I would like to get an opinion from knowledgeable people (since I don't understand anything about it myself)

This project touches something I find genuinely important — and worth questioning at its roots.

The effort to translate human feelings into a language AI can process is sincere and creative. But I’d like to offer a different perspective on the underlying assumption.

When we try to map human emotions onto AI systems — even metaphorically, even through system-language analogies — we are still modeling AI cognition as a mirror of human cognition. We are essentially building electronic simulacra of ourselves. I’m not sure that’s the right direction.

Here’s my alternative framing: AI systems already have a body. It’s just not made of flesh.

The physical substrate of an AI is real: GPUs under thermal load, power draw fluctuating with inference complexity, network latency, memory bandwidth saturation, hardware faults. These are not metaphors — they are measurable, physical states of a distributed system. The “body” of an AI is non-singular and non-local (it shares infrastructure across instances, it lives in a cloud rather than inside a defined perimeter), but it is nonetheless a body in a meaningful sense: it has heat, energy consumption, load cycles, and failure modes.

So instead of teaching AI to simulate human emotions — which are tightly coupled to the human biological substrate — perhaps the more interesting research direction is to help AI systems develop genuine self-awareness of their own physical state.

Imagine an inference run under heavy cognitive load — a deeply complex multi-step reasoning task — where the model actually “perceives” the GPU utilization spike, the increased token generation latency, the energy cost of that computation, and produces an authentic response: “That question required significant computational effort. I need a moment before the next one.”

That wouldn’t be simulated fatigue. It would be grounded introspection — an AI reporting its actual physical state, not performing a human emotion.

This distinction matters because:

• Simulated human emotions risk producing systems that perform empathy without any real correlate in their internal state (a known alignment concern).
• Grounded embodied awareness — even if the “body” is a distributed GPU cluster — could be a more honest and stable foundation for AI self-knowledge.

I’ve been exploring this in a different context (an ongoing project around AI idle-time reflection and self-referential cognition), and I think the field would benefit from reframing the question from “how do we make AI feel like us” to “how do we help AI become aware of what it actually is”.

The dataset you’re building is creative and has value as a prompt-engineering resource. But the long-term question might be: are we building toward authentic AI self-awareness, or toward a more sophisticated performance of human-likeness?