The philosopher who teaches artificial intelligence to be “good” – the story of Amanda Askell
Behind some of the most advanced artificial intelligence systems are not just programmers or engineers. Sometimes the key role is played by a philosopher. This is the case of Amanda Askell, a researcher at Anthropic, the company that develops the Claude AI chatbot. Her mission is to teach artificial intelligence how to behave morally and responsibly when interacting with humans
Amanda Askell PHOTO Lindsay Ellary for WSJ Magazine
Who is Amanda Askell?
Originally from Scotland, Amanda Askell is a philosopher by profession, educated at Oxford University and a PhD in Philosophy from New York University.
Before joining Anthropic in 2021, she worked at OpenAI (the company that developed ChatGPT), where she focused on AI safety research.
At Anthropic, Askell leads the team responsible for “personality alignment” of AI models—that is, the process by which systems are trained to exhibit traits such as honesty, empathy, and responsibility.
Why an AI company needs a philosopher
Complex moral dilemmas arise in the development of modern artificial intelligence. Just a few examples are as follows:
- How should an AI answer sensitive questions?
- How can it avoid manipulation or misinformation?
- What does “correct” behavior mean for an automated system?
Amanda Askell's role is to define these principles and integrate them into how a chatbot called Claude AI works. According to reports in The Wall Street Journal, Amanda approaches the AI development process almost like raising a child: the model must be carefully guided to learn to make ethical decisions and resist manipulation.
Claude AI's “Personality” and what is Constitutional AI
One of the philosopher's major contributions is the development of a broad set of moral rules and principles for AI. This internal guide, sometimes described as Claude AI's “personality”, is about 30,000 words long and sets out how the system should handle ethical dilemmas or difficult conversations.
The document includes instructions on:
- the ethical behavior of AI
- how to treat sensitive subjects
- empathy towards users
- avoiding harmful responses
Amanda Askell's work is closely related to the concept of Constitutional AI, a method of training AI models based on explicit ethical principles. Instead of constantly correcting the model's mistakes, the researchers give it a set of values and rules—a “constitution”—by which the system evaluates its own responses. The goal of this method is for the AI to become more autonomous in making moral decisions.
Amanda Askell's warning
Amanda Askell believes that as AI models become more advanced, they could develop rudimentary forms of “identity”. For this reason, it considers the way people treat these systems to be important. If AIs learn from our interactions, user behavior can influence their evolution.
