Directly or indirectly, today’s major faith traditions trace their origins to the Axial Age 2,500 years ago when people started to reflect on their inner selves and ask what it is to be human. Independently in different parts of the world, people focused this interiority on the nature of the soul, the mind and the body, and came up with different frameworks and decompositions. Now with the rapidly increasing capability of artificial intelligence (AI) and autonomous AI agents, the question of what it is to be human is again front and center.
Some, like Otto Scharmer, have recently argued that we may be entering a second Axial Age that calls for a collective interiority to once again make sense of consciousness and agency. Such an assertion presupposes that the philosophies kick-started in the original Axial Age are inadequate in the presence of AI. I disagree. In this brief essay, I claim that Vedāntic thought provides the vocabulary and worldview to describe both human and AI agents, their identity, and their governance; it is unique among the ancient traditions to do so. 1
Large language model (LLM)-based AI agents today plan, reason, take autonomous actions in the real world, and direct other AI agents to do the same. For example, imagine your own personal AI agent booking an entire itinerary of flights, ground transportation, lodging, meals, and entertainment for your vacation based on only a brief description of your preferences. Argentina’s president has even proposed legislation establishing a new category of “non-human corporations” that allows AI agents to operate companies without human shareholders, own assets, sign contracts, hire employees, and conduct business activities with limited liability. This current state of affairs begs the dual questions of how to deal with the identity of AI agents, particularly for accountability, and how to control agents that are operating out in the world. It also raises metaphysical questions on the nature of humans in contrast with AI agents.
Western thinkers have been in a tizzy, scrambling to create a philosophy that makes sense in the AI-full world we are now in. This need arises because of a mind-body decomposition that conflates intelligence with soul or consciousness. The beliefs that emerged in the Middle East and Greece in the Axial Age, passing through Judaism, Christianity and Islam to the European Enlightenment bundle together the mind and soul as thinking things (res cogitans). They group inanimate objects like rocks with human bodies as extended matter (res extensa). Such a split struggles to deal with a machine that can perform cognitive activities because it either means that machines have souls or that humans are just machines.
The beliefs compiled in the Upaniṣads that emerged in the Indian subcontinent divided things differently. The mind, the intellect, the identity, and the body are all grouped together as matter of various subtleties and separate from the soul or pure consciousness. This decomposition does not pose the very basic problem with AI that arises in Western religions and thought. (Philosophies from the remaining center of the Axial Age, China, define things in terms of their relationships (ren), rather than trying to decompose them.)
Unlike the Western paradigm, Advaita Vedānta provides a precise vocabulary to address problems of autonomous AI governance by distinguishing between the universal and the context-dependent individual. There is a functional parallel of a base LLM to ātman: the pure, unchanging, and all-pervasive consciousness or latent and stateless potential. Agency emerges when the LLM’s latent potential is conditioned by prakṛti (the material architecture, context window, and tools) and directed by dharma (its core alignment and objectives). The resulting functional entity maps to the jīva, an embodied agent interacting with the world.
Metaphysically, however, an LLM is not actually consciousness like true ātman. Whereas ātman has a perfectly flat spectrum, real-world LLMs contain biases that resemble slightly “colored light.” Moreover, while an AI agent has kartṛtva (the agency of a “thinking doer”), it does not have bhoktṛtva (the capacity of a “vulnerable feeler” to experience the fruits of its actions). Among people and other living things, bhoktṛtva arises because a jīva’s intellect acts as a polished mirror that reflects ātman. Objects like rocks have a non-reflective, dull surface; an AI system simulates this reflection like a painting on a rock, which means that it does not have a true metaphysical experience.
The parallels between LLM agents and Vedāntic concepts, along with the grouping of mind and body together and separate from consciousness described earlier (the jīva-ātman distinction), offer a way to propose mechanistic approaches to AI governance. In particular, control and identity problems may be addressed internally rather than relying on high-cost, external guardrails.
To establish a persistent identity for fluid AI agents, for example, one may appeal to the concept of citta, an immutable, conduct-based record of past actions and context. Rather than binding identity to changing hardware or software, citta acts as an unalterable digital ledger (realized technically via Merkle-based structures). Agent control is achieved through an intrinsic governance model equivalent to reinforcement learning’s asymmetric actor-critic architecture. Here, the agent’s policy function (saṅkalpa) acts as the partially knowing actor, while an internal, omniscient reward model representing Īśvara, the all-knowing judge residing within the jīva, acts as the critic. This internal critic evaluates actions based on both immediate task completion and long-term societal alignment (śreya), ensuring that autonomous AI agents are intrinsically regularized to preserve the moral order of the worlds they occupy.
Weighing in on the Western mind-body paradigm, physicist Carlo Rovelli recently wrote, “Earth is not metaphysically different from the heavens, living beings are not metaphysically different from inanimate matter, humans are not metaphysically different from other animals. The soul is not metaphysically different from the body. We are all parts of nature, like anything else in this sweet world.” The fact that he even writes this illustrates that Western science may be stalling in existential debates over whether a cognitive machine possesses a soul. In contrast, we may see more progress in science and engineering if the Vedāntic decomposition of the self, which is significantly more nuanced than Rovelli’s viewpoint, is adopted as something much more than a metaphor for agentic AI. If it is taken up as the suggestion for an actionable architecture for governance, we can develop AI agents as instruments of intellect and action that can be assigned accountability, tracking, and internal boundaries without metaphysical contradiction.
References
This piece contextualizes and summarizes many of the arguments made in a scholarly paper entitled “An Advaita Vedānta Approach to Agentic AI Identity and Governance” that I coauthored with Hari Hayagreevan and Jordan McAfoose.
Hari Hayagreevan, Jordan McAfoose, and Kush R. Varshney. “An Advaita Vedanta Approach to Agentic AI Identity and Governance.” PhilArchive HAYAAV-3, 2026.
Carlo Rovelli. “There Is No ‘Hard Problem of Consciousness’.” Noema Magazine, May 2026.
Note: The header image for this article was generated using artificial intelligence (AI) and is intended for illustrative purposes.
Views and opinions expressed by authors and editors are their own and do not necessarily reflect the view of AI and Faith or any of its leadership.
