Why Cortana doesn’t work at work

Microsoft is betting that Cortana will bring AI to the workplace. Here’s why that won’t happen.

Cortana is an intelligent agent  that is supposed to act as a personal assistant. You can interact with her (notice I said “her”? More on that in a minute) via voice or text, on mobile devices or on desktop computers. Given that Microsoft’s mobile market share has fallen below 1%, it’s pretty much a certainty that most people would interact with Cortana in their offices.

We know that most Windows 10 computers are in workplaces, so there’s a very strong likelihood that people will talk to Cortana in an office. This is very different place than where people might interact with Siri on their phones, or Alexa in their homes. Siri and Alexa t are called upon in private, controlled places (in fact, just 3% of iOS users report using Siri in public).

Let’s walk through that interaction of Cortana as a member of a workplace.

Microsoft encourages you to command Cortana by saying, “Hey Cortana…” and then giving her a command. A typical office scenario might be, “I wonder if I should book a vacation for the first week of August. Hmm. I’ll ask Cortana.”

This is how Cortana is supposed to work:

User: Hey Cortana, should I book a vacation for the first week of August?

Cortana: Let me check your calendar. Looks like you have a meeting on Monday, August 1st. Should I move it for you?

User: Yes, that’d be great.

Cortana: Okay, I’ve moved that meeting to Monday August 8th. Would you like to see some vacation suggestions?

User: Yes, please!

This is exactly how it plays out on a demo video one Microsoft’s site.

But let’s face it: there are a lot of contextually dependent reasons why this is completely unrealistic. Leaving aside Cortana’s technical limitations for the moment (and there are many), let’s take a look at what a real office and real user might look like.

Most offices are either open concept without even the suggestion of walls. As many as 70% of us work in open concept offices. As anyone who’s worked in such an office can tell you, hearing a neighbor on the phone can be excruciatingly annoying or excruciatingly awkward, depending on your neighbor’s TMI quotient.

cortana

So there’s a good chance that everyone in the user’s office will hear this idealized scenario.  There are two clear disincentives against this happening. First, Cortana will make more “boundary work” for office workers. The mere act of trying to keep your private life private at work is turns out to be, well, work. Recent research  has found that keeping work and life private actually causes cognitive overload. If people use Cortana as intended, she is poised to make that much worse.

Second, Cortana demands office workers treat their workplaces as if they were kings and queens, instead of pawns and rooks. Voice interactions require workers to own their workspace, something that we know they do not do. Typical workers share their workspaces with others, and because we are apt social animals, we tend to comply with unwritten rules of workplace etiquette. Bosses’ calendars take precedence over workers’ calendars. Bosses talk more than workers. Men talk more than women. In other words, people with power talk out loud more than people with less power.

Which brings me to the fact that Cortana is a woman. Is it any coincidence that most intelligent agents today are anthropomorphized as women? One of the most striking changes in the twentieth century workplace was the almost total elimination of support staff, which were typically women. Only the most senior executives have assistants nowadays, and other mid-level white collar workers are on their own for scheduling  and administrative work.

cortana 02

Let’s not forget that Cortana is actually based on a supportive AI character in a video game. Cortana provides these workers with a sense that they can indeed recoup the times of Mad Men and have a compliant, supportive, and self-abnegating assistant who has no needs of her own. Practically, this promises white-collar workers with a huge productivity boost, but the symbolic nature of this is even more interesting. When white-collar workers have a virtual assistant, they have re-claimed a sense of hierarchy, of control, and power (even if it is completely imaginary).

And this is why Cortana will not work in the workplace. Today’s typical office worker does not have power enough to command the space around her, and bark orders to anyone out loud, even if just to an intelligent agent. This office worker has been stripped of her ability to occupy a rung on the ladder higher than admin or support staff, because there is no admin or support staff. This typical office worker is embedded in a physical space that reflects this lack of hierarchical position — she has no command over it.

Scholars of gender and technology have described some ill-advised approaches to gender equality as “add women and stir.” The same applies to Cortana and other intelligent agents. You cannot “add Cortana and stir” and expect to see productivity improvements that somehow negate the existing organizational and physical structures of contemporary workplaces.

 

What we know about gender and performance reviews

Update December 23, 2017: New findings from an AI-based experiment on gender and performance reviews has been added, and a new evidence-based improvement strategy called the “small wins” model.

The evidence is mounting that our existing feedback systems have inherent biases that penalize women. I have a personal commitment to help improve this, and in this post, I make specific recommendations on how to do so. But first, what do we know about gender and performance reviews?

  • Women receive more criticisms of their personalities in performance reviews: A linguist did study on performance reviews. Men and women provided positive performance reviews. She found that in 83 performance reviews, men received personality feedback in only 2 cases. In 94 critical reviews, women received personality criticisms 71 times. [1] Words like “abrasive” and “strident” were present in these performance reviews.
  • Women receive less helpful feedback than men. In a study of 200 performance reviews in a tech company, researchers found that women were more likely to receive vague praise than were men (57% and 43%, respectively), which included unhelpful comments like “You had a great year.” Men were more likely to receive developmental feedback, and linked specifically to business outcomes. When women did receive developmental feedback, it tended to relate to their personalities rather than to their performance [2]
  • Women are less likely to be rewarded for good ideas. Men who provide revenue-generating ideas are given higher performance ratings. Women who provide revenue-generating ideas see no improvement in their ratings. [3]
  • Men reward other men more highly than women for achieving the same goals: 70% of men rate men more highly for achieving the same goals as women, while an algorithm rate men and women equally (as did other women).
  • The “glass ceiling” is the result of many tiny obstacles. Researchers at Harvard Business School found that there is no specific point where women face a “ceiling,” but many small instances of discrimination lead to their careers stalling over time. [4] This may be why after only 2 years with a company, women’s aspirations for career advancement fall an astounding 60%, while men’s aspirations fall negligibly.
  • Ostensibly “meritocratic” reward systems favor men over women, and whites over minorities. Researchers experimentally tested whether managers would reward people differently when explicitly creating a system of “merit.” Over 3 experiments and 445 participants, they found men were rewarded with more money than women in this supposed “meritocratic” system. They also found that ethnic minorities and non-American born people were given lower raises, even when using the same evaluative criteria. [5] In another experiment, researchers switched male and female professors of an online course in the middle of the term. Students consistently rated the male professor higher, even though they actually had a female professor without knowing it. [6]
  • Women are penalized for asking for raises: Researchers found that people judge women more harshly when they ask for a raise; women don’t ask for raises because they realistically assess the social cost of asking. [7]
  • Men are rated more highly for helping colleagues, and women are rated more negatively for not helping. In a series of experiments, researchers asked participants to rate the performance of men and women who either agreed to stay late to help colleagues, or refused to stay late and help. Men who offered to stay were rated 14% more positively (women’s rating remained the same). Women who refused to stay were rated 12% more negatively (men were not rated more negatively). [8]
  • Women pay a penalty for motherhood, while men reap a bonus for fatherhood: Researchers have found women who become mothers pay a minimum penalty of 4% decline in income. [9] The penalty is larger for more educated women. [5] By contrast, fathers reap a bonus [11]. This is not due to a lack of commitment by women, by biased perceptions of their commitment. [10]  In other words, the cost of being a parent depends on your gender, not your performance.
  • Men are penalized when they ask for family accommodation. In a study of a management consultancy, one researcher found that men who ask for flexibility to care for their families are punished in performance reviews. [12] Men who did not openly ask but made private, covert arrangements got better performance reviews.
  • Keeping track improves fairness. Just keeping track of how people get rewarded, broken down by race and gender, was enough to reduce inequality over 5 years in a single company. So know your data! How are people doing relative to each other? [5]
  • Just pointing out bias actually increases its incidence! Other research has found that pointing out bias actually increases its negative consequences. [13] This effect disappeared when researchers noted that discrimination is not desirable.
  • Training managers on potential work/life conflicts decreases employee stress. Researchers trained a single company’s managers on how to deal with work/life conflict. They found reduced employee stress, and no increase in employee hours. [14]

 

Opportunities for Improvement

  1. Examine the words you use in your performance feedback. Is it related to personality or performance? Are the words very gendered, such as “bossy”? Consider what words you might have used to describe the opposite gender.
  2. Is your feedback helpful and specific? Did you provide vague praise like “Great job this year” or did you say, “Your work on the launch plan led to greater sales”? Make sure you link the feedback to specific business goals.
  3. Did you reward fairly? Is there a systematic difference between the genders? Could this be unconscious bias?
  4. Is this a “tiny obstacle”? How many barriers has your direct report experienced in her career? Is this performance review a chance to create a “tiny ladder” through the glass ceiling?
  5. Do you expect women to be more altruistic? Reward men and women equally for the same behaviors. Consider if you expect women to be more giving of their time than men.
  6. Do you expect men to be less family-oriented? Recognize that men have families too. Are you penalizing a man for being a caring father?
  7. Is your direct report asking for something…and are you evaluating that ask fairly? Consider what kinds of requests your direct report has made. Are you judging those requests fairly, or are you penalizing the person for speaking up?
  8. Unconscious bias is wrong. Make sure you point out that bias is wrong, not just that it exists. Norms are powerful, especially for senior leaders.
  9. Keeping track improves fairness. Just keeping track of how people get rewarded, broken down by race and gender, was enough to reduce inequality over 5 years in a single company. So know your data! How are people doing relative to each other? [5]
  10. Pursue “small wins“: researchers at Stanford found that they could improve outcomes if they worked directly with managers on reducing bias. Introducing a new scorecard reduced personality-based feedback to zero.  [15]

 

 

 

References

[1]       K. Snyder, “The Abrasiveness Trap: High Achieving Men and Women Are Described Differently in Reviews,” Fortune, New York, Aug-2014.

[2]       S. Correll and C. Simard, “Vague Feedback Is Holding Women Back,” Harvard Business Review, no. April, 2016.

[3]       A. Grant, “Rocking the Boat but Keeping It Steady: The Role of Emotion Regulation in Employee Voice,” Academy of Management Journal, vol. 56, no. 6, 2013.

[4]       A. Eagly and L. Carli, “Women and the labyrinth of leadership,” Harvard Business Review, no. September, pp. 62–71, 2007.

[5]       E. J. Castilla and S. Benard, “The Paradox of Meritocracy in Organizations,” Administrative Science Quarterly, vol. 55, no. 4, pp. 543–576, Dec. 2010.

[6]       L. MacNell, A. Driscoll, and A. N. Hunt, “What’s in a Name: Exposing Gender Bias in Student Ratings of Teaching,” Innovative Higher Education, vol. 40, no. 4, pp. 291–303, 2015.

[7]       H. R. Bowles, L. Babcock, and L. Lai, “Social incentives for gender differences in the propensity to initiate negotiations: Sometimes it does hurt to ask,” Organizational Behavior and Human Decision Processes, vol. 103, no. 1, pp. 84–103, May 2007.

[8]       M. Heilman and J. Chen, “Same Behavior, Different Consequences: Reactions to Men’s and Women’s Altruistic Citizenship Behavior,” Journal of Applied Psychology, vol. 90, no. 3, pp. 431–441, 2005.

[9]       T. Street, A. Arbor, and P. O. Box, “Has the Price of Motherhood Declined Over Time ? A Cross-Cohort Comparison of the Motherhood Wage Penalty,” Journal of Marriage and Family, vol. 65, no. August, pp. 597–607, 2003.

[10]     J. a. Kmec, “Are motherhood penalties and fatherhood bonuses warranted? Comparing pro-work behaviors and conditions of mothers, fathers, and non-parents,” Social Science Research, vol. 40, no. 2, pp. 444–459, Mar. 2011.

[11]     G. Hundley, “Male/Female Earnings Differences in Self-Employment: The Effects of Marriage, Children, and The Household Division of Labor,” Labor Relations Review, pp. 95–114, 2000.

[12]     E. Reid, “Embracing, Passing, Revealing, and the Ideal Worker Image: How People Navigate Expected and Experienced Professional Identities,” Organization Science, vol. 0, no. 0, p. null.

[13]     M. Duguid and M. Thomas-Hunt, “Condoning Stereotyping?: How Awareness of Stereotyping Prevalence Impacts Expression of Stereotypes,” Journal of Applied Psychology, no. October, 2014.

[14]     E. L. Kelly, P. Moen, J. M. Oakes, W. Fan, C. Okechukwu, K. D. Davis, L. B. Hammer, E. E. Kossek, R. B. King, G. C. Hanson, F. Mierzwa, and L. M. Casper, “Changing Work and Work-Family Conflict: Evidence from the Work, Family, and Health Network,” American Sociological Review, vol. 79, no. 3, pp. 485–516, May 2014.

[15]   Correll, S. J. (2017). SWS 2016 Feminist Lecture: Reducing Gender Biases In Modern Workplaces: A Small Wins Approach to Organizational Change. Gender & Society, 31(6), 725–750. https://doi.org/10.1177/0891243217738518