“How Can She Help You?” won 2nd place for the 2017 DeWitt Wallace Prize for Science Writing for the Public.
I was having dinner with friends a few years ago when our conversation landed on the topic of Apple’s voice-activated personal assistant on the iPhone – Siri. Announced as an artificially intelligent personal assistant, Siri was intended to aid users with tasks like scheduling appointments, calling friends, and locating nearby restaurants. However, being college students with a certain taste in humor, we all proceeded to pull out our phones and began provoking Siri into giving us responses to ridiculous questions.
“What’s your favorite color, Siri?” asked one of my friends, holding down the home button on her iPhone, which called Siri into action.
“I don’t know how to say it in your language, it’s sort of greenish, with more dimensions,” came the somewhat robotic woman’s voice through the phone’s speakers.
“Hey Siri, can you tell us a joke?”
“Two iPhones walk into a bar … I forget the rest.”
“What’s your favorite song?”
“I always enjoy more cowbell.”
Then, one of my male friends asked Siri what she was wearing, to which she replied: “why do people keep asking me this?” and who her “daddy” was, to which she said, “you are. Now can we please get back to work?”
Two things struck me about the interaction – first, I felt a bit of disgust at my friend who had thought to take our Q&A session with a smartphone feature in such a sexual, gendered direction, and second, that the developers anticipated these questions enough to include responses to them in her code. Though she has a woman’s voice, name, and personality, Siri is ultimately just a program running through the computing unit of a cellphone, technology which is inherently sexless and genderless. Why did developers assume their artificial intelligence would face the same sexism as a female human being?
Siri is not the first computer to listen, nor the first computer to speak. Speech-recognition and text-to-speech programs have been the subject of research since the beginning of automated, electronic computing in the 1950s. The premise is fairly simple – the human voice makes vibrations in the air, sensors in the computer sample and record these vibrations at regular intervals to produce a digitized copy of the sound. Next, the computer matches each imprint of sound with that of known phonemes, the building blocks of all human language, and uses a set of models to match the patterns of phonemes to a library of known words and phrases. Over the second half of the 20th century and into the New Millennium, computers got much better at recognizing and reproducing spoken languages leading to new technologies like voicemail, GPS for cars, computer dictation, language-learning and translating software, automated response on military aircraft, automatic subtitling, and hands-free computing.
In 2008, three years before Siri hit the scene, Yahoo and Microsoft both released smartphones with voice-activated speech tools, software that could take a user’s voice, translate it into a task, execute the task, and speak the answer back to the user. Shortly thereafter, the rise of touch-screen technology and advanced operating systems opened the smartphone floodgates, and in 2016 there were an estimated 2.1 billion users worldwide, according to an eMarketer survey. With every company vying for users’ attention, a deluge of features for the phones were also released, including, of course, voice-activated intelligent personal assistants (IPAs).
Microsoft came out with Cortana; Apple, Siri; Amazon, Alexa; and Samsung, Viv. There’s a trend here – they all have female names. In Old Norse, Siri means “a beautiful woman who leads you to victory.” Cortana was originally the name of a bodacious character from the video game Halo. Amazon stated that Alexa comes from the library at Alexandria in ancient Egypt, which was dedicated to the Muses. It comes as no surprise then that the default setting for all of these IPAs for users in the U.S. is a low female voice with a slight Mid-Atlantic accent. Even IPAs with only job titles, like Google’s Assistant for the Android or Blackberry’s Assistant, respond by default in a feminine voice.
Smart device makers are always trying to differentiate their product – so how come they can all agree that their IPAs should be female? The first justification that jumps to mind is sexism. Women have historically been relegated to support roles in the workforce (when they weren’t being shut out altogether) as secretaries, personal assistants, administrators, nannies, flight attendants, the list goes on. IPAs like Siri perform functions historically found in these jobs: they schedule appointments, look up information, entertain, and are designed for communication.
Women’s representation in popular media is also slanted. A recent study from San Diego State University found that women accounted for only 22 percent of protagonists in the top-grossing films of 2015, a figure that drops to 13 percent in male-directed films. In movies and television shows that deal with the subject of artificial intelligence, the AI tends to fall into two categories: malevolent and subservient. These categories also fall along gender lines. HAL, the homicidal computer in 2001: A Space Odyssey, and Auto, the evil robot pilot in Disney’s Wall-E are both male. Obedient, sexualized characters like Samantha, the operating system in Her, and Ava, the android in 2015’s Ex Machina are female.
Maybe the outdated stereotype of women as subordinate staff – always working, always helpful, anxious to please – has become so entrenched in our cultural knowledge that it has begun to manifest itself in the technology we create to automate certain processes. As we hand over the tasks of secretaries to AI, do we also hand over a historical gender?
Representatives from the tech companies offer a different rationale: maximizing customer satisfaction. Voice technology has been pursued so aggressively by the tech world because it marks the next frontier of user engagement. By allowing users to speak directly to their devices (and for their devices to answer them) even the thin barrier of a touchscreen melts away. The interface between human and machine disappears, giving way to seamless communication (or so they assure us). People seem to agree, as over 6 million Americans invited Amazon’s Alexa into their homes last year via a cylindrical device that syncs up to their phones and computers, with whom they are meant to have friendly conversation. In the end, if customers feel more connected with their devices, they spend more time with them or even begin to use them ubiquitously, leading to more money spent on the devices and their associated features.
From the companies’ perspective, they’re just giving users what they want. A 2008 study lead by Karl MacDorman, a professor at Indian University, found that men and women both find female voices to be more “welcoming” and “understanding,” and women implicitly put more trust in female voices. A study at Stanford University in 1997 found that people prefer a male voice when learning about computers, but like a female voice better when hearing about love and relationships. In an article on Vocativ, a Microsoft spokesperson agreed, saying “when creating Cortana … we did extensive research and found there is a certain warmth to a female voice that is associated with helpfulness. We are going for a helpful digital assistant and we made the choice based on this research.”
In an interview with CNN in 2011, the late Stanford University Professor Clifford Nass, author of The Man Who Lied to His Laptop: What Machines Teach Us About Human Relationships, said “it’s a well-established phenomenon that the human brain is developed to like female voices.” Nass claims that voice preference starts as early as the womb, citing a study in which fetuses were found to react to the sound of their mother’s voice but not to other female voices or their father’s.
Biology and business aside, it is only becoming more apparent that technology and society have become engaged in a feedback loop of instilling their politics and value in each another. Just as our expectations for real-life assistants inform our expectations for virtual ones, our virtual impressions of gender roles can have damaging consequences for women everywhere.
Many students have demonstrated that people tend to prefer men in leadership positions. A study at the University of Michigan in 2013 found that in teams of engineering students, the men lead presentations and the women tended to be given the role of “supporter of the male expert.” A 2013 Gallup survey showed that 35 percent of Americans preferred a male boss over a female one, while 23 percent preferred the opposite.
Gender expectations can also be prescriptive. When women are expected to behave a certain way, they incite criticism when they fail to do so or behave in ways “reserved” for men. A study published in Fortune in 2014 of performance reviews in tech companies found that women, but not men, were rated lower for being “abrasive” and not “letting others shine.” Traits that workers valued in men such as decisiveness and confidence were coded as negative when applied to women.
Maybe society will never be able to shake the preference for female voices in intelligent personal assistants. Some companies, like Apple, have given users the option to customize their IPA’s accent and gender. But the industry default remains female, and defaults inform our sense of the standard, the normal state of being. The growing ubiquity of smart devices in our world gives tech developers unprecedented power in shaping the interactions we have with artificial intelligence. Given that this AI is moving closer and closer to becoming indistinguishable from human beings, tech companies need to think carefully about how the values and personalities they hard-code in their products will affect the values those products could hard-code into our offline lives.