Learn how to make more responsible data connections. I help educators, researchers and practitioners align data polices, practices and products for equity. Sign up for my Rebel Tech Newsletter!
The Rebel Tech Newsletter is our safe place to critique data and tech algorithms, processes, and systems. We highlight a recent data article in the news and share resources to help you dig deeper in understand how our digital world operates. DataedX Group helps data educators, scholars and practitioners learn how to make responsible data connections. We help you source remedies and interventions based on the needs of your team or organization.
“In a tweet posted on Saturday (July 15), artist Lana Denina bitterly complained about being oversexualized by Remini, an AI photo enhancer and editor. She submitted fully clothed reference photos of herself and requested professional headshots. The resulting headshots of Denina showed her with a considerable amount of cleavage. AI’s innate biases concerning race and gender aren’t new. Ethics activists like Joy Buolamwini, have in the past said, “These systems are often trained on images of predominantly light-skinned men,” and these training practices lead to “the coded gaze” or “the bias in artificial intelligence that can lead to discriminatory or exclusionary practices.”
Let’s dig deeper into the origins of oversexualized content on the interwebs – 1-2 steps further than this article mentions. The training data, as Dr. Joy Buolamwini indicates, is designed with white men users first in mind (then white women) so this data overwhelmingly includes white men in the best light – as people in authority, physical features are made distinguishable by algorithms and so on. But there’s the unspoken design process: people who are members of historically excluded groups are digitized as opposite of white men/women – as people not in authority, e.g., criminals, sex workers, etc., similarities in physical features are grouped together and so on.
This “white men first elevated and everyone else demoted” computational design flaw indicates that the training data including people from historically excluded groups is demoralizing. One such abundant place where training data including people from historically excluded groups comes from is pornography. According to a 2016 Distributed Computing Systems research study, adult content constitutes nearly 12% of websites. And upon that even deeper dive, top pornography searches by U.S. states have a large focus on women and Black and Asian people (see PornHub’s The 2022 Year in Review article). The longstanding fetishing of historically excluded groups has created a wealth of digitized racist content. So to emphasize, there are two important mutually reinforcing conditions happening simultaneously that fuels AI bias: systems, tools and platforms prioritizing white men/women users AND systems, tools and platforms debasing the humanness of everyone else.
It’s not hard to now connect the dots on how and why Lana Denina received oversexualized headshots. The reference training data she provided was deemed in congruent with the training data in the AI system so her submitted photos were largely ignored. Understanding the fuller context allows us to implement remedies to some of these disparities. We can’t remove all the -isms in our digital infrastructure until we remove them from our society, imho. What we can do is more intentionally root it out and mitigate harmful AI scaling.
Click Here to Read the Entire Article |
"Discriminatory practices are embedded within database architecture. DBAs and those with intimate knowledge of database architecture and design currently bear the weight of these decisions without many viable avenues to implement alternatives or counteract destructive structures." pg. 144
As a first step in identifying the data pipeline’s contribution to discriminatory practices, consider which data tools your organization and team uses in each phase of the data pipeline. I commonly reference 5 data pipeline phases so here’s a quick recap: data acquisition and cleanup, data storage, data analytics, data visualization and data storytelling. With each phase, you’re trying to polish that “messy” data into more specific, measurable, actionable, relevant and timely insights. Oh and we’re crossing our fingers that we don’t create mis/disinformation. This means we require a data equity strategy to intentionally integrate inclusionary practices.
By discussing your team’s data tool stack, you reveal frequently used data tools in order to start establishing more standardized data equity practices. You can begin to identify ethically-weak data process criteria so that you can fill harmful data/AI gaps and implement alternative data structures – as a team.
Get Your Copy of Data Conscience Here! |
The Black Women in Data Summit 2023 just wrapped up this past weekend. For Year 2, it was a SOLD OUT event filled with:
A great inspiring and informative Summit indeed. Learn more about this year’s Summit at blackwomenindata.com. A big thank you to our sponsors – Cox Coaching Co., DAIR, Dataiku, Elastic, Moxy Analytics, Posit and Taye Johnson Co. – and community partners – BlkWomenHustle, Data Science Connect, Diversity Tech, EMOIR Magazine and WiMLDS!
See y'all next September!
Subscribe for Updates on All Things BWD Here! |
AIAI Network Kickoff | October 4th, 2023
DataedX Group, as a founding member of the AIAI Network, is excited that AIAI Network is partnering with the Science Gallery Atlanta for our kickoff event. Please join us on October 4th 4-7PM at the Science Gallery to learn about ways to get involved in the AIAI Network, tour the JUSTICE exhibit, and meet like-minded members of the Atlanta AI research community. Refreshments provided by Meraki Soul.
Oct 4, 2023 | 4-7 PM | In-person event only
AIAI Network Kickoff RSVP |
Follow us on social
Stay Rebel Techie,
Brandeis
Thanks for subscribing! If you like what you read or use it as a resource, please share the newsletter signup with three friends!
Learn how to make more responsible data connections. I help educators, researchers and practitioners align data polices, practices and products for equity. Sign up for my Rebel Tech Newsletter!
February 20th, 2024 The Rebel Tech Newsletter is our safe place to critique data and tech algorithms, processes, and systems. We highlight a recent data article in the news and share resources to help you dig deeper in understand how our digital world operates. DataedX Group helps data educators, scholars and practitioners learn how to make responsible data connections. We help you source remedies and interventions based on the needs of your team or organization. IN DATA NEWS “Don’t let...
February 6th, 2024 The Rebel Tech Newsletter is our safe place to critique data and tech algorithms, processes, and systems. We highlight a recent data article in the news and share resources to help you dig deeper in understand how our digital world operates. DataedX Group helps data educators, scholars and practitioners learn how to make responsible data connections. We help you source remedies and interventions based on the needs of your team or organization. IN DATA NEWS “Wisconsin’s...
January 23, 2024 The Rebel Tech Newsletter is our safe place to critique data and tech algorithms, processes, and systems. We highlight a recent data article in the news and share resources to help you dig deeper in understand how our digital world operates. DataedX Group helps data educators, scholars and practitioners learn how to make responsible data connections. We help you source remedies and interventions based on the needs of your team or organization. IN DATA NEWS “Concerns about...