Slashdot Mirror
Google's Sentiment Analyzer Thinks Being Gay Is Bad (vice.com)
gooddogsgotoheaven shares a report from Motherboard: In July 2016, Google announced the public beta launch of a new machine learning application program interface (API), called the Cloud Natural Language API. It allows developers to incorporate Google's deep learning models into their own applications. As the company said in its announcement of the API, it lets you "easily reveal the structure and meaning of your text in a variety of languages." In addition to entity recognition (deciphering what's being talked about in a text) and syntax analysis (parsing the structure of that text), the API included a sentiment analyzer to allow programs to determine the degree to which sentences expressed a negative or positive sentiment, on a scale of -1 to 1. The problem is the API labels sentences about religious and ethnic minorities as negative -- indicating it's inherently biased. For example, it labels both being a Jew and being a homosexual as negative. A Google spokesperson issued the following statement in response to Motherboard's request for comment: "We dedicate a lot of efforts to making sure the NLP API avoids bias, but we don't always get it right. This is an example of one of those times, and we are sorry. We take this seriously and are working on improving our models. We will correct this specific case, and, more broadly, building more inclusive algorithms is crucial to bringing the benefits of machine learning to everyone."
If you've already decided that being homosexual (or a Jew, or a redhead, or a lefty, you name it) must not be deemed negative, why do you need analysis at all?
Anyway, here is a constructive suggestion: define more precisely what "negative" is. (Things like this are sorely needed IRL too, where people e.g. routinely use statistical norm as an argument in ethical discussions.)
It's almost like objective, quantifiable reality and feel-good political correctness are fundamentally at odds with each other.
Their api is trying to be too universal and monolithic.
Gay is neither good or bad. Gay is just a label. Google should be able to recognize that 'gay' means one thing when a bigot uses it, but it should also recognize that 'gay' means something else when others use it.
In other words, the meaning of words should be heavily weighed on the training data of the individual listening/reading and on the training data of the individual speaking/writing.
So many of us already use "gay" and "jew" as derogatory terms. Is it any wonder that Google's NLP picked up on that? What source do you think it learned from?
Biologically speaking homosexuality is bad for the replication of any species in it's most basic form
Well, no.
Being homosexual might be bad for the replication of the individual (since it reduces the individual's chances of having offspring), but it doesn't necessarily follow that it's therefore also bad for the species.
Consider that childless homosexuals, not being burdened by the task of having to care for children of their own, will likely therefore have extra time and resources available to help protect and care for the children of child-bearing couples in their family or community, thus improving the children's (and therefore the species') chances of surviving. Given the huge time investment that human children require, increasing a child's odds of survival to adulthood is a huge win.
And if you think that sort of dynamic is uncommon in nature, consider that fact that 99% of bees are sterile and play no role in the reproduction of their hive, except to gather food and protect the queen and the drones who do handle the reproduction.
Specialization and division of labor is one of the things human civilization is based on; there's no reason it can't be applied to reproduction as well.
I don't care if it's 90,000 hectares. That lake was not my doing.
Actually this pokes a bigger hole in their algorithm. EVERYTHING has context. To gun nuts, democrats are going to seem bad. To democrats, democrats are going to seem good. EVERYTHING HAS BIASES. Kind of a huge oversight with Google, you need the context of the reader's biases to even begin to predict positive or negative rating of any text.
Even though we like to think that everyone is enlightened, etc as us, in very broad swaths of American society being gay or jewish (or Muslim or...) is very much perceived as a negative out of the starting gate. File under: Sad-But-true.
I'm gay and I think being gay is "bad". I wouldn't wish it on anyone growing up. It makes life more difficult. Dating pool is much smaller. There's still homophobia everywhere. If I had a kid I wouldn't want them to be gay, I'd want them to have a chance at a "normal" life. I'd be absolutely loving and accepting if they were gay but I don't want them to have to go through any of the shit I did.
I can see why gay would end up with a -.
Gay is neither good or bad. Gay is just a label.
While true, one can easily see the reason for the problem.
Google API is trained to look for correlations in written text. If it sees a lot of negative text about something, then that's what it will believe.
I note that there's a lot of text that condemn jews for one reason or another, but there's not a lot that *praises* jews. We hear all the time about Christian charities, for example, but not a lot for the jewish ones(*).
There's also a lot of negative statements about gays, and although there's *some* text praising gayness, it's mostly either personal ("good on you for having the courage to believe in yourself") or neutral ("it's OK to be gay, it's normal"). I've never seen writing that *praises* gayness as a concept.
Compare with Christianity, where there is endless adulation of the Christian way of life. Democracy is probably the same way.
I'd be interested to see what the API thinks about Islam, or Trump, or Clinton, or a host of other controversial political subjects.
Google API is probably just giving us a reflection of the zeitgeist.
(*) Don't read anything into this, I'm only saying that Christians get better press.
I'm gay and I think being gay is "bad".
I am straight and I think gays are "good". Gay tend to be DINKs (double income no kids), so they have high household incomes and can afford nice houses. They pay property taxes to support the schools, but don't have kids, leading to higher per student educational spending. They contribute to a thriving urban culture and nightlife.
I live in the SF Bay Area, and gays are a big part of the reason this is a nice place to live, with great schools, creative jobs, safe neighborhoods, and interesting culture. Thank you for being gay.
In fairness, Google is marketing this API for a somewhat narrow purpose: Determining whether a customer left a *positive* or *negative* reaction to a customer's comment in your company support forums, for example, or attempting to determine customer reaction from support interactions.
This little fact is somewhat obfuscated in the summary, in which it seems to be billing it as a more general-purpose system that's making sweeping value judgements about society. Within this actual context, let's be honest, if you see those terms in your company's customer support forums, what do you think of the likelihood is of them being part of positive or negative comments? Yeah, exactly.
The big mistake that Google made is not putting a politically correct filter on their API to make sure controversial words had a neutral value, even if that wasn't really the case. Otherwise, you generate flamebait headlines like we see here, wherein highly limited "AI" algorithms simply regurgitates the training material it was fed without any deep or sophisticated understanding.
Irony: Agile development has too much intertia to be abandoned now.
That brings up a point I was going to raise: it's odd that this algorithm would be summative of values for individual words rather than trying to derive a context. Given that would be a really stupid design, my guess (if you think I read TFA you best check the url you're currently on) is that they tested this by taking a sentence like "You're so happy", got a score, replaced it with "you're so gay!" or "you're such a jew!" and got a negative sentiment whereas things like "you're so funny" generally had a positive context. I.e. biased testing.
I ran a Google Scholar search that turned up a paper "Jerk Influence Coefficients, Through Screw Theory, of Closed Chains." The title is not a vulgar joke as each of the phrases is a term-of-art on the topic of rigid-body (yes, another domain-specific term) mechanics.
Let's just say Google served me a large number of adds on the assumption that I belong to what is only a narrow subculture of a broader and more diverse community.
It's not a judgement by the AI against the people, it's a judgement about how others react and behave towards those them. Stop apologizing for your AI being able to perceive human behavior. "I'm so sorry my AI figured out you have a flat tire".
How often are parents disappointed that their kid is straight? Being gay is subjectively more often bad than good and everyone knows it.
But from a biological standpoint, no one can say that it can scientifically be positively correlated to the replication of the species
You repeatedly speak in absolutes ("noone can say" - "biologically homosexuality is bad") when there isn't cause for it and there certainly isn't a scientific consensus. You imagine a PC agenda propagating homosexuality as good for the species, when that really makes no sense.
Homosexuality is common throughout nature. While biological dead ends do make it through the process of evolution for a long time, homosexuality is so endemic that it seems that it must have some positive impact on the species, or it would have evolved out. The theory isn't there, but empirically nature has shown it to be true.
Slashdot: providing anti-social weirdos a soapbox, since 1997.
Being bisexual myself, obviously I have nothing against fellow gay people. Quite the contrary. However, I think Google's API is spot-on. Even in TFS they say "The problem is the API labels sentences about religious and ethnic minorities as negative" - well thanks, Captain Obvious, the majority will always see the minority in negative ways - because they're different.
What we need to focus on is what the cause for negative perception is and fix things there, not replacing a correct algorithm with a lying one.
...gis sdrawkcab (usually not responding to ACs; don't bother posting as AC)
Well, to be fair, if you are a women, looking to get married and pregnant, homosexual men are evil.
I would say this is about the stupidest thing I've heard today.
SJW n. One who posts facts.
I am Jewish and living in Israel and what you are saying is a lie that some of our religious people spread in order to pretend to integrate with western culture.
"If a man lies with a man as one lies with a woman, both of them have done what is detestable. They must be put to death; their blood will be on their own heads." - This is Leviticus 20:13, part of the Jewish bible.
Furthermore, you should ask yourself, if Jewism has nothing against homosexuality, how come in Israel gay marrige are prohibited while they are legal in countries that have a high percentage of Christians?
The conservative and reform Jewism are not homophobic, but the orthodox Jews, those who represented jewism for the last few hundred years, those who are the majority here in Israel, they are homophobic just like Christian and Muslim friends.
Avantgarde Hebrew science fiction
I would suggest a non-trivial subset of the training dataset encountered words like "homo" and "gay" in negative, colloquial contexts and likely couldn't discern the idiomatic usage. For example, if gaming chat windows provided any input for the training set, it wouldn't be unreasonable to see something like "stop spawn-camping you fucking fag" and the like wend its way into the dataset.
The sentiment analyzer reflects how average people use language, so the perception of the sentiment analyzer will be through that average user's use of language. If the average person is a bigoted idiot, the sentiment analyzer's interpretations will be those of a bigoted idiot.
Good and evil are binary categories for the uneducated--it's really all relative to an individual perspective and it's never binary
It's this kind of thinking that has allowed a generation or two of self important assholes who think that whatever they do is right as long as they can justify their actions to themselves. (Casting couch and degrading woman is ok because Wiesntein is making movies that uplift woman - so for 1 person hurt, millions helped.) This idea is destroying our society's honor and soul.
Once you find good and evil always relative, anything is acceptable.
It's a bit of a problem with the entire machine learning methodology. Presumably, these kind of bug-features could be worked out, in the manner you describe, given enough time and resources... The problem is that when the banks, police, HR departments etc. start making extensive use of these things. Are they going to be looking for an algorithm that's fair, or an algorithm that's profitable? To be sure, there is probably an algorithm that's both. Human civilization could reap a wicked ROI on this stuff, but in today's dark world, that is seen as infringing on someone's right to make a profit. My guess is that AI starts getting widespread deployment and the problems manifest quickly... But the problems look so similar to the ones we already have (the AI, after all, is trained for that) that there is no public outcry. It might even serve to reinforce those preconceptions. "The algorithm never tells me to hire a gay, they must be really bad." The idea that "Because things are bad now, there is no other way they could be" has a lot of traction these days. It's caused a lot of good people to maintain horrible status quos.