[spacehacker]

In this talk touches upon an interesting hypothesis that the purpose of junk DNA might be evolvability: https://www.youtube.com/watch?v=ZcD_L6_iBLU

The idea is basically that junk DNA contains a memory in shape of a distributed representation of the past of the organism and its environment, akin to how neural networks encode information. It basically provides a basis for fast adaptability by introducing noise into the gene expression and morphogenesis process so as to have more versatility and robustness to explore alternatives (very similar to dropout in neural networks).

[jonathansizz]

You're looking for purpose, but as the article states at the outset, don't ask 'what is this for'; ask 'how has this sequence evolved?'.

90% of our genome is unconserved, meaning that it is not under selection. Most of this consists of dead viruses and mobile elements. Such DNA was present for its own purposes while it was active but is long since dead. A tiny proportion of this junk is later co-opted by the host organism.

The null hypothesis is that junk DNA is junk. It survives in the genomes of species with small effective population size because its selection coefficient is too small for it to be purged.

The alternative hypothesis you gave would need evidence to support it, otherwise it's another 'just so' story. Ask yourself, if this junk is beneficial for adaptation as you hypothesize, why don't bacteria have any? More broadly, why is the amount of junk DNA indirectly proportional to the effective population size (as the null hypothesis predicts)?

[spacehacker]

But the null-hypothesis remains that it is junk. The hypothesis is that it is not junk, borrowing evidence from neural networks in which (1) things that look like noise are actually distributed representations and (2) in which noise improves robustness and facilitates exploration. This evidence is possibly transferable because both processes, neural network training and evolution can be formalized as a high-dimensional optimization problem (one being informed by gradient information while the other just randomly mutates and exploits ensemble effects of recombination).