There's been a lot of discussion on how machine learning neural nets are at all like our brains or not - but wouldn't DNA likely be more similar? Lots of random factors that in a roundabout way increase/decrease trait propagation (a new generation before death). And splicing genes would be a bit like assuming you could transplant the bit of a neural net that you think recognizes zero, to be used to write better letter/glyphs for "o"s and/or "ø"s?

It might work with good, continuous testing (ie, with bacteria were you can test x generations in n specimen) - but it's hard to see how it would ever work reliably in higher animals?