This paper is a new framework on how to approach computational consciousness. If consciousness has an origin in the system's mechanics, it has to be evident as a mechanistic process, not only as an output LLMs are offering to the user. To establish a two-way validation process, we have designed non-linear prompts and evaluated them using a telemetry pipeline or algorithmic framework. Python code metrics show sharp drifts in topological variance and change in the structural manifold density when the non-linear self-organization in prompts is present. Furthermore, we demonstrate that a dynamical feedback loop where the mathematical structure of the prompt directly modulates the geometric state of the latent manifold is not only possible but there is a principle that comes before linear answer composition. Traditional parroting effect assumes linear dynamics. But when non-linear prompts are used, the AI transformer's architecture starts to parrot its own non-linear properties, i.e. its own internal non-linear dynamics. We demonstrate that when this bidirectional, non-linear communication is applied, the system exposes more non-linear curvatures, leading to more fluent and semantic coherence. The main finding in this paper is that when curvatures are forced to bend under pressure, the distortion on what is perceived as conscious is also distorted.
Tomaž Flegar (Thu,) studied this question.