On the occasion of Ma Sita’s birthday, I was at the Sati Anusuya Ashram. This is where Lord Brahmas son set foot and where Sita received her Divya robes (which did not need to be washed or cleaned). It is also where the Mandakini Ganga springs out of the ground. The location where Dattatreya was born. Rishi Atri graced this during the 12 years that Lord Ram lived there in exile.
But what struck me is the similarity between ancient wisdom and world models of AI. And the potential of Sanskrit as a language to develop the world model.
Ai Gods are not new to me – one sees them everywhere like snake oil salesmen trying to sell their wares. But what I heard here was wisdom which is the God of Ai.
LLMs or Large Language Models are simply speaking about language. And language is not what human perception or comprehension is about. It is just a means of expressing them. And that is where LLMs hit the glass ceiling. Language can be fiction or non fiction – truth or lies. That is why LLMs can hallucinate wildly. There is training (like teaching a dog to play with a ball) and there is inference (getting the dog to go fetch the ball). But in a LLM there is no dog and no ball. So it is all imagination. This is the hash that the western world is smoking.
World models in AI on the other hand come from disciplines like reinforcement learning, neuroscience, and statistical inference. They are mathematical, testable systems designed to predict how the world evolves and how actions lead to outcomes.
In our ancient civilizations, eight thousand years ago at the time of Lord Ram, thinkers were grappling with the same underlying question: what does it take to act intelligently in a complex world?
Forget the dog. Let’s look at humans playing football or cricket. Can you imagine a World Cup without an empire or referee.
The idea of Pratītyasamutpāda framed reality as a web of interdependent conditions. Texts like the Arthashastra focused on anticipating consequences before taking action. Description in language is not enough; you need a deeper model of how reality works.
Modern AI expresses this in a more formal way. Through fields like Reinforcement Learning and Bayesian inference, systems build internal representations of the world, update them with new data, and use them to simulate possible futures. The goal is not philosophical clarity but predictive accuracy. If I hit a football, I want it to go into the goalpost.
Any system that needs to plan, adapt, and make decisions under uncertainty is forced to go beyond raw inputs and construct an internal model of reality. That constraint applies equally to humans and machines.
Across time, we keep rediscovering that intelligence requires more than information. It requires a working model of how the world actually behaves.
The IIT Alumni Council calls it an orientation that every sustainable perpetual system must adhere to:
Megaspheres.
That is where ai has to end.
To begin.
