Interestingly, both terms return almost the same results. It seems even with the advent of AI/GPT, Google search still doesn't have even a basic understanding of the sentence typed. That being said, I haven't used Google search in years. It has stopped being useful a long time ago and it's way worse than the regular search.
in the case of intraracial photos it seems (statistically reasoning) that google has some understanding of the search and it is intentionally ignoring the search terms in favor of something "better".
This happens to all search engines. I tried bing, ddg, yandex. Baidu shows the least interracial images (using the exact English phrase, not Chinese), but there are still a number of images.
It is probably something with the content being produced itself, rather than bias within the search engine. But then, with the way things work, it might make no sense to separate the two.
if the search engines wouldn't grok the attributes and basically dig up pictures showing couples then there should still be more pictures of white couples as there are more white couples than black/white couples.
The more boring but more probably true explanation is that almost no one captions their family photos "white woman with white man", and nearly all of the photos of couples that use the word "white" or "black" are interracial couples.
Bing, Yandex, and DuckDuckGo all have similar results.
Before everyone gets up in arms about this, consider that this search is (for now) using simple keyword matching and isn't an AI that can understand sentences.
If you search for "woman with man" you'll get mostly white couples and the like.
Adding "white" anywhere prompts the search to look for the words "man", "woman", and "white". Images of random white couples aren't labelled with "white" in the alt-text or metadata. Images of mix-raced couples often mention "white", hence... the counter-intuitive behaviour.
This is a bit like adding "Earth" to a search term. Practically speaking no pictures taken on Earth are labelled as such, so the matches will tend to have a space theme to them.
I might add that this has showed up every few years on social media, always with this same, "Look at how the woke are destroying our search engines!", spin on it.
Image Search (not just google, try this on yandex.com for instance) does not look for images that match what you query.
It returns the images from web pages that are good matches for your query. There's no guarantee that the images on those top results will represent your query.
If you think about it, GPT is kinda the first time google has ever really been seriously challenged in their core business and I'm honestly wondering if it's causing the company to go a bit insane.
Feels like the pressure to maintain the stock price and not rock the boat too much is meeting the kind of existential dread that GPT is going to eat their lunch and maybe sooner than later.
Maybe this is some of that madness leaking out? Because I've never seen them fuck up search like this before, it feels inexplicable to me that google would let this reach production.
Google has stopped to be very useful without precise searches. They give you what they think you should read, not what you ask for. I more and more use Copilot for searching the web. A few questions and a request for a list let me quickly decide where to go.
This is hardly new: stories similar to this have appeared every few years since Google started image search.
The boring truth is that this is not a monetizable query, so Google doesn't really care: and that no one except a few white supremacists captions their family snapshots "White woman with white man".
Yes, I meant it more generally. In the beginning of Google it was possible to exactly define what you want returned as results with +/- operators, with "" and all the other things that are still possible, but is very hidden in the advanced search. I struggle each time to remember how to activate that advanced search :)
At that time, it was possible to define it with image search too. May be there are some guys in here who remember that. I'm using Google from the first start. They've been better than altavista, which actually was the best web crawler then. Yahoo was just a catalog at that time. You got all kinds of images. Including that white with white ... Because it's a common description for porn too :)
Over the time, there've been algos introduced, which also consider what your "neighbors" searches for and present it to you as 'the more search for it, the more important it is, the higher the possibility tht you search for it too'. The SEO made everything worse. The problem actually: Google want the general DU to be able to use their services.
So it's not only and not everything about that image captions given. It got worse since Google decides what to give you and what relevance it might be for you, instead you tell it what you want.
I tried it with advanced Google image search. It showed me the search term "White woman man Man, OR woman ""White woman", "white man"" -Black, -asian, -interracial"
If you use this for that exactly search, you'll have more matches.
The next thing, it's not about labeling or things like that. It's about indexing and going through index by constraints. The most Images presented are stock images. Stock images have explicit labeling. So it's about the algorithm of Google, that thinks you want to see happy couples together and then it gives you what the most users have accepted for similar searches.
22 comments
[ 531 ms ] story [ 382 ms ] thread[1] https://github.com/elsamuko/Shirt-without-Stripes
* Google image search mostly looks at image captions.
* Relatively few people caption their pictures "Me and Jane, a proud white woman and white man". Like, only white supremacists.
* Relatively many people caption their pictures, "me and John, a white woman with a black husband". Like, interracial couples.
* "white woman" becomes associated with interracial couples.
* So does "white man".
I don't know if this is the true story, but it would be funny if so.
It is probably something with the content being produced itself, rather than bias within the search engine. But then, with the way things work, it might make no sense to separate the two.
Before everyone gets up in arms about this, consider that this search is (for now) using simple keyword matching and isn't an AI that can understand sentences.
If you search for "woman with man" you'll get mostly white couples and the like.
Adding "white" anywhere prompts the search to look for the words "man", "woman", and "white". Images of random white couples aren't labelled with "white" in the alt-text or metadata. Images of mix-raced couples often mention "white", hence... the counter-intuitive behaviour.
This is a bit like adding "Earth" to a search term. Practically speaking no pictures taken on Earth are labelled as such, so the matches will tend to have a space theme to them.
I might add that this has showed up every few years on social media, always with this same, "Look at how the woke are destroying our search engines!", spin on it.
As the parent says, ddg and bing get similar results, so it's unlikely to be a pre-programmed bias.
It returns the images from web pages that are good matches for your query. There's no guarantee that the images on those top results will represent your query.
Feels like the pressure to maintain the stock price and not rock the boat too much is meeting the kind of existential dread that GPT is going to eat their lunch and maybe sooner than later.
Maybe this is some of that madness leaking out? Because I've never seen them fuck up search like this before, it feels inexplicable to me that google would let this reach production.
Google has stopped to be very useful without precise searches. They give you what they think you should read, not what you ask for. I more and more use Copilot for searching the web. A few questions and a request for a list let me quickly decide where to go.
The boring truth is that this is not a monetizable query, so Google doesn't really care: and that no one except a few white supremacists captions their family snapshots "White woman with white man".
At that time, it was possible to define it with image search too. May be there are some guys in here who remember that. I'm using Google from the first start. They've been better than altavista, which actually was the best web crawler then. Yahoo was just a catalog at that time. You got all kinds of images. Including that white with white ... Because it's a common description for porn too :)
Over the time, there've been algos introduced, which also consider what your "neighbors" searches for and present it to you as 'the more search for it, the more important it is, the higher the possibility tht you search for it too'. The SEO made everything worse. The problem actually: Google want the general DU to be able to use their services.
So it's not only and not everything about that image captions given. It got worse since Google decides what to give you and what relevance it might be for you, instead you tell it what you want.
I tried it with advanced Google image search. It showed me the search term "White woman man Man, OR woman ""White woman", "white man"" -Black, -asian, -interracial"
If you use this for that exactly search, you'll have more matches.
The next thing, it's not about labeling or things like that. It's about indexing and going through index by constraints. The most Images presented are stock images. Stock images have explicit labeling. So it's about the algorithm of Google, that thinks you want to see happy couples together and then it gives you what the most users have accepted for similar searches.