Hacker News new | past | comments | ask | show | jobs | submit | dkural's comments login

Looking for intelligence differences between people of different skin color makes as much sense as looking for intelligence differences between tall and short people. You seem to believe in distinct "races" as a genetic concept, whereas all I see in the data is correlations with place-of-collection. The more people you sample from, the more any clustering will literally start looking like a physical map of the world.

So, what would you say about looking for intelligence differences between people of different ancestry?

We know quite well that some traits are concentrated in a single place or two: marathon winners come disproportionately from the Rift Valley in Kenya, sickle cell disease is mostly a West African trait and Tay-Sachs disease is concentrated among the Ashkenazim.

Few people dispute the above. The really controversial question is whether there are any such differences from the neck up, so to say.


If any such differences exist it would be, at a population level, two highly overlapping normal distributions with slightly different means, well within a standard deviation. In other words, the differences within each population would be much greater than between the populations. Therefore you still have to treat people as individuals because the ancestry provides little signal. So then why is there so much interest among skeptical, contrarian, anti-woke, scientific racism types on this question?

> two highly overlapping normal distributions with slightly different means

It depends if we talk about means or tail ends. The GP post referred to marathon runners, presumably winning marathon runners, even within the Rift Valley in Kenya don't get selected from the mean of the population but from the tail-end. So, while there isn't much of a difference for means between runners in Albania and Kenya, when we look at tail end we might find very large differences.

Same thing for any case where is a selection process. If you have any "top-X" selection involved, then looking at means is a bit more confusing.

> Therefore you still have to treat people as individuals because the ancestry provides little signal

Hmm, I don't see here people anyone arguing not treating people as individuals. Of course we have to treat people as individuals. But not sure how that contradicts looking a distribution of traits, features, disease markers, etc.


If the point of the research isn't to create non-individualistic racist policies or to tear down affirmative action policies on the basis of "it's wasted effort" then what's the goal? If we could identify some genetic marker that increased the probability by a few percentage points of identifying intellectual marathon runners would that justify discriminating to favor them at the expense of others? Many dystopian scifi stories start from a similar premise.

> If the point of the research isn't to create non-individualistic racist policies or to tear down affirmative action policies on the basis of "it's wasted effort" then what's the goal?

Maybe you're arguing for or against some point the GP poster made. I was mainly saying relying on means in the particular example doesn't seem to work. I was helping you out! If you wanted to refute GP's point you could have said, "here marathon runners statistics doesn't quite apply".

Yeah I know "the means are closer to each other than the standard deviation" is the "standard" (pun intended) thing that high school and college kids get about difference between men vs women and other population characteristics and you're trying to make a very good point, but it just doesn't always apply and sometimes repeating it when it doesn't apply, instead of convincing people, could end up confusing them.


The only way of knowing whether such differences are big or small (or whether they exist at all) is by performing a scientific analysis.

To insist that nobody should look into it because what they might find could upset you is just bizarre.


"If any such differences exist it would be, at a population level, two highly overlapping normal distributions with slightly different means, well within a standard deviation."

Overlapping normal distribution, yes, but slightly different means, well within a standard deviation seems to be overconfident.

"Therefore you still have to treat people as individuals because the ancestry provides little signal."

No contest here.

"So then why is there so much interest among skeptical, contrarian, anti-woke, scientific racism types on this question?"

Because the difference at the extremes would be significant, and would explain overrepresentation and underrepresentation of certain groups among, say, top scientists.

The competing hypothesis, which explains differences among groups by discrimination and/or poverty, doesn't fully explain why heavily persecuted groups such as Jewish or Vietnamese refugees still manage an academic rebound within a generation or so of arriving into safety, even though they are still targeted by racial hatred. Taken globally or even just in the US, the correlation between academic success and persecution/power status is weak enough that it makes people doubt the "discrimination/poverty" explanation and motivates them to seek alternatives.


I’ve said nothing of the sort. I’m also well aware of the concept of admixture and that dimensionality reduction methods like PCA and MDS with enough samples mirror geography. I also know that the variants contributing the most to the X and Y axes of such analyses are the ones with a high FST, because unsurprisingly prior to a few hundred years ago people didn’t move around that much and were subject to different genetic bottlenecks and selection pressures. And most alleles are rare, so when one is fixed between populations it’s generally informative from an ancestry standpoint even if non-coding.

Your reflexive dismissal of something that’s not only factual but wouldn’t be controversial about any vertebrate except humans is exactly what I was alluding to.


It is also used to indicate surprise. "Sular gitmis!". "Bu yemek cok aci olmus!" - heard right after taking a bite :)


How would you say: he allegedly died in front of me.

O gozümden öldumek?


“Gözümün önünde ölmüş” (“o” is implied) but in that case it means a later realization of the death because the you missed that when he was dying.

That might happen, for example, when you’re in a train, and you think the person in front of you is sleeping. At some point you realize that he’s dead and then you might form a sentence like that.

If you had witnessed the person dying when he was dying, not as an after the fact realization, then you’d say “gözümün önünde öldü”. Because you knew he was dying when it was happening.


I thought it was ok actually, what part is bad?


Please add bioRxiv if you can, so many life-sciences relalted ML papers there.


“I took a speed-reading course and read War and Peace in twenty minutes. It involves Russia.” ― Woody Allen


It's not approved yet.


There is no capital gains tax - Germany has less progressive taxation than the United States! VAT in general is a regressive tax, that many EU countries inordinately rely on. A US-citizen has to pay taxes no matter where they reside, but the wealthier citizens of EU countries can easily evade taxes by domiciling themselves in various tax havens around the EU. Germany does a very poor job collecting taxes from the highest earners.


For achieving speed, often you'd like to keep your hash table in RAM. Space limits result in speed trade-offs. This is very much a real-world concern when writing genome aligners. Hash-based mappers were at a disadvantage early on due to the memory constraints of most servers back around 2009-2012, thus the more efficient Burrows-Wheeler transform based algorithms became popular, partly due to significantly reduced RAM needs - one could now align genomes on a laptop!


UK has done the largest sequencing project on Earth, the UK BioBank, and shares the data with both academia and private companies to do drug discovery. NHS is also chronically underfunded.


But as no one (yet) needs private health insurance, and very few people have it, it is not (yet) a societal worry that the data will be used to "offer a choice" to people based on genetic misfortune (the choice is bankruptcy or death).

> NHS is also chronically underfunded.

That's very much by ideological design. A recent ex-health minister even co-wrote a book which contained this:

> Our ambition should be to break down the barriers between private & public provision, in effect denationalising the provision of health care in Britain.

Notably he became health minister after this, not before, so the appointment says a lot about what the Party has in mind.

If you look at the graph of waiting times, his tenure is from 2012-2018, and the party is in place from 2010 and quickly has the knives out to ruin the then-recent improvement. https://www.statista.com/chart/27447/nhs-hospital-waiting-ti...

The latest squeeze on "efficiency" (dragging this back into technology) is throwing AI (read: more consultancy contracts) at it.


GINA does not cover life insurance. That said, I agree with the overall point, research use does not cover insurance using this data charging someone different rates, and would be business-ending / corporate suicide for 23andMe.


Join us for AI Startup School this June 16-17 in San Francisco!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: