5 ms·

Google has another 'woke' problem with Gemini and it's going to be hard to fix

24 points by neverrroot 3y ago

bobse 3y ago
[dead]
whywhywhywhy 3y ago
What fumble of AI are we in with Google now? 10? 15? Like how many more fumbles on this tech that they had a decade head start on are they gonna make before someone does what needs to be done and trim the fat, fire the people causing the future of their company to slip extremely deep into irrelevance likely.
- neverrroot 3y ago
  When you don't focus anymore on performance, merit and excellence, I would argue, this is what you get.
- pyrale 3y ago
  > before someone does what needs to be done and trim the fat, fire the people Ah yes, the typical MBA answer. I don't understand the problem, I don't even know the company inner workings, but we can probably turn this around with a round of layoffs, agressive deadlines and a healthy scapegoating culture.
tirpen 3y ago
So their AI refused to actively contribute to increasing global warming, and people are calling this "woke" and say it's a bad thing? Some people just don't like any kinds of ethics and morals at all I guess. When I was young and watched Saturday morning cartoons, charterers with such beliefs were referred to as "evil", but I guess the preferred term nowadays is "anti-woke".
- UweSchmidt 3y ago
  1. Can you think of other things you would want to query an AI on that "increases global warming" (i.e. ultimately most things humans can do)? Should it be possible to learn about things that could increase global warming? 2. Do you want Google decide ethical or moral questions or should that be left to the individuals? How far should those corporate AIs interfere with human thinking and decisionmaking?
darkwater 3y ago
So, Gemini has a "woke" problem, and on of the problem is that it wasn't criticizing enough Adolf Hitler's actions compared to Elon Musks tweets? [1] Defending the literal biggest nazi's boss doesn't seem very "woke", but I don't live in the US so maybe I'm wrong... OTOH I agree that not willing to write a job description for an oil company [2] is way, way too much. [1] https://twitter.com/NateSilver538/status/1761800684272308302 https://twitter.com/NateSilver538/status/1761800684272308302 [2] https://twitter.com/TheStalwart/status/1761832472357335068 https://twitter.com/TheStalwart/status/1761832472357335068
- true_religion 3y ago
  Google has trained it AI to take no positions on moral topics. If it was permitted to then I could easily imagine it eventual generating a response where it counts up the number of people harmed by person A and the amount harmed by person B, then writing Steve Jobs is worse than Hitler because Apples anti competing practices have harmed (in a very minor way) more absolute numbers of people than what’s recorded in the holocaust.
dom96 3y ago
There is being ‘woke’ and then there is being factually inaccurate. Google’s Gemini was the latter with the images it generated. We really need better words to describe what is going on. ‘Woke’ and leftist employees aren’t it, it’s employees that have done a poor job in ensuring the model doesn’t discriminate. I’d expect far better from Google.
- belorn 3y ago
  There is a common phenomenon that when viewing people belonging to other social groups, people get described using generalized labels. When viewing people belong to ones own social group, they get described as non-generalized individuals. Is it systemic racism, or individual poor performing police employees who do a bad job? Are there systemic sexism, or individuals in managements that act unprofessional? Usually it is both. In the wrong culture individual people are allowed to do a bad job, and in many case rewarded for the wrong metrics which only further reinforce the wrong culture.
  - raxxorraxor 3y ago
    I disagree. To prove systemic issues within a culture is quite difficult. And people are justifiably defensive because they have become suspects. Ironic, considering the fight against racism and sexism is a fight against prejudice. So the cure is actually worsening the problem. But no, people in power in corporations don't need the consent of anyone working under them. There is also a wrong culture where people get on witch hunts towards allegedly blasphemous behavior.
happytiger 3y ago
> But it's also going to be a problem for Google because it has already said it is trying to influence the way its AI produces results. Goodness. Instead of building a digital twin to be of service and partnership with society, it seems abundantly clear that Google seems more aimed at reshaping society to their specifications and with no input or oversight from the public that receives them. This would be a “wrong use” of this technology. We do not want to create subservient “receiving classes” of people and controlling information people who run the thinking machines and decide how society should be. I think Google is getting kickback because their AI is presciently invoking the plot of Gattaca. Google feels like they are getting caught running an agenda with the technology, not having a technical problem. Now obviously everyone on here is going to be commenting about how hard it is to get these systems right, but this version was shipped after it was reviewed internally. How can these kinds of issues go unnoticed by an entire testing staff of some of the smartest people on the planet prior to their most important product release? Does that make sense? Is there something I’m missing with that?
- neverrroot 3y ago
  This, exactly this.
- raxxorraxor 3y ago
  If you work at google as an engineer or tester and you notice that the newest AI does output ridiculous levels of "corpo diversity", would you report it to higher ups? I believe anything involving diversity in the workplace or in this case advertising is extremely toxic and people might think twice. It is a culture of no tolerance (against intolerance supposedly) towards diverging thoughts that it is in stark contrast with the message they are trying to send in their products.
  - true_religion 3y ago
    Would anyone’s job be at risk if the framing of the error were “it seems our AI is whitewashing history by only generating black and Asian Nazis” or “our AI is incapable of generating European Jewish people, it always generates another race instead”?
raziel2p 3y ago
I really dislike the gut reaction to blame "left-leaning workers". It's not that long ago that AI models were accused of refusing to draw people of color [1], or other forms of pro-white racism [2]. It's obvious that creators of these models are trying to find the balance between biases - and that we may never succeed. 1. https://www.washingtonpost.com/technology/interactive/2023/ai-generated-images-bias-racism-sexism-stereotypes/ https://www.washingtonpost.com/technology/interactive/2023/a... 2. https://twitter.com/ronawang/status/1679867848741765122/photo/1 https://twitter.com/ronawang/status/1679867848741765122/phot...
UweSchmidt 3y ago
The public is asking for quite a lot of manual "fixing" for an AI that was supposed to be an entirely technical process. This has to be an impossible task, given the range of contentious issues and the difficulty of addressing them to the liking of the loud voices on social media. This is exacerbated by the current style of articles that don't primarily report on the issue, but on the reactions to the issue and thus preventing the readers from forming their own opinions. An article like this could give the reader a little excercise for the comments to try and create a better answer for some of the problematic queries - not easy! I hope we get to have our own AIs on our own machines, raw powertools instead of walled gardens carefully curated by scared corporations.
neverrroot 3y ago
And flagged. Honestly wondering why.
- manuelabeledo 3y ago
  Because we are seeing the same kind of article posted on HN, multiple times a day. It’s annoying.
  - neverrroot 3y ago
    This is not a reason to flag it. Furthermore it appears that the submissions that do not receive attention/votes are left unflagged. > What does [flagged] mean? > Users flagged the post as breaking the guidelines or otherwise not belonging on HN.
    - manuelabeledo 3y ago
      Spamming the same subject over and over, I would think doesn’t belong to HN.
      - neverrroot 3y ago
        Spamming the same subject over and over as the situation evolves is what historically has belonged and keeps belonging to HN, just hit the search button and see how many submissions there are on the same topic, that do not get flagged. That is the proof, that this belongs here. They of course do not all get the votes, but don't get flagged.
        manuelabeledo 3y ago
        > Spamming the same subject over and over as the situation evolves... I wouldn't say that opinions from random people are "developments". To me, the influx of articles about Gemini, look like classic Internet drama that will end nowhere. I'll happily flag these submissions.

24 points by neverrroot 3y ago

[dead]

What fumble of AI are we in with Google now? 10? 15? Like how many more fumbles on this tech that they had a decade head start on are they gonna make before someone does what needs to be done and trim the fat, fire the people causing the future of their company to slip extremely deep into irrelevance likely.

neverrroot 3y ago
When you don't focus anymore on performance, merit and excellence, I would argue, this is what you get.
pyrale 3y ago
> before someone does what needs to be done and trim the fat, fire the people Ah yes, the typical MBA answer. I don't understand the problem, I don't even know the company inner workings, but we can probably turn this around with a round of layoffs, agressive deadlines and a healthy scapegoating culture.

tirpen 3y ago

So their AI refused to actively contribute to increasing global warming, and people are calling this "woke" and say it's a bad thing? Some people just don't like any kinds of ethics and morals at all I guess. When I was young and watched Saturday morning cartoons, charterers with such beliefs were referred to as "evil", but I guess the preferred term nowadays is "anti-woke".

UweSchmidt 3y ago
1. Can you think of other things you would want to query an AI on that "increases global warming" (i.e. ultimately most things humans can do)? Should it be possible to learn about things that could increase global warming? 2. Do you want Google decide ethical or moral questions or should that be left to the individuals? How far should those corporate AIs interfere with human thinking and decisionmaking?

darkwater 3y ago

So, Gemini has a "woke" problem, and on of the problem is that it wasn't criticizing enough Adolf Hitler's actions compared to Elon Musks tweets? [1] Defending the literal biggest nazi's boss doesn't seem very "woke", but I don't live in the US so maybe I'm wrong... OTOH I agree that not willing to write a job description for an oil company [2] is way, way too much. [1] https://twitter.com/NateSilver538/status/1761800684272308302 https://twitter.com/NateSilver538/status/1761800684272308302 [2] https://twitter.com/TheStalwart/status/1761832472357335068 https://twitter.com/TheStalwart/status/1761832472357335068

true_religion 3y ago
Google has trained it AI to take no positions on moral topics. If it was permitted to then I could easily imagine it eventual generating a response where it counts up the number of people harmed by person A and the amount harmed by person B, then writing Steve Jobs is worse than Hitler because Apples anti competing practices have harmed (in a very minor way) more absolute numbers of people than what’s recorded in the holocaust.

dom96 3y ago

There is being ‘woke’ and then there is being factually inaccurate. Google’s Gemini was the latter with the images it generated. We really need better words to describe what is going on. ‘Woke’ and leftist employees aren’t it, it’s employees that have done a poor job in ensuring the model doesn’t discriminate. I’d expect far better from Google.

belorn 3y ago
There is a common phenomenon that when viewing people belonging to other social groups, people get described using generalized labels. When viewing people belong to ones own social group, they get described as non-generalized individuals. Is it systemic racism, or individual poor performing police employees who do a bad job? Are there systemic sexism, or individuals in managements that act unprofessional? Usually it is both. In the wrong culture individual people are allowed to do a bad job, and in many case rewarded for the wrong metrics which only further reinforce the wrong culture.
- raxxorraxor 3y ago
  I disagree. To prove systemic issues within a culture is quite difficult. And people are justifiably defensive because they have become suspects. Ironic, considering the fight against racism and sexism is a fight against prejudice. So the cure is actually worsening the problem. But no, people in power in corporations don't need the consent of anyone working under them. There is also a wrong culture where people get on witch hunts towards allegedly blasphemous behavior.

happytiger 3y ago

> But it's also going to be a problem for Google because it has already said it is trying to influence the way its AI produces results. Goodness. Instead of building a digital twin to be of service and partnership with society, it seems abundantly clear that Google seems more aimed at reshaping society to their specifications and with no input or oversight from the public that receives them. This would be a “wrong use” of this technology. We do not want to create subservient “receiving classes” of people and controlling information people who run the thinking machines and decide how society should be. I think Google is getting kickback because their AI is presciently invoking the plot of Gattaca. Google feels like they are getting caught running an agenda with the technology, not having a technical problem. Now obviously everyone on here is going to be commenting about how hard it is to get these systems right, but this version was shipped after it was reviewed internally. How can these kinds of issues go unnoticed by an entire testing staff of some of the smartest people on the planet prior to their most important product release? Does that make sense? Is there something I’m missing with that?

neverrroot 3y ago
This, exactly this.
raxxorraxor 3y ago
If you work at google as an engineer or tester and you notice that the newest AI does output ridiculous levels of "corpo diversity", would you report it to higher ups? I believe anything involving diversity in the workplace or in this case advertising is extremely toxic and people might think twice. It is a culture of no tolerance (against intolerance supposedly) towards diverging thoughts that it is in stark contrast with the message they are trying to send in their products.
- true_religion 3y ago
  Would anyone’s job be at risk if the framing of the error were “it seems our AI is whitewashing history by only generating black and Asian Nazis” or “our AI is incapable of generating European Jewish people, it always generates another race instead”?

raziel2p 3y ago

I really dislike the gut reaction to blame "left-leaning workers". It's not that long ago that AI models were accused of refusing to draw people of color [1], or other forms of pro-white racism [2]. It's obvious that creators of these models are trying to find the balance between biases - and that we may never succeed. 1. https://www.washingtonpost.com/technology/interactive/2023/ai-generated-images-bias-racism-sexism-stereotypes/ https://www.washingtonpost.com/technology/interactive/2023/a... 2. https://twitter.com/ronawang/status/1679867848741765122/photo/1 https://twitter.com/ronawang/status/1679867848741765122/phot...

UweSchmidt 3y ago

The public is asking for quite a lot of manual "fixing" for an AI that was supposed to be an entirely technical process. This has to be an impossible task, given the range of contentious issues and the difficulty of addressing them to the liking of the loud voices on social media. This is exacerbated by the current style of articles that don't primarily report on the issue, but on the reactions to the issue and thus preventing the readers from forming their own opinions. An article like this could give the reader a little excercise for the comments to try and create a better answer for some of the problematic queries - not easy! I hope we get to have our own AIs on our own machines, raw powertools instead of walled gardens carefully curated by scared corporations.

neverrroot 3y ago

And flagged. Honestly wondering why.

manuelabeledo 3y ago
Because we are seeing the same kind of article posted on HN, multiple times a day. It’s annoying.
- neverrroot 3y ago
  This is not a reason to flag it. Furthermore it appears that the submissions that do not receive attention/votes are left unflagged. > What does [flagged] mean? > Users flagged the post as breaking the guidelines or otherwise not belonging on HN.
  - manuelabeledo 3y ago
    Spamming the same subject over and over, I would think doesn’t belong to HN.
    - neverrroot 3y ago
      Spamming the same subject over and over as the situation evolves is what historically has belonged and keeps belonging to HN, just hit the search button and see how many submissions there are on the same topic, that do not get flagged. That is the proof, that this belongs here. They of course do not all get the votes, but don't get flagged.
      - manuelabeledo 3y ago
        > Spamming the same subject over and over as the situation evolves... I wouldn't say that opinions from random people are "developments". To me, the influx of articles about Gemini, look like classic Internet drama that will end nowhere. I'll happily flag these submissions.