Stacy Blake-Beard was 29 years old when she was starting out as a professor at Harvard University’s Graduate School of Education. Not only was she the youngest faculty member, but she was also younger than most of her students. One day, one of her doctoral students came into her office to discuss a research project. “[The student] looked over at me and asked, ‘How old are you, anyway?’” Blake-Beard recalls. “I think I did not fit her image of what a doctoral advisor would be.”

The idea that Blake-Beard’s age somehow prevented her from being an effective professor was a bias she often faced as an advisor, she said, even though she had the knowledge and experience necessary for the job. Since that first job, Blake-Beard has gone on to study the dynamics of such unconscious, or “implicit,” biases, and how they can affect diversity in the workplace. She’s focused on how unconscious biases can shape who mentors whom, recently publishing a book with the University of Pittsburgh business professor Audrey Murrell called Mentoring Diverse Leaders.

The idea that deep-seated biases can rule whom one gets close to in a corporate setting may help explain why typical forms of mentorship can sometimes work to the detriment of groups that are already historically disadvantaged or underrepresented. “Like attracts like,” says Vernā Myers, a Baltimore-based diversity consultant. “There is something the mentor sees in another that clicks or reminds them of themselves, and then they want to invest in the potential they see in this protégé.” But when leadership across companies are “fairly white, fairly male, and fairly straight,” she says, this presents a problem, since those leaders will reflexively try to cultivate young professionals who remind them of themselves.

Murrell says that getting these relationships right is incredibly important. If companies don’t, they’ll struggle to diversify their workforces, since the biases that shape who mentors whom can limit mentorship opportunities for people who come from different backgrounds than people higher up in the company. David A. Thomas, a professor of business administration at Harvard’s Business School, says studies show that women of color are least likely to find mentors in large corporations. That’s in part because women of color have the least in common, demographically speaking, with those who are traditionally in positions of power and authority, Thomas says.

The data backs up his point: According to research from LeanIn.org and McKinsey, women of color—though they make up 20 percent of the U.S. population—hold just 12 percent of first-level manager positions, compared with the 45 percent held by their white male peers. Moving up to the C-suite level, women of color make up just 3 percent of corporations’ top senior executive positions, compared with the 71 percent of those jobs held by white men. In that same study, women also reported fewer interactions with senior leaders than their male counterparts did, with such opportunities dwindling further moving up the corporate ladder.

Myers, the diversity consultant, argues that people often don’t realize they pursue sameness—that the brain automatically sorts through people based on stored presumptions and perceptions, noting those who are like us and those who are not. That search for sameness can keep people from getting to know others they might have a strong commonality with, she says. “Someone could be your person and they could be just like you in many ways, but their packaging isn’t like yours,” Myers says. “You can miss them.”

Both Myers and Thomas also note that unconscious bias can go both ways. That is, potential mentors aren’t the only ones who make subconscious judgements about their ability to relate to each other. Prospective mentees can do it, too: They might be quick to make assumptions about their compatibility with a mentor or they may be mistrustful of a mentor’s intentions. This can make for a less fruitful mentoring relationship for pairs that cross demographic barriers, whether that’s race, gender, or some other difference.

One attempt to quantify (and then, often, combat) these biases is the Implicit Association Test, or the IAT, which researchers have used for about 20 years to try to measure the strength of a bias. The test measures how strongly test-takers link together demographic characteristics such as “male” with stereotypes like “athletic” or “executive,” and the results are frequently extrapolated to see how they might affect things like education or employment.

While most psychologists don’t question the existence of unconscious biases, some wonder about the ultimate value in quantifying them in hopes of improving workplace dynamics. Greg Mitchell, a lawyer and a social psychology professor at the University of Virginia, says it’s tough to say whether unconscious bias can undermine the success of mentorships, since it’s difficult to measure something that cannot be observed.

Moreover, in recent years, many corporations, several of them in Silicon Valley, have tried to address issues created by unconscious bias by implementing diversity training exercises, many of which have participants take the IAT. Intel, Facebook, and Google are among some of the companies that put such training programs in place. Yet it’s one thing to identify the existence of a bias; it’s quite another to actually mitigate its ill effects. Some unconscious-bias training may even make things worse by normalizing stereotypes—for example, spreading the message that everyone has them—inadvertently creating a collectively indifferent attitude about fixing a universally shared aspect of human nature. What’s more, there is research showing that making people aware of their biases may just create an uneasy dynamic among people from different backgrounds: If they’re self-conscious about their biases, that might lead to more strained interactions, or even avoidance.

Perhaps the path to a more equitable mentorship requires not only drawing attention to biases, but by establishing systems that take mentorship relationships outside of the informal social networks in which they normally develop. Experts such as Myers say that companies can have employees across the company volunteer to participate in a mentorship program, and then match interested participants together and provide them with full support—from various training exercises to ensuring there is a process in place for feedback. Others, like Murrell, point to peer-to-peer mentoring as an unexplored avenue for companies interested in making sure all employees’ growth is accounted for. They might both be right—it’s likely that the diversity of companies will require a diversity of approaches.