[marginalia_nu]

This isn't really a problem with computers so much as with taxonomy. You pretty much always end up with sorites-esque delineation paradoxes when imposing taxonomies onto nature. No matter how refined your system of categories is, it's never complete.

[mdeck_]

This seems to be precisely what the article says.

> The attempt to use computers to assist in racial classification tasks has helped sharpen the issues because computers can't deal with fuzzy concepts. If you try to define an ethnic code that is logically complete, consistent, and determinable for every person using current technology, you find that you can't.

> There seems to be a silent conspiracy to deny the existence of mixed racial groups in the United States. Most such people have acquiesced to this conspiracy and don't even think of themselves in those terms. Instead, they go along with the idea that they are members of one of the races recognized by the government. In fact, they often identify with a traditional race that represents only a small fraction of their genetic heritage!

[wongarsu]

Computers can deal with fuzzy concepts just fine, the problem is that we don't want fuzzy answers.

Any reasonable way to get a computer to determine race will yield a fuzzy answer: based on your family tree you are 36% European, based on facial recognition the algorithm is 60% certain you are white, based on genetic sequencing you are 4% of African descent. All those are perfectly valid fuzzy answers.

If you then try to match those to ridiculous categories like "Black, not of Hispanic origin" it obviously goes wrong quickly. What exactly is "Black"? Does one Spanish great-grandfather disqualify you from that category? Or only if that great-grandfather was from Brazil instead of Spain?

[User23]

> What exactly is "Black"? Does one Spanish great-grandfather disqualify you from that category?

Interestingly you might get a very different answer to that question in Brazil than you would in the USA. I saw an interview with a Brazilian footballer that looked black to Americans sounding surprised when asked by an interviewer asked how racism affected him. His answer was, paraphrasing from memory, “what are you talking about I’m not black.”

Meanwhile race in the USA is of major political importance. It matters for college admissions, government loans and contracts, and so on. By and large it’s on the honor system. So far at least there haven’t been many white presenting Americans identifying as black or mixed race when their DNA test shows 4% African ancestry.

[ZeroGravitas]

> So far at least there haven’t been many white presenting Americans identifying as black or mixed race when their DNA test shows 4% African ancestry.

This used to be more common when it was legally enforced by the state.

See the fascinating history of "Walter White":

https://en.wikipedia.org/wiki/Walter_Francis_White

[marginalia_nu]

There is a tension between the utility of categories, which is very much real; and their limitations, which are also real and inevitable.

We absolutely need to categorize the world to make sense of it, even if the categories themselves are never going to be 100% complete or accurate. I think the takeaway should be awareness of the inherent limitations of categorization, not its attempted abolition.

There may be a point in revisiting categories though, as they tend to ossify and become less useful with time; you can't step in the same river twice, and that map your grandfather drew may not quite be accurate anymore.

[darkerside]

Why doesn't k means clustering help? Maybe we'd even invent new racial groups for everyone to get mad about!

[snovv_crash]

The problem isn't the clustering. The problem is choosing which individual points belong to which clusters, because every time people have mixed-race (or mixed-mixed-race) or whatever kids, which has been happening for millennia, a new 'cluster' gets invented.

Basically, there are no nice separations between the clusters, there are just denser and sparser regions in the feature space, so assignment of a point to a single cluster doesn't make sense.

[dekhn]

it's almost a textbook case for unsupervised learning. Like, almost as if you could train a manifold that separated out people by genetic history (much of which was super-regionalized, and did lead to large phenotypic differences that clustered spatially)... and in that manifold "mixed-race" people would probably fall on points between two well-separated clusters.

In fact there are entire scientific papers (mainstream) doing exactly this. We've learned all sorts of interesting things, like the total genetic diversity in africa is larger than the african/eurasian diversity split.