It seems to me that, in the final analysis, a living thing doesn't so much as "take its instructions" from a process of selection pure and simple, either through "learning or evolution" or in any combination of the two, globally speaking; more is required to produce a valid (i.e., life promoting and life sustaining) selection than mere orientation in a "time scale," coupled to opportunistic potentialities. (Of course, this assertion depends on the validation of the idea that truth has universal application irrespective of time scale problems.) It seems notable that, among other things, the "living thing" is not a machine.
Do you suggest that a living thing can receive its instructions only from "a process of selection?" Well, fine. A question remains, begging for an answer: according to what criterion is this "natural selection" to be interpreted, or judged, as a "natural" standard by which all other existents in the universe are to be qualified and/or determined??
Which is the "bloviated version" of a simple observation that can be formulated in much simpler language: What is the standard of judgment applicable to the decision of what lives and what does not? Have you any bright, innovative ideas of your own that would clarify this ancient problem?
From there, it seems the next question we must answer is: What is truth (Pilot's question), such that it can provide a criterion or touchstone by which the truth of anything else, anytime, anywhere, must be measured?
Natural selection is simply a statement of an observed phenomenon.
Here's my point. There is simply no way to know in advance what new and untested traits will contribute to survival. It is possible to develop a list of traits that will prevent survival, but such a list has limited utility.
Survival and reproduction are moving targets. The environment is not static. Nature's answer is variation, overproduction and selection. This would be true even in a "WestWorld" universe where designers in the backroom are consciously making new designs. It would be true even if the designer is onniscient.
The "information" about what works is after the fact.