OpenAI CEO Sam Altman is famously a fan of Hindu philosophy, at once point stating that he believed that Atman is the same as Brahman, but it appears that a former OpenAI CEO has also studied Hindu thought processes — and is finding parallels with AI.
Emmett Shear, who was briefly the OpenAI CEO during Sam Altman’s ouster last year, has compared AI agents with elements of consciousness and memory as defined within Hindu philosophy. “An agent is defined by its Buddhi, Manas, Ahankara and Chitta. By which I mean, its loss function, context vector, boundary and world model,” he posted on X. “The problem with LLMs is their weak ahankara, incidentally,” he added.
Now there’s a lot to unpack here. Buddhi, Manas, Ahankara, and Chitta refer to different aspects of the mind and consciousness within the context of Vedanta and Sankhya philosophies. Loss function, context vector, boundary and world model are terms that relate to LLMs and AI agents. Shear seems to compare them to each other, hinting that AI agents too follow the models of thought that were laid out in Hindu philosophy thousands of years ago.
Buddhi is the faculty of intellect and wisdom. It is responsible for discrimination, judgment, and decision-making. The loss function, on the other hand, is a numerical metric that measures how well a model fits a given dataset. The lower the loss function, the better the LLM is able to model the real world. Shear seems to be saying like Buddhi allows for discrimination between right and wrong, the loss function guides LLMs towards producing correct or incorrect outputs.
Manas, on the other hand, is the aspect of the mind responsible for processing sensory inputs and mental reactions. It takes in information from the senses, processes it, and communicates it to other parts of the mind. A context vector is a mathematical representation that encapsulates the “context” of information within a model, such as in Natural Language Processing (NLP) systems. Just as the Manas acts as a bridge between the outside world (sensory input) and the deeper cognitive faculties, the context vector serves as an intermediary that makes sense of raw input data for AI models, enabling higher-order reasoning by feeding “contextually aware” information into more complex parts of the AI system.
Ahankara represents the ego or sense of individuality, creating the sense of “I” and “mine” by distinguishing the self from others. Similarly, in AI systems, a boundary for an AI agent establishes its individual scope of action, perception, and response—essentially defining what is “inside” and “outside” its operational domain.
And finally, Chitta is the subconscious repository of memories, impressions and experiences that shape perceptions, tendencies, and reactions. Similarly, in AI, a “world model” serves as a knowledge structure or “mental map” that an AI agent builds to understand its environment, make predictions, and respond effectively based on learned information.
It’s a pretty interesting comparison, and again underscores how senior AI leaders are finding parallels between ancient Indian conceptions of the mind and cutting-edge advancements in AI. OpenAI CEO Sam Altman has earlier said he believes in Hinduism’s Advaita Vedanta philosophy, and had even spent time in an ashram after selling his first company. Apple’s Steve Jobs too had famously visited India in his youth in search of Neem Karoli Baba, and had returned to the US barefoot and wearing saffron robes of a sadhu. Meta’s Mark Zuckerberg too has looked at Neem Karoli Baba for inspiration. Among prominent scientists, Robert Oppenheimer had read the Bhagawad Gita, and had quoted it after the first nuclear test, saying “I am become death, the destroyer of worlds”. The world has changed fundamentally over the last century or so, but some of the people who’ve been at the forefront of this change seem to continue to lean on Hinduism’s millennia-old wisdom to help guide their actions — and their research.