Google’s AI-powered ‘inclusive warnings’ feature is very broken • Hacker News Board

Re: Google’s AI-powered ‘inclusive warnings’ feature is very broken

#251

Post by Auracle » Fri, Apr 22, 2022, 5:26 PM UTC

duxup wrote: post #154

dekhn wrote: post #13

One of the reasons I left google is that their "content moderation team" (the folks who make you take down wrongthink memes) is so far out of touch, that somebody had to explain to them there are people in the world who are discriminated against, but aren't black. They simply didn't know that was the case! And if that's the people who are moderating content... I said motherboard all the time in meetings and chats, ne…

One of the challenges with moderation teams like that or similar is that the folks who REALLY want to do that job .. are the folks who absolutely should NOT be doing it.

For anyone reading this, I volunteer. My moderation level would be based on words that were considered bad in the year 2005. I won’t have to do much work and you still get to say that you have someone that’s doing the job. Win-win.

Re: Google’s AI-powered ‘inclusive warnings’ feature is very broken

#252

Post by dogleash » Fri, Apr 22, 2022, 5:26 PM UTC

madamelic wrote: post #166

olalonde wrote: post #42

Earlier quoted context omitted.

Reminds me of when GitHub stopped using "master" for the default branch because it was somehow offensive.

My opinion: `main` is better. It's shorter, it's more descriptive, and costs nothing to remove something someone might find offensive. It's pretty much an all around win.

>costs nothing

The communication cost of the change is not free. Do you think the internal wiki and new-hire git training materials all updated themselves?

It wasn't a large process change. Most of our git users are competent git users. But I billed hours dealing with it.

I still have people here who haven't touched a repository that uses "main" instead of "master". They have better things to do with their lives than lurk social media focused on programming. They don't know yet. Eventually they'll pull something with "main" instead of "master". Can I get your phone number so you can be the one to explain to them?

Re: Google’s AI-powered ‘inclusive warnings’ feature is very broken

#253

Post by foolfoolz » Fri, Apr 22, 2022, 5:26 PM UTC

orangecat wrote: post #155

jacobsenscott wrote: post #82

Earlier quoted context omitted.

whitelist and blacklist are lazy language, even if you ignore the fact that you personally are not bothered by them and you don't have the mental capacity to understand why someone else might be. They carry no intrinsic meaning, and if you can't come up with more descriptive language you are the problem. In all cases you can come up with a better word - include, deny, allow, block, ban, accept, etc. It all depends on…

Your ableist bigotry against people with lower intellectual capacity has been noted.

You will now be limited to 1 inclusive language dismissal per week. All inclusion warnings past the first one will be made for you

Re: Google’s AI-powered ‘inclusive warnings’ feature is very broken

#254

Post by riedel » Fri, Apr 22, 2022, 5:26 PM UTC

meetups323 wrote: post #11

I wonder if this is an example of not knowing your audience (or rather assuming one's self represents a wider audience than it does). The people vouching for this at Google are likely product marketing managers, public relations folks, social media managers, etc. All they do is write corporate garbage all day, and much like we have "nit"s in PR's for formatting, variable names, etc; they likely have similar reviews t…

IMHO OK if the feature would actually provide explainations and it would be based on some sort of rulebook rather some random decision of individuals. I also would be fine if there is a warning if I use the word 'property owner' because some random internet user says that is deeply capitalistic. In the end I could decide if I want to follow the argument. Just nudging people to get away with a warning is bad and will lead to no warning. I doubt even that it will lead to a more inclusive world because no reflection is involved.

The problem for me is particularly that the combination of monopolies combined with AI that will learn from data largely filtered by those monopolies will generate some questionable gradients. So, yes, this might ultimately change language very quickly without much of human discourse over it. This will lead to language with less variation and arguably to a world that does not encourage variation and will be in effect less inclusive.

Re: Google’s AI-powered ‘inclusive warnings’ feature is very broken

#255

Post by wincy » Fri, Apr 22, 2022, 5:27 PM UTC

blast wrote: post #142

karaterobot wrote: post #20

I dislike this proposed feature as much as the next person, but it's worth noting that the main complaint of this article is fairly unique: they work on a channel called Motherboard, and this feature is going to cause the name of their channel to be considered a problem. Most people won't experience that specific error case. It's valid, and points to the kinds of unforeseen consequences you run into when trying to ma…

It's not valid to censor the word "mother". It's either laughable or sinister, depending on whether the people trying to do it actually have power.

This is literally Brave New World stuff. Not exaggerating, the word mother was hugely offensive to the people in that book. Unfortunately we don’t even get the Orgy Porgy parts. It’s like the worst parts of 1984 and Brave New World mixed together.

Re: Google’s AI-powered ‘inclusive warnings’ feature is very broken

#256

Post by CuriouslyC » Fri, Apr 22, 2022, 5:27 PM UTC

vimy wrote: post #204

mgraczyk wrote: post #9

In the article they use the example of "Motherboard" and "Landlord" and seem to suggest that these are areas where the Google AI is making mistakes or being overly strict. As a Google employee expressing my own opinion and observations of company culture, I can say that these are 100% not mistakes. Many Google employees are just so out of touch with the real world that they believe it is the duty of Google Docs to ch…

This is why I’m pessimistic about the West winning the second cold war. While Chinese engineers are working hard on world dominating AI models and other hard tech, Western engineers are wasting man hours on crap like this.

The west will win the second cold war because the communist party will not be able to release its iron grip on its populace. The Chinese people only accept that iron grip now because it's carrying them up from a low place, but once that stops the constriction will become suffocating, leading to civil unrest and diaspora of the upper tier of Chinese society.

Re: Google’s AI-powered ‘inclusive warnings’ feature is very broken

#257

Post by throwaway0x7E6 » Fri, Apr 22, 2022, 5:27 PM UTC

jedimastert wrote: post #78

kache_ wrote: post #22

I've been privy to people trying to ban words like "whitelabel" at my company. Thankfully, management & executives at my company never stood for that BS so I still get to use whitelist & blacklist in documentation, and don't worry about those non-issues get in the way of me actually doing work. Google might be too far gone, and working with those people would be so exhausting that it's actually stopped me from applyi…

Silly question: ignoring any kind of history, why fight for terms like "whitelist" and "blacklist" when terms like "allowlist" and "blocklist" are objectively clearer, especially for folks who's first language is not English?

>why fight

because you give them an inch and they'll take a mile.

look up "menstruators" and "birthing people" to see what else that kind of people fight for.

>objectively clearer

then it wouldn't be necessary to force the change. nobody had a problem with these words until very recently, and only in very narrow circles of very loud people with disproportionate amount of power

>especially for folks who's first language is not English?

that's a very slippery slope

Re: Google’s AI-powered ‘inclusive warnings’ feature is very broken

#258

Post by simion314 » Fri, Apr 22, 2022, 5:28 PM UTC

madamelic wrote: post #166

olalonde wrote: post #42

Earlier quoted context omitted.

Reminds me of when GitHub stopped using "master" for the default branch because it was somehow offensive.

My opinion: `main` is better. It's shorter, it's more descriptive, and costs nothing to remove something someone might find offensive. It's pretty much an all around win.

I had a script that broke because someone changed master into main, put in an equation the 100 people that got satisfaction from this change and the tousands of people getting frustrated because of it.

Re: Google’s AI-powered ‘inclusive warnings’ feature is very broken

#259

Post by Auracle » Fri, Apr 22, 2022, 5:28 PM UTC

londons_explore wrote: post #182

xg15 wrote: post #83

Earlier quoted context omitted.

Out of couriosity: Why then didn't flag it the actual n-word of all things? > A transcribed interview of neo-Nazi and former Klan leader David Duke—in which he uses the N-word and talks about hunting Black people—gets no notes. Edit: Also interesting that this comment is dropping like a stone in HN's comment ranking, even though currently the comment score is at 1. If I accidentally triggered some anti-flame or anti-…

Perhaps certain words disable the feature entirely? The feature only seems to trigger when it has a suggestion. I can't imagine what suggestion would be suitable to replace the n word.

We need Clippy back, with webcam functionality. That way he can suggest ending the word in an “a” if you’re black. If you’re not he’ll report you to HR.

Re: Google’s AI-powered ‘inclusive warnings’ feature is very broken

#260

Post by IAmWorried » Fri, Apr 22, 2022, 5:28 PM UTC

londons_explore wrote: post #214

ElFitz wrote: post #62

And now the rest of the world will be very happy learn that they will once more have to silently deal with the aftermath of America’s latest political trend. After banning world-famous paintings from Facebook because they dared show a breast, a few well-off engineers and product managers will now pick which English words should remain, and which should be let go. How nice of them. Don’t know what we’d do without them…

This isn't unique to SV. There are plenty of examples of words and behaviours that were commonplace and then over time evolved to be unacceptable to use in polite company. I'd like to give some examples, but I fear I'd end up swiftly banned for even using some of the best examples. One interesting case is "idiots, imbeciles and morons" - once technical terms to describe a mental health scale. The 'most insulting' end…

> There are plenty of examples of words and behaviours that were commonplace and then over time evolved to be unacceptable to use in polite company.

Cmon man, there's a big difference between the natural evolution of language and some radical-run company forcibly jamming their desired changes down our throats through an instantaneous and global software update.