7 ms·

>Tools like Dall-E-2 and Stable Diffusion have been criticized for reproducing racial stereotypes even when a text prompt doesn’t imply the subject’s race. Such

by GaggiX 4y ago

>Tools like Dall-E-2 and Stable Diffusion have been criticized for reproducing racial stereotypes even when a text prompt doesn’t imply the subject’s race. Such biases were why Google decided not to release Imagen, which had been trained on LAION. This is so ingenuous and naive to think that the reason Google didn't release Imagen is because of racial stereotypes.

GuB-42 4y ago
Google absolutely don't want to be accused of racial stereotyping. Remember, Google is in the business of advertising, and the last thing most brands want is to be labeled as racist. Anything Google does that can be interpreted as racist is going to cost them, as advertisers pull away to avoid the association.
atleastoptimal 4y ago
It's what Google claimed and thus the eternally true correct reason and its very problematic if you think otherwise
- astrange 4y ago
  If there's a reason Google doesn't want to talk about, it's that the biggest legal risk of AI image generation is that it can be used for revenge porn and CSAM. I don't think it's problematic to be worried about that!
  - ImprobableTruth 4y ago
    They could just filter it like Midjourney or DALL-E.
    - astrange 4y ago
      How do they know they can "just" do something that came out a year after they invented Imagen? Especially when getting it wrong once might mean Europe makes AI illegal. Making things illegal is their favorite hobby.
      - ImprobableTruth 4y ago
        Midjourney and DALL-E were both announced and in closed beta a month before Imagen was announced, with filters already in place.
  - anonylizard 4y ago
    The biggest legal risk, empirically, is massive copyright lawsuits. The AI art models are rapidly eliminating most artist/photography jobs. Revenge porn and CSAM are trivially solved by community moderation methods like Midjourney, simply force every generation to be tied to a paid account, and made in the public eye (posted on discord channels). So in reality, its not Google's risk conservatism, its Google's utter lack of imagination and creativity, and the lack of organisational incentive for anything other than ad money.
    - dontupvoteme 4y ago
      You realize stable diffusion has been out for 6 months and is more powerful than midjourney?
      - yellow_postit 4y ago
        Maybe in some technical sense but in practice the majority of quality (subjective) work I see shared on Twitter is Midjourney backed. Which is crazy to me because I don’t think I could design a more frustrating interface than Discord.
        nickthegreek 4y ago
        Midjourney gives you good images with near zero work. SD makes you work for the output but gives you FAR greater control but also needing to be technically skilled, and have pricey hardware. Given this, it is easy to understand why the majority of what you see is done in midjourney.
      - janekm 4y ago
        Was, when it came out, since then MJ have released 3 new models (plus variants) and most users assess the MJv5 model as more powerful. Of course new SD models are also on the horizon…
        astrange 4y ago
        > Of course new SD models are also on the horizon… SDXL is available at https://beta.dreamstudio.ai/ https://beta.dreamstudio.ai/ though they say they're going to release more variants. I think ControlNet is a lot more interesting than just "better tuned models"; it means there's no line between creating something yourself and asking an AI to do it anymore.
klodolph 4y ago
Google previously had problems with Google Photos categorizing dark-skinned people as “gorillas” or something like that. This would show up when you were searching for photos in your personal Google Photos account. Everyone talked about it. Nobody wants to release a product like that. Removing bias is hard; you can’t just decide to get an unbiased training set or something magical like that. It is NOT surprising that larger companies would behave with an abundance of caution. Leave it to the startups to rush in head-first, cause a bunch of problems, make a cool product, get a consent decree, and capture the market. Large companies that act that way get PR problems and lawsuits.
- anonylizard 4y ago
  In retrospect, those concerns were pretty much drummed up by the "AI ethics" people, who needed an easy topic to demonstrate their importance. When actually powerful AIs got released, the 'bias' angle got overwhelmed 100x by the "trained on copyright data" and "Are we going to lose our jobs" angle. Your average black artist doesn't care if SD defaults to drawing white people, because A: he cares about not being able to draw for a living, permanently, 100 times as much. B: He can always add 'dark skinned' to his prompt The bias problem is also relatively easy to solve (Midjourney has already made massive improvements), while the copyright/job loss problem is extremely hard. The AI ethics people have had shockingly little to say on copyright/job loss issues. Which is why they got fired en masse. Large organisations systemically overestimated the risks from bias, and underestimated the actual AI risks that society actually cares about. I think the answer is also simple, because accused of racism, will cost any executive in a large company their job. Being accused of automating millions of jobs, will earn them prestige and a promotion. Smaller companies can ignore those accusations because they aren't vulnerable to activist pressure, they answer only to their customers.
  - klodolph 4y ago
    > The bias problem is also relatively easy to solve (Midjourney has already made massive improvements), while the copyright/job loss problem is extremely hard. It only seems easy to solve on the surface, it’s a deep problem. It’s also not just the bias thing, Bing and ChatGPT have been saying some truly unhinged things. > I think the answer is also simple, because accused of racism, will cost any executive in a large company their job. It takes more than an accusation, otherwise you could go around accusing executives you don’t like of racism and getting them fired. Ethics is a tenuous job position at best, even in a large company. It’s seen as a cost center. I don’t think there’s much to read into why AI ethicists would get laid off.
andresgottlieb 4y ago
"such biases"

by GaggiX 4y ago

GuB-42 4y ago

Google absolutely don't want to be accused of racial stereotyping. Remember, Google is in the business of advertising, and the last thing most brands want is to be labeled as racist. Anything Google does that can be interpreted as racist is going to cost them, as advertisers pull away to avoid the association.

atleastoptimal 4y ago

It's what Google claimed and thus the eternally true correct reason and its very problematic if you think otherwise

astrange 4y ago
If there's a reason Google doesn't want to talk about, it's that the biggest legal risk of AI image generation is that it can be used for revenge porn and CSAM. I don't think it's problematic to be worried about that!
- ImprobableTruth 4y ago
  They could just filter it like Midjourney or DALL-E.
  - astrange 4y ago
    How do they know they can "just" do something that came out a year after they invented Imagen? Especially when getting it wrong once might mean Europe makes AI illegal. Making things illegal is their favorite hobby.
    - ImprobableTruth 4y ago
      Midjourney and DALL-E were both announced and in closed beta a month before Imagen was announced, with filters already in place.
- anonylizard 4y ago
  The biggest legal risk, empirically, is massive copyright lawsuits. The AI art models are rapidly eliminating most artist/photography jobs. Revenge porn and CSAM are trivially solved by community moderation methods like Midjourney, simply force every generation to be tied to a paid account, and made in the public eye (posted on discord channels). So in reality, its not Google's risk conservatism, its Google's utter lack of imagination and creativity, and the lack of organisational incentive for anything other than ad money.
  - dontupvoteme 4y ago
    You realize stable diffusion has been out for 6 months and is more powerful than midjourney?
    - yellow_postit 4y ago
      Maybe in some technical sense but in practice the majority of quality (subjective) work I see shared on Twitter is Midjourney backed. Which is crazy to me because I don’t think I could design a more frustrating interface than Discord.
      - nickthegreek 4y ago
        Midjourney gives you good images with near zero work. SD makes you work for the output but gives you FAR greater control but also needing to be technically skilled, and have pricey hardware. Given this, it is easy to understand why the majority of what you see is done in midjourney.
    - janekm 4y ago
      Was, when it came out, since then MJ have released 3 new models (plus variants) and most users assess the MJv5 model as more powerful. Of course new SD models are also on the horizon…
      - astrange 4y ago
        > Of course new SD models are also on the horizon… SDXL is available at https://beta.dreamstudio.ai/ https://beta.dreamstudio.ai/ though they say they're going to release more variants. I think ControlNet is a lot more interesting than just "better tuned models"; it means there's no line between creating something yourself and asking an AI to do it anymore.

klodolph 4y ago

Google previously had problems with Google Photos categorizing dark-skinned people as “gorillas” or something like that. This would show up when you were searching for photos in your personal Google Photos account. Everyone talked about it. Nobody wants to release a product like that. Removing bias is hard; you can’t just decide to get an unbiased training set or something magical like that. It is NOT surprising that larger companies would behave with an abundance of caution. Leave it to the startups to rush in head-first, cause a bunch of problems, make a cool product, get a consent decree, and capture the market. Large companies that act that way get PR problems and lawsuits.

anonylizard 4y ago
In retrospect, those concerns were pretty much drummed up by the "AI ethics" people, who needed an easy topic to demonstrate their importance. When actually powerful AIs got released, the 'bias' angle got overwhelmed 100x by the "trained on copyright data" and "Are we going to lose our jobs" angle. Your average black artist doesn't care if SD defaults to drawing white people, because A: he cares about not being able to draw for a living, permanently, 100 times as much. B: He can always add 'dark skinned' to his prompt The bias problem is also relatively easy to solve (Midjourney has already made massive improvements), while the copyright/job loss problem is extremely hard. The AI ethics people have had shockingly little to say on copyright/job loss issues. Which is why they got fired en masse. Large organisations systemically overestimated the risks from bias, and underestimated the actual AI risks that society actually cares about. I think the answer is also simple, because accused of racism, will cost any executive in a large company their job. Being accused of automating millions of jobs, will earn them prestige and a promotion. Smaller companies can ignore those accusations because they aren't vulnerable to activist pressure, they answer only to their customers.
- klodolph 4y ago
  > The bias problem is also relatively easy to solve (Midjourney has already made massive improvements), while the copyright/job loss problem is extremely hard. It only seems easy to solve on the surface, it’s a deep problem. It’s also not just the bias thing, Bing and ChatGPT have been saying some truly unhinged things. > I think the answer is also simple, because accused of racism, will cost any executive in a large company their job. It takes more than an accusation, otherwise you could go around accusing executives you don’t like of racism and getting them fired. Ethics is a tenuous job position at best, even in a large company. It’s seen as a cost center. I don’t think there’s much to read into why AI ethicists would get laid off.

andresgottlieb 4y ago

"such biases"