Constructing fictional eugenics (LW edition)

by Eliezer Yudkowsky 7y29th Oct 20124 min read175 comments

9


Yvain asked:

So if you had to design a eugenics program, how would you do it? Be creative.

I'm asking because I'm working on writing about a fictional society that practices eugenics. I want them to be interesting and sympathetic, and not to immediately pattern-match to a dystopia that kills everyone who doesn't look exactly alike.

My reply was too long for LiveJournal, so I'm posting it here:

1.  The real step 1 in any program like this would be to buy the 3 best modern textbooks on animal breeding and read them.  (My grandfather is a researcher in this field so I'm unusually aware that it exists.)

2.  If you give me genetic selection on multiple possible embryos where I can read off the genome of each one, I can do much better, much faster, than if I'm only allowed to look at the mother and father's genome and predict on that basis.  If I can only look at the mother and father's relatives and life achievements, I do worse, but modern tech is very rapidly advancing to be able to read off the parents' genome cheaply.

3.  If society's utility has a large component for genius production, then you probably want a very diverse mix of different high-IQ genes combined into different genotypes and phenotypes.  (Although some recent research suggests that the most important thing for IQ may be avoiding mutational load, i.e., the modal genome would be super-von-Neumann.  Even so, we'd want a diverse mix of everything else cognitive that wasn't about modality.)

4.  Doing a Bayesian value-of-information calculation on rare alleles and potentially interesting allele combinations will automatically include a value for diversity into your eugenic program, based on the value of promoting a gene / combo in much larger numbers if that gene or gene combo is found to be successful.  You would get much *more* interesting diversity in the next generation automatically, as many previously low-frequency alleles were combined in greater numbers and greater diversity than before.  *Not* doing a value-of-info calculation accounts for a lot of the dystopic load of alleged dystopias.

5.  The obvious basic instrument in a society depicted as well-intentioned would be an economic policy of trying to internalize the externalities of a child, just like a well-intentioned society might try to internalize the externalities of e.g. carbon dioxide emissions, instead of regulating/capping them directly, in order to maximize net social welfare.  There would be a tax or benefit based on how much your child is expected to cost society (not just governmental costs in the form of health care, schooling etc., but costs to society in general, including foregone labor of a working parent, etc.) and how much that child is expected to benefit society (not lifetime tax revenue or lifetime earnings, but lifetime value generated - most economic actors only capture a fraction of the value they create).  If it looks like you're going to have a valuable child, you get your benefit in the form of a large cash bonus up-front (love that hyperbolic discounting) and lots of free childcare so you can go on having more children.  The marketed social goal would be to avert the modern trope where parenthood is this dreadful burdensome inconvenience compared to playing video games, and this is bad for society because society runs out of valuable future workers whose benefits-to-society the parents mostly don't capture.  Probably the hard part from a marketing standpoint would be the proposal to do actual genetic calculations, even if it's to allegedly increase social benefit and prevent the system from being "exploited" (i.e. going dysgenic-Malthusian).

6.  As suggested in an earlier comment, financializing progressive shares of future income (as diverted from tax streams, maybe) is an obvious way to privatize prediction, but only of tax streams, or at best revenue earned by the prospective individual.  (I hadn't thought of this until I read that comment, so credit where it's due.)

7.  Taxes on expected-negative kids are more icky but would still have the obvious economic justification.  A nicer-sounding way of framing it would be requiring parents to post bond corresponding to the baseline government cost of each child in schooling and healthcare, with expected value potentially helping to make up the bond.  An interesting question is whether anyone would really work out to expected-net-negative under this system, which question is isomorphic to asking whether it ever makes selfish sense for a country to restrict immigration.  But adding at least some burden here makes sense from a cognitive perspective, because adding a cost is better at shaping behavior than adding a potentially foregone benefit.

8.  The incentive for e.g. taking advantage of sperm banks is automatic in this system - you can either pay a bunch of money to have a kid with your current husband, or you can be paid thousands of dollars and get free child care to be inseminated by the sperm of a Nobel winner who never had to diet.  I think that, in practice, the basic test of a system like this would be whether it could get people to go over the inconvenience threshold of actually using sperm banks and egg donors.

9.  More interestingly, there's a built-in incentive for most people to have daughters rather than sons under this system.  If we take the expected externalities of grandchildren into account in calculating the expected externalities of a child, then daughters can bear children using the best sperm via gene banks, while men have a harder time getting at the best eggs, making the grandchildren of daughters much more valuable if you'll assume they'll all be Nobel-laureate-descendants.  Daughters also add more marginal children to society than sons, since adding another son does not increase the marginal reproductive capacity of society unless single women aren't willing to reproduce using sperm banks (even taking into account subsidized childcare) and the polyamory factor has gone over what women with children are willing to tolerate.  So if grandchildren are net positive, daughters are more marginally valuable to society until the sex ratio has gone well over 1:1.  This is leaving aside generally larger criminal downsides of men, the fact that men do worse in school (which may be a mere artifact of our horror of a school system), and so on.  However, if the sex ratio becomes very extreme and the system is supposed to stick around for many generations, then most of the males generated will be by people defying system incentives; and unless very few women reproduce with those males, there will be a large selective advantage for having sons outside the system.  I.e., the system will be selecting for those who defy its incentives, which is a key design criterion for avoiding.  (Though on yet further reflection, if there are many males with suboptimal genetics being produced and then reproducing, child-value calculations would rapidly yield the social advice to start birthing more above-average males even if they won't win the sperm-bank contest; and if women have a strong preference for present fathers, you could directly calculate that as social value as well as a factor in calculating expected genetic impact of males.)

10.  In the end, all of this just adds up to, "If you can correctly internalize these externalities, the following social welfare factor will be increased..." and the key part is of course that "If".

(See Yvain's post and comments as well.)

9