🚨 New preprint () w/
@jiank_uiuc
&
@YizhouSun
.
We dive deep into the question: why do GNNs perform better for high-degree nodes? Finding diverse hypotheses in the literature, we offer many theoretical & empirical insights into the origins of degree bias.
I am once again asking researchers to not list "predictive policing," "criminal risk assessment," "recidivism prediction," etc. as applications of ML that need to be made more fair. They are applications of ML that should not exist.
undergrads that I previously mentored: “I used to think that machine learning was magical, but now it feels like glorified math with questionable assumptions and oppressive tendencies”
with a deadly virus & threat of a mass shooting, ucla not guaranteeing remote learning, despite the student strike and recognizing that students "are terrified," or even informing the entire student body of the threat is a new level of fucked up, even for ucla
My therapist was trying to better understand some of my anxiety, and I don’t think that “terrible uses of machine learning” was a can of worms she expected to open
In 1 week of classes, I’ve seen these “toy” examples of ML:
- using BMI to predict heart attack risk
- using “intelligence” (??) to predict SAT score
- crime prediction
So much to learn to about what ML researchers deem acceptable based on the examples they use in the classroom.
Excited to share our new preprint on a critical review and reimagination of Intersectionality in AI fairness! Work with
@ovalle_elia
,
@DippedRusk
, Gilbert Gee, and
@kaiwei_chang
🌈 (1/n)
lately, when writing or giving talks about queer AI harms, I’ve asked myself, “do I actually need to discuss why sexuality classifiers are pseudoscientific and harmful? isn’t it obvious?” turns out it’s not 🥲
there is no such thing as brain correlates of homosexuality. this is unscientific with a disastrous implication in countries where homosexuality is illegal.
just let people be or let people identify their own sexuality
Today I talked to someone interested in applying to AI PhD programs to do research with "real-world impact." While I encouraged them to do a PhD for other reasons, I emphasized that expecting to make real-world impact *solely* through an AI PhD can be dispiriting. 1/n
.
@uclanlp
is researching the harms of treating gender as binary in NLP tasks, as seen and experienced by non-binary folks. As part of this, we are conducting a participatory survey () of non-binary folks with any level of familiarity with AI.
This weekend, I walked after graduating summa cum laude from
@UCLAengineering
(in 2.67 yrs) as the school-wide Outstanding Bachelor of Science and a recipient of the Chancellor's Service and Engineering Student Welfare Awards. I want to address my privilege and express gratitude.
Love my advisor! She asked me a few months back about how she can better support and include LGBTQIA+ students, and this quarter, in her data mining class, she’s committed to discussing the ethical implications of treating gender as a binary variable.
Hey, you! Are you a queer grad school applicant in AI? Whether you're waiting for an acceptance or have already been accepted to an amazing program, I'm so proud of you 💜
I also can't recommend joining
@QueerinAI
enough, for a supportive community of queer friends & mentors! 1/2
Excited to start a CS PhD this Fall with
@YizhouSun
at
@UCLAengineering
:D
Much love for everyone who answered my questions and gave me advice during the admissions and decision processes 💜
Thinking of the time that my high school math teacher started class on Indian Republic Day with “Today’s a very special day in India. Who knows what day it is?” and a ton of students in the class responded “Arjun’s birthday” 😭
Accepted to
#AIES
2023! This was such a fantastic, healthy collaboration that centered justice and relationships at every stage of the research process 💜 So grateful for my collaborators and to be able to present this work 😊
Excited to share our new preprint on a critical review and reimagination of Intersectionality in AI fairness! Work with
@ovalle_elia
,
@DippedRusk
, Gilbert Gee, and
@kaiwei_chang
🌈 (1/n)
🚨 New preprint () w/
@DippedRusk
, Dietrich Klakow, and
@ZeerakTalat
!
We seek to clarify how democratization is understood in NLP and ML publications. 🧵 (1/n)
The non-binary students who reach out to me for advice on getting involved with AI research often profusely thank me for taking the time to chat with them, but they don’t understand 😭😭 talking to them & learning about their interests is genuinely the best part of my day
- Arjun (they/them)
- Research interests: graph machine learning, fairness
- Hobbies: hiking, birding, ukulele
- Fun fact: 2nd dan black belt in ITF Taekwon-Do
🏳️🌈🏳️⚧️🎉Happy Pride month! 🎉🏳️🌈🏳️⚧️
Let's do a queer in computer science (and related fields) roll call! 📜Please QT with:
-name/pronouns (if any) 🙋
-research area, specialization, or interests 🔬
-hobbies ⚽️🚵🎮📖
-fun fact ⁉️
Concerned about the inequity of the NeurIPS high school track? Valid.
Let's also channel our concern towards the same, numerous inequities that already pervade the conference: wildly disparate funding, resources, & mentorship; visa restrictions; language barriers, etc.
Announcing the
#ICLR24
@iclr_conf
workshop on
🌏 Global AI Cultures 🌍, a hybrid event with scholars in AI/ML and social sciences discussing the possibility of globally inclusive AI systems.
Submit your papers and provocations by 🚨 Feb 24 🚨
Read on for CFP details... 🧵
Few NLP friends said reviewers asked them to evaluate on more languages despite their work centering challenges of specific marginalized languages/dialects, and I have to wonder… would reviewers ask the same if the work only considered English?
Our workshop on non-binary representation in language technologies was rejected from the 2022 *CL conferences. However, many NLP papers don't consider the harms and erasure of modeling gender as binary. This workshop is important, and we'll work hard to host it soon.
.
@uclanlp
and
@QueerinAI
are putting together the first workshop on Non-Binary Representation in Language Technologies at a *CL conference in 2022!
If this interests you and/or you're likely to attend, please indicate so in the attendance survey: :)
🚨 New preprint () w/
@DippedRusk
@anne_lauscher
@farbandish
Are you associating personal names with sociodemographic characteristics like gender and race in your NLP research? Read this and reconsider. 🧵
It makes me so happy to see all the ACL papers on queer NLP harms 💜 So much more work being done in this space than even two years ago!! Your research is so critical and valuable.
Will likely organize a panel on experiences of queer neurodivergent individuals in academia + industry at this year's NeurIPS
@QueerinAI
Workshop. Any recs for individuals (esp people of color) you'd like to see featured? Not required that they work in AI. Feel free to reply/DM!
Vulnerability 😭
@QueerinAI
has fundamentally shaped my values, research, & career. I wouldn't have done grad school if it weren't for QAI. QAI has bolstered my commitment to justice, given me the best community ever, and keeps me optimistic. Hope you enjoy our
#FAccT2023
paper💜
After six years of hard work + research + organizing, we could not be prouder to present our
#FAccT2023
paper on Queer in AI as a case study for community-led participatory design in AI! . Read on to learn more about our endeavors 🌈🏳️⚧️ (1/10)
feels so weird and amazing that 3 years ago, i was a freshman at ucla anxious about making friends and learning what a neural network is from andrew ng, and now i'm a phd student with the best friends *ever*
Exhaustion! from being inundated with news of "progress" in harmful AI while, as an early-stage PhD student, knowing that people with power in my field do not value community work or actually dismantling algorithmic injustice
I started my PhD two days after a particularly difficult breakup. I didn't give myself enough time to grieve and overworked to distract from sadness. For the first time, I found myself miserable doing research. I am currently so much happier and confident in myself.
Can’t wait to see my friends and make new ones at
#FAccT
next week!
(If I previously told you I wasn’t going, oops, my situation changed and I’m going now.)
Passed my W21 classes, so can say that I’ve graduated summa cum laude from
@UCLAengineering
with a BS in CS! It was an incredibly fun and transformative 3 years, thanks to my friends,
@ACMUCLA
family,
@qwerhacks
besties, and research (grateful to
@YizhouSun
for her mentorship!)
.
@uclanlp
and
@QueerinAI
are putting together the first workshop on Non-Binary Representation in Language Technologies at a *CL conference in 2022!
If this interests you and/or you're likely to attend, please indicate so in the attendance survey: :)
Feels like 2021 started yesterday yet everything happened this year. Through it all, Queer in AI provided me with a community of friends & mentors, with whom I worked on so many things I’m proud of. QinAI feels so integral to me that it’s odd to think I wasn’t part of it in 2020.
Please RT! We are investigating barriers NLP researchers face in using large language models (LLMs), as well as their conceptualization & theories of democratization of LLMs. We invite NLP practitioners to fill out our survey:
w/
@ZeerakTalat
,
@dippedrusk
As someone who has lived in California nearly their entire life (Bay Area, LA, San Diego), I can’t overstate how much the Paris Métro has blown my mind re: what I thought was possible with public transportation.
Also received a fellowship review today that told me that my application could benefit from “demonstrating engagement with communities that are underrepresented.” I spend 10-15 hours every week on
@QueerinAI
.
Had a reviewer claim my fellowship application that heavily focused on my work with LGBTQ+ representation, equity, and inclusion in STEM showed “no evidence of sustained engagement with UR groups in academia”. Are the gays no longer UR? Have we finally taken over??
Great article, please read! Many people have spent countless years educating Google & others about how Scholar harms trans researchers, drafting precise demands for how Scholar should be fixed, incessantly advocating for their demands to be addressed, & Google has done NOTHING
Google policy continues to violate the privacy, dignity, and safety of transgender researchers, according to people who have urged change in recent years
Much of the discourse on responsible ML concentrates irresponsibility in individual researchers or particular design choices, while neglecting the academic socialization, cultural denial, & social power relations that push our field to oversell and shirk accountability.
Wishing all trans people (especially trans people of color) some happiness on this Trans Day of Visibility 💜 And fuck transphobic structures and people everywhere who make trans folks feel safest when they’re not visible.
New preprint () w/
@leventsagun
&
@YizhouSun
. We theoretically & empirically show that Graph Conv Nets can have a preferential attachment bias in link prediction. We analyze how this bias amplifies degree/power disparities & propose an alleviation strategy.
Attending
@NeurIPSConf
#NeurIPS2022
? We have an amazing lineup of affinity workshops, socials, & poster sessions on Mon, Nov 28 (in-person) & Mon, Dec 5 (virtual) 🎉 Check out this meta-thread for more details and RT this & the affinity groups' threads for wider reach 😊
Vinyl adventures at the MakerSpace 😈
(Sorry to those who have seen this shirt already, I’m just veryyy excited.)
With permission from
@Abebab
Demon twink idea:
@willie_agnew
Vinyl help:
@dieeantha
My heart repeatedly breaks for the people in Tigray, Ukraine, Yemen, and so many other places in the world. To all the people in tech who are saying people should care about what's going on in Ukraine because of the products developed there, what the hell is wrong with you.
In a new blog post, we reflect on how
@QueerinAI
's award-winning FAccT paper can inform participatory governance in AI, and why U.S. government leaders, agencies, and stakeholders should pay attention to the organization's insights:
I am annotating some ChatGPT-generated Tamil sentences, and one of them translates to "I am waiting for the samosa. I am getting hot and crispy." So... in good news, ChatGPT-enabled surveillance of Tamil speakers seems unlikely...
I'm often peeved by fairness papers that claim that their algorithms are important in "high-stakes applications" of machine learning. I don't know, if an application is high-stakes (esp. if there's a history of discrimination), perhaps it's best to just... not automate it??
Lovely time opting out of FRT at SFO:
1) was told I wasn’t allowed to opt out, then asked twice why I wanted to
2) armed border security took pictures of my passport & DL, checked every item in my backpack, & even tested my vitamins for THC
3) gates almost closed before I boarded
I am SO excited for the panel on Labor Organizing and Queer Activism in Academia and Tech at our
#ACL2023
Queer in AI Workshop! We have folks from UMich, IIT Delhi, the NYT, and the Universidad Central de Venezuela 💜
Uploaded my
@uclanlp
guest lecture on Bias in NLP! I discuss the multiplicity & sociotechnical nature of bias; sources of bias in the NLP lifecycle; harms; bias metrics, interventions, & their pitfalls; & the power structures in which NLP is embedded.
Currently, the AI community incentivizes general, portable solutions to complex problems with minimal discussion of their limitations. It is lauded to create algorithms that "work" off-the-shelf without any understanding of the data or their context. 2/n
I love
@QueerinAI
and everyone in it so much. I get to work with the *coolest* people, and I truly appreciate the platform it provides me to be an advocate for causes dear to my heart. We wrapped up our
@icmlconf
workshop today, & our speakers and panelists were phenomenal.
PhD student in the Computer Science Department at UCLA: $25k/yr
Want to note that “With the average apartment renting for $4,883, [Westwood]’s 90024 ZIP Code ranks third on RENTCafe’s list of the 50 priciest ZIP Codes in the nation”
To increase transparency around grad school stipends, retweet this tweet with your department, university, and annual stipend. I'll go first: I'm a PhD student in the Computing and Mathematical Sciences (CMS) department at Caltech, and I'm paid $36k/year.
#StipendTransparency
Frustrated by the extractive "diversity leads to better products" rhetoric behind inclusion initiatives in tech. Why doesn't everyone simply deserve an inclusive, healthy work environment?
Looking forward to briefing the NAIAC (on behalf of Queer in AI) about how AI harms queer communities and advancing equity, rights, and justice for LGBTQIA+ people in AI design, development, and deployment!
We look forward to briefing the National AI Advisory Committee, which advises the President and NAIIO, on June 27th, alongside other leaders. Find out more details at , and register to watch at
#NAIAC
Intersectionality is rarely incentivized by conferences, though it’s necessary to combat inequalities & shift power. QAI exercises intersectionality as inquiry & praxis to inform areas of collective action. Our paper reflects this by embedding critical self reflexivity throughout
After six years of hard work + research + organizing, we could not be prouder to present our
#FAccT2023
paper on Queer in AI as a case study for community-led participatory design in AI! . Read on to learn more about our endeavors 🌈🏳️⚧️ (1/10)
We desperately need more trans people of color and trans folks from the Global South researching gender and AI and have to do a much better job giving them platforms to share their work.
Many brilliant women and non-binary researchers I know are getting faculty interviews and job offers, and it makes me so happy 😊 They're going to positively impact so many students!
I recently learned that the professor who runs the CS teaching assistant training seminar at UCLA really committed to the content in this guide, even thanking people (who feel comfortable) for introducing themselves with their pronouns during practice sections 🤗
Please attend the amazing
#NeurIPS2023
affinity workshops taking place all-day on Monday, December 11 💜
Schedule: (accessible with registration)
Blog post:
UCLA Police Department is aware of a concerning email and posting sent to some members of the UCLA community today and actively engaged with out-of-state law enforcement and federal agencies. We will keep the community informed as we learn more.
This afternoon at
#NeurIPS2022
, I will present our work on how graph feature imputation affects the fairness of models applied to the imputed data. We theoretically analyze how imputation affects disparate impact, and which structural factors contribute to imputation unfairness.
Some (disorganized) thoughts about how more authors have been including statements resembling "we acknowledge that gender is not binary" or "extending our work to non-binary genders is a promising direction for future work" in their papers (1/n)
Had a great time talking with
@TheGoodRobot1
about why I joined
@QueerinAI
, how
@QueerinAI
is shaping AI towards more inclusive and queer futures, why ‘bias’ cannot be seen as a purely technical problem, and why
@QueerinAI
rejected Google sponsorship 😊
I just recalled feedback I got on a research and personal statement submission a while back; it said: “I love your diversity and fairness work! But make sure to take a facts-based approach to these things.” It still bothers me... (1/3)
Want to make fairness and bias mitigation tools accessible to all? Want to contribute to a cutting-edge NLP library? Contribute to allennlp.fairness! I’m happy to answer any questions you have, especially if you’re a newcomer, student, and/or part of a marginalized community.
@arjunsubgraph
did some fantastic work on fairness and bias in AllenNLP, and summed it all up in one big blog post:
Highlight: Use the new BiasMitigatorApplicator to wrap your existing model and check if the new built-in fairness metrics improve.
Super excited about this! The team reached out to
@QueerinAI
and incorporated some of our recommendations for asking for pronouns (), to make the feature more inclusive 💜
🆕 Authors now have the option to add their pronouns to their page on Semantic Scholar.
To set this up, first search for and claim your author page, and then visit Settings to add your pronouns.
We're here to help! Reach out with any questions or feedback.