Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

AI Recognises Race in Medical Images

explainthispaper.com

Post reply on HN
Print view View original

331–340 of 357 posts

Re: AI Recognises Race in Medical Images

#331

Post by soundnote »

lostlogin wrote: post #255
soundnote wrote: post #83

This caused glorious meltdowns on Twitter. Some people just don't want to face reality.

What reality is that? I haven’t seen the thread sorry, have you a link?

That race is a physically real thing that goes beyond skin color and means pervasive differences in all parts of human biology, some meaningful, some not so much. The AI community is very invested in making a colorblind AI that can't or won't use racial characteristics in doing its job, which at least in medicine seems completely idiotic to me. We know that even at the rough proxy level of race that populations sometimes need different treatment for best results, and not building for that just leaves easy value at the table. But the ideologues building things really, really, really want it all to be totally arbitrary categories and all differences to stem from society's treatment of people. Actual biological differences are verboten.

Re: AI Recognises Race in Medical Images

#332

Post by roflc0ptic »

pessimizer wrote: post #188
roflc0ptic wrote: post #22

Earlier quoted context omitted.

As someone with a scientific bent who is of the left, I always find it incredibly frustrating when people say “x is a social construct”, because it’s technically true, but also utterly elides the dual nature of the category. Race is a social construct that can be used to infer true things (probabilities) about the real world! Other social constructs that have this property: sociology, economics, physics… This isn’t t…

You can't say anything about someone's bone density by deciding that they are black. Averaging the bone density of everyone that you've defined as black might be predictive of distributions of the bone densities of large numbers of people you've defined as black. But it's just because of the metrics you've chosen. If you started defining race by bone density, and ignored ahistorical half-biblical half-mystical 19th-c…

I happily grant much of what you’re saying; I think the difference here in my view is that you can define groups in one way, and find that _other things_ can be predicted. No one said, “we will define African Americans by whether or not they have problems with high sodium causing heart issues” and then determined that African Americans have this medical problem. Race is certainly a suboptimal way to predict this medical problem; much better to have statistically determined genetic markers, or better yet, a complete rendering of the mechanisms of action. But we don’t.

I’m all for “variation with groups is larger than variation between groups”, both as a literal reality and also as a practical answer to the question of how do I go through the world without being a jerk to people.

Racial categories are squishy and permeable. So are ideas of species, sub species. They’re useful tools with sharp limitations. Again, some people certainly oversell the utility of the tools, often for malign ends. But it doesn’t mean it’s an entirely bankrupt view

Re: AI Recognises Race in Medical Images

#333

Post by shadowgovt »

0-_-0 wrote: post #323
shadowgovt wrote: post #304

Earlier quoted context omitted.

I cannot, but I don't understand why the question is asked. I'm not a convolutional neural network. And recognition of my mother's voice isn't going to have impact on, say, medical treatment for people of a particular race, or who gets a loan granted to them, or whether an autonomous vehicle successfully recognizes a person crossing the street, or whether a drone's auto targeting system decides that this blob of sens…

We rely on human judgements that can't be described all the time in many of the above situations, so why would it be a "major concern" with machines? Machines can be much more thoroughly tested than humans, so they should be more statistically predictable

Three reasons: one practical, two psychological.

The practical one is that errors in a machine system scale, as do most things with machines. If I have a single bad X-ray tech who is applying the wrong medical process because I have a different race, for some reason, the damage that tech is doing is limited to whatever specific set of patients they are seeing. If a similar error occurs in a popular machine classification tool used widely by a hospital network, the damage is widespread. It is a plus that the machine can be corrected and the correction also scales, but with the (relatively speaking) stone tools we use to understand why a CNN makes its decisions these days, every fix risks breaking something else we're not testing for.

The first psychological reason is that machine learning systems break in "alien" ways. They don't make the kind of mistakes humans make... They make mistakes as a product of their machinery, which means it's much much harder to predict what those mistakes will look like for an average operator. As a frequent example, it's pretty rare for humans to misclassify human beings in photographs as apes, or to fail to recognize a face in an image because the skin is too dark. That's a failure mode that happens over and over again with image recognition systems.

And the second psychological reason is that humans don't trust machines to make human decisions yet. And that mistrust doesn't extend to other humans, even though we're incapable of cracking open another human's mind and understanding their thought process at the mechanistic level. It doesn't matter... we are the same organism and have a shared experience and empathy with them that we lack with machine recognition systems. It's semi-irrational, but it can't be wished away. A system for understanding why a machine makes decisions would be a step in the direction of addressing those concerns.

Re: AI Recognises Race in Medical Images

#334

Post by goto11 »

JamesBarney wrote: post #209
andreyk wrote: post #156

Earlier quoted context omitted.

From what I understand, the definition of a social construct is a bit nuanced, and it's possible to both believe race is a social construct and that it correlates with things. As a metaphor, let's say we create two categories in the world for all people - tallers (people above six feet in height) and shorters (people below six feet in height). Human height objectively exists, but these categories are social construct…

So if you were going to create a test for "Is this a social construct?" What would it be? It seems like the definition from your comment would be "Could we take this labeling system, and define different labels or the labels differently?" I watched some of the youtube, and in the thought experiment she proposes she take a continuous trait(height) and arbitrarily splits it into two buckets. And talks about how this is…

A lot of the angst around "social construct" is because people think "X is is a social construct" means "X is imaginary" or "X can be whatever you want it to be". Just as an example, legislation and the judicial system is clearly a social construct, but that does not mean it is not real!

Re: AI Recognises Race in Medical Images

#335

Post by naasking »

burkaman wrote: post #315
naasking wrote: post #183

Earlier quoted context omitted.

The imprecise definition of "species" is exactly what I was describing. See: https://en.wikipedia.org/wiki/Species_concept

I agree we don't have a perfect definition of species, but I'm wondering if we have literally any definition of race. Has anybody even tried? Is there any criteria more specific than "if two people look very different then they are different races"?

Can't say to be honest, I don't have that much interest in the topic of race beyond its informal use to define cohorts in studies. More than likely someone could start from ethnicity and expand it from there, but the whole endeavour is likely doomed from the start given the sensitivity of the subject.

Re: AI Recognises Race in Medical Images

#336

Post by vintermann »

exporectomy wrote: post #251
runnerup wrote: post #227

Earlier quoted context omitted.

"never" is too strong here. Two black people occasionally produce a white baby without albinism. There's a lot of genes that code skin color and you can have a bunch of recessive "white skin" genes and still be very black-skinned yourself. Then sometimes your child gets all the recessive white-skin genes from both parents. I had a personal friend who was white with two black parents, not albino, and genetic tests pro…

Skin color isn't race. I would expect this AI to detect your white-skinned friend as black because all his other traits would be black-like.

Well, you would be wrong. Because this AI predicts self-reported race, not ancestry.

Think about what this means People who have, say, three black great-grandparents and five white ones, will vary widely in how black they look, and this likely affects how they identify. But there are also invisible markers of ancestry - of the sort you would expect maybe were visible only on an X-ray if you looked carefully. Those could easily break the other way. You could easily get someone who looked white (and thus, probably identified as white), but were "black on the inside", in terms of their less visible characteristics.

Yet the algorithm sees through all that, and manages to see what you feel like? Correctly classifying Shaun King as black and Tom Jones as white? From a 8x8 X-ray picture?

The people who insist that "race is real" should be the most confused by these results, since we know how fuzzily identity is coupled with ancestry, especially in the main groups studied here, black and white Americans.

I'm much more prepared to believe, for instance, that there is something catastrophically bad going on in medical image dataset collection, than I am that self-ID race is nebulously predictable from almost nothing.

Re: AI Recognises Race in Medical Images

#338

Post by romwell »

oh_sigh wrote: post #317
romwell wrote: post #309

Earlier quoted context omitted.

OK, I'll speak in terms you might get. Humans have a myriad of visible characteristics: height, weight, skin color, eye color and shape, hair color and type, shapes of facial features, and so on. All these characteristics vary continuously. If you see that data as points filling a high-dimensional cube, there's not going to be an empty space there. Some areas are going to be denser than the others, but there are no g…

Do you believe that life begins at conception? Or at birth? Or is it somewhere in between? Do you feel comfortable marking the line where life possibly begins? If you mark it before birth, aren't you just giving ammunition to anti-abortion advocates to take away freedom from women? Should we just say that life begins at birth, and shut down anyone who asks otherwise, because anything else is dangerous and possibly ar…

You seem to be confused.

What's the continuous variable here that you are measuring to assign the discrete label (alive/not alive) to?

If you want to do that based solely on one variable, time from conception, then you get bad science. People can become dead at any point in time.

Yes, it's not possible to assign the "alive" label based on time from conception alone.

We do have plenty of discrete characteristics based on which to assign the alive/not alive label.

The Bible, for example, assigns the discrete label "alive" based on discrete label "breathing" in at least one case[1].

The definition I use is "does not need to be within a human body to sustain activity in brain cells".

Hope that makes things clear.

[1] https://www.biblegateway.com/passage/?search=Genesis%202%3A7...

Re: AI Recognises Race in Medical Images

#339

Post by iSnow »

soundnote wrote: post #330
iSnow wrote: post #327

Earlier quoted context omitted.

No, between the two end of the gender spectrum, there is all kinds of shades of grey, starting with persons with intersex genitals and all the way up to women with a more male personality, body hair and so on. Nature doesn't care about categories and will happily produce all kinds of distributions. It's us humans who try to bin them cleanly.

Intersex conditions are disorders of physical sex development, and the specific disorders themselves most of the time sex-specific. It is a physical medical condition, and they certainly do not create a spectrum of sex: There is not a third or an inbetween kind of breeding setup, just bugs in building the proper things. Bugs that are insanely rare. Women with more male-like personality goes into that "variation in a…

Sorry, you are falling for a category system that is entirely human-made. Nature doesn't care about "disorders" but produces the whole spectrum. Nature doesn't even care about that the evolutionary fitness of many variations is zero.

Re: AI Recognises Race in Medical Images

#340

Post by zepto »

tpm wrote: post #322
zepto wrote: post #306

Earlier quoted context omitted.

According to many proponents of CRT or wokeness, absolutely he becomes white. People from typically non-white ethnicities are routinely accused of being white if they adopt cultural traits associated with white people. And it’s entirely possible that if he was adopted at a young enough age, whatever this AI is detecting would read him as white too, assuming his habits and diet affected his development, as they might.…

My claim is that race is not a meaningful concept.

Do you think that racism is real?
Post reply on HN