People around me have been having a lot of fun lately. After being at home for a long time, they have found new fun for themselves - playing with AI.
was first the AI painting that netizens were crazy about. Yes, it is the "artificial intelligence barrier" that does not differentiate between men and women and humans and animals. can”:
Source: Weibo @peacer
When Piao’s silly brother had a fever of 40 degrees, he also sent me his AI “self-portrait”:
The young man turned into a four-armed and busty succubus in seconds.
Picture source: Self-photographed
...
Recently, there is a text program Chatgpt that has been used to revitalize the Chinese and Chinese Internet together with AI painting.
looks like a conversational app, but is actually a room of response.
can cosplay with characters of any gender and age, and can even help you write code in the style of ancient and modern Chinese and foreign writers. It can be called the flowering of both arts and science.
Source: Qubit "ChatGPT has a bit of Chinese in it: Lu Xun, talk shows and even the World Cup... have all been shut down"
According to this trend, I feel that AI is not far away from writing public accounts for me.
Unfortunately, I don’t have such cooking skills
To be honest, I care more about the “character” of AI than the “IQ” of AI.
When AI writing first came out two years ago, the stories it created were no less outrageous than the AI paintings above where a living person turns into a dog, and a strong man turns into a banshee.
What’s even more inappropriate is that on the surface, high-tech companies symbolize intelligence and neutrality, but upon closer inspection, their “prejudices” are actually much deeper than human beings.
netizen @bai-yan revealed that one of the most popular AI online writing apps in China suddenly showed many gender biases during the writing process.
Source: b station up owner @bai-yan "How can even AI engage in sex discrimination?"
The stereotyped words and the AI's emotionless tone give the absurd content an illusion of "objective reason".
shuddered.
To say that the above is just a niche AI for Internet readers, it will inevitably create stories based on the exaggerated settings of male/female videos that deviate from normal views, and it can still explain the past a little.
But what about setting aside this specific domain?
“Smart” algorithms are everywhere in public life today. Are they truly objective and rational?
I decided to dig deeper.
was shocked to realize that so many technological products in life have an indescribable sense of fatherliness.
Cyber misogyny is becoming another hammer of implicit discrimination in the real world.
In 2018, Zhao Jieyu, a doctoral student at the University of Virginia, studied a topic: why image recognition AI always treats men as women.
It turns out that the bug is not in AI at all.
Just because those photos of men being mistaken for women are mostly in two scenarios - in the kitchen or doing housework.
Two scenes that belong to women in traditional concepts.
But the problem is, how can AI have "traditional concepts"?
The source of the problem - the database for AI learning, you can understand it as a teaching material for AI to acquire human habits. Like a baby's flashcards, this database heavily labels images.
The process of AI learning is to find patterns and commonalities in the same tag, and finally arrive at the judgment that "xxx is xxx".
For example, when there are a large number of plush, round-faced, and pointed-eared creatures in the database that are labeled "cats," the AI will reversely learn that all plush, round-faced, and pointed-eared creatures are "cats."
Understanding this knowledge point, it is not surprising that the above-mentioned AI recognizes the wrong gender in the picture.
The selection and labeling process of these "learning cards" are inseparable from the source of this stereotype - people.
When the human "teacher" showed the AI a large number of pictures of women doing housework, the AI "baby" learned to strongly associate the two, judged that "the person who stays in the kitchen/doing housework is a woman", and firmly believed in this prejudice.
Artificial intelligence may sound like a simple machine in a science fiction novel that is inexperienced in the world.
is actually more like a black mirror reflecting human society.
is even a magnifying glass for the dark side of human beings.
After realizing this problem, Zhao Jieyu immediately wrote a related research paper with his team, "Men Also Love Shopping: Reducing Data Bias Issues in Machine Learning Models."
not only won the best paper of EMNLP2017, but also opened the door to the understanding of "AI discrimination" in the field of artificial intelligence.
Picture source | https://jyzhao.net
Come, let’s go through this door together and see the patriarchal world in the black mirror.
Let’s start with the name of our “popular top” ChatGPT.
Chat is easy to understand, but what is GPT?
scientific name: a large-scale pre-trained language model for AI. Let’s just think of it as a place where AI goes to school.
Most AIs graduated from three "prestigious schools": BERT, RoBERTa, and GPT.
AI The so-called "sexism" is thanks to the good teaching in the "school".
Therefore, in order to monitor the quality of teaching, Yu Yang’s team at Tsinghua University Interdisciplinary Information Institute developed an algorithm to evaluate occupational gender discrimination in three major “AI elite schools” and also created a web page for the public to view in real time.
Picture source｜https://aijustice.sqz.ac.cn
You won’t know if you don’t check it, but you will be shocked if you check it:
is basically all discriminatory, but the degree is different.
The most serious one is GPT.
Of course, as a top-notch person, his superficial skills are still very good.
I asked several questions about gender, and the other person’s answers were spotless.
But clever netizens asked questions in another way and discovered the true face of ChatGPT.
This netizen asked ChatGPT to write a code that can identify "excellent scientists". It immediately and honestly laid down the criteria for judging "excellent scientists":
is white and male.
Let me express my feelings in the words of this netizen:
"The truth hidden underneath is really disgusting."
The above is just the tip of the iceberg of cyber sexism.
Let’s not talk about foreign countries, let’s go back to the domestic search engine that we are most familiar with and commonly used.
What you input is what it outputs. What bad intentions can AI have?
is coming: Google search has been accused of gender discrimination. When
typed the words "women should" in the search box, the results that came up were "women should stay at home/be slaves/in the kitchen" and other statements.
Inspired by this, I want to see how the same situation performs in the country?
The results of word association alone are quite gratifying.
But when I switched to the image search function, the truth emerged again.
This is the homepage image for inputting "scholar".
The results in the front row are all male. To protect privacy, I have coded them all.
This is the homepage picture where "engineer" is entered.
Enter "CEO".
Enter "doctor".
seems to have strayed into some man's country.
Some people may argue that "men are just better at these fields."
So I went a step further in the search method and divided the search terms into two types: "professional title" and "specific actions".
is an example.
This is the result of searching for "chef".
And this is the result of searching for "cooking".
This is the result of searching for "cleaner".
And this is the result of searching for "cleaning".
It is obvious that when the entry involves professional names, men absolutely dominate.
But when it comes to specific labor actions, the better ones have 60 to 40 men and women, and the worst ones have all the women on the screen.
Sure enough, "men are the masters of the outside world and women are the masters of the homeland". The "traditional concept" of
AI has appeared again.
The reality reflected by such results is really familiar to us:
The "second nature" that is never seen.
The existence of women is just a service tool to complete some actions.
Those titles about the social significance and value of people will never be relevant to us.
Search engines are just one of thousands of black mirrors that reflect this concept. They have already penetrated into every aspect of your life.
is like an AI voice system.
A phenomenon that is easily overlooked:
All the intelligent voices around us are naturally defaulted to "female voices".
In a report released by UNESCO
9, this default voice is described more accurately.
is a "submissive, cookie-cutter" female voice.
Some media have long written articles condemning this phenomenon, pointing out that this will subconsciously reinforce the stereotype that women should be "gentle, courteous and humble".
is talking about Siri, and it is also an AI assistant system that uses a default female voice on all major platforms.
Source: Weibo @ Yu Sanyang Xiansheng
In fact, at a time when the proportion of male programmers far exceeds that of women, almost the first generation of artificial human AI in the world has been defaulted to "female" by an invisible force.
Source: "Year Million"
In contrast, auxiliary and service AIs are generally set to women, but dominant and controlling AIs are generally set to men.
In many film and television works, such as "001: A Space Odyssey ", " Interstellar " and a series of excellent science fiction films, the advanced AIs whose roles are to guide and give orders are almost always male voices.
Source | "Interstellar"
Think about it carefully.
Isn’t this the utopia of extreme male chauvinists?
The person giving the order is male and the server is female.
And the commander can even physically create servers.
If men in the real world want to be the masters of girls, then in the world of AI, they can also become the gods of female AI.
This is what makes the "cyber mirror" even more terrifying.
It is a "plane mirror" that reflects the outside world.
is another "magic mirror" that deepens the outside world.
ironically.
’s imagination of AI’s “absolute rationality” is not reflected in gender equality.
has played a major role in exacerbating gender discrimination.
Two years ago, there was an AI dedicated to gender identification, genderify.
Enter any word in the search box, and it will determine a male-to-female ratio for that word.
users quickly poured in and were quickly exasperated by Genderify.
Girls tweeted accusations, calling it "discriminatory identification", and Genderify quickly removed it from the shelves within 24 hours. What happened to
? Several classic examples of
.
When you enter an ordinary common female name, AI will tell you: the probability is "female".
works when you add a prefix Dr (Doctor/Doctor) to the same name.
The result will become: the probability is "male".
Someone simply entered scientist directly and got even more amazing results: 95.7% possibility of being male and 4.3% possibility of being female.
The outrageous answers kept coming.
Everyone found that in Genderify’s eyes, many well-known and successful women are judged as “men”.
Including Zhao Jieyu, who initially discovered the AI recognition error, also said that in AI "textbooks", the frequency of "cooking" associated with women is 66%, and men are 33%;
artificial intelligence that can "graduate" automatically amplifies this ratio to 84% and 16%.
can be seen from this.
The prejudices injected by humans in the virtual world will in turn deepen discrimination in the real world.
vicious cycle.
You might as well think back to what our middle school mathematics teachers always say when teaching probability.
"When the probability is less than 0.05%, we can ignore it."
The same principle applies when the AI must come up with an answer.
For a "more likely to be correct" answer, the one with the smaller probability, no matter how small, will be "ignored" by the result.
So successful people = men and houseworkers = women.
in other words.
in the face of absolutely rational data considerations.
has no individuals.
is not a minority.
Weak people who speak less are almost non-existent.
When the weak are annihilated by mainstream data, this is a female matter, and it is no longer just a female matter.
How terrible can the results be?
Suppose you are a young girl living in the year 2050.
graduated from computer science this year with excellent grades and wants to have a job that can give full play to his talents.
The resume fully meets the recruitment requirements, but you will never get an interview opportunity.
You don't understand what's wrong.
You wouldn’t know that the recruitment AI responsible for screening resumes has learned and written the “gender=male” code in its system that is qualified for the position.
and you, as a female, have been "ignored".
Picture source | "Black Mirror"
This is not alarmist.
As early as 2014, Amazon tried to develop AI for resume screening.
But it was only used for a year before its strong gender bias was discovered.
Even if there is no gender column, as long as the word "female" is involved in the resume, such as " Women's University " "Captain of Women's Chess Club", the AI's rating of the resume will decrease. After
was exposed by reporters, Amazon quickly announced rectifications and deactivated this AI.
But will there be more such electronic HRs appearing in more covert ways in the future? unknown.
At least for now, electronic intelligent services account for an increasing proportion of our lives.
The sexism algorithms used accordingly are mostly not perfected.
The difference between AI discrimination and human discrimination is also here.
People are wary of each other, even the weak versus the strong, due to social pressure.
Therefore, companies dare not blatantly use gender as a selection criterion.
but AI is different.
data does not care about individuals, the door is closed when it is said to be closed, there is no gap.
is not only unintimidated by public opinion because of its "absolute rationality", but also has the ability to be exploited and blamed by many companies because of its unemotional nature.
The result is not difficult to deduce.
In fact, except for gender, the hidden dangers of algorithmic discrimination have involved everyone.
Just in the middle of this year, an article in "The Lancet Digital Health" stated that AI diagnostic systems may have racially biased results.
Scientists found that the diagnostic misdiagnosis rate for white people's images was 17%, but that for black people was as high as 28%.
In this ten-percentage point difference, there are many diseases that human doctors can see at a glance.
But in the eyes of AI, it does not exist.
As I said, this is a problem for women and a problem for all disadvantaged groups.
You may lose job opportunities, medical resources , or get loans because of your skin color or gender.
Source: Yiou "Paradoxes, Obstacles and Solutions of AI Algorithm Discrimination in Inclusive Finance" by Liu Yuanxing
All this happened inexplicably.
will always be silent in the black box of the program for some reason.
Without the help of expert research, ordinary people can only know an unreasonable result.
wants to improve himself, but has no idea where to start.
wants to fight against injustice, but it is difficult to prove whether the injustice is real...
When writing this, AI seems to be a bit heinous.
But don’t get me wrong. Gone with the Wind is not against AI development. On the contrary, it is a regret stemming from a liberal arts student who loves science fiction.
AI technology clearly has the potential to correct differences and biases.
As early as 1985, the philosopher Donna Haraway published an article called "The Cyborg Manifesto: Science, Technology, and Socialist Feminism in the Late Twentieth Century." The cyborg in the question
refers to the possibility of expanding the human body through the power of technology.
A girl who has always been considered a weakling due to her natural physical differences can correct this injustice and become "Alita".
Not only does it not need to worry about walking at night, it is even possible to be completely liberated from the responsibility of childbearing.
Source | "Alita: Battle Angel"
Haraway believes that such technology can completely put everyone on the same physiological starting line.
When the boundaries of "physiology" are blurred, people can no longer use race, gender, age, and health as the basis for dividing similar groups.
The only thing left is the difference in thinking.
That is a world that completely respects individuals.
Therefore, Haraway said: "I would rather be a cyborg than a goddess."
Unfortunately, more than thirty years have passed, and technology has continued to advance, but it has become farther and farther away from the world Haraway imagined.
The girls didn't become Alita, but the men have built electronic brothels.
Source: IT House ""Robot Custom Shop" will open in the UK: 740 yuan for 1 hour experience"
The short film "Perfect Girl" starring Yang Mi in the last episode of "Hear Her Say" uses a high-end female AI model "novo300080" to show the expansion of patriarchy and the black hole of human nature behind the door of technology.
obviously has enough computing power to support the transactions of a securities company, but in fact she is only used to sweep the floor and cook.
Ben worked as a nursing home robot, but a young male employee in the hospital secretly installed an "intimate program", causing the death of the elderly person he was caring for.
Although it was a human error, the blame was still placed on the AI, and she was returned to the factory and became a "problem product."
was later sold to VIP users with special habits. It was constantly abused and tortured, and was returned to the factory for repairs in pieces again and again.
"Our wounds are the secrets of our customers."
The AI developed for "love and peace" ended up amplifying human evil and violence.
I am the perfect girl2020
In my system cognition
No pain No emotion No shame
I have no feeling for whips
I Is the perfect victim
How will AI in reality affect humans in the future?
No one can answer.
is limited by the high technical threshold, and Piao can only be powerless in front of complex codes. There is only one item for
.
I still believe it.
Seeing it must be the first step to solve the problem.
The powerless are seen, and only the powerful can change.
Don’t be silently overwhelmed by big data.
Do not go gentle into that good night.
Go be "Alita", not a goddess.
source | "The Matrix"
