[ExI] 23andSingularity

Anders Sandberg anders at aleph.se
Sun Jul 7 18:55:07 UTC 2013


The trope that "most people have not yet made a phone call yet" is alive 
and well, despite having been wrong for several years (there are 6 
billion phones by now), yet it hangs on. Same thing with a lot of other 
assumptions about just how backwards the rest of the world is - this is 
why Hans Rosling is doing important work in debunking these myths.

Clearly personal genomics is done by a super minority today - 23andme 
are hoping for one million signed up at the end of the year, which is an 
order of magnitude increase. That might well be very optimistic. But as 
the price comes down exponentially, people find uses for the data 
(whether sane or crazy uses) we should expect an exponential increase in 
people getting it - and as cellphones have demonstrated, even very high 
tech can, if packaged well, be spread in fairly poor societies. Personal 
genomics looks like it could do that (send in samples, get access to 
results - no need for local hightech infrastructure). So at some point 
we are going to see a genomics transition as a sizeable fraction gets 
sequenced.

How large does the fraction need to be to allow inferences about 
relatedness? One million SNPs is about one million bits of 
distinguishing information, one should be able to discern relatedness 
out to thousands of generations that way (caveats about non-50% 
distributions here). In practice there are plenty of statistical 
problems - see 
http://www.biostat.washington.edu/~bsweir/BIOST551/WeirAndersonHepler.pdf <http://www.biostat.washington.edu/%7Ebsweir/BIOST551/WeirAndersonHepler.pdf> 
for some examples, although a lot of this discussion is already obsolete 
since we can get so many SNPs. 
http://www.nist.gov/mml/bmd/genetics/upload/PIIS1872497311001426.pdf 
suggests that current methods work well out to second cousins, but 
becomes shaky at third cousins. So that would suggest that if X and Y 
are within 5 family links (sibling/offspring/parent) from each other 
they can be linked. Each person has about 5 people within 1 step (2 
parents, 1 sibling, 2 offspring), so there is about 4^5=1024 people 
within that radius (4 because we are tracing the tree outwards). So as 
soon as 1 on 1024 people has their genome we should expect a big 
percolation transition as the probability of getting somebody within 
this radius jumps. That will happen at about 306,000 US members, or 6.8 
million worldwide.

Now, if 23andme are successful in their ambitions, that transition will 
happen for the US within a year.

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




More information about the extropy-chat mailing list