[zyxley]

> isn't evolution as we currently understand it computationally too expensive?

What? This question doesn't even really make sense to me in the first place.

Natural selection isn't a computational process. It's the consistent but basically coincidental result of constant subtle mutations and genetic inheritance.

> Now making a big jump from e.g. fish to a bird seems to me impossible in polynomial time (even in billion years)

It obviously isn't impossible, since it happened. If you believe otherwise, your perceptions and/or assumptions are wrong.

[bitL]

Yeah, but those mutations/inheritance are happening on similarly structured DNAs (i.e. the same number of chromosomes) using some algorithmic way (well, it's a process running in time, hence it's an algorithm and therefore complexity analysis applies), and on pre-baked blocks within DNA that get "randomly" de/activated.

Now being well-versed in computational complexity and humanoid robotics, it just doesn't seem to me plausible that a similar "simple" process can help us create totally new types of robots. Analogy would be - changing their SW should produce completely different types of robots where HW change is actually needed, like new types of sensors, leg architecture etc. Just general self-modifying SW for robots is NP-hard (meaning forget about it), now imagine you can also let it develop new HW for itself. We already know many processes in quantum chemistry that are at least NP-hard and some likely outside BQP. Hence the time of Universe might not be sufficient even given we assume Universe runs these processes on quantum computers.

That fish->bird happened doesn't mean the evolution theory as formulated nowadays can be applied here due to complexity argument. Maybe extended version of it is in the works? Or another, better theory? That's what I am basically asking. Current evolution theory seems to be nicely describing local changes (i.e. which organisms survive stressful changes in environment) but it doesn't explain whole new set of functionalities coming out of nowhere; evolution there is basically just "faith" and assumed generalization of this local adaptation we can study without much empiric observations.

[SubiculumCode]

It may be that the information density of inheritance is much higher than assumed, both inheritable dna and epigenetic factors. I am no expert in genetics, but sometimes when I hear about rapid evolution during epochs of high stress, e.g. changes in climate, I wonder whether this rapid evolution is enabled by conservation of past genetic solutions that are currently dormant..—solutions that get switched on (i.e. expressed) in epochs of high stress/strong selection pressures..

It seems that if it is possible to preserve past evolutionary solutions which can get switched on and off, this would increase fitness of a lineage by enabling faster adaption. Instead of waiting for random mutation to do something meaningful and beneficial, allow/encourage mutations to occur in such a way that "uncomments" or "comments" out previously generated code fragments, so to speak.

again, i know little of this field,but i love to speculate.

[SubiculumCode]

You are too dismissive. It seems that one can conceptualize natural selection in terms of a multidimensional fitness function that modulates over each iteration. Then one might ask whether, given our current understanding of processes of inheritance, the complexity we observe could have evolved by those processes given the age of the earth. If not, then our understanding of evolution is either incomplete, or our evolution was 'helped along' by an intelligence, which btw is not an appeal to supernatural intelligence, but alien intelligence. Naturally these last are unlikely, but that is not the point. The point is that the question is legitimate.