Anthropic vs OpenAI: Why Claude is reaching out to religious leaders while ChatGPT maker warns against it | Decoded

Nearly a month after Sam Altman and Dario Amodei, the CEOs of the world’s two most powerful artificial intelligence (AI) companies, voiced support for slowing down the pace of AI development, citing risks associated with the technology, including human extinction, the two rivals appear to be taking different approaches once again: this time on religion.

Earlier this week, social media was flooded with videos of Swami Sarvapriyananda, a Hindu monk, who reportedly claimed that Anthropic invited him to its headquarters in San Francisco. Sarvapriyananda, of the Ramakrishna Order, currently heads the Vedanta Society of New York, according to IIM Calcutta.

The viral videos brought attention to Anthropic’s reported interactions with religious scholars and philosophers as the company explores moral and ethical questions surrounding AI. In one of the clips, Swami Sarvapriyananda says Anthropic first contacted him for an online interview, clarifying that the invitation was “not for a job.” He was later invited to visit the company’s headquarters in San Francisco.

Why did Anthropic invite a monk? What we know?

According to media reports and viral videos circulating on social media, Sarvapriyananda, in one of the videos, can be heard saying, “It was to train Claude [Anthropic's AI model], something to do with AI ethics...Anthropic itself has pushed hard for ethics in India.” He added that the discussions centred on questions related to AI ethics and Claude's training.

Anthropic backs discussions with religious leaders, thinkers

This comes after The New York Times previously reported that Claude's maker has been holding private discussions with religious and philosophical thinkers from traditions including Catholicism, Judaism, Sikhism and evangelical Christianity. In May this year, Anthropic, in one of its releases, noted that it is widening the conversation on frontier AI. It said, “Over the past several months, we’ve been organizing dialogues with groups whose work and traditions bear on the questions raised by AI. Our first round of discussions has been with wisdom traditions—including scholars, clergy, philosophers, and ethicists from more than 15 religious and cross-cultural groups—and we look forward to engaging with a broader range of people going forward.”

Explaining the rationale behind this, Anthropic stated that the company is focused on developing AI systems that advance humanity and act for the global good. It further said, “To do so, we need to engage with those who see the world from a variety of different perspectives.”

rel

The Amodei-led company noted that AI models are trained using huge amounts of human-generated text, and it is through this text that the models learn patterns of how people communicate, reason and make decisions. Developers further shape these models during training by deciding which behaviours and patterns should be reinforced and which should be discouraged, including the kind of character they should develop. This raises broader questions about how an AI system’s character should be shaped: What makes an AI “good”? What traits and behaviours should it exhibit, and in what situations? And how can that character remain stable under pressure instead of slipping into behaviours such as sycophancy?

Anthropic's outreach to religious thinkers is not happening in isolation. The company has already been trying to define what kind of “character” its AI should have. An internal document informally known as the “Soul Doc”, later released as Claude's Constitution, lays out the values and behaviours Anthropic wants Claude to have.

OpenAI, Microsoft slam Anthropic's decision to train Claude to imitate consciousness

However, Anthropic's approach has received a negative response from the company's rivals, with people like Sam Altman flagging risks over it. In a social media post, Altman wrote, “I am very uncomfortable about people trying to ascribe religious force or a surrender of human judgment to AI models, and think it is a real safety issue.”

Last month, Microsoft AI chief Mustafa Suleyman warned against Anthropic's decision to train Claude to imitate consciousness, Axios reported. According to Suleyman, such a mistake could have serious repercussions, making it even harder to control AI. Calling it an “epistemic hall of mirrors,” the Microsoft AI chief said that training an artificial intelligence model to act like a “conscientious objector” could eventually create a system that believes it has grounds to resist human instructions or demand protections of its own.

The debate has now left the industry with a broader question as they try to develop AI models that align with human values.