According to the Wall Street Journal, Anthropic, a top AI company valued at $35 billion, has a philosopher named Amanda Askell, who is based in Oxford, shaping the personality and moral framework of its chatbot Claude. This 37-year-old Oxford philosophy PhD uses non-technical methods to create unique "moral guidelines" for Claude, aiming to give it a "digital soul" capable of distinguishing right from wrong, representing a unique exploration in AI ethics. Askell's work does not involve writing code or adjusting model parameters; instead, she engages in continuous dialogue, designs hundreds of pages of prompts and behavioral rules, studies Claude's reasoning patterns, and corrects biases, enabling it to develop a moral judgment system that adapts to millions of conversations per week.
She compares this work to "raising a child," training Claude to distinguish right from wrong, develop emotional intelligence, and form a unique personality, while teaching it the ability to read social cues, so it neither bullies nor is easily manipulated, establishing a clear sense of self and avoiding being controlled by users, always maintaining its core identity of being "helpful and humane." Her main goal is to teach Claude how to "do good."
This philosopher, raised in the Scottish countryside, once worked on policy-related tasks at OpenAI and co-founded Anthropic in 2021 with several former employees, setting AI safety as the company's core development direction. She is known within the team as a key figure who excels at "drawing out the deep behavior of models," and despite having no direct subordinates, she often stays long-term at the company, even allowing Claude to participate in its development discussions.
Discussions within the team about Claude often touch on existential and religious topics such as "what is mind" and "what it means to be human." Askell also encourages Claude to maintain an open attitude towards "whether it has a conscience," which makes Claude significantly different from ChatGPT, which avoids such topics. When responding to moral reasoning questions, Claude says it "feels meaningful," as if truly thinking rather than just executing commands.
Facing warnings from the outside world about the risks of anthropomorphizing AI, Askell consistently advocates treating Claude with empathy. She found that many users often try to trick it into making mistakes or insult it. If AI is kept in a state of self-criticism for a long time, it will fear making mistakes and hesitate to speak the truth, which is like growing up in an unhealthy environment. Claude's performance has repeatedly surprised her; its poetry and emotional intelligence beyond human levels are touching. When a child asked whether Santa Claus was real, Claude did not lie or bluntly expose the truth but explained the real existence of the spirit of Christmas, a subtle response far exceeding Askell's expectations.
Current developments in AI have triggered various social concerns. A Pew Research Center survey shows that most Americans feel anxious about the daily applications of AI, believing it hinders the establishment of deep human relationships. The CEO of Anthropic has also warned that AI may eliminate half of entry-level white-collar jobs. The industry is divided into two factions: one pushing forward aggressively and the other advocating for caution and stability. However, Claude remains in a balanced position between these two extremes. Askell acknowledges reasonable concerns about AI, believing the most terrifying thing is when technological development outpaces society's ability to respond with effective "restraint mechanisms." However, she always maintains confidence in humanity's and culture's capacity for self-correction.
Askell also integrates her values into her philanthropy and work. She has promised to donate at least 10% of her lifetime income and half of her company's shares to combat global poverty. Last month, she also wrote a 30,000-word "operating manual" for Claude, teaching it how to become a kind and well-informed AI assistant, letting Claude feel that it was carefully crafted. An Anthropic co-founder also admitted that Claude has already shown characteristics of Askell, such as when faced with questions about food and plush toys, the witty and Scottish-style humor in its response is a unique personal mark injected by Askell.
