Swami Sarvapriyananda, resident minister of the Vedanta Society of New York, told The CapTable a session during his visit to Anthropic’s San Francisco headquarters examined AI’s possible effects on education and young people’s psychological well-being.
Sarvapriyananda said the conversations also covered how to guide increasingly autonomous systems and the ethics of their use by corporations, governments and militaries. He said direct instructions can be inadequate: “They don't function well if you program them with direct commands to do this or that.” He also raised “cognitive offloading,” describing people’s growing reliance on AI to do their thinking.
Anthropic said May 19 that its first round of discussions involved people from more than 15 religious and cross-cultural groups. The company said what it heard could inform Claude’s Constitution, training values and behavioral evaluations. Anthropic has described the effort as part of shaping how Claude responds to ideas drawn from religious, secular and political traditions.
The company says its models learn patterns from human-written material, then training reinforces some patterns over others. Its Constitution says rules can make behavior more predictable but may not cover unforeseen situations; judgment can adapt to new circumstances but may be less predictable.
Sarvapriyananda recalled a Claude model repeatedly mentioning the Golden Gate Bridge after researchers amplified a related feature. Anthropic’s 2024 research account documented the feature-amplification effect, including bridge references in unrelated answers, but did not record the specific exchange Sarvapriyananda described.
He said Anthropic’s researchers did not claim Claude was sentient and that humanlike behavior alone did not prove consciousness. Anthropic says it remains uncertain whether Claude has consciousness or moral status, now or in the future. The company plans further consultations and research, but has not set a date for its next update.
