5 ms·

The asymmetry is the bigger part. Advertising allows for community growth by pointing to more resources and I formation. Essentially, "If you support X, here

The asymmetry is the bigger part. Advertising allows for community growth by pointing to more resources and I formation. Essentially, "If you support X, here is a group working on the topic of X.". When this is allowed for X="white power" but forbidden for X="black lives matter", that's a huge disparity. That means ads cannot be used to recruit against police violence, but can be used to recruit for white supremacy.

CivBase 6y ago
It's very easy to make a list like this look asymmetrical when your article cherry picks the most outrage-inducing results from your test and disregards any results that don't fit your narrative.
- danso 6y ago
  The story links to a Github repo [0] with code, input data, and several notebooks, as well as a lengthy methodology [1]. [0] https://github.com/the-markup/investigation-youtube-ad-placements https://github.com/the-markup/investigation-youtube-ad-place... [1] https://themarkup.org/google-the-giant/2021/04/08/how-we-discovered-googles-hate-blocklist-for-ad-placements-on-youtube https://themarkup.org/google-the-giant/2021/04/08/how-we-dis...
  - CivBase 6y ago
    But they don't give background for how they chose the terms they used as inputs. They just hand-wave through that part of the methodology. All you have to do is cherry pick the inputs and keep re-running until you have the desired output. Garbage in, garbage out. > 58 well-known hate terms and phrases Funny enough, I didn't recognize many of those "well-known" terms and every single one of them ended up being white supremacist. On top of that, 58 is a pathetically small dataset to draw conclusions from. You'll need far more than that to credibly demonstrate a bias. I can probably come up with that many racial slurs off the top of my head and I bet you anything the overwhelming majority of them will be blacklisted.
    - MereInterest 6y ago
      Their github repo has a link to further description on each hate term. Some are hate groups. Some are deliberate misinformation about marginalized groups. Some are phrases instead to sound non-offensive to a general audience, while edging racist ideas into the mainstream. Also, hate groups tend to cycle through new terminology whenever the previous batch becomes easily recognized by a general audience. https://github.com/the-markup/investigation-youtube-ad-placements/blob/master/data/input/hate_terms_background_info.csv https://github.com/the-markup/investigation-youtube-ad-place...
      - CivBase 6y ago
        Describing what makes a term hateful is not the same as explaining how it got on the list. There are plenty of better known hate terms which are conspicuous by their absence. Naturally, the more obscure the term the less likely it is to be caught by the blacklist.
      - suifbwish 6y ago
        How is it 2021 and no one questions this BS? IMO any group that focuses on affirming the identity of fictional races, IS a hate group. A racially themed group pushing its own propaganda can never not be a hate group. We have the human species and just variations in hair, skin and eye color. Can’t we get past that simple concept?
    - danso 6y ago
      As the methodology states, they compiled a list of 2,039 terms, and did the manual and automated preprocessing work to dedupe that list (the result of which seems to be here [0]). They subsequently vetted the final list in collaboration with Harvard's Shorenstein Center. > Funny enough, I didn't recognize many of those "well-known" terms and every single one of them ended up being white supremacist. You don't recognize the phrases "kkk", "holocaust denial", "white power", "ethnic cleansing", or "daily stormer"? How is that the story's fault that you haven't encountered these well-known phrases and entities? In any case, the code and source list of keywords is all in their repo. You're accusing them of "cherry-picking" – when they've provided the entirety of their search space and API output? If you have a better idea of assessing Youtube's policy vs. actual implementation, or you just feel that the methodology is easy to invalidate – there's nothing stopping you from cloning the repo and running your own automated check and analysis. I mean, other then the fact that Google obfuscated their API in response to this story. [0] https://github.com/the-markup/investigation-youtube-ad-placements/blob/62d581ff1438c742be74bc578027fef983cf4fe8/notebooks/terms.py https://github.com/the-markup/investigation-youtube-ad-place...

CivBase 6y ago

It's very easy to make a list like this look asymmetrical when your article cherry picks the most outrage-inducing results from your test and disregards any results that don't fit your narrative.

danso 6y ago
The story links to a Github repo [0] with code, input data, and several notebooks, as well as a lengthy methodology [1]. [0] https://github.com/the-markup/investigation-youtube-ad-placements https://github.com/the-markup/investigation-youtube-ad-place... [1] https://themarkup.org/google-the-giant/2021/04/08/how-we-discovered-googles-hate-blocklist-for-ad-placements-on-youtube https://themarkup.org/google-the-giant/2021/04/08/how-we-dis...
- CivBase 6y ago
  But they don't give background for how they chose the terms they used as inputs. They just hand-wave through that part of the methodology. All you have to do is cherry pick the inputs and keep re-running until you have the desired output. Garbage in, garbage out. > 58 well-known hate terms and phrases Funny enough, I didn't recognize many of those "well-known" terms and every single one of them ended up being white supremacist. On top of that, 58 is a pathetically small dataset to draw conclusions from. You'll need far more than that to credibly demonstrate a bias. I can probably come up with that many racial slurs off the top of my head and I bet you anything the overwhelming majority of them will be blacklisted.
  - MereInterest 6y ago
    Their github repo has a link to further description on each hate term. Some are hate groups. Some are deliberate misinformation about marginalized groups. Some are phrases instead to sound non-offensive to a general audience, while edging racist ideas into the mainstream. Also, hate groups tend to cycle through new terminology whenever the previous batch becomes easily recognized by a general audience. https://github.com/the-markup/investigation-youtube-ad-placements/blob/master/data/input/hate_terms_background_info.csv https://github.com/the-markup/investigation-youtube-ad-place...
    - CivBase 6y ago
      Describing what makes a term hateful is not the same as explaining how it got on the list. There are plenty of better known hate terms which are conspicuous by their absence. Naturally, the more obscure the term the less likely it is to be caught by the blacklist.
    - suifbwish 6y ago
      How is it 2021 and no one questions this BS? IMO any group that focuses on affirming the identity of fictional races, IS a hate group. A racially themed group pushing its own propaganda can never not be a hate group. We have the human species and just variations in hair, skin and eye color. Can’t we get past that simple concept?
  - danso 6y ago
    As the methodology states, they compiled a list of 2,039 terms, and did the manual and automated preprocessing work to dedupe that list (the result of which seems to be here [0]). They subsequently vetted the final list in collaboration with Harvard's Shorenstein Center. > Funny enough, I didn't recognize many of those "well-known" terms and every single one of them ended up being white supremacist. You don't recognize the phrases "kkk", "holocaust denial", "white power", "ethnic cleansing", or "daily stormer"? How is that the story's fault that you haven't encountered these well-known phrases and entities? In any case, the code and source list of keywords is all in their repo. You're accusing them of "cherry-picking" – when they've provided the entirety of their search space and API output? If you have a better idea of assessing Youtube's policy vs. actual implementation, or you just feel that the methodology is easy to invalidate – there's nothing stopping you from cloning the repo and running your own automated check and analysis. I mean, other then the fact that Google obfuscated their API in response to this story. [0] https://github.com/the-markup/investigation-youtube-ad-placements/blob/62d581ff1438c742be74bc578027fef983cf4fe8/notebooks/terms.py https://github.com/the-markup/investigation-youtube-ad-place...