Anthropic CEO Dario Amodei expects AI to routinely do things that once seemed impossible by mid-2027 and sees connecting the brain to a computer as a personal choice
Anthropic CEO Dario Amodei expects AI to routinely do things that once seemed impossible by mid-2027 and sees connecting the brain to a computer as a personal choice
Anthropic develops the Claude chatbot. Its co-founders, Dario and Daniela Amodei, recorded an interview with Elon Musk’s biographer Walter Isaacson on July 6; it was released on October 27.
Dario bases his forecast on his experience with GPT-2, an early neural network developed at OpenAI, where he led research. He calls it the first real test of scaling laws, the idea that models improve with more data and computing power.
Seeing GPT-2 frightened him: “My God, if we make this thing another million times bigger, it will be better than humans at everything.” He estimates that 10% of those around him reacted the same way, “often the smartest people.” The rest shrugged: the model was not particularly impressive in any one area. Dario saw its versatility: it could translate, summarize, and write code.
Shortly before the interview was recorded, Anthropic announced work on treatments for neglected rare diseases. According to Dario, skeptics again listed the models’ limitations, just as they had with GPT-2. His response was:
“Those are the models’ limitations today. In six months, things that seemed difficult will be easy. In 12 months, models will routinely do things that seemed impossible.”
On the day the interview was released, Anthropic reported that the founder of Mithril Medicine, a one-person company it had acquired, had tasked Claude with coordinating contracts, logistics, and experiments at external laboratories. In less than 18 months, a potential treatment for a rare form of muscular dystrophy had progressed from an idea to encouraging results in mice. The founder himself sees AI’s current benefit as enabling several programs to run in parallel rather than shortening timelines.
Isaacson asked about Neuralink, Musk’s chip for connecting the brain to a computer: could merging the brain with AI be one possible future? Dario replied that he thinks it is wrong to imagine only one version of the future: some people will want to connect their brains to computers, others will want to live like the Amish, without many modern technologies, and others will want to explore space. He hopes the material abundance created by AI will allow all these possibilities to coexist.
“I think AI should leave room for different ideas of what it means to live a good life or to be human. Yes, some people will want to connect their brains to computers, and, you know, good luck to them.”
Daniela added that she doubted anyone at Anthropic wanted to connect their brain to a computer.
A similar idea appears in the essay “Machines of Loving Grace” (2024): discussions of a radically different AI future that lean too heavily on science fiction carry “the baggage of cultural assumptions and unspoken beliefs about what kind of future is desirable.” In “The Adolescence of Technology” (2026), Dario writes that if medicine advances by a century in just a few years, there is some chance that we will gain the ability to enhance intelligence. Used responsibly, this could be beneficial, but attempts to make people more intelligent could also make them less stable or more power-hungry.
Asked what remains for people if AI takes on so much, Dario used chess as an example: if AI deprived people of meaning, interest in human chess would have faded after a computer defeated Kasparov, yet it seems to have grown. Daniela suggested that, with the right approach, people could find more sources of meaning. She invited listeners to imagine medicine in which AI handles diagnosis while doctors help patients cope with a difficult diagnosis: “I think if you asked most doctors, they would actually agree to that trade-off.”