I would like to share an article, published by Forbes, exploring the theme of unconscious bias training which is being broadly implemented by many silicon valley companies.
As the diversity agenda continues to be a key focus across organisations, HR teams and the leadership community, unconscious bias training has come to the fore as a potential “magic bullet” solutionto building a more diverse workforce. One of the principles around unconscious bias training revolves around a primary assumption that people recruit in their own likeness and therefore any lack of diversity across race and gender is perpetuated through the hiring process. Unconscious bias training is designed to identify trends and help Managers adapt their approaches accordingly but the debate as to its effectiveness is still ongoing.
It is likely unconscious bias training will become more of a theme across industry so hopefully, this article provides some useful insight into this growing topic.
Rise Of The Bias Busters: How Unconscious Bias Became Silicon Valley’s Newest Target
Jon Bischke couldn’t have timed it better. He was traveling in New York City to pitch the newest product to come out of his start-up, Entelo: a service that scrapes the web to help companies find a diverse set of job candidates, released just three days earlier. When he stepped out of his hotel room that morning, he glanced down at the USA Today and gaped at its front page. It was May 30, 2014, and Google had revealed its pitiful workforce diversity numbers and pledged to improve them. Hours later, his phone was buzzing with calls from NPR and the Washington Post.
Bischke couldn’t have known it that day, but that was just the beginning. Before Google’s move, most Silicon Valley companies resisted the idea, calling their diversity reports “trade secrets.” (Intel and HP were notable exceptions.) Once Google changed its mind, though, everyone else tripped over themselves to follow suit. Facebook and Yahoo opened up in June, Twitter in July, Apple in August. In the past year and a half, the ritual has even spread down to hot private companies like Slack, Pinterest, Pandora and Indiegogo. Most had the same skew: women held about a third of all jobs, and even fewer in technical and leadership roles. Asian workers were far overrepresented with about a third of jobs, while black and Hispanic workers had just a percent or two. The accompanying promises were so similar it’s almost a joke: Here are our not-great stats. We’re not where we want to be. We still have work to do.
The clock started ticking. “Everyone expected Google to release its numbers the next year and the next year,” Bischke said. “So it changed the conversation from, ‘We’re doing all these nice things around diversity’ to, ‘If we announce 20% this year, and a year from now we announce 19%, can you imagine the press storm?’ … So all these companies now have a huge incentive to improve.”
But how? Google provided another trend-setting answer to that a few months later: unconscious bias. The search giant revealed it had put more than half of its then-50,000 employees through workshops on how to understand and stop unconscious bias, which is the set of deep-in-the-brain automatic preferences that almost all humans have. Unconscious bias influences your decisions in way you can’t notice and can’t control. It’s what makes people who consider themselves champions for women in the workplace rate a man’s resume as more qualified than the same resume with a woman’s name on top. It explains, in part, how a company of well-educated, well-intentioned people could still end up hiring mostly young, white and Asian men.
The idea that you can methodically stamp out unconscious bias (also called implicit or hidden bias) has caught fire with tech companies because it’s relatively new, data-driven and blameless – everyone is told they have it and can’t avoid it, so no one is singled out or gets defensive. It’s a vast improvement on the years when diversity training amounted to lawyers telling managers what to do to cover their backs from lawsuits, said Freada Kapor Klein, a longtime diversity advocate and partner at the Kapor Center for Social Impact. In truth, most of the old methods don’t work. A 2009 review of hundreds of studies showed that the effects of most diversity efforts, including trainings, remain unknown, and a 2006 studylooking at data from 708 private companies found that diversity trainings didn’t produce more diverse workforces. It was time to try something new.
In the past year and a half, demand for bias-busting solutions, in the form of consulting firms and anti-bias hiring software, has shot through the roof. Vying to become the new leaders in the $8 billion-a-year diversity training industry are hot consultants with waitings lists and at least a dozen software startups selling tech tools that promise more diverse ranges of candidates and less biased job descriptions. Venture capitalists have poured almost $50 million into the sector, and many companies are just getting started. In the unconscious bias training video that Google released a year ago, its director of people analytics Brian Welle said, “We’re probably the vanguard of what’s going to happen in this space.” He can drop the “probably” now.
Critics of this new boom worry that companies are jumping on unconscious bias training without making sure it’s being deployed correctly. A recent study even shows that trainings can backfire and make people more likely to stereotype because they’re told everyone has bias. That makes it seem more socially acceptable and lessens the motivation to avoid it. And companies are paying millions of dollars for consultants and software products but aren’t doing controlled tests to see whether they’re producing the advertised effect. In its eagerness to embrace the unconscious bias lifestyle, is Silicon Valley ignoring the possibility that it doesn’t work — or might make things even worse?
A paper version of the implicit association test, which measures unconscious bias. Most people finish this test faster (or make fewer errors) when pleasant words are put on the same side as white faces. People have taken the online version of the test more than 16 million times, and it can be tweaked to measure bias against certain gender, races, ages and other groups. (Photo: Blindspot)
Tony Greenwald, a University of Washington psychology professor, started conducting unconscious bias research in 1994, when diversity programs were mostly point-and-blame. Some employees were bigoted while others were enlightened, trainings said. No one believed themselves to be racist or sexist, though, and overt discrimination was becoming rarer. Greenwald wanted to measure subtle bias instead. He developed the implicit association test, a five-minute speed exercise that cuts through your conscious ideas of your own biases to show you what is happening below the surface.
The implicit association test was revolutionary at the time, and it’s still the cornerstone of Google and Facebook‘s workshops, which they released to the public. Both use the test as the opening technique to demonstrate to a room full of people that they’re biased, even if they believe they’re not.
If you’ve never taken it before, take one now. It’s deceptively simple: you’re given two categories of words to sort into left or right buckets as quickly as you can. The innocuous example gives test-takers a mix of words that are either insects or flowers, and either pleasant (“heaven, cheer”) or unpleasant (“hurt, poison”). Test-takers can quickly put flowers and pleasant words to the left and insects and unpleasant words to the right. When they’re told to sort flowers with unpleasant words and insects with pleasant words, however, they trip up – they take longer or make more errors. Because most people associate flowers with pleasant feelings, the first setup is easier. When switched, there’s “no mental glue available” to hold insects and happy feelings together, as Greenwald put it.
When the test is applied to sorting across race and gender instead of flowers and insects, the results are revealing and disheartening. About 75% of people who have taken the IAT online complete the test faster when white faces are sorted alongside pleasant words, when male words are sorted alongside career terms and when women are sorted with liberal arts studies, not science and tech. It matters only a little whether you’re white or black, a woman or a man, or the order in which you sort the categories. It’s a reflection of the messages the world gives you about the way things are.
Can unconscious bias be the key to making diversity efforts work this time around? Kapor Klein is optimistic but also remembers how diversity efforts in past decades quickly became a chore. She worries that the tech industry is “on a headlong path that looks eerily familiar to previous diversity efforts, mostly in Fortune 1000 companies,” she said. “I’ve seen this movie before.”
Even Greenwald, thrilled as he is that his research is making its way into tens of thousands of workplace trainings, is skeptical. The trainings and software solutions use strategies that have been proven in lab research settings, but no one has shown yet whether they actually make workplaces more diverse in practice. “Diversity training does not have a good track record in producing effectiveness,” Greenwald said. “And there’s no indication yet that this new wave based on implicit bias will do any better.”
If any tech companies can heal themselves of soft bigotry, it would be Facebook and Google, which have more PhDs than most small colleges and have designed their own internal bias workshops. Everyone else is scrambling for outside help – even startups that would have previously scoffed at pursuing anything but growth. “It went from zero interest, ‘we have a business to run,’ to the newest check-the-box, me-too thing,” said the Kapor Center’s Kapor Klein, who has given customized talks at a couple dozen tech companies including Square and Yelp. Venture capital firm Kleiner Perkins Caulfield and Byers, having just emerged from Ellen Pao’s month-long gender discrimination trial, brought Google’s Brian Welle to train 150 of its portfolio CEOs this summer. The Stanford Clayman Institute for Gender Studies’ executive director Shelley Correll has been doing trainings since 2003 but said demand in the last year and a half has “skyrocketed.” She gets at least two requests a week for trainings and has done in-person workshops with 10,000 managers in the last two years.
Paradigm, a diversity consultant company, is less than a year old and already is in full partnerships with Slack, Pinterest and a half-dozen other companies.
Their unconscious bias trainings cost $3,000 to $5,000 a pop for 30 to 50 people, though companies can get a volume discount. Months-long partnerships with Vaya, another consulting firm, can cost $50,000 to $100,000. All the consultants say they are selective about whom to work with and careful not to do trainings that will be treated as a one-and-done solution. But the demand is there. Vaya “could have made thousands of dollars a month if we had agreed to just do trainings, 75% of which would have been unconscious bias trainings, but we were fundamentally not interested in helping companies check a box,” said Nicole Sanchez, who founded Vaya and joined GitHub in May.
A rash of startups, buoyed by the same rising interest in tackling bias, is betting that software can address some of training’s weaknesses – usually for less money, so smaller companies can afford it. These bias-buster products focus on one step of the employee lifetime, like recruiting, hiring or retention. Want to broaden your pipeline and search specifically for candidates that are women or underrepresented minorities? Entelo’s diversity product will scrape public data and make educated guesses about whether a candidate falls into that pool for about $12,000 per recruiter per year. Piazza Careers will let you comb through college students to nab fresh hires in particular categories as well, for anywhere from $15,000 to $150,000, depending how many you’re aiming for. Want to make sure your job descriptions aren’t turning away certain groups with words like “rockstar” or military analogies like “mission critical”? Textio and Unitive will flag them and suggest better options. Unitive, which wants to be a full-service product, also has a slick resume randomizer that prompts hiring managers to reiterate what they want most to see in a candidate before it shows them resumes – stripped of names and non-essential information that studies show can trigger bias. Want blind coding interviews, so you can recreate the famous study where blind orchestra auditions led to a huge jump in the number of female musicians?GapJumpers and Interviewing.io are already on it.
Kapor Klein and others see these startups as the first wave of powerful sea change in using artificial intelligence and big data to make hiring less biased. “The best of these tech startups … get a new set of candidates through the first hurdle,” she said. “Then, the impression is set that this person aced the first one or two sets of the hiring process.” It doesn’t solve everything. That impression still has to live side-by-side with someone’s appearance and background once they move further in the process. But it can be an important head start – one that the startups hope will translate into many paying clients.
As tantalizing as unconscious bias training is, it faces serious limits – ones that companies might be choosing to ignore. Google and Facebook released their trainings publicly despite saying there’s no evidence yet they led to increased diversity. Sanchez said she’s had many firms come and ask her just to have a training session with no follow-up – “that’s a red flag,” she said. Google’s Welle admitted that the company’s unconscious bias training is more explanatory and “not very practical.” Google has built a second workshop that trains people to step in when they see biased interactions, but it’s just getting rolled out – only about 5% of their employees have gone through it.
The central contradiction of hidden bias training is that you can’t train something you can’t control. The classes suggest that you can become more objective just by learning about and thinking about your unconscious biases, but it’s not that easy. “Understanding implicit bias does not actually provide you the tools to do something about it,” said Greenwald, the University of Washington psychologist. He thinks there may be another reason driving companies to do trainings: publicity. “Perhaps the main value of this training to Google and Facebook is to put a desirable appearance on their personnel activities by indicating their (commendable) awareness of problems and implying that they’re doing something to effectively address the problems,” he wrote in an e-mail.
Worse than doing nothing, hidden bias training may even backfire and cause more prejudiced behavior. A 2014 study from professors at Washington University in St. Louis and University of Virginia showed that telling people everyone is biased makes them more likely to act on those biases. “People tend to do whatever other people are doing,” said Melissa Thomas-Hunt, one of the paper’s authors. “If we say that everyone stereotypes, the norm has become that people stereotype. Now I’m not that motivated to change my behavior because everyone’s doing it.” She said the findings left diversity advocates and scholars surprised and disarmed. “This has been seen as the cure to all the ills of stereotyping and why people are not advancing,” she said. “And now to suggest that this might be flawed as a mechanism, people are caught off guard and don’t know where to turn.”
Looming over this growing practice is an even bigger problem: very few firms are measuring the effects of trainings or software, and even fewer are doing it in controlled settings. Many of the software startups are so young they haven’t had time to field-test their products. Instead, services like Unitive are targeting clients with 60,000 employees or more to be able to do large-scale testing within the workplace — but results won’t come for months. For companies who have released their diversity numbers – and feel a pressure to show some improvement in a year — testing each new tool or training is a luxury they don’t have time for. Instead, they try several tactics at once and see if numbers improve.
“I don’t know that we’ll ever be able to know exactly what did it for us,” said Anne Toth, vice president of people and policy at Slack, which recently started doing company-wide unconscious bias training alongside several other initiatives. “But if the data indicate we’re trending in a way that’s positive for our company, then I think we’re going to be happy with that outcome.”
Some firms can’t even point to a correlation between training and improvement, let alone causation. This summer, as companies started to release a second round of diversity numbers, the changes were often 1% — or no change at all. Google has been able to measure and show, using control groups, that Googlers who went through unconscious bias training felt more aware of unconscious bias and felt more motivated to stop it. But does it mean that they – or any others who are being taught about bias — are less likely to make biased decisions? Beyond a handful of positive anecdotes, there’s not much evidence yet that the answer is “yes.”