A couple of years ago, I was enjoying dinner at a family gathering, loudly chiming in between bites of salad and veggie burger. In a quiet moment, my Nana’s significant other leaned in and looked at me closely. ‘I can’t pinpoint your accent,’ he said. Surprised, I wondered out loud if I had developed some strange hybrid of Virginia and Texas—my home state and the state where I was attending graduate school, respectively. ‘No,’ he said. ‘It’s more like California.’ He then faux flipped his hair, batted his eyelashes, and repeated something I’d said earlier using exaggerated uptalk. The table broke into laughter, jokes about nail chipping and mall shopping, and ironic air-quoted references to “Dr. Davis.” It was funny because this particular speech pattern—deeply classed and gendered—connotes “ditziness,” which sits in (apparently comedic) contrast to my position as an adult in general, and an academic in particular. And indeed, my speech patterns growing up, like those of many girls I knew, followed the stereotypical “valley girl” inflection and cadence. I have learned to temper this over the years, but in relaxed moments, the excessive “likes” and statements-that-sound-like-questions slip back in.
I was not offended by this dinner table exchange. On the contrary, I put my gender activist hat away for a bit and joined in, asking people to pass various food items in my best Alicia Silverstone (from clueless, obvi) voice. It was totally funny!! It would be less funny, however, if I found myself unable to get a job because of these speech patterns.
Jobaline is a company that helps hourly wage employers obtain and sort job candidates using mobile applications. It promises to both expand the applicant pool, and then produce the strongest candidates from that pool through prescreening. Here’s how it works: Employers give Jobaline the criteria that they’re looking for—education, skills, language fluency, etc.—and in some cases provide specific questions to ask applicants. Jobaline collects this information from applicants, and uses an algorithm to provide the employer with candidates of best fit. Although the employer has ultimate decision-making power, Jobaline’s algorithm significantly increases the odds of some applicants getting the position, while decreasing the odds for others.
Jobaline’s latest innovation is Voice Analyzer, a program that identifies the emotional response a particular voice is likely to elicit. From the website:
Advanced algorithms will measure their voice attributes: inflection, pitch, wave amplitude and predict if the voice will have a particular emotional impact on the listener.
Identify candidates whose voices are calming or soothing; perfect for customer service jobs.
Identify candidates whose voices create a positive engagement with your customers; perfect for sales and frontline employees.
Jobaline touts this tool as a way around human prejudice. The website displays the following quote by CEO Luis Salazar:
There are so many sources of bias when you’re dealing with humans. The beauty of Voice Analyzer algorithm is that it’s blind.
Similarly, writing about Voice Analyzer on NPR, Aarti Shahani states:
The benefit of computer automation isn’t just efficiency or cutting costs. Humans evaluating job candidates can get tired by the time applicant No. 25 comes through the door. Those doing the hiring can discriminate. But algorithms have stamina, they do not factor in things like age, race, gender or sexual orientation (emphasis added).
I think we can all get behind hiring practices that seek to minimize prejudice and discrimination. Voice based algorithms, however, inherently miss the mark. The framing of this technology as objective falsely assumes that 1) algorithms are not human made and 2) what “calms,” “sooths,” and “engages,” is biologically determined rather than socially produced.
Algorithms sort in the way that humans tell them to sort. They are necessarily imbued with human values. Hidden behind a veil of objectivity, algorithms have a powerful potential to reinforce existing cultural arrangements and render these arrangements natural, moral and inevitable.
When applied to desirable voice, algorithms necessarily rely on normative values that intersect race, class, gender, sexuality, geographic locale, and neurotypicality. My uptalk, for example, emanates of white, heterosexual, American, middle-class, girlhood. My male-raised counterparts would therefore likely have a built-in advantage, their voice algorithmically preferred for its relative professionalism. Imagine what this does to a person who speaks with black English vernacular, affected femininity, or the cadence of a deep-southern drawl. The odds are not in this person’s favor.
Technological processes are, always, human processes.
Follow Jenny Davis on Twitter @Jenny_L_Davis