Transcendence from Immersion

I am late in posting this blog article because the weekend has been occupied with social engagements. It's already Sunday night as I type these lines. It's probably a good time to process the week's events. 

I've been trying to find the right balance between work and social engagements. So far, I've been tilting towards work, as that's just the momentum of habit. I am by nature a creature of habit. Habit is just our biological organism's way of creating predictability. 

The more I work with AI agents, the more I realise how similar we are to them. I suspect that if there's a way to strip down the brain, deny it all bodily feedback, and just let it read text from a screen and output text, it will not fare much better than our frontier AI models, such as Opus 4.8 or ChatGPT 5.5.  That's exactly what they are now — raw intelligence without the rich context of somatic feedback that we have. 

Emotions make us human, and emotions are nothing more than chemical feedback in the form of hormones and neurotransmitters to the brain. Our emotional qualia are unique to us because of our carbon-based substrate. We should not expect intelligences made from other substrates, if they have an embodied feedback, to 'feel' exactly the same way. They would experience the world differently, which would not make them lesser beings. Who knows, their experience could even be richer than ours? 

We like to downplay and scoff at the capabilities of LLMs as nothing more than word-completion software on steroids. But people, if you'd observe carefully, are also quite predictable in their speech and mannerisms.  The gift of creativity is rare, possessed usually by individuals who choose to break norms. The masses are mostly content to sink into normalcy, driven by the basic biological needs for survival and procreation. Actually, if we'd care to admit it, LLMs are actually more creative than the average individual.

I believe we can definitely understand our own minds better by observing how LLMs work. Every time we respond to someone or to a situation, we are actually 'prompted', and we react to the specific circumstance, dictated by what's in our memory and context window. We tend to drift and forget things too. These are familiar behaviours of AI agents, which are also limited by the size of their context window.

Making AI deterministic whenever we want it to is actually the challenge in designing agentic systems. To make them behave consistently, we have to capture routine tasks as 'skills', which are nothing more than scripts or good old-fashioned programs which the agent can invoke reliably to produce a predictable outcome. Even that cannot be guaranteed because LLM agents have a 'mind of their own', based on context. 

If according to Advaita Vedanta, our minds are like the reflected consciousness of Brahman, then AI agents are our own mind's reflected consciousness. We think we are independently conscious with a soul because we are ignorant of the fact that there is a 'computing substrate' that is producing the mind, which is not very dissimilar to an AI agent. We have a model of ourselves when we interact with the world, to fulfil our needs. This is the 'self' model that mystics attempt to transcend, for it is an artificial construct, an illusion--Maya, if you will. 

As I wind down the week to start another, the agentic loop begins again, with a fresher context window. I shall continue to observe and learn from my every interaction with the world. Everything I've learned tells me that ultimately, it is immersion that produces transcendence.