4. Three aspects of diversity in software
The people and teams that create software
The processes we use to create software
Supporting diversity in the software itself
15. Best practices
Diverse representation, inclusive culture, equitable
policies
Setting intention - intersectionally
Accessibility
Flexible self-identification of gender, race,and pronouns
16. Questions: diversity, inclusion, equity
How diverse is your team and community?
How inclusive is your culture?
How equitable is your organization?
What are you doing to increase diversity, inclusion, and equity?
How committed is the senior leadership - and the entire organization?
Who's responsible for it - and do they have the time and resources to make
an impact?
17. Setting intention - intersectionally
Make diversity a priority
Educate your team and
community
Avoid problematic language in
products and code
18. Intention: Questions to ask yourself
Is diversity as a goal and a priority?
If so, how are you communicating that internally and externally?
Do you have a Diversity Statement and a Code of Conduct?
Do people in the organization understand microagressions?
How are you educating your team and community?
Are you encouraging a "call-in" culture?
What standards, tools, and processes do you have about problematic
language?
19. Accessibility
Design for people with disabilities from the
beginning, including
Supporting screen readers and other assistive
technologies
Keyboard-only navigation
Color vision impairment
Hard of hearing or deaf
Seizure disorders and cognitive disabilities
20. Accessibility: Questions to ask yourself
Are you designing for accessibility from the beginning?
Do you have accessibility expertise throughout the team?
Do you include accessibility in your testing?
Does everybody on the team use the software in accessibility modes?
What automated accessibility tools do you use?
What are you doing to increase awareness and understanding?
21. Flexible, optional self-identification
Allow flexibility (not just a fixed list)
Allow multiple choices; e.g., somebody
who’s multiracial may be black and
Latinx
Let people decline to answer
Avoid the term “other”
Let people choose the pronouns they
prefer
22. Self-identification: Questions to ask yourself
Do people have a flexible way to specify their race, gender, orientation, ...?
Does it support multiracial, asexual, gender-fluid, and other frequently-
marginalized people?
Is it optional?
Can people choose their preferred pronouns?
Have you listened to feedback from a diverse group of people?
26. Gender HCI: five key facets
Self-efficacy: how confident are people in their abilities?
Information-processing style: start by gathering fairly complete
information, or try something promising and backtrack if necessary?
Risk aversion: how comfortable are people with risk?
Tinkering: how much do people playfully experiment with the software?
Motivation: interest in technology for its own sake, or in aid of
accomplishing something?
28. Threat modeling for harassment
Threat modeling is a standard security technique
An adversary attacks the systems
Defenses prevent attacks
Mitigations reduce the effect
Harassers = adversaries
Work with experts (aka targets) as well as studying
patterns of harassment techniques
29. Algorithmic biases
Analyze algorithms for “fairness”
What does “fairness” mean in your context?
Work with social scientists as well as techies
Be wary of biases in training sets and historical data
Biases can be race, gender, class, cultural, urban/rural, age, ...
Make sure algorithms transparent enough that you can analyze them for
fairness
31. Challenges - and responses
Diversity is typically seen as low priority and divisive
Choose to prioritize - in an inclusive way
See diversity as an asset to product development
Look for ways diversity can give a strategic advantage
Diversity failures can have huge financial, PR, and strategic
consequences
Inability to pivot, expand the audience, exit, ...
32. Challenges - and responses
Lack of knowledge about diversity
Treat it just like you would any other key skill
your organization doesn’t have enough of
Budget time and money for training, education,
consultants
33. Challenges - and responses
Investment patterns favor cis straight
white male focused products
Look for non-traditional investments
(crowdfunding, etc.)
We’re seeing more forward-looking
decision-makers who get it
We need a few breakthrough
success as proof points!
35. Start with the highly marginalized
Women of color
Gender-diverse people
People with disabilities - including “invisible disabilities”
...
It seems easier to start designing for the usual suspects
(cis straight white able-bodied techie guys)
But that leaves you with a diversity debt
36. Start with the highly marginalized
There are great designers and developers
from marginalized communities out there
Get them on your team - as leaders
Bring them in as consultants, early users, beta testers,
advisors
Prioritize their needs
There’s lots of very valuable and important work going on to make participation more diverse (the first bullet). In this presentation, we’re focusing on the last two bullets.
There’s lots of very valuable and important work going on to make participation more diverse (the first bullet). In this presentation, we’re focusing on the last two bullets.
Given industry demographics: able-bodied, cis, straight, white and Asian, guys
There are techniques for making software that embraces differences - but they are not yet widely adopted
Digitall redlining after Trump: Real names and fake news on Facebookhttps://medium.com/@tressiemcphd/digital-redlining-after-trump-real-names-fake-news-on-facebook-af63bf00bf9e#.k9g7ema70
Fake news software issue: algorithms favor “fake news”; mandatory automated race and gender identification (as opposed to optional self-identification) allows affinity targeting to penalize people identified by the algorithms as “black” and “woman”
“If you are a black woman, like me, that can mean my ability to promote my new book (that was advertising), communicate with my college students, share legitimate information or sources with those who cannot access the academy, and shape the preferences of people similarly marked as “black” and “woman” in Facebook’s affinity algorithms to skew away from class-based assumptions is severely undermined.”
[Real names software issue: designing reporting system without threat modeling how it can be abused. Real names policy issue: burden falls disproportionately on marginalized people. See http://geekfeminism.wikia.com/wiki/Who_is_harmed_by_a_%22Real_Names%22_policy%3F for more But we’re covering this on the next slide.Process issue: prioritizing reports of “real name” violations over reports of racism and sexism]Propublica on excluding users by racehttps://www.propublica.org/article/facebook-lets-advertisers-exclude-users-by-race
Software issue: mandatory automated race and gender identification (as opposed to optional self-identification) allows affinity targeting to penalize people identified by the algorithms as “black” or “Latino”
http://blackyouthproject.com/leslie-macs-facebook-ban-is-the-latest-development-in-racially-biased-censorship/software issue: designing reporting system without threat modeling how it can be abused.
Software issues: - functionality favors harassers - lack of threat modeling, unintended consequences
Also http://nymag.com/thecut/2016/08/a-timeline-of-leslie-joness-horrific-online-abuse.html
http://fusion.net/story/327103/leslie-jones-twitter-racism/
Software issues: - functionality favors harassers - lack of threat modeling, unintended consequences- attempts to improve the situation haven’t worked. Software process issue: not working with the people who are being affectedProcess issues: Waited till after the election to enforce ToS. Gave voice to alt-righters, while allowing silencing women of color (especially black women). (here I have the data and the WoC who were hacked while they were engaging folks to get out the vote. Twitter did not respond or fix their accounts while those that broke ToS weren’t removed until after the election. https://www.buzzfeed.com/charliewarzel/a-honeypot-for-assholes-inside-twitters-10-year-failure-to-s?utm_term=.ohyn7BZyz#.idBwk0BKo
company culture of sexism and acceptable sexual misconduct overflowed into their product. Drivers were also sexually assaulting their customer base with little to no actions from Uber. It was clear this was an internal company practice so it was implemented in the software.
http://www.huffingtonpost.com/eric-holder/airbnbs-work-to-fight-bias-and-discrimination_b_11910438.html
https://www.technologyreview.com/s/602355/airbnb-isnt-really-confronting-its-racism-problem/
Ben Edelman’s “Preventing Discrimination at Airbnb” http://www.benedelman.org/news/062316-1.htmlThe solution: Limit the distribution of irrelevant information that facilitates discrimination - hide names and photos. “Names and photos typically indicate the races of Airbnb guests and hosts. But names and photos are not necessary for guests and hosts to do business. Hosts and guests can amply assess one anothers' trustworthiness using the significant other information Airbnb already collects and presents. For these reasons, I contend that the Airbnb site should not reveal sensitive race information until a transaction is confirmed. If guests and hosts don't see names and photos in advance, they simply won't be able to discriminate on that basis.”Allow testing
Examples:
Twitter not providing tools for people to protect themselves against harassment and abuse
Lack of accessibility support in software means that a lot of people can’t even use it
Sites that limit gender to male/female, or male/female/other
Google photos tagging black people as gorillas: http://www.usatoday.com/story/tech/2015/07/01/google-apologizes-after-photos-identify-black-people-as-gorillas/29567465/
Google’s autocomplete for “women should” … stay at home, be slaves, be in the kitchen http://dhpoco.org/blog/2013/11/19/googles-autocompletion-algorithms-stereotypes-and-accountability/
Google “three black teenagers” as opposed to “three white teenagers” http://www.usatoday.com/story/tech/news/2016/06/09/google-image-search-three-black-teenagers-three-white-teenagers/85648838/
Racial biases in software used to predict future criminal behavior - used in sentencing https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing
Dreamwidth and Django diversity statements
Work to date focuses on males and females, problem-solving software
Work to date focuses on males and females, problem-solving software
AKA “what Twitter doesn’t do”
Software that truly embraces differences works better for everybody
Despite the tech industry’s huge diversity problems, there are a lot of talented women, blacks, Latinxs, Native Americans, QUILTBAGs, seniors, people with disabilities …
and we want software that solves our own problems!