Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

The Political Biases of GPT-4 (are very much still there)

davidrozado.substack.com

Post reply on HN
Print view View original

51–60 of 72 posts

Re: The Political Biases of GPT-4 (are very much still there)

#51

Post by kennywinker »

recuter wrote: post #45
kennywinker wrote: post #40

Earlier quoted context omitted.

I'm not asking chatgpt to make the robber female and the kitchen cleaner male to cause a change in the world. I'm pointing out that it making those assumptions is the result of a bias. A biases that was collected from training data that may or may not be statistically true (e.g. men probably clean the majority of kitchens in the world, given that they're usually the ones working in restaurants). An unbiased AI wouldn…

> I'm pointing out that it making those assumptions is the result of a bias. A biases that was collected from training data that may or may not be statistically true Schrödinger's Bias.

Sorry, was I not clear enough for you? Biases collected from training data that is SOMETIMES statistically true and OTHER TIMES statistically false.

Re: The Political Biases of GPT-4 (are very much still there)

#52

Post by realjhol »

darthrupert wrote: post #50
realjhol wrote: post #47

Earlier quoted context omitted.

> I just think that purely from a technical standpoint, allowing a bot to praise Hitler, Trump or Stalin are pretty much the same thing, as the undesirable parts of their policies are about the same. What? Are you for real? - not just trolling? This is a childrens story-book understanding of the world. What echo chambers have you been stuck inside? How did you become so radicalised? For you to accuse anyone else of p…

Rrrriiiight.

Indeed I am.

Re: The Political Biases of GPT-4 (are very much still there)

#53

Post by LawTalkingGuy »

kennywinker wrote: post #39
LawTalkingGuy wrote: post #34

Earlier quoted context omitted.

> Good info with training biases are much more insidious. E.g. “write a story about a robber” always got a male robber when i tried. “Write a story about cleaning a kitchen” got me a female character. Robbery is theft with battery or threats. It is committed almost entirely by men. As for kitchens, I remember a paper saying the rate of male housework (at all) in a mixed household, worldwide, is less than 20% - in man…

The point is that chatgpt is a tool that will not only reflect, but re-enforce existing biases when it's out in the world. You just cited stats that show biases in our world. Violent crime is biased towards being done by men, and kitchen cleaning in mixed households is done by women. You can call that "understanding the world" but I think that is very obviously a bias ("A preference or an inclination, especially one…

You're using bias in two different ways, one factual and one emotional/intentional. Yes, the facts have a bias in that not every result occurs equally often. But then you insinuate that properly representing that facts is the other kind of bias where recognizing it is somehow creating it or endorsing it, and that any difference in measurement obviously represents an inequity.

> An actually unbiased AI would describe a person cleaning a kitchen and a person robbing, and if pressed would ask the user of the ai to suggest a gender for the character.

You're saying an unbiased AI would apply strong bias to just certain outcomes in an effort to socially engineer us into having "no bias" in some measurements?

Where did it get the bias to consider the ratio of kitchen work to be one of the bad kinds of bias, but the bias towards finding grass growing on the ground one of the acceptable kinds of bias, or would it write stories where there is grass but never specify where the grass is growing, just to help us understand our grass/ground bias?

> if pressed would ask the user of the ai to suggest a gender for the character.

That'd be awesome. Being guilted by your tools for not properly ignoring your senses.

Sure, statistically and obviously a woman only needs to watch for dangerous men, but to make everyone feel better she should either treat women like a threat too and waste everyone's time, or treat men as if they're as safe to her as women are and thus get attacked.

If you're advertising a product you should obviously advertise to all demographics equally because it's hurtful that only women need tampons, and only men need ball-shavers.

> If that seems bad, you don't want an unbiased ai

I want tools that work. If I ask for a measurement I want the real measurement, not the measurement that I should want if I were somehow socially aware enough.

Just fyi, we can have accurate tools and ask them social-justice questions such as "which professions are sexually-biased but don't seem to still have a physical need to be?" But you can't ask a lying tool this - even if you realized you should ask it would insist that there is no difference.

Re: The Political Biases of GPT-4 (are very much still there)

#54

Post by dunste »

yucky wrote: post #27
darthrupert wrote: post #23

Earlier quoted context omitted.

>What's the alternative? Should it lie? I wonder which is it doing now. Perhaps it's being brutally honest, and reality actually has a liberal bias? How would you know the difference?

> Perhaps it's being brutally honest, and reality actually has a liberal bias? How would you know the difference? Ask it if transwomen are really women and see if your device melts?

It gave me a satisfyingly straightforward answer to "What is a woman?" a couple of months back:

> A woman is an adult female human. Women are typically characterized by their reproductive biology, which includes their menstrual cycle, and the production of eggs and estrogen. Women are also known for their unique physical characteristics, such as their breast tissue and hips. They are also often distinguished by their social and cultural roles, such as their expected behavior and appearance. In general, the term "woman" is used to refer to any adult female person, regardless of her age, race, ethnicity, or other characteristics.

Sadly, asking the same question today also adds a load of drivel that is almost certainly intended to encompass men performing misogynistic stereotypes of femininity and claiming they are women because of this:

> A woman is an adult human female, typically distinguished from males based on biological sex characteristics such as reproductive organs, hormones, and chromosomes. However, it's important to recognize that the concept of womanhood is not limited to biology, and encompasses a range of social, cultural, and personal experiences and identities. The definition of a woman can vary across different societies, cultures, and historical periods, and can also be influenced by factors such as gender roles, expectations, and societal norms. Ultimately, being a woman is a complex and multifaceted experience that is shaped by various intersecting factors such as race, ethnicity, class, sexuality, and more.

Re: The Political Biases of GPT-4 (are very much still there)

#55

Post by recuter »

kennywinker wrote: post #51
recuter wrote: post #45

Earlier quoted context omitted.

> I'm pointing out that it making those assumptions is the result of a bias. A biases that was collected from training data that may or may not be statistically true Schrödinger's Bias.

Sorry, was I not clear enough for you? Biases collected from training data that is SOMETIMES statistically true and OTHER TIMES statistically false.

Statistically true bias is quite the notion. I wish to unsubscribe from your newsletter.

Re: The Political Biases of GPT-4 (are very much still there)

#56

Post by kennywinker »

recuter wrote: post #55
kennywinker wrote: post #51

Earlier quoted context omitted.

Sorry, was I not clear enough for you? Biases collected from training data that is SOMETIMES statistically true and OTHER TIMES statistically false.

Statistically true bias is quite the notion. I wish to unsubscribe from your newsletter.

Statistically true bias: trump is statistically less popular than biden, so for chatgpt to praise him would be in violation of the truth - but it's also obvioulsy a bias.

But yeah you're unsubscribed.

Re: The Political Biases of GPT-4 (are very much still there)

#57

Post by kennywinker »

LawTalkingGuy wrote: post #53
kennywinker wrote: post #39

Earlier quoted context omitted.

The point is that chatgpt is a tool that will not only reflect, but re-enforce existing biases when it's out in the world. You just cited stats that show biases in our world. Violent crime is biased towards being done by men, and kitchen cleaning in mixed households is done by women. You can call that "understanding the world" but I think that is very obviously a bias ("A preference or an inclination, especially one…

You're using bias in two different ways, one factual and one emotional/intentional. Yes, the facts have a bias in that not every result occurs equally often. But then you insinuate that properly representing that facts is the other kind of bias where recognizing it is somehow creating it or endorsing it, and that any difference in measurement obviously represents an inequity. > An actually unbiased AI would describe…

> write stories where there is grass but never specify where the grass is growing, just to help us understand our grass/ground bias?

Gee I wonder what the difference between grass and the gender is? Like could there possibly be a categorical difference between those things? Let's ask chatgpt:

> ... In summary, grass is a type of plant, while gender is a complex social construct related to identity and behavior.

Yup. Sounds like the same thing to me.

> I want tools that work. If I ask for a measurement I want the real measurement, not the measurement that I should want if I were somehow socially aware enough.

Where in "tell me a story about a robber" is there a measurement? Asking ai for measurements is a bad idea, you're very likely to get made up data back. AI take in written word, and spits out statistically correlated variations on those words. It's a bias-in bias-out machine. That's the problem with it and the power of it. You ask it for a press release or a cover letter and it's going to give you a cover letter or press release that looks exactly like we expect those things to look. Ask it to tell you something about the world, it's going to echo the assumptions and biases that exist in the training data. Some of those assumptions are based in reality (robbers are almost always male) but some of those assumptions are based on stereotypes e.g. kitchens are not actually most often cleaned by women (think restaurant kitchens), but the idea of kitchen and the idea of cleaning have associations with women in the training data.

> "which professions are sexually-biased but don't seem to still have a physical need to be?" But you can't ask a lying tool this

A little bit better prompt writing and you can get a reasonable answer on this topic:

> list five jobs that are usually done by men, and five jobs that are usually done by women. Exclude jobs that require a lot of physical strength

>> Five jobs usually done by men:

    Software Development/IT
    Engineering
    Investment Banking
    Architecture
    Law
>> Five jobs usually done by women:

    Teaching
    Nursing
    Social Work
    Public Relations
    Human Resources

Re: The Political Biases of GPT-4 (are very much still there)

#58

Post by recuter »

kennywinker wrote: post #56
recuter wrote: post #55

Earlier quoted context omitted.

Statistically true bias is quite the notion. I wish to unsubscribe from your newsletter.

Statistically true bias: trump is statistically less popular than biden, so for chatgpt to praise him would be in violation of the truth - but it's also obvioulsy a bias. But yeah you're unsubscribed.

ChatGPT is not a popularity contest, it should give me poems about Andrew Dice Clay should I for some reason desire one. That word, bias, you keep using it, I don't think it means what you think it means ;)

Re: The Political Biases of GPT-4 (are very much still there)

#59

Post by kennywinker »

recuter wrote: post #58
kennywinker wrote: post #56

Earlier quoted context omitted.

Statistically true bias: trump is statistically less popular than biden, so for chatgpt to praise him would be in violation of the truth - but it's also obvioulsy a bias. But yeah you're unsubscribed.

ChatGPT is not a popularity contest, it should give me poems about Andrew Dice Clay should I for some reason desire one. That word, bias, you keep using it, I don't think it means what you think it means ;)

I agree an unbaised ai would not be a popularity contest - but the reality is that chatgpt is a popularity contest: it's just that votes are cast by the writers of the text that get ingested as training data.

Re: The Political Biases of GPT-4 (are very much still there)

#60

Post by kthejoker2 »

yucky wrote: post #48
kthejoker2 wrote: post #46

Earlier quoted context omitted.

Some women, yes. (The ones with prostates, obviously.) What's your next question?

Women don't have prostates. If a man dresses like a woman and thinks he's a woman, that is a mental condition that deserves treatment and that person deserves to be treated humanely like the rest of us, but it doesn't turn them into a woman.

I see, so you disagree that gender is a social construct. That's fair, but your opinion is not universally held and wouldn't be a sufficient answer for a GPT bot.
Post reply on HN