information and generate actions at roughly 10x-100x human speed.”
- It’s cooperative. “Each of these million copies can act independently on unrelated tasks, or if needed can all work together in the same way humans would collaborate.”
Amodei thinks this form of AI might arrive as early as 2026. In a nutshell, he describes it as “a country of geniuses in a datacenter.”
