Dario Amodei, CEO of $19 billion AI startup Anthropic, doesn’t think the evolution of artificial intelligence models and tools poses any immediate threat to humanity. But he has a problem with the justification his colleagues use to deal with threats.
Speaking at an AI conference in San Francisco on Wednesday, Amodei took issue with the idea that AI models are, and always will be, mere chatbots with limited capabilities. He particularly criticized a version of this mindset espoused by Marc Andreessen, the famous venture capitalist and champion of unfettered AI, who has famously dismissed concerns by arguing that AI is really just math. “Limiting AI means limiting math, software and chips,” Andresen tweeted in March.
According to Amudi, this logic does not hold. Because everything in the world can be classified as mathematics, he said.
“Isn’t your brain just math? A neuron fires and sums up calculations, that’s math, too,” he said on stage Wednesday during Eric Newcomer’s Cerebral Valley conference. “Like, we shouldn’t be afraid of Hitler, it’s just math. The whole universe is math.”
Amodei, a former vice president at OpenAI who left in 2021 to start rival LLM firm Anthropic, is among a group of AI executives who are outspoken about the technology’s potential risks, from rogue models to bad actors. Be warned. The CEO supports some regulation of the AI industry and Anthropic even supported a controversial California bill to that end, which was ultimately vetoed.
Andreessen, whose VC firm has invested billions of dollars in several AI companies, including OpenAI and Elon Musk’s Xai, is on the other side: an AI “boomer,” who calls for unfettered development of AI technology by individual companies. is doing “The ‘rule’ of AI (mathematics) is the foundation of a new totalitarianism,” VC wrote last year. He also called AI safety critics, or domers, “a cult.” A representative for Andreessen declined to comment on Amoudi’s criticism.
Amodei acknowledged during the conference that today’s AI models are “not smart enough… not autonomous enough” to pose much of a threat to humans. But he noted that the technology is advancing rapidly, with AI “agents” capable of acting autonomously on behalf of human command. As those still-nascent tools emerge, Amodei said the public will get a deeper sense of what AI is capable of and the potential pitfalls.
“Today, people laugh when chatbots say something unexpected,” Amudi said. “But we have to do a better job of controlling agents.”
Are you an employee at a tech company or have an insight or tip to share? Connect securely through KaliHaze. Signal at +1-949-280-0267 or at kali.hays@fortune.com.
