Exciting and scary. Consider CRISPR: you need 100+ edits for a serious intelligence boost, and you'll want to do edits for other traits like schizophrenia. It's not at all obvious that it's safe to do that many edits to one embryo, or that that many edits won't have unintended consequences you can't catch as simply as doing a whole-genome sequence afterwards to look for mutations. So if you want to do a lot of edits... why not synthesize the entire genome as desired? Now the marginal cost of an edit is $0. You can make every edit you want, even the SNPs with posterior probabilities of increases barely >50%. Hsu's 'flip all the switches' scenario for IQ1000+ suddenly becomes doable, potentially within a decade or two, without depending on CRISPR scaling to hundreds of edits, multi-generational accumulating effects from embryo selection, or iterated embryo selection. Oh, and you can 'flip all the switches' on as many SNPs or rare mutations as you want, and if you don't know what you want, simply delete the entire 80k+ mutational load and produce the modal human on everything else.
"But George Church, a professor of genetics at Harvard Medical School and one of the organizers of the new project, said that if the changes desired are extensive, at some point it becomes easier to synthesize the needed DNA from scratch. “Editing doesn’t scale very well,” he said. “When you have to make changes to every gene in the genome it may be more efficient to do it in large chunks.”"
No kidding.
Yes, it'll cost $1b for the first genome, but it's easy to replicate genomes. Once you have the optimized genomes with the equivalent of thousands of edits, you can easily replicate it into eggs or sperms, and offer it for near-free to anyone who wants them for fertility procedures; half of 1000 is still an incredible game-changer. At 10k+ parents a year in the USA alone, the $1b will be amortized almost immediately ($1000 an IQ point NPV, IQ500+, 1000 * 500 * 10000 = paid off the first year based on IQ alone). Heck, there may be plenty of people willing to adopt/be surrogate parents to a pure-bred offspring. And if DNA synthesizing follows a cost curve remotely like DNA sequencing has...
Scary and fascinating. I don't know which is going to happen first: widespread embryo selection, embryo editing with CRISPR, iterated embryo selection, or whole-genome synthesis, or some combination of all of them (iterated embryo selection to pick up the tagged causal variants, creating an optimal-selected genome which is then sequenced and edited in arbitrary ways for a final genome synthesis and CRISPR spotchecks?).
Fulltext: http://science.sciencemag.org/content/early/2016/06/01/science.aaf6850.full
"But George Church, a professor of genetics at Harvard Medical School and one of the organizers of the new project, said that if the changes desired are extensive, at some point it becomes easier to synthesize the needed DNA from scratch. “Editing doesn’t scale very well,” he said. “When you have to make changes to every gene in the genome it may be more efficient to do it in large chunks.”"
No kidding.
Yes, it'll cost $1b for the first genome, but it's easy to replicate genomes. Once you have the optimized genomes with the equivalent of thousands of edits, you can easily replicate it into eggs or sperms, and offer it for near-free to anyone who wants them for fertility procedures; half of 1000 is still an incredible game-changer. At 10k+ parents a year in the USA alone, the $1b will be amortized almost immediately ($1000 an IQ point NPV, IQ500+, 1000 * 500 * 10000 = paid off the first year based on IQ alone). Heck, there may be plenty of people willing to adopt/be surrogate parents to a pure-bred offspring. And if DNA synthesizing follows a cost curve remotely like DNA sequencing has...
Scary and fascinating. I don't know which is going to happen first: widespread embryo selection, embryo editing with CRISPR, iterated embryo selection, or whole-genome synthesis, or some combination of all of them (iterated embryo selection to pick up the tagged causal variants, creating an optimal-selected genome which is then sequenced and edited in arbitrary ways for a final genome synthesis and CRISPR spotchecks?).
Fulltext: http://science.sciencemag.org/content/early/2016/06/01/science.aaf6850.full
All the folks involved need to go read some Nancy Kress novels, and maybe think twice about those 1000+ IQ kids they wanna cook up20w
SF is not a good guide to the future. For example, in Kress's novels, most of the work is being done by the intelligence rather than the titular sleeplessness - it's the intelligence which gives the Sleepless their edge over everyone else, not the extra time. We already see short-sleepers with mutations cutting their sleep needs to 4 hours or less, and while it helps a lot, it doesn't visibly lead to ruling the world.20w
When I was on John Stossel's Fox show he told me that the Clintons told him that they needed little sleep. This part of our conversation never made the show.20w- Yeah, but if it made a big difference, even 1SD on average, by the logic of thin tails, pretty much all movers and shakers should have the short-sleeper mutation, and while 'not needing sleep' does tend to come up in a decent number of bios (however inflated by boasting and attempting to look special), it doesn't come up in all of them, and plenty of people in SMPY/TIP have been very successful while almost surely only a handful of them are short-sleepers.20w
- How common is high IQ + the short-sleeper mutation? If rare then most movers and shakers wouldn't have both even if short-sleeper mutation doubled chances of being a mover and shaker.20w
- Short-sleepers are maybe 1% of the population; multiple variants have been linked to it. But I think you are going backwards here: extremes have counterintuitive behaviors, where a doubling actually implies an extremely small, almost non-existent original advantage.
Let me give a specific example from http://drjamesthompson.blogspot.com/2016/06/some-characteristics-of-eminent-persons.html - SMPY/TIP selects for kids which are 1-in-10,000 for intelligence; they are, of course, ridiculously successful. But the paper also gives the ethnic breakdown: almost all Asian and white (again, unsurprising). Specifically, the Asians are overrepresented by a factor of 31x: they are 22% of the high IQ sample, even though they were only ~0.7% of the population at that time. Using the normal distribution, one can work backwards to figure out what the mean IQ of Asians and whites must be to produce these particular ratios when you use a cutoff as stringent as 1-in-10000. What sort of IQ advantage must these Asian samples have to have a 31x rate of high intelligence? It must be very large - maybe 3 SDs? Surely at least 1 SD? Nope; it is actually more like 0.7SD - and could easily be as low as 0.2SD (111, 95%CI 103-114). What about if a group was overrepresented by 2x in TIP/SMPY? What would that imply? A 2x overrepresentation implies an advantage of 0.17SD. (At 1-in-100,000, 0.15, 1-in-1,000,000, 0.13, and so on.) Calculations in http://www.gwern.net/Statistical%20notes#inferring-mean-iqs-from-smpytip-elite-samples
This is the thin tails effect: as you go further out on the tails, the density of the distribution drops dramatically; and if you are drawing from 2 distributions with the same SD but different means, the further out you go, the more 1 distribution will dominate. SMPY/TIP isn't far out enough to make Asians the majority (perhaps because the white sample here is increasingly Jewish?) but the US black distribution is shifted downwards by ~half a standard deviation, so as far out on the tail as SMPY/TIP, they hardly exist. (You'll recognize this as a variant on the observation that male advantages on ever-more stringent testing/thresholds may stem from higher variability/SDs.)
That's just one trait, IQ. But productivity, like research productivity, is often non-normal, and modeled as a lognormal, which I believe show the same tail effects.
Hence, if short-sleeping caused a productivity increase even under 1SD, we would expect to see a large fraction or majority of world-class figures to be short-sleepers (1% * >30x ~> >30%). But I don't find that plausible since it's not mentioned enough in bios, so working backwards, the advantage must be quite small - like, 0.2SD or lower. Maybe it's not all that useful or maybe it does come with some cognitive harm which, while too subtle to be detected in available studies, still negates it enough to reduce its tail effects.20w - OK, I see why you are right.19w
