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

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

businessinsider.com

24 pointsneverrroot36 comments

> 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?

This, exactly this.

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.

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”?

Of course you could frame it in a way that doesn't offend those that created the problem, but question would be if that is worth it to you tiptoeing around these issues. I wouldn't expect people to solve the underlying issue here anyway.

James Damore spoke up and Google made an example of him.

The company already changed a lot before, but it was a strong signal that speaking up will be punished no matter how good engineer you are .

This contributed in a big way to the culture at Google, culture that can now be seen flowing in their products.

Let’s be honest here: how many workplaces would allow a random engineer to broadcast a gender critical rant on their internal network?

People get fired for less than that. Damore was, and rightly so.

You should look at what he actually said - it was not some anti-woman screed like you may think. The overall goal of the paper AFAIR was to examine why (then) current diversity programs were not achieving the desired results and how to address that from beyond the lens of inherent bias. Its very short, you could read it over a single cup of coffee!

https://s3.documentcloud.org/documents/3914586/Googles-Ideol...

It does not matter what the contents are.

I could post the most scientific accurate post about IQ differences correlated to ethnicities, and get fired for it, because it's just not work appropriate.

For many years before that we (Googlers) were incentivized to mix work and life:

Do you really think playing pool,foosball, guitar hero, eating dinner at 18:30 when work hours end at 17:00, bringing kids to workplace on weekends to play, sleeping at the workplace, bringing laptops with you / being oncall when not working are work appropriate?

The problem was specifically not agreeing with democrat US politics, which is relevant to what's happening, nothing else.

> Do you really think playing pool,foosball, guitar hero, eating dinner at 18:30 when work hours end at 17:00, bringing kids to workplace on weekends to play, sleeping at the workplace, bringing laptops with you / being oncall when not working are work appropriate?

Since it is the company the one setting all that up, why don't you bring it up with them?

Regardless, US companies are more than happy to have employees working for them 24/7, and they seem to be getting away with it. In general, they dislike when employees bring inflammatory rhetoric into the workplace.

I invite you to join any Fortune 500 company and try what this guy did. Rest assured, they will not fire you because of who you vote, but they will fire you nevertheless.

See, the indignation of him being allegedly sexist is a pre-judgement (it is unlikely for that matter, but irrelevant). This is the reality in the business world, I get it.

But then you cannot turn around and try to school people about prejudice. That is a severe contradiction.

That the rules of PR demand his firing, is to fulfill emotional needs of people that formerly maybe would go witch hunting. Nothing better to do, I guess. So yeah, let us be honest indeed.

If something needs work, it isn't Damore, it is indeed this indignant culture. And it would be trivial to change too.

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.

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.

When you don't focus anymore on performance, merit and excellence, I would argue, this is what you get.

> 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.

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.

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.

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.

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".

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?

And flagged. Honestly wondering why.

Because we are seeing the same kind of article posted on HN, multiple times a day. It’s annoying.

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.

Spamming the same subject over and over, I would think doesn’t belong to HN.

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.

> 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.

Could it be that the reason for you flagging them is related to your beliefs/topic being discussed?

That would be irrelevant, even if true. Same could be said about people upvoting it.

Beg to differ. The counterbalance to mass upvoting is downvoting not flagging. Flagging removes a submission entirely from the listing and impacts future submissions from the same user. Flagging is what people use when they want to silence something or to prevent abuse/junk from staying listed.

There are no downvotes for stories.

> Flagging is what people use when they want to silence something or to prevent abuse/junk from staying listed.

"Junk" is a good qualificative.

Interesting, you allocate so much time to junk… can’t help but see you rather often in my flagged stories. Then there is your history

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

[2] https://twitter.com/TheStalwart/status/1761832472357335068

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.

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/a... 2. https://twitter.com/ronawang/status/1679867848741765122/phot...

Reply on news.ycombinator.com