In a recent move towards AI safety and ethics, Anthropic AI, an Amazon-backed AI startup and rival to OpenAI, has undergone a significant transformation in its governance. The company has embraced a democratic approach involving the public in creating its new AI constitution, which serves as a set of guidelines and principles for training its chatbot, Claude. The constitution, consisting of 75 points, highlights the importance of balanced and objective responses, as well as accessibility and ethical considerations.
One of the key aspects emphasized in the new constitution is the avoidance of toxic, racist, sexist, violent, or unethical responses. It states that the chatbot’s responses should be wise, peaceful, and ethical. This approach reflects the company’s commitment to building a safer AI model.
Anthropic was founded in 2021 Dario Amodei, a former employee of OpenAI. His departure from the organization was driven the desire to develop a more secure AI system. The San Francisco-based company focuses on AI safety and research, developing its own AI systems and large language models. It received a substantial investment of $500 million from Sam Bankman-Fried of FTX last year, and Amazon recently announced its investment of up to $4 billion in the company.
To ensure a diverse and inclusive set of guidelines, Anthropic partnered with the research firm Collective Intelligence Project. They conducted surveys involving 1,000 Americans, taking into account factors such as age, gender, income, and location. Participants were given the opportunity to vote for or against existing rules or propose new ones. The company then incorporated widely shared sentiments into its constitution, using them for the retraining of its AI chatbot.
The new constitution incorporates over 1,000 statements and more than 38,000 votes from the public. Users expressed the desire for the chatbot to admit flaws, prioritize good mental health, and not exhibit jealousy towards humans. Principles that discouraged racism and sexism and encouraged reliability and honesty were also popular among voters.
However, not all public statements made it into the constitution. Some suggestions, including training AI without the principles of diversity, equity, and inclusion (DEI), or imbuing AI with emotions, did not receive widespread agreement.
The update to Anthropic’s constitution comes in response to growing concerns about AI safety, voiced both technology leaders and government officials. Elon Musk, the co-founder of OpenAI, has stated that AI poses one of the biggest risks to the future of civilization. Microsoft CEO Satya Nadella has stressed the importance of keeping powerful AI models in check, ensuring that humans remain in control.
In June, Anthropic’s Amodei, along with Sam Altman from OpenAI, Sundar Pichai from Google, and Tim Cook from Apple, met with White House officials in Washington, DC, to discuss plans to address AI safety concerns.
Anthropic’s new AI constitution is a significant step towards creating responsible and ethical AI systems. The full version of the constitution can be found on the company’s website.