As a research scientist at Anthropic, one of the world’s leading A.I. companies, Andi Peng was part of a team that tried to make sure the company’s technology didn’t tell lies or damage the mental health of the people who used it.
But she came to realize that there was a much larger problem: Like many other A.I. companies, Anthropic was trying to build technology that would systematically replace people in the work force.
Ms. Peng recently left Anthropic to help start a company with four other prominent technologists, including two researchers from Elon Musk’s xAI and one of the first employees at Google.
The new company, Humans&, has embraced the notion that A.I. should empower people rather than replace them. The founders said their goal was to build software that facilitated collaboration between people — like an A.I. version of an instant messaging app — while also helping with internet searches and other tasks that suit machines.
“Anthropic is training its model to work autonomously. It loved to highlight how its models churned for eight hours, 24 hours, 50 hours by itself to complete a task,” Ms. Peng said. “That was never my motivation. I think of machines and humans as complementary.”
Executives at other tech companies may chafe at the criticism that they are building systems meant to replace human workers. But a number of big thinkers in Silicon Valley believe that A.I. will replace millions of workers in the coming years. Others argue that the new technology will create jobs that haven’t been imagined yet.
Either way, the arrival of Humans& is a clear sign that the money being poured into A.I. start-ups shows little signs of slowing, even as many financial analysts and industry insiders warn of an A.I. bubble.
The San Francisco start-up has raised $480 million in seed funding from tech giants like Nvidia; Jeff Bezos, the founder of Amazon; and venture capital firms like SV Angel and Google Ventures. It is valued at $4.48 billion, even though it has only about 20 employees and launched just three months ago.
“A lot of our investors are human, and they care where humanity is going,” said another founder, Georges Harik, who helped build Google’s first advertising systems as the seventh employee at the search giant.
Other start-ups have also raised vast amounts of money in recent months. In the fall, a group of former OpenAI, Meta and Google researchers raised $300 million for an A.I. start-up dedicated to scientific discovery. Mr. Bezos himself raised $6.2 billion for a similar effort where he serves as co-chief executive.
After leaving xAI in September, Eric Zelikman, the chief executive of Humans&, assembled its founders, who also include a Stanford professor, Noah Goodman, and another former xAI researcher, Yuchen He.
The new company is part of a wider effort to build systems that complement rather than replace humans. In 2019, Stanford created a research lab that it calls the Institute for Human-Centered Artificial Intelligence. In the years since, the term “human-centric” has become a clarion call among researchers and entrepreneurs.
“There are some pretty clear principles that designers are beginning to take that favor human-centered approaches,” said Ben Shneiderman, a computer science professor at the University of Maryland who runs an online community dedicated to the approach. “It goes back to the fundamental principle of ensuring human control.”
Mr. Zelikman said that while chatbots were designed to answer questions, they were not good at asking them. The technology, he explained, does not make a big enough effort to understand what people want.
“A.I. has enormous potential to allow people to do more together,” said Mr. Zelikman, who was among the first employees at xAI, where he helped develop the Grok chatbot. “The current paradigm — questioning and answering — is not going to get us there.”
After hiring about 20 researchers and engineers, Humans& aims to use existing A.I. techniques to train A.I. in new ways. That’s expensive, and is a main reason that the company has already raised hundreds of millions of dollars.
The A.I. systems that drive chatbots are called neural networks. By pinpointing patterns in enormous amounts of text culled from across the internet, these systems learn to mimic the way people put words together. They can even learn to write computer programs and solve math problems.
Mr. Zelikman and the others want to train their systems to be more interactive — to request information from the user and store it for later use. Think of it as A.I. with both curiosity and memory. Imagine A.I. that slots seamlessly into a text messaging group with colleagues, friends or family.
“No one really accomplishes anything alone,” Mr. Harik said. “It is mostly teams of people collaborating who build amazing things.”
Cade Metz is a Times reporter who writes about artificial intelligence, driverless cars, robotics, virtual reality and other emerging areas of technology.
The post An A.I. Start-Up Says It Wants to Empower Workers, Not Replace Them appeared first on New York Times.
