I've worked with enough legacy code to know that playing around with seemingly unused areas can have unexpected consequences.

The trouble here is that we can't see those consequences until they end up hurting someone.

The genome is the ultimate legacy code:

- it's been written for billions of years (https://en.wikipedia.org/wiki/Abiogenesis#Earliest_biologica...)

- without any sort of upfront architecture or planning

- the development methodology can be summarized as "change characters randomly and see if it compiles"

- incredibly bloated (https://en.wikipedia.org/wiki/Non-coding_DNA)

- contains random snippets copy/pasted from other codebases (https://en.wikipedia.org/wiki/Endogenous_viral_element)

- the only success criteria is that the "program" shouldn't "crash" before it can make a copy of itself

> incredibly bloated

This would actually be preferable to how DNA seems to evolve. In most systems, there's actually quite an acute constraint to genome length due to the Error Threshold; when a genome gets too large, it has too many mutations and therefore the 'fittest' genome, often called the 'master' genome, cannot be sustained in the population. Therefore, there's quite a lot of evolution where the genome becomes essentially minified. Most often, this is done by overlapping genes on the DNA, (mostly by shifting the reading frame by one or two nucleotides for one of the two genes).

As someone who works on software for DNA analysis, this is spot on

I wonder if one were to clean up the code, removing all bloat, keeping only the relevant pieces, how would that human turn out?

Fixing our inverted retina would be one of the low hanging fruits:

>The vertebrate retina is inverted in the sense that the light sensing cells are in back of the retina, so that light has to pass through layers of neurons and capillaries before it reaches the rods and cones. In contrast, in the cephalopod retina the photoreceptors are in front, with processing neurons and capillaries behind them. Because of this, cephalopods do not have a blind spot.

https://en.m.wikipedia.org/wiki/Retina

I want this comment as a poster.

Genetic code in large, multicellular organisms is quite different; in mammals, for certain functions you have multiple pathways to achieve similar things through different vehicles (proteins). In this sense, it is more akin to building or blowing up a bridge in relation to a transportation network; usually, other paths still exist, but sometimes and in certain critical locations it really fucks up the infrastructure.

>> I've worked with enough legacy code to know that playing around with seemingly unused areas can have unexpected consequences.

Imagine trying with million or tens of million year code.

There are many people that are CCR5 null (largely in Africa due to positive selection for HIV resistance), so it's unlikely to have much effect, but still...

edit: I'll also amend the statement, as the gene is implicated in responses to other viruses: https://en.wikipedia.org/wiki/CCR5#Potential_costs

I think the question is not so much the effect of a lack of CCR5 but rather the accuracy of the genetic procedure and whether or not anything else was unintentionally affected. If it was accurate and only desired change was implemented, we have opened pandora's box.

In the short term I think the risks are overstated. Vastly increased health and intelligence looks like it will be possible to select and/or edit for, as well as the elimination of Mendelian disease. The scale of this good is so large it overwhelms most scruples.

"Vastly increased health and intelligence"... again, you keep making assertions without any reference.

The source is idealism. I've noticed a lot of tech people tend to hand wave away implementation details and talk about things as a theoretical exercise on a very wide time frame. So eventually, at some point, this technology will be able to do all the sci-fi stuff of dreams. Probably right around when we achieve singularity.

What kind of references do you want for a procedure that was just done for the first time? We already know that some diseases are affected by genes. It is also clear that intelligence is at least in part inheritable. It's a reasonable assumption that we can improve both if we get good at editing genes.

No, it is not. Heritability does not mean that there is a simple correlation between a trait and a small number of genes that could be exploited for targeted manipulation.

Take height: Height is extremely heritable. If your parents are both tall, you are very likely to be tall as well. But height is influenced by thousands of genes, and each individual gene has only a minuscule influence. And each of those genes is also responsible for many other unknown things.

It's not at all a reasonable assumption that genetic modification will allow us to influence a complex trait like intelligence. That's just wishful thinking.

On the other hand, we can (today!) predict height from raw sequence with correlation 0.65: http://www.genetics.org/content/210/2/477. It does not mean we can do such prediction for designed sequence (predictor likely will fail to generalize), but it can be used to select from natural variation. I don't see any wishful thinking here.

Yes, selecting the "tallest" embryo for implantation sounds somewhat feasible. It's similar to selective breeding with a shorter feedback loop. The only downside is that you are only selecting for genotype, instead of phenotype, so your model might accidentally select for some other things that you didn't expect.

But that's very different from designing an embryo by direct genetic modification.

Agreed, there's a big difference between what's possible with selection and what's possible with editing. Selection is already available (as increasingly sophisticated PGD). It's a lot trickier to do anything useful with editing.