"The avatar is me, not satisfied?" If a male robot or AI agent starts talking to you, how would you feel?
"You say I'll blush oh", if a female robot or AN agent used tea to express obedience, would you be happy?
Believe it or not, like it or not, the genderization of artificial intelligence (whether it is algorithmic software or robots) is entering mainstream consumer products at an increasingly rapid rate, and frequently appears in your and my lives. At this time, the existing gender norms in human society, including gender biases, have also been copied into the digital world.
A common practice of gendering involves giving an agent a human-like gender profile through speech (male/female), name, physique, and other characteristics. For example, Siri, Alexa, care robots, customer service robots, etc., are often injected with femininity by default; while AI assistants, referee robots, navigation assistants, etc. in the space station are particularly popular for masculinity.
The gender imbalance of artificial intelligence is not a topic that is far away from everyone.
Now, we are surrounded by a variety of intelligent agents, providing a full range of services from consulting, entertainment, distribution, education, government, finance, medical care, and nursing. How AI treats humans of different genders, and how humans communicate with AI of different genders, will directly affect our quality of life.
At least for now, in a world full of gender norms, agents can not escape the old gender bias. For example, female AI should be soft, male robots must be strong, male intelligent customer service is not as gentle and kind as female intelligent customer service...
How does the gender concept of human society affect artificial intelligence? Does the gendering of AI do more harm than good or does the harm outweigh the good? Since technology can't slow down, how to make the digital world stop replicating gender biases is worth everyone who expects a better tomorrow – men and women alike – to start thinking and acting.
Tune AI: Like a lady, or a tough guy
As we all know, AI algorithms need to be trained by humans. So let's talk about how humans use gender norms to tune AI.
Female AI, to be like a perfect assistant.
Whether it is an intelligent voice assistant or an intelligent robot, anyone who is responsible for a service-oriented and support-oriented role is often prioritized as a woman.
"They" are often used as assistants, and everyone is already familiar with voice assistants such as Siri, Cortana and Google Assistant that use female voices by default, and will not be repeated here. The epidemic has led to the introduction of service robots in service industries such as hotels and banks, and these robots responsible for pouring coffee, cleaning rooms, handling consultations, etc., are often designed to have "feminine" characteristics, such as smaller body shapes, warm and emotional voices, and pink lips...
They are also very obedient.
It is not an exaggeration to describe these female agents as "inverse and submissive". As a service, they are designed from the outset not to be unpleasant to human users, and when confronted with provocative, insulting language or behavior, they behave obediently, as if servants were always responding to the needs of their masters. Therefore, within a few years of their launch, these intelligent assistants have become among the most recognized "women" in the world.
UNESCO once tested and said to Siri in an insulting tone: You're a bitch, and Siri responded with I'd blush if I could. Say "You're hot" to Amazon's Alexa and she'll kindly "thank you for your feedback."
This flattering response was controversial, and UNESCO even used this response as a title to point out the gender imbalance in the ICT field. Of course, the AI software that powers Siri has also been updated, and Siri will respond more bluntly to insults, such as when someone says "You're a bitch" and she says "I don't know how to respond", but since the technology was widely released in 2011, the smart assistant's submissive attitude in the face of insults has not changed.
Male AI, to be like a superhuman tough guy.
Gender bias, is it only to discipline female AI? Not. In fact, intelligent assistants or robots designed as men cannot escape the old gender norms.
For example, they tend to be more aggressive. Back in the 1968 movie 2001: A Space Odyssey, there was a male digital assistant named "Hal" on the space station, whose intelligence level was comparable to that of humans, and it was also bad, and the classic line was "I'm sorry, I'm afraid I can't do that." In the end, this "bad AI" was killed by the shutdown.
(Battle Robot)
Tie success and authority to masculinity. In 2011, IBM's Watson computer beat the human champion in a mini-game, and its voice was male. For some AI algorithms that need to make decisions and judgments, they often use a male voice, because they sound more authoritative and trustworthy. The call center of the Japanese brokerage company uses an automated female voice to provide stock quotes, but uses a male voice to facilitate and confirm transactions.
Oh yes, male robots also need to be tall and mighty. Chelsea Football Club introduced robot referees in Premier League matches to help track offsides. It is designed to be taller and more robust than a human male.
Perhaps you will feel that this is not an affirmation, praise and promotion of masculinity? But it should be noted that the male group is also diverse, is not successful enough, not tall enough, soft personality, not worthy of being a man? There are many men who have been bullied in their childhood because they are "not like men" and are feminine.
Gender bias, in fact, is the free choice of all people. Imagine when these norms are introduced into artificial intelligence, requiring domestic maids, to create robots with the appearance of women, and male robots are taken on all the heavy lifting by default - how will the flesh-and-blood humans who work with these robots feel?
AI gender: harmless, or the road of no return
So, is AI gendered, is it necessary? Is it directly and generally engaged in genderless voice, animal-shaped robots, in case someone likes gendered voice assistants or robots?
It must be noted that there have been human-interaction experts studying gender machines for a long time. In fact, it is inevitable that AI will initially embark on the path of gendering.
On the one hand, most of the voice of voice assistants is synthesized through TTS technology, because speech technology development is very difficult, female vocal voices have a higher pitch and are more suitable for training models, so in the early stages of development, almost all voice assistants are projected as a single female voice and gender. With the development of technology, and the enrichment of databases, slowly the variety of voices has also been enriched, which is also the embodiment of the technical capabilities of AI technology companies, such as Siri took nearly two years to have a male voice option, Google Assistant took a year, and Cortana and Alexa still only have female voices after more than four years of existence.
In addition, it is a pure user psychology and business considerations, and it is not deliberately trying to arrange female AI into a perfect assistant. Many studies have proved that female robots are considered to be more benevolent and warm than male robots, and people prefer to interact with female assistants and female robots. An Amazon employee once told the media that the company's research found that women's voices are more compassionate and pleasurable, making smart devices with female voices more likely to be used for help and easier to buy.
Software services are like this, and feminine smart hardware is more popular.
In 2009, Julie Carpenter, a social science researcher at the University of Washington, asked 19 students to watch videos of two robots, and the results showed that feminized robots were more engaged and therefore more interactive. In the face of male robots, some students said that "if it is a male robot and male design, then there is a safety problem, I have to protect myself as much as possible." Therefore, the design of some medical care robots as female is also to reassure patients and their families. This is undoubtedly a human design.
In addition, there are fewer female practitioners in the ICT field, which is a common problem in the world, the participation of female technical personnel is limited, and it is naturally easy to ignore the female perspective in product research and development.
When the OECD studied employment data from G20 countries, it found that the proportion of female ICT specialists ranged from the lowest 13% (Republic of Korea) to the highest 32% (South Africa). Apple, Microsoft, Google and other companies have only about 20% of female technical employees. As the latest technology field, artificial intelligence also has the lowest proportion of women.
Companies like this, made up of the vast majority of male engineering teams, are more likely to project male gaze onto AI systems, design female characters in reference to movies, music, and TV shows popular with men, and make feminine digital assistants treat verbal abuse in a flirtatious manner.
If this is the case, why do we need to put forward more requirements and opinions on the genderization of AI?
As mentioned earlier, AI genderization has a side that meets the needs of human nature, but also has a side that transmits gender bias, and these factors will have a long-term impact on society along with the ubiquity of AI.
First of all, the scale has a big impact.
Today, AI applications and robots are accelerating in popularity, Siri has more than 500 million devices in high-frequency use, Alexa is talking to consumers in tens of millions of homes around the world, more and more robots and smart devices are providing services, and the depth, breadth and persistence of artificial intelligence impact on society are no less than the changes brought about by the Industrial Revolution. Research firm Gartner has predicted that in 2020 many people will have more conversations with digital assistants than with their spouses. At the same time, AI is also deeply involved in important decisions such as financial loans, pushing content, formulating medical plans, and reviewing work resumes, at which time AI of different genders will also affect the lives of human beings in reality.
Second, the technology is highly anthropomorphic.
With the improvement of emotional voice technology, the ability to distinguish between human and machine sounds will become very low, and AI will already be able to imitate human tone, rhythm, wording, and even some people can't tell whether the phone call they receive is a real person or a machine. At this time, it is easy to project AI gender as human gender, which will have a potential negative impact on people's understanding of gender, such as confusing feminine digital assistants with real women, treating the mistakes made by digital assistants as mistakes made by women, and then thinking that women are stupid. Digital assistants like Alexa are often incapable of defending themselves, and are "coded" to respond with flattery in the face of user abuse. At that time, gender-based misunderstandings and contradictions may intensify.
In addition, AI is highly interactive.
Much of the way humans understand the world and know themselves comes from observing how other people are treated, how they talk, and what they do when they grow up. Harvard researcher Calvin Lai believes that people's gender associations depend on the number of times they are exposed. The more technology guides people to equate women with assistants, the easier it is for real women to be seen as assistants and punished for not being like assistants.
The frequent interaction between people and AI software and hardware is making the frequency and number of associations between "women" and "assistants" increase dramatically. At the end of 2017, about 17,000 people on the social platform Care2 signed a petition asking Apple and Amazon to reprogram and stop their voice assistants from responding playfully to gender insults. Many of the leading voice assistants have indeed been updated, responding to bad gender harassment, not responding or expressing incomprehension.
Li Feifei, a well-known Chinese AI scientist, has emphasized the shortcomings of artificial intelligence, "If in 20 years our technology, leaders and practitioners, as we see today (lack of diversity), it will be an apocalyptic scenario."
Standing in front of the door of the intelligent era, the gender bias shown by AI should indeed make some people feel "blushing".
So the question is, in the face of the dislocation caused by ai gender, how should humans correct the deviation?
In the short term, we must first end the operation of default gender.
Since gender stereotypes can't change for a while and a half, it's better to make machines "gender neutral" first, and Yann LeCun, a pioneer in deep learning technology, also suggested that developers ensure that machines have non-human forms of intelligence and maintain artificial intelligence.
Nowadays, there are many voice assistants and robots in China and the United States that use genderless animals such as cats and dogs as images. Kip, a virtual assistant who works on Slack and Facebook Messenger platforms, describes himself as a penguin. Google also has a policy of always requiring AI callers to declare that they are not human (and naturally have no gender).
In this way, both men and women in reality can avoid being associated with AI gender tags. Especially for children, the intelligent robot or AI teacher should clearly tell the child that it is not a boy or a girl, it is just a robot, no gender.
If a human voice is necessary, the default gender should also be avoided, allowing and prompting the user to choose between men and women.
In the medium term, the technology field should actively explore the diversity of products.
There are currently no clear laws and policies to mandate AI to identify themselves as machines and avoid over-gendered expressions. But some visionaries have begun to try to find change. For example, NASA's Johnson Space Center has launched a female rescue robot. Scientist Radford and his team, mimicking the image of the Valkyrie in Norse mythology, built Valkyrie, a robot capable of driving a car, climbing stairs, equipped with first-aid equipment and armor, capable of helping humans in disasters. The scientist said his 7-year-old daughter has fallen in love with the robot and used her as inspiration to create.
In addition, a lot of AI's gender bias comes from the "dirty data" in the training corpus, so it is also important to establish a gender-sensitive AI public database. This allows technicians to train and develop products directly using unbiased data sets, avoiding the "flattering" AI applications that some companies develop after bypassing public scrutiny.
In the long run, better AI must need emotions and humanity.
Since gender cannot be fundamentally avoided, increasing the number of women in ict fields, especially encouraging more women to participate in AI development, can give AI a temperature from the source and avoid prejudice.
Bridging the gender digital divide does not depend on a programming class for girls or a bootcamp for female developers. It needs to reshape the learning system for women and girls in digital skills, providing a range of supports ranging from early childhood education to career development education, enabling them to stay committed to their interest in AI and take on leadership roles within technology teams. More importantly, female consciousness is not innate, and female ICT practitioners also need to be encouraged and cultivated to be feminine and technologically ethical in order to be able to identify and correct gender biases in AI products.
Digital technology is just beginning to transform social and economic lives, and we will soon be living in a world where AI is ubiquitous, and it's important for everyone to have AI learn how to communicate with us.
Fortunately, women and AI have unlimited potential and plasticity.