Comment by ggggtez | Hacker News Reader

Comment by ggggtez | Hacker News Reader

Parent Full thread

>why life is evolvable

Sorry, could you explain why you think life is not evolvable exactly? Assuming you take the existence of a single celled organism with DNA as a given (we still don't know the origin of life), evolution gets you the rest of the way rather nicely. Notably, "life" usually contains the assumption that it is evolvable as part of the definition. If the children of the organism can't adapt to the environment, we don't consider those things to be "alive" (e.g. a 3d printer that can print a copy of itself isn't alive).

As for the origin of life, all serious scientists are onboard with abiogenesis, though we don't know the mechanism. Every year, new science comes out showing how microfluid droplets with organic compounds + the natural environment, can result in behavior that looks similar to a cell.

For example, this one shows fairly interesting "cell like" movement without any life, and there was another last year that proposed a possible abiogenesis of cell walls through evaporation and organic compounds that suck up large molecules into the interior when evaporated.

https://qz.com/487712/why-these-colored-water-droplets-seem-...

We know that life is evolvable because life exists and we know the biochemical mechanisms involved (DNA + cellular biochemistry).

Evolution implies a relatively smooth path through "DNA space" from, say for example, an early single cell eukaryote to a mushroom. However the search space is enormous. Even if we account for billions of years of evolution and a trillions of evolutionary experiments each year, a simple random walk with selection through DNA space should go nowhere because of the numbers involved. The curse of dimensionality[0] means there has to be some other principle of nature to make the search space yield a path from one viable life form to another. The search space of life would have to be 'smooth' in some sense. That 'smoothness' is something we don't understand.

If DNA space is just 256 bits (as a dramatic simplification), then 2^256 is a very very big space to search just by chance [1]. Now imagine a space orders of magnitude bigger.

[0] https://en.wikipedia.org/wiki/Curse_of_dimensionality

[1] https://youtu.be/S9JGmA5_unY?t=22 (3Blue1Browns wonderful illustration of how large 2^256 is)

This is an interesting way of framing the idea, but it's not a question of traveling in DNA space from some point (eukaryotic cell) to another specific point (mushroom): that would be very difficult in the way that you're talking about.

Imagine flipping a fair coin 256 times. The particular outcome ('HTTTTHHTTTTTTTTHTHHHTHHTHTHHHHH...') is extremely difficult to replicate, but getting any outcome is very easy: just flip the coins again. In this case we also have a lot of selection bias: all the paths through DNA space that don't result in intelligent life don't result in anyone having this conversation.

Regarding the curse of dimensionality: it's a statement about the available data rapidly becoming sparse in high dimensional spaces. It doesn't really say that high dimensional spaces are necessarily sparse, it's just hard to "fill" them in with the amount of data available.

Comparing evolution to a random walk with selection doesn’t quite sit right with me. In practice much of evolution occurs via gene duplication and recombination. At that point you can evolve complex changes very quickly. Evolving novel phenotypes is much easier if your starting material is an existing functional gene. Many motifs can be reused and reapplied.

Comparing a mule with it’s parents shows how much novelty can be produced in a single generation (in this case an evolutionary dead-end of course)

It's not really a random walk in any way. Having designed artificial evolutionary systems, even if you screw up the implementation and the search space is really bumpy, it usually still makes progress, albeit very slowly.

Reply on news.ycombinator.com