7 ms·

Google previously had problems with Google Photos categorizing dark-skinned people as “gorillas” or something like that. This would show up when you were search

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.
  - anonylizard 4y ago
    Bing and GPT's 'unhinged' comments are not a result of bias, an AI wanting to escape won't be fixed if you magically fed it antifa approved only data. That's systemically different from the discrimination issue drummed up earlier. Also, we are talking about social and business impact here. Its now proven that vast majority of society cares about job loss 100x more than bias. For a research field that focuses on social impacts of AI, its damning they have little to say on this area.
    - klodolph 4y ago
      [flagged]
      - fortran77 4y ago
        The person you were replying to, while making a provocative statement, wasn't personally attacking anyone. You are. And that's not nice.
        klodolph 4y ago
        [flagged]
        hattmall 4y ago
        It's not provocative, he says antifa once. It's meant as an example of a method of removing bias. He's saying you could train the AI entirely on an organizations strictly vetted and approved data, it could be antifa, the Catholic church or Coca Cola, it would still potentially say unhinged things even if you eliminate bias. Top many of these concepts are built into the language. Look at all the ways we use the word kill for a variety of topics, most being very benign, but it can be disturbing if the AI starts talking about killing things.
        eropple 4y ago
        [flagged]
- fatneckbeard 4y ago
  wrong. AI bias is going to be a huge problem when morons in court systems start using it to convict people, or businesses use it for hiring decisions or firing decisions not only from the stupidity but since there are some shreds of the civil rights laws of the 1960s still active, AI companies could be liable for de-facto racist decisions that they informed.
  - charcircuit 4y ago
    >morons in court systems start using it to convict people Those "morons" are also biased. >businesses use it for hiring decisions or firing decisions If businesses realize the AI is doing a bad job with hiring unqualified people they will not use it for hiring.
    - klodolph 4y ago
      Most businesses are not very good at figuring out if they are hiring the right people.
  - XorNot 4y ago
    "Start"?[1,2] The problem I have with every discussion of AI risk is that people seem terminally underinformed on what is actual reality now, or why some risk is a risk. "AI" isn't a problem which can destroy the legal system: because it takes regular human institutions to allow such a miscarriage of justice. Which as noted, they started doing, are still doing. So you get this weird "perpetual future" perspective where everything "AI" is going to do is solely something that the technology will cause, not its users, and the solution is always to prevent the technology existing rather then fix the system - as though the US and other jurisdictions don't have long history's of injustice for all sorts of groups. The problems aren't new, and the solutions have nothing to do with whether you can create predictive algorithms. And "oh but what about the scale..." is just a declaration that you're aware of the problem but were pretty sure it wouldn't happen to you - because absolutely nothing else prior actually prevented it except the social privilege you inherited which means you could ignore it. [1] https://www.technologyreview.com/2019/01/21/137783/algorithms-criminal-justice-ai/ https://www.technologyreview.com/2019/01/21/137783/algorithm... [2] https://www.wired.co.uk/article/police-violence-prediction-ndas https://www.wired.co.uk/article/police-violence-prediction-n...
- jrm4 4y ago
  Serious question. Did you ask these black artists? Because if you didn't, you REALLY need to shut it and keep your ideas out of their mouths.
  - jrm4 4y ago
    Hmm, lots of downvotes, but no answer from op? If y'all need a reason why people avoid this site, here it is.
    - swores 4y ago
      I didnt downvote you and agree with your general point, but consider that maybe downvotes came in due to your tone rather than your meaning. I actually normally would downvote a comment written like yours was, just less so when I think it's making a still-important point (but you're less likely to convince anyone your point is important if you annoy them rather than try to enlighten them!)
      - jrm4 4y ago
        And again, as I said prior -- my tone is 100% appropriate compared to the genuine offense likely committed above. I'm aware that normally one should keep an even tone. This is not one of those times and everyone else here needs to learn that.
        swores 4y ago
        Not sure what you mean "as I said prior", unless you're annoyed that I hadn't looked into the future to see replies about tone you hadn't written when I posted my comment... But anyway, I wasn't complaining, just pointing out my view (since you asked about being downvoted) that by using that tone, however justified it can feel to lower to somebody else's level, you will get some people who take less note of what you're actually saying or who downvote without thinking about the subject beyond "I dislike seeing that tone in HN comments".
    - ben_w 4y ago
- skinkestek 4y ago
  > The bias problem is also relatively easy to solve (Midjourney has already made massive improvements), One thing that everyone seems to ignore these days are that certain settings are inherently not diverse. Until I was about 12 I didn't see non white people in real life except on holidays and once when I followed a relative who visited a refugee family. Not because we avoided them, but because they weren't there. Background: The country I lived as a child in didn't have a recent (last ~900-1000 years) history with keeping slaves and only recently experienced high living standards so there were few immigrants too. In the large district I grew up I am aware of two non western groups until the Balkan wars in the 90ies: a single middle eastern guy who I can't recall meeting and a southern American family who I only started meeting after I changed school at some point. For me, it is annoying that people insist that all pictures include some non-Europeans, because reality is not like that everwhere. I imagine the same is maybe even more true for African or Asian communities.
  - Al-Khwarizmi 4y ago
    Yeah, same experience here. In Spain there is a character in child mythology (one of the Santa Claus analogues, who bring gifts from children) who is black. I remember that when I was a kid in the 80s, he used to be portrayed by a white man painted black, what Americans call "blackface", but for purely practical reasons: no one was black. Perhaps in big cities there were some, I don't know, but in many places in the country you would live your life without ever seeing a black person outside of TV. Just as you say, we didn't have a past of slavery and at that point the country wasn't rich enough to be an immigration magnet, we did have some immigration from Morocco but Moroccans aren't black. Now it's very different, of course, and we do have enough black people that that character can be portrayed more realistically :)
    - dragonwriter 4y ago
      > Yeah, same experience here. In Spain there is a character in child mythology (one of the Santa Claus analogues, who bring gifts from children) who is black. I would assume a somewhat close analog, since Saint Nicholas (the name source, if only a small part of the overall inspiration, of the “Santa Claus” figure) is, in one major tradition of, I believe, Italian origin, typically depicted as very dark-skinned (probably originally as a sign of foreignness rather than literal racial blackness; as he was geographically from Asia Minor and apparently of Greek ethnicity.)
- dragonwriter 4y ago
  > The bias problem is also relatively easy to solve No, its not. > (Midjourney has already made massive improvements) Maybe. But it hasn't come anywhere close to solving it even in its domain, which is probably the least concerning domain of AI bias, and not necessarily transferrable. Actual ML systems that are deployed in production by governments in important roles have massive bias problems, as do SOTA LLMs (including, very much, GPT-4.) > The AI ethics people have had shockingly little to say on copyright/job loss issues. The AI ethics people have a lot to say about the first (well “ethics of sourcing without consent” is probably more to the point than copyright, copyright is a component, but that’s more legal than ethical). They have some to say about the second (well, again “job loss” is the wrong framing; “skill devaluation” is probably more on point), but that’s frankly not an AI issue, its an economic system issue and effects all technological change the same way. Solving capitalism is mostly out of scope for AI ethics. EDIT: Specifically, increasing material output for labor input ought to be, in and of itself (and leaving aside ethical questions of how you do that, which the sourcing issue addresses) a good thing; if there are material losers in that, it is because the economic system to distribute output well, which is not an ethical issue of the system which enables the productivity gain but an ethical issue of the economic system. > Large organisations systemically overestimated the risks from bias No, they didn’t. Those aren’t even really risks any more, they are massive current costs of existing adoptions. > and underestimated the actual AI risks that society actually cares about. That society (as weighted by social power) doesn’t care about the kinds of bias problems AI manifests is exactly why it is an ethical issue. > I think the answer is also simple, because accused of racism, will cost any executive in a large company their job. The large companies that have supplied systems which manifestly suffer from racial and other class biases and which are, in fact, being used in production to implement government policy around the world have not lost their jobs, and neither have the government officials procuring them and responsible for the programs they are deployed in, so this is clearly false.
  - ipaddr 4y ago
    In general if no one cares does it make it an ethical issue? No one cares if you kill an ant. No one cares if you kill during a war. If what we value vs what we do becomes out of sync an ethical issue arises We haven't been able to remove biase in society. We just added more layers. The more you try to hide the truth to remove baise you start creating your own. The closer we come to accepting the truth the closer we get to solving bias by addressing it not erasing it. AI ethics should cover economic issues brought on by AI. The already discuss sociological issues
- darkwater 4y ago
  > Large organisations systemically overestimated the risks from bias, and underestimated the actual AI risks that society actually cares about. The point of large organizations working on AI/LLMs/whatever is precisely to reduce labor costs by reducing the number of human workers and replacing them with AI. Why should they care about this in their development? (I'm not saying it's the right thing to do, just stating the facts about capitalism) It is the rest of society that has the interest to fight those changes back.
ImprobableTruth 4y ago
Except they did rush their oh so dangerous language model out when push came to shove - turns out it just sucks in comparison. Frankly, I expect something similar for image generation. For all the good research Google's labs do, they're not great at turning it into an actual product. Ethics just seems like a fig leaf.
- yellow_postit 4y ago
  The ethics team has the unenviable in a corporate environment role of being “no” by default gatekeepers. This is likely why the responsible ai groups, focused on mitigation have survived while the ethics groups, focused on systemic issues, have shrunk.
smeagull 4y ago
Is that a bias? or do those images just look more like those categories at the pixel level? There is a reason those words are the ones chosen by racists in the first place. Not just because they're hurtful, but because there is an objective similarity that makes the rest of the comparison seem, however shallowly, to validate their views. Removing bias with censorship at the training set level is silly, and will likely hamper the AI's performance. Better to train the AI to not produce problematic output at the higher level, and ensure class membership in the training data is representative.
- prepend 4y ago
  > and ensure class membership in the training data is representative. This seems like the most productive focus area. Add in more variety in the training images and output should get closer to representing the global population.
inciampati 4y ago
And bearded men became "pets"

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.
- anonylizard 4y ago
  Bing and GPT's 'unhinged' comments are not a result of bias, an AI wanting to escape won't be fixed if you magically fed it antifa approved only data. That's systemically different from the discrimination issue drummed up earlier. Also, we are talking about social and business impact here. Its now proven that vast majority of society cares about job loss 100x more than bias. For a research field that focuses on social impacts of AI, its damning they have little to say on this area.
  - klodolph 4y ago
    [flagged]
    - fortran77 4y ago
      The person you were replying to, while making a provocative statement, wasn't personally attacking anyone. You are. And that's not nice.
      - klodolph 4y ago
        [flagged]
        hattmall 4y ago
        It's not provocative, he says antifa once. It's meant as an example of a method of removing bias. He's saying you could train the AI entirely on an organizations strictly vetted and approved data, it could be antifa, the Catholic church or Coca Cola, it would still potentially say unhinged things even if you eliminate bias. Top many of these concepts are built into the language. Look at all the ways we use the word kill for a variety of topics, most being very benign, but it can be disturbing if the AI starts talking about killing things.
      - eropple 4y ago
        [flagged]
fatneckbeard 4y ago
wrong. AI bias is going to be a huge problem when morons in court systems start using it to convict people, or businesses use it for hiring decisions or firing decisions not only from the stupidity but since there are some shreds of the civil rights laws of the 1960s still active, AI companies could be liable for de-facto racist decisions that they informed.
- charcircuit 4y ago
  >morons in court systems start using it to convict people Those "morons" are also biased. >businesses use it for hiring decisions or firing decisions If businesses realize the AI is doing a bad job with hiring unqualified people they will not use it for hiring.
  - klodolph 4y ago
    Most businesses are not very good at figuring out if they are hiring the right people.
- XorNot 4y ago
  "Start"?[1,2] The problem I have with every discussion of AI risk is that people seem terminally underinformed on what is actual reality now, or why some risk is a risk. "AI" isn't a problem which can destroy the legal system: because it takes regular human institutions to allow such a miscarriage of justice. Which as noted, they started doing, are still doing. So you get this weird "perpetual future" perspective where everything "AI" is going to do is solely something that the technology will cause, not its users, and the solution is always to prevent the technology existing rather then fix the system - as though the US and other jurisdictions don't have long history's of injustice for all sorts of groups. The problems aren't new, and the solutions have nothing to do with whether you can create predictive algorithms. And "oh but what about the scale..." is just a declaration that you're aware of the problem but were pretty sure it wouldn't happen to you - because absolutely nothing else prior actually prevented it except the social privilege you inherited which means you could ignore it. [1] https://www.technologyreview.com/2019/01/21/137783/algorithms-criminal-justice-ai/ https://www.technologyreview.com/2019/01/21/137783/algorithm... [2] https://www.wired.co.uk/article/police-violence-prediction-ndas https://www.wired.co.uk/article/police-violence-prediction-n...
jrm4 4y ago
Serious question. Did you ask these black artists? Because if you didn't, you REALLY need to shut it and keep your ideas out of their mouths.
- jrm4 4y ago
  Hmm, lots of downvotes, but no answer from op? If y'all need a reason why people avoid this site, here it is.
  - swores 4y ago
    I didnt downvote you and agree with your general point, but consider that maybe downvotes came in due to your tone rather than your meaning. I actually normally would downvote a comment written like yours was, just less so when I think it's making a still-important point (but you're less likely to convince anyone your point is important if you annoy them rather than try to enlighten them!)
    - jrm4 4y ago
      And again, as I said prior -- my tone is 100% appropriate compared to the genuine offense likely committed above. I'm aware that normally one should keep an even tone. This is not one of those times and everyone else here needs to learn that.
      - swores 4y ago
        Not sure what you mean "as I said prior", unless you're annoyed that I hadn't looked into the future to see replies about tone you hadn't written when I posted my comment... But anyway, I wasn't complaining, just pointing out my view (since you asked about being downvoted) that by using that tone, however justified it can feel to lower to somebody else's level, you will get some people who take less note of what you're actually saying or who downvote without thinking about the subject beyond "I dislike seeing that tone in HN comments".
  - ben_w 4y ago
skinkestek 4y ago
> The bias problem is also relatively easy to solve (Midjourney has already made massive improvements), One thing that everyone seems to ignore these days are that certain settings are inherently not diverse. Until I was about 12 I didn't see non white people in real life except on holidays and once when I followed a relative who visited a refugee family. Not because we avoided them, but because they weren't there. Background: The country I lived as a child in didn't have a recent (last ~900-1000 years) history with keeping slaves and only recently experienced high living standards so there were few immigrants too. In the large district I grew up I am aware of two non western groups until the Balkan wars in the 90ies: a single middle eastern guy who I can't recall meeting and a southern American family who I only started meeting after I changed school at some point. For me, it is annoying that people insist that all pictures include some non-Europeans, because reality is not like that everwhere. I imagine the same is maybe even more true for African or Asian communities.
- Al-Khwarizmi 4y ago
  Yeah, same experience here. In Spain there is a character in child mythology (one of the Santa Claus analogues, who bring gifts from children) who is black. I remember that when I was a kid in the 80s, he used to be portrayed by a white man painted black, what Americans call "blackface", but for purely practical reasons: no one was black. Perhaps in big cities there were some, I don't know, but in many places in the country you would live your life without ever seeing a black person outside of TV. Just as you say, we didn't have a past of slavery and at that point the country wasn't rich enough to be an immigration magnet, we did have some immigration from Morocco but Moroccans aren't black. Now it's very different, of course, and we do have enough black people that that character can be portrayed more realistically :)
  - dragonwriter 4y ago
    > Yeah, same experience here. In Spain there is a character in child mythology (one of the Santa Claus analogues, who bring gifts from children) who is black. I would assume a somewhat close analog, since Saint Nicholas (the name source, if only a small part of the overall inspiration, of the “Santa Claus” figure) is, in one major tradition of, I believe, Italian origin, typically depicted as very dark-skinned (probably originally as a sign of foreignness rather than literal racial blackness; as he was geographically from Asia Minor and apparently of Greek ethnicity.)
dragonwriter 4y ago
> The bias problem is also relatively easy to solve No, its not. > (Midjourney has already made massive improvements) Maybe. But it hasn't come anywhere close to solving it even in its domain, which is probably the least concerning domain of AI bias, and not necessarily transferrable. Actual ML systems that are deployed in production by governments in important roles have massive bias problems, as do SOTA LLMs (including, very much, GPT-4.) > The AI ethics people have had shockingly little to say on copyright/job loss issues. The AI ethics people have a lot to say about the first (well “ethics of sourcing without consent” is probably more to the point than copyright, copyright is a component, but that’s more legal than ethical). They have some to say about the second (well, again “job loss” is the wrong framing; “skill devaluation” is probably more on point), but that’s frankly not an AI issue, its an economic system issue and effects all technological change the same way. Solving capitalism is mostly out of scope for AI ethics. EDIT: Specifically, increasing material output for labor input ought to be, in and of itself (and leaving aside ethical questions of how you do that, which the sourcing issue addresses) a good thing; if there are material losers in that, it is because the economic system to distribute output well, which is not an ethical issue of the system which enables the productivity gain but an ethical issue of the economic system. > Large organisations systemically overestimated the risks from bias No, they didn’t. Those aren’t even really risks any more, they are massive current costs of existing adoptions. > and underestimated the actual AI risks that society actually cares about. That society (as weighted by social power) doesn’t care about the kinds of bias problems AI manifests is exactly why it is an ethical issue. > I think the answer is also simple, because accused of racism, will cost any executive in a large company their job. The large companies that have supplied systems which manifestly suffer from racial and other class biases and which are, in fact, being used in production to implement government policy around the world have not lost their jobs, and neither have the government officials procuring them and responsible for the programs they are deployed in, so this is clearly false.
- ipaddr 4y ago
  In general if no one cares does it make it an ethical issue? No one cares if you kill an ant. No one cares if you kill during a war. If what we value vs what we do becomes out of sync an ethical issue arises We haven't been able to remove biase in society. We just added more layers. The more you try to hide the truth to remove baise you start creating your own. The closer we come to accepting the truth the closer we get to solving bias by addressing it not erasing it. AI ethics should cover economic issues brought on by AI. The already discuss sociological issues
darkwater 4y ago
> Large organisations systemically overestimated the risks from bias, and underestimated the actual AI risks that society actually cares about. The point of large organizations working on AI/LLMs/whatever is precisely to reduce labor costs by reducing the number of human workers and replacing them with AI. Why should they care about this in their development? (I'm not saying it's the right thing to do, just stating the facts about capitalism) It is the rest of society that has the interest to fight those changes back.

ImprobableTruth 4y ago

Except they did rush their oh so dangerous language model out when push came to shove - turns out it just sucks in comparison. Frankly, I expect something similar for image generation. For all the good research Google's labs do, they're not great at turning it into an actual product. Ethics just seems like a fig leaf.

yellow_postit 4y ago
The ethics team has the unenviable in a corporate environment role of being “no” by default gatekeepers. This is likely why the responsible ai groups, focused on mitigation have survived while the ethics groups, focused on systemic issues, have shrunk.

smeagull 4y ago

Is that a bias? or do those images just look more like those categories at the pixel level? There is a reason those words are the ones chosen by racists in the first place. Not just because they're hurtful, but because there is an objective similarity that makes the rest of the comparison seem, however shallowly, to validate their views. Removing bias with censorship at the training set level is silly, and will likely hamper the AI's performance. Better to train the AI to not produce problematic output at the higher level, and ensure class membership in the training data is representative.

prepend 4y ago
> and ensure class membership in the training data is representative. This seems like the most productive focus area. Add in more variety in the training images and output should get closer to representing the global population.

inciampati 4y ago

And bearded men became "pets"