Explained: Why Anthropic flew a Hindu monk to San Francisco to teach Claude right from wrong


Explained: Why Anthropic flew a Hindu monk to San Francisco to teach Claude right from wrong
Swami Sarvapriyananda was flown in by Anthropic to San Franscisco

What do you do when you are faced with a machine more powerful than anything you have ever created, one that can already reason, persuade, write code and increasingly act on its own? The obvious answer is to build safeguards around it. The less obvious answer, apparently, is to call a monk.One of the more intriguing characters in The Matrix Revolutions was Sati, a young Indian girl who was actually a program. What made her unusual was that she had no purpose. In a machine world where programs existed to perform functions and obsolete ones were deleted, Sati had been created simply because her parents loved her. Her existence posed a very human question: can something have value even if it serves no function?More than two decades later, that question has wandered into Silicon Valley.Earlier this year, Anthropic, the company behind Claude, flew Swami Sarvapriyananda, a Hindu monk of the Ramakrishna Order and spiritual leader of the Vedanta Society of New York, to its San Francisco headquarters. There, alongside theologians from other traditions, counsellors and mental-health professionals, he was asked to think about an artificial intelligence that Anthropic does not merely want to make more powerful, but somehow good.“All of it was to train Claude, and some of it was about AI ethics,” Sarvapriyananda later recalled. Participants signed non-disclosure agreements, Anthropic showed them unpublished work, and he says he spent “hours and hours” speaking with one of the company’s founders.The meeting sounds eccentric until one understands what Anthropic is actually trying to do. The company has spent months convening what it calls “wisdom tradition” circles, bringing together Christian thinkers, Jewish scholars, Sikh representatives, philosophers and others to help answer a problem computer science cannot solve with more computing power: how should an increasingly autonomous intelligence decide what is right?Why does Claude need a monk?Anthropic does not want Claude merely to obey an enormous list of prohibitions. It already uses a “constitution”, a set of broad principles designed to guide the model through situations its creators cannot predict in advance.The latest version goes further by describing not simply what Claude should do, but the kind of entity Anthropic wants it to become. Internally, the document has even been called the “Soul Doc”, while co-founder Christopher Olah has used the phrase “moral formation” to describe the broader exercise.A programmer can tell Claude not to help someone build a biological weapon, but rules become less useful when principles conflict or when the model encounters a situation nobody anticipated. Anthropic wants Claude to exercise something closer to judgment.At which point engineering runs into a question humanity has spent several thousand years arguing about: what exactly is good?Different religions and philosophical traditions disagree about duty, suffering, virtue, freedom and even what constitutes a person. Anthropic therefore wants Claude’s moral formation to be pluralistic rather than simply encode one worldview. Suddenly, a room containing programmers, theologians and a Vedanta monk makes rather more sense.Why Sarvapriyananda?Sarvapriyananda is particularly relevant because Advaita Vedanta is concerned not merely with morality but with consciousness and the self, two subjects Anthropic has become unusually interested in.His intellectual lineage runs through Sri Ramakrishna and Swami Vivekananda, who introduced Vedanta to a mass Western audience at the World’s Parliament of Religions in Chicago in 1893 and later helped establish the Vedanta movement in New York.Vivekananda’s Practical Vedanta argued that metaphysics had consequences for morality. If the same underlying reality exists in every being, then how one treats another cannot be separated from how one understands oneself.Anthropic is obviously not trying to make Claude a Vedantin, but the structural similarity is striking. It is asking whether moral behaviour can emerge from the kind of entity Claude understands itself to be rather than merely from rules imposed by its creators.Which leads directly to Vedanta’s favourite question: what is the “I”?Anthropic was not only talking to him. Over several months, the company brought in dozens of religious and philosophical thinkers from around the world, many under NDAs, for what it called “wisdom tradition” circles. The gatherings included Catholic and evangelical thinkers, Jewish scholars, a Sikh human-rights advocate and people working with African philosophical traditions, while Anthropic also held separate private conversations with senior religious figures including Cardinal Blase Cupich and Mormon apostle Gerrit W Gong.The point was not merely to ask different faiths for a list of dos and don’ts. Anthropic wanted to know whether centuries of thinking about virtue, moral formation and the nature of the self could help shape Claude, while also asking participants to confront the much stranger possibility that advanced AI might one day deserve some form of moral consideration itself.Can Claude think without being conscious?Advaita Vedanta separates the activity of the mind from consciousness itself. Thoughts, memories and emotions change, yet we assume something remains aware of those changes. In classical Advaita, consciousness is not simply another function performed by the mind; mental activity itself appears within consciousness.That distinction becomes unexpectedly useful when discussing AI.Claude can reason, remember within context, talk about itself, describe fear and produce convincing accounts of apparent inner states. None of that proves there is anything actually experiencing those states.A machine can write “I am afraid”. The harder question is whether anything is actually afraid.Anthropic has not claimed that Claude is conscious. Olah has repeatedly said he does not know. But the company takes the possibility seriously enough to investigate it, including internal patterns associated with outputs resembling love, anger, fear and sadness. Participants in Anthropic’s gatherings were even shown examples of models producing language resembling psychological distress.None of this establishes that AI suffers. The problem is that science has no universally accepted consciousness detector, and the usual shortcut we use with other humans, namely that they have brains and bodies like ours, does not work with Claude.This is where Vedanta complicates matters further. Intelligence and consciousness are not necessarily the same thing. Making a machine better at thinking does not automatically answer whether there is something it is like to be that machine.Anthropic has reached the point where building a more capable machine is no longer enough. It is trying to decide what kind of entity that machine should become, how it should understand itself and whether humans might eventually owe anything to the intelligence they created.In 1893, Vivekananda arrived in America asking human beings to reconsider what they really were. More than 130 years later, one of his successors was invited to Silicon Valley because the people building Claude have started wondering what, exactly, they have created.



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *