[ExI] A Gedanken Rational Eugenics Experiment (AGREE)

Anders Sandberg anders at aleph.se
Thu Oct 24 22:50:03 UTC 2013

On 2013-10-24 17:33, rex wrote:
> Anders Sandberg <anders at aleph.se> [2013-10-24 07:29]:
>> Fertility is surprisingly changeable through culture and 
>> institutions. The points made about higher education being 
>> parent-unfriendly are just the tip of the iceberg; the big birth 
>> differential between northern and southern Europe is largely due to 
>> institutional affordances... and *soap operas* have been a major 
>> vector in changing fertility patterns in Brazil and India.
> Soap operas changed fertility patterns? How would a causal relationship
> be proven?

Ah, it is a classic set of studies. Television arrived in some villages 
but not other comparable villages, fertility went down in the television 
villages. A lot is ascribed to viewing family patterns among the soap 
opera people (anthropologists found that soap operas were the major 

> The payments would make some fraction think it's a good idea.

Yes, but if that is 5% (still a lot of people) the selection pressure 
will be very low.

> What
> fraction that would be depends upon the amount of payment and the IQ
> required to qualify (the sim used -1 SD, but -1.5 SDs or -2 SDs)
> would also work.

If you select against the extreme tail you will have a far weaker 
selection pressure. The "wall" the distribution moves away from will be 
further down.

>> This is why I think liberal eugenics, which works by parents 
>> selecting themselves what traits to go for, might be more robust. And 
>> cheaper...
> It's merely a Gedanken experiment to illustrate the ideas, and not
> intended to be a policy to be compared with others which may well be
> more socially viable. And, it's not an exclusive policy.

I get it. But I still think there is a problem with a policy that 
requires a sustained centralized effort. Getting people to pay for their 
own selection is distributed, cheaper and unlikely to stop just because 
of a change in government.

>> Note that you would need to pay *a lot*. The below 1 SD population is 
>> around 15%, so in the US that would be about 47 million people you 
>> need to pay off. I am not sure what the going price for getting 
>> sterilized is, but at least I expect it to be on the order of a few 
>> thousand dollars - we are easily talking hundreds of billions here. 
>> Not quite as much as is already spent on elementary schools, but 
>> still a lot.
> But only half that 47 million are females

Your model only uses females, but half of the genome is from the males.

> In politics 15 generations is an unthinkably long period. In
> evolutionary biology it's an eyeblink.
In technology one human generation is more than a millennium of Internet 

I am often dissing genetic enhancement for this reason: many 
technologies for direct manipulation have a good chance of being 
developed and refined much faster than designer babies can grow up. 
Genetics has the advantage of actually working to some extent, and we 
might be wrong about the other technologies. But generically it will be 
slower than a lot of other factors.

> La Griffe du Lion's smart fraction theory is not about genius; it's
> about a critical fraction with the modest IQ required to be useful to
> society. (I don't know who he really is, but he's clearly an incisive
> thinker who uses statistical inference fluently.)
> http://www.lagriffedulion.f2s.com/sft.htm
>  "Thus, for a technologically sophisticated society, SFT asserts that a
>  nation's per capita GDP is determined by the population fraction with
>  IQ greater than or equal to some threshold IQ. Consistent with the
>  data of Lynn and Vanhanen, that threshold IQ is 108, a bit less than
>  the minimum required for what used to be a bachelor's degree. Figure
>  3 illustrates the fit of (3) to the data of Lynn and Vanhanen."

108+ is about 30% of the population. If we assume actual useful 
performance in a knowledge heavy world is some kind of concave function 
of IQ (at least close to the middle of the bump) one should not be too 
surprised that this subset is responsible for the majority of the useful 

Still, if the usefulness grows as a polynomial (or even exponential!) 
function of intelligence the main effect when multiplied with the 
available population is merely to shift the Gaussian towards the right: 
there are so few of the real extreme performers that the bulk of useful 
work is done by the mediocre high performers.

> I'm more interested in varying parameters in simulations to see how
> they behave than in comparing potential policies. For example,
> assortative mating and the high heritability of IQ cause speciation
> pressure. As these values large enough to eventually cause speciation
> in humans? If not, how large must the correlation between the IQs of
> mates be to cause speciation?

That would be fun to run. I think the correlations in assortiveness we 
see enable mixing of genes over rather few generations, precluding 

I suspect that a population with some sort of assortive mating but a 
single fitness function will have a distribution mostly determined by 
the fitness, not the mating. But it deserves simulating.

Dr Anders Sandberg
Future of Humanity Institute
Oxford Martin School
Oxford University

More information about the extropy-chat mailing list