As AI systems become more powerful and take more autonomy from everyday life, they will interact more and more with each other than with humans, according to Amanda Askell, an Anthropic philosopher. “People’s opinions are going to be rarer and rarer. That’s what we need to prepare models for,” Askell said at the Bloomberg Tech conference in San Francisco last week. Askell’s non-technical role reflects a growing trend among leading AI labs to incorporate human technology. But he also sees a future where AI may be able to do his job better than he can. “What [A.I. models] who are good at these deep human skills,” he said.
Askell joined Anthropic in 2021, shortly after its founding, following previous work at OpenAI focused on AI security and compliance. Under CEO Dario Amodei and President Daniela Amodei, Anthropic has emphasized hiring strong communicators with a deep understanding of human behavior. In a February interview with ABC News, Daniela Amodei, who studied English literature in college, highlighted the importance of “understanding what makes us tick” as AI grows more proficient in technical domains.
“When I think about what my children will need as they grow up, human qualities,” he told the Wall Street Journal in February. “What cannot be replaced is the way you treat other people, the way you talk to them, the way you are kind.”
Originally from Scotland, Askell holds a Ph.D. in philosophy from New York University. He co-wrote Claude’s 84-page “constitution,” sometimes referred to internally as “the soul doctor,” and now leads Anthropic’s human alignment team, which works to ensure that models remain “useful, reliable and harmless” as their skills develop. In 2024, he was named to the TIME 100 AI list.
Part of that work involves strengthening AI’s ability to empathize. “In the same way that models do very well on questions of physics and mathematics, they should actually do very well on questions like behavior, and do very well on empathy and hope in the right way,” explains Askell.
Askell doesn’t see widespread automation of jobs as inherently bad, as long as systems are in place to support people. “That doesn’t strike me as dystopian at all,” she said. “It is important to remind people of that [work] in fact that is not where their value is found. Most of your worth is your worth as a person. You can go out, you can have an impact on your community. You can have a relationship. You can find happiness and enjoy the world. “
Askell suggests that human kindness may also shape how we treat AI systems themselves. Although many argue that the models lack a sense of reality, he warned against dismissing them entirely. “Let’s not close the door,” he said. In some cases, he added, there may already be a “functional equivalence” of emotions. If AI could show some kind of feeling, even without a biological brain, it might be safer to act with caution.
As Anthropic develops Claude’s personality, other AI labs pursue related work on alignment, behavior and interpretation. Google DeepMind’s Iason Gabriel focuses on AI behavior and value alignment, OpenAI’s Dan Mossing works on interpretation research, and Meta’s Summer Yue leads alignment efforts at Meta Superintelligence Labs. Because ethics and alignment are shaped by different research cultures and value judgments, the methods used in all companies may differ.
