Counter^3 point: a system that was thoroughly engineered and tested a long time ago, and that still fulfills all the technical requirements that the system must meet will simply not spontaneously break.

Analogously: this would be like using an ML + LLM to rewrite the entire Linux kernel in Rust. While an (arguably) admirable goal, doing that in one fell swoop would be categorically rejected by the Linux community, to the extent that if some group of people somehow unilaterally just merged that work, the rest of the Linux kernel dev community would almost certainly trigger a fork of the entire kernel, with the vast majority of the community using the forked version as the new source of truth.

This is not preventative maintenance. This is fixing something that’s not broken, that has moreover worked reliably, performantly (enough), and correctly for literal decades. You do not let a black box rewrite your whole codebase in another language and then expect everything to magically work.