2021-02-17 10:41:07 the cognitive bros strike again. just to be clear, a "universal", "single" or "standard" notion of "the mind" is synonymous with the Western, cisgendered, white male's mind https://t.co/jje3hFQ7WH 2021-02-16 22:47:18 @tage_rai 2021-02-16 21:33:59 @llaisdy @dimitrisbolis i didn't say "all humans are the same" and I don't get your point about oppression 2021-02-16 21:24:57 just finished recording our @FAccTConference presentation on "Narratives and counternarratives on data sharing in Africa" (paper coming soon) with @Geobaido and just want to share this slide https://t.co/90aZFUqjAD 2021-02-16 21:20:55 RT @blairfrost: @Abebab Currently working on a discursive-performative theory of cognition that addresses exactly this :) 2021-02-16 19:32:49 @IasonGabriel okay, maybe not "every meeting" but more than I should 2021-02-16 19:31:47 I'm tired of theories of cognition that treat cognition in abstract as if all humans are the same regardless of background, culture, political beliefs, race, class, economic status and so on 2021-02-16 19:25:05 @djnavarro I'm so sorry, Danielle sending you much love 2021-02-16 16:50:13 @fisseha_tegegn 2021-02-16 16:49:50 "consensual images" https://t.co/hLYGOuivzQ https://t.co/C5tzB53SWd 2021-02-16 16:31:26 @malk_zameth that or i bang my head on my desk constantly 2021-02-16 16:26:27 welp, I'm in danger of becoming that asshole that drops their work at every meeting 2021-02-16 16:22:35 @Iwillleavenow 2021-02-16 16:18:02 RT @vinayprabhu: (2/n) The hagiography of Google's researchers who have peddled peer-reviewed papers based on these opaque datasets reads l… 2021-02-16 16:17:44 RT @vinayprabhu: A on some of the worries that @Abebab & 2021-02-16 15:11:42 RT @maiheurem: Very excited to share that the project I worked on during my internship @DeepMind got awarded a Distinguished Paper Award @R… 2021-02-16 11:45:04 RT @MogensAgerbo: " I propose to look at human knowing through the lens of loving. We then see that both knowing and loving are existential… 2021-02-16 09:49:38 RT @JMateosGarcia: The principle of requisite variety suggests that ML systems can't grasp social realities that are relational, complex an… 2021-02-16 07:52:37 RT @jackclarkSF: So when Google inevitably has to disclose more details about JFT and tries to make a big song and dance about its work on… 2021-02-16 07:52:30 RT @jackclarkSF: two of the world experts in documenting models and their underlying datasets work(ed) on Google's Ethical AI team - @timni… 2021-02-16 07:52:27 RT @jackclarkSF: It's ironic to me that more and more of Google's papers reference JFT, a secret in-house image dataset. JFT is going to be… 2021-02-16 02:11:59 RT @negar_rz: 11 weeks after my first manager got fired and 5 weeks after my second manager is locked out of work! 2021-02-16 01:30:00 RT @alexhanna: Not sure how Google really expects to rebuild trust in folks working in ethics when we're still dealing with this. Mind you… 2021-02-16 00:47:38 RT @FrankPasquale: Harassment of Black feminists became a blueprint for what would eventually become Gamergate, “which became a blueprint f… 2021-02-16 00:20:32 RT @mmitchell_ai: In other words, even after being a world expert with a long and productive career, I still have not made it to a stage in… 2021-02-16 00:20:09 RT @mmitchell_ai: My career is possible in part because of an intensely honed ability to see and work with patterns. This is arguably why I… 2021-02-15 23:30:54 @TGyateng 2021-02-15 23:30:32 RT @TGyateng: Eloquently written paper which makes clear the need to centre the voices of people most likely to be harmed (marginalised/ lo… 2021-02-15 20:37:54 the number of times i said "fuck" today, mostly while trying to record myself doing a presentation https://t.co/3O88XWQmJJ 2021-02-15 18:50:13 RT @kharijohnson: This https://t.co/NSUofAv7LI 2021-02-15 17:13:25 RT @undersequoias: Roses are red Violets are blue Phrenology was wrong then Your algorithm is too 2021-02-15 17:13:18 @kachangnut omg, I hate it 2021-02-15 17:12:58 @RisuToInu 2021-02-15 17:12:47 @MissIG_Geek 2021-02-15 17:12:40 RT @MissIG_Geek: Required reading for anyone in a DPO role, everyone in ‘AI’ tech, and anyone formulating tech policy on a large scale 2021-02-15 15:48:36 RT @cloudquistador: 5./Quoting from the paper... https://t.co/LTSUpuBhG7 2021-02-15 15:48:04 RT @cloudquistador: 1./Pre-slumberland reading has been devoted to @Abebab's fantastic paper, Algorithmic injustice: a relational ethics ap… 2021-02-15 13:36:06 Trying to record yourself doing a presentation staring into the void makes speaking in front of a large audience (which is already very daunting) feel like a piece of cake 2021-02-15 12:27:37 RT @MogensAgerbo: " [ the Newtonian mechanistic] view fared poorly, however, when it came to capturing the messy, interactive, fluid, and a… 2021-02-15 12:27:33 RT @MogensAgerbo: " automated and standardized solutions to complex and contingent social issues often contribute more harm than good—they… 2021-02-15 12:26:54 RT @iancdavieseval: Note to evaluators https://t.co/wl1j9MXk7n 2021-02-15 12:26:12 this is so key and should be repeated again and again esp within the fields of AI/ML which are based on the very premises that 'the answer to every complex question lies in the data' https://t.co/qunV2GQKHX 2021-02-15 00:46:23 RT @ncgreyy: Happy Valentine’s Day twitter fam! Here’s ‘Ready For Love’ by @indiaarie on the #HatSessions https://t.co/im3EGCMDmW 2021-02-14 21:10:40 RT @mtlaiethics: @Abebab Fixed https://t.co/HUlnarOWPZ 2021-02-14 20:03:48 "Do stupid things faster with more energy" is such an apt phrase that captures the AI field https://t.co/lrHAbctrKW 2021-02-14 19:19:14 Plus, the "left/right leaning" framing is not something that translates into political ideologies globally (it seems very US centric to me) https://t.co/QwaB8z30XT 2021-02-14 18:18:06 This level of blind faith in facial recognition tech is disturbing and there's plenty in my mentions https://t.co/3qmvyGJkZn 2021-02-14 16:37:58 The answer to that is the same as "is it ever okay to bash a toaster?" https://t.co/sMArBnIGwN 2021-02-14 16:35:11 RT @cloudquistador: @Abebab It’s truly a symptom of a type of desperation to retain or revive a supremacist worldview. This ‘research’ and… 2021-02-14 12:47:35 Blocked me. Good! Anyone defending AI phrenology should be ashamed of themselves and go run to hide. https://t.co/BpniDAW11k 2021-02-14 12:37:46 RT @IrisVanRooij: It is SO important that problems without solutions can be published. The alternative is that we think the world has no re… 2021-02-14 12:37:08 This is how discredited pseudoscience moves from the margin to the centre -- by comparing it to "dissent text" & 2021-02-14 12:17:22 @farbandish @rachelcoldicutt @michaelzimmer 2021-02-14 11:58:18 @rachelcoldicutt This makes me soooo happy 2021-02-14 10:37:14 @Anandstweets You're too kind 2021-02-14 10:35:02 @tristanbergh thank you! much appreciate it! 2021-02-14 10:34:27 @rachelcoldicutt thank you, Rachel! 2021-02-14 10:34:12 RT @rachelcoldicutt: “Relational ethics as a down-to-earth practice” is v clear and practical and there’s no excuses, basically. Thanks @Ab… 2021-02-14 10:34:08 RT @rachelcoldicutt: I am obsessed with the Donna Haraway concept of “making good compost” right now, and something about this paper gives… 2021-02-14 10:34:04 RT @rachelcoldicutt: This is a brilliant paper, expressing the needed shift in data ethics from “rational to relational”. And I this: “E… 2021-02-14 09:54:26 RT @AngRaymond: Thank you for this wonderful, insightful paper! https://t.co/wA4YxAXbrc 2021-02-14 09:08:01 I don't think people are trying as hard doing anything as they are trying to resurrect physiognomy. You can't prevent history from repeating if you have no idea of it in the first place & 2021-02-14 00:23:57 RT @GenderRadboud: So many interesting views in this paper. Not only for AI, but for many fields 2021-02-14 00:17:40 actually "nah it don't" is a totally legitimate response to AI phrenology. It's also totally okay to trash it, that's what phrenology deserves AI aided or not https://t.co/WkIT5aO3Ac 2021-02-14 00:06:31 This is why racism is often unquantifiable and why looking for data to measure and quantify racism misses crucial points https://t.co/i39MSnVIus 2021-02-13 23:45:17 RT @black_in_ai: If you're black and are building an AI company or intend to start one in the next 2 years, fill this anonymous survey by F… 2021-02-13 23:35:04 no https://t.co/yP6fKvXeyD 2021-02-13 22:10:13 @alexhanna I'm so sorry Alex 2021-02-13 22:06:41 RT @_CharlesPreston: I wish this was getting national attention. You have eight former child slaves of the Ivory Coast taking on western ca… 2021-02-13 21:08:57 The fact that my the above tweet is full of replies like this is truly disconcerting and depressing https://t.co/GLLY1qkWMX 2021-02-13 20:26:18 This thought is astounding and scary https://t.co/KqYTgA4pLK 2021-02-13 20:23:34 @athundt 2021-02-13 20:23:18 RT @cdemwell: This is an underrated perspective and contains useful language. 2021-02-13 20:22:57 RT @psicopeix: Injusticia algorítmica: un planteamiento sustentado en una ética relacional. El "Machine learning" no es "puro", requiere de… 2021-02-13 20:21:54 @SergeThill @tim_elmo @IgorBrigadir, @ThomasL56304720 and @Ellen_Rushe thought (jokingly) about making a ML model for palm reading 2021-02-13 20:16:54 @quiteclare Thanks, Clare. Xx hope you're well. Xx 2021-02-13 17:54:17 RT @Blackamazon: YESSSSSZZZSSSS I LOVE IT 2021-02-13 17:54:05 @DocDre 2021-02-13 17:53:51 RT @vavacolor: @Abebab “Although such work can constitute part of the remedy, a fundamentally equitable path must examine the wider picture… 2021-02-13 17:53:35 Facial recognition can do no such thing https://t.co/Z2zwiKRKx3 2021-02-13 15:42:44 RT @sniphist: There's a really particular frustration about being a medical historian right now, as so many people say "well history will t… 2021-02-13 15:41:20 RT @daly66: This is incredibly thought provoking about its own discipline, about use of models to interpret complex patterns and processe… 2021-02-13 13:41:02 The major source of challenge with getting this paper accepted was that it mainly offered criticisms and pointed out problems without proposing actionable solutions. But seeing it in print makes me hopeful that the field might start to see critique as a valuable contribution. 2021-02-13 13:39:08 Also, over the last year, this paper has received the most devastating and harsh reviewer comments (I am grateful for editors and friends that helped me deal with it) as well as some loving ones. This version is a result of so many iterations. 2021-02-13 13:31:04 Okay, technically speaking this is not a new paper but the longer version of the paper that won the best paper award at last year's @NeurIPSConf @black_in_ai workshop. It's taken over a year but the full paper is here! https://t.co/SL94zrO1LF 2021-02-13 13:26:49 End Algorithmic injustice: a relational ethics approach https://t.co/5IanGvliFG 2021-02-13 13:25:59 Rethinking ethics is about undoing previous and current injustices to society’s most minoritized and empowering the underserved and systematically disadvantaged. https://t.co/ILlJt3boX6 2021-02-13 13:25:12 Relational ethics encourages us to view data science in general, and the tasks of developing and deploying algorithmic tools that cluster and predict, as part of the practice of creating and reinforcing existing and historical inequalities and structural injustices. https://t.co/uYM89HHEpK 2021-02-13 13:23:20 A data practice that prioritizes understanding over prediction is one that interrogates prior beliefs instead of using the evidence to confirm such belief and one that seeks alternative explanations by placing the evidence in a social, historical, and cultural context. https://t.co/HpubtoR2N8 2021-02-13 13:20:51 The common phrase ‘‘bias in, bias out’’ captures deeply ingrained reductive thinking. Although datasets are part of problem, this commonly held belief relegates deeply rooted societal & 2021-02-13 13:18:02 Given that harm is distributed disproportionately and that the most marginalized hold the epistemic privilege to recognize harm and injustice, relational ethics asks that for any solution that we seek, the starting point be the individuals and groups that are impacted the most. https://t.co/oXVwaW2uSG 2021-02-13 13:16:47 Since knowing is a relational affair, it matters who enters into the knower-known relations. Within the fields of CS and data sciences, the knower is heavily dominated by privileged groups of mainly elite, Western, cis-gendered, and able-bodied white men. https://t.co/YdZwOhJnVF 2021-02-13 13:15:15 What might a version of ethics—in the context of data practices and algorithmic systems— that takes the core values of enactive cognitive science and Afro-feminist epistemology (described in the two preceding subsections) as its foundations look like? https://t.co/eyg3htgys5 2021-02-13 13:13:45 In a similar vein to Afro-feminist thought, the enactive cognitive science theory of participatory sense-making advocates for an active and engaged knowing rooted in our relating. https://t.co/g4KLRfW6sA 2021-02-13 13:12:17 From the core arguments of Afro-feminist epistemology, it follows that concepts such as ethics & 2021-02-13 13:09:23 Data science, & 2021-02-13 13:06:40 The rational worldview, the quintessential orthodoxy for Western thought, is exemplified by the deep contention that reason & 2021-02-13 13:04:08 At the heart of relational ethics is the need to ground key concepts such as ethics, justice, knowledge, bias, & 2021-02-13 13:01:41 Relational ethics, at its core, is an attempt to unravel our assumptions and presuppositions and to rethink ethics in a broader manner via engaged epistemology in a way that puts the needs and welfare of the most impacted and marginalized at the center. https://t.co/nsPRRBzSGr 2021-02-13 13:00:18 Algorithmic injustice: a relational ethics approach https://t.co/5IanGvliFG New paper fresh off the press! https://t.co/2mkAY9Gp6P 2021-02-13 08:48:04 RT @_omalicha__: https://t.co/oZZJMu2dYK 2021-02-13 08:44:57 RT @Mehtabkn: CFP for 2021 CVPR workshop on “Beyond Fairness: Towards a Just, Equitable, and Accountable Computer Vision” organized by @cep… 2021-02-13 07:07:42 @shengokai https://t.co/fW3sXAejnu 2021-02-12 20:39:13 @ruha9 https://t.co/geTClqTGnF 2021-02-12 08:28:12 It was so great to hear @ruha9 speak again. Every word she uttered was pure gold! Thanks for sharing your wisdom, professor Benjamin! https://t.co/WYlBMgRJMd 2021-02-12 00:08:39 RT @carolecadwalla: Not. A. Single. News. Outlet. Has. Reported. This. 2021-02-11 23:55:26 "Google’s advertising system allowed employers or landlords to discriminate against nonbinary and some transgender people" https://t.co/z2Fui6hAAg 2021-02-11 23:19:34 RT @Limerick1914: When Black abolitionists such as Frederick Douglass and Charles Lenox Remond toured Ireland and Britain in the mid-19th c… 2021-02-11 22:13:23 on the 1 hand these translations make me laugh b/se they are often way off the mark by a mile (pls don't ask me for the actual translation) & 2021-02-11 21:32:07 @ZemenuYA I clicked on "translate" out of curiosity and I died https://t.co/X4dFXY6nHq 2021-02-11 21:30:14 RT @TrendsParasitol: Happy International Day of Women and Girls in Science! Smashing the stereotypes against African women scientists @wome… 2021-02-11 21:27:34 @timnitGebru i'm sorry 2021-02-11 21:24:40 RT @newscientist: It's International Day of Women and Girls in Science - so let's celebrate some of our scientific heroes. #womeninscience… 2021-02-11 21:11:54 today is #InternationalWomenInScienceDay and shout out to all the Black women who supported and uplifted me and created the ecology that allowed me to pursue the research that I do (you know who you are!) and here's to paying it forward https://t.co/IXHzbtKri5 2021-02-11 21:03:11 the essence of toxic culture https://t.co/Z9odh5wnjB 2021-02-11 20:57:58 @somesheep 2021-02-11 20:57:26 RT @somesheep: @madplatt @zeroXten Didn't know of identity theory, thank you for sharing. Will explore. I would reference Bourdieu and Ubun… 2021-02-11 20:56:00 RT @ucdcswomen: On this #InternationalDayOfWomenAndGirlsInScience the @UN will address the theme “Women Scientists at the forefront of the… 2021-02-11 20:55:38 @o_guest 2021-02-11 20:55:02 @jessegene2 it's not science 2021-02-10 23:47:25 RT @datasociety: Please join us in welcoming @bsmith13 as @datasociety's first Policy Director! https://t.co/QE83qsq6FE 2021-02-10 23:22:56 @ErinWestgate @cconrymurray Both 2021-02-10 22:31:01 RT @ChikeJeffers: Journey with us to the early part of the 20th century as we look at the American Negro Academy in order to appreciate the… 2021-02-10 21:39:52 @lastpositivist Oh dear God 2021-02-10 20:59:15 @k8em0 @timnitGebru @TwitterSafety I went to report him again and found he's blocked me. Still reported him again anyway https://t.co/5mRcTrlSJL 2021-02-10 20:44:02 RT @ChinasaTOkolo: My recent work is featured in @VentureBeat! Happy to have such a great group of collaborators who provided significant i… 2021-02-10 20:41:24 RT @WideBlackEyes: fellas is it gay to survive a car crash? https://t.co/XazXqXolTp 2021-02-10 20:39:44 This! https://t.co/69K0z6soRv 2021-02-10 20:27:59 @heciritoglu @Mirko_Torrisi I like the idea of power napping for a week or two 2021-02-10 11:17:55 i have so much to do, i don't know what to do https://t.co/xKKZDTftYi 2021-02-10 00:45:57 RT @MairavZ: Almost 60 percent of Israelis have received one shot of the vaccine. Practically no Palestinians have. No matter what your pol… 2021-02-10 00:31:37 @alexhanna https://t.co/RuLQWUjeqM 2021-02-10 00:30:25 RT @alexhanna: But have you told the people about the benefits of large language models???? 2021-02-10 00:21:09 RT @FrankPasquale: Part of a great tweet thread on a critically important paper on large language models (LMs). 2021-02-10 00:20:47 RT @BiancaHatin: As a white, western, able-bodied woman who benefits from existing hegemonic structures implicit in academia this is a very… 2021-02-09 12:57:05 word! https://t.co/5P9UFAGdzS 2021-02-09 12:56:51 RT @o_guest: My favourite part about our paper [@Abebab] is the potential for a better future by "grappling with our fields’ histories and… 2021-02-09 11:13:35 RT @wsisaac: So excited this is now public! This paper offers an excellent survey of the harms LGBTQ+ communities face from AI systems as w… 2021-02-09 11:12:55 @o_guest And you, Olivia 2021-02-09 10:13:52 Me and @o_guest's paper 'Towards Decolonising Computational Sciences' is out today with 'Women, Gender & 2021-02-09 10:07:51 RT @rohanjay: Beautifully referenced & 2021-02-09 10:07:35 RT @abmakulec: @Abebab @Twitter I would like a strong arm with a bandaid on it please! Much more upbeat (and less fear inducing) than a dri… 2021-02-08 23:38:14 RT @IgorBrigadir: @Abebab @Twitter ^^^ @TwitterDev @TwitterOSS who does Twemoji maintenance? Anyone interested in updating the emoji? cc… 2021-02-08 23:07:05 RT @honestari: Marge Simpson is a black woman. https://t.co/jPOhSj6Xfp 2021-02-08 23:05:39 Hey @Twitter, this is such a horrible emoji for a vaccination. Please change it to something that's more pleasant. 2021-02-08 23:03:31 @cyetain Yeay 2021-02-08 22:54:26 RT @saraflocks: DoorDash funded Prop. 22 to buy their own law so they wouldn't have to give drivers #paidsickdays, #minimumwage, or anythin… 2021-02-08 21:25:33 RT @sdopoku: @zararah @aeonmag @Abebab Such a great read!!! I also recently learned about this paper by @sabelonow which makes a similar c… 2021-02-08 17:45:20 @zararah @aeonmag 2021-02-08 17:44:35 RT @zararah: From 2017, but always relevant – fantastic @aeonmag essay from @Abebab on how ‘a person is a person through other persons’ htt… 2021-02-08 17:44:22 @CaroRowland that annoys the hell out of me 2021-02-08 15:22:57 @fisseha_tegegn @imruwa እንዴት እንደምወድህና እንደማከብርህ ታውቃለህ መቸም 2021-02-08 15:08:29 empty (you can tell from the way they carry it) take-away coffee cups https://t.co/37scz2jKiD 2021-02-08 15:07:01 heartbreaking https://t.co/DvJbrGtnW3 2021-02-08 15:04:06 Since hiring algorithms perpetuate stereotypes, individuals at the top of societal hierarchy will be deemed as "successful" applicants while marginalized folk who are historically excluded from some jobs as "unfit". The crux of automated hiring is automating historical patterns. https://t.co/yjbmqgmRi8 2021-02-08 13:35:50 @wsisaac sent you a message :) 2021-02-08 13:31:21 I have been doing nothing but trying to resolve tech issues with my laptop today and I'm at the end of my wits https://t.co/xdNlHVESCd 2021-02-07 23:56:34 RT @BlackSocialists: This is the most in-depth Twitter thread on Black American x Irish relations of all time. It is extremely long, but w… 2021-02-07 23:53:30 Someone should do a version of how it actually IS: the darker your skin colour, the less accurately "self-driving" cars recognize you https://t.co/YUcOQRWToM https://t.co/Skhe03kuYr 2021-02-07 19:19:29 @giacomotesio Calm down, Giacomo 2021-02-07 17:42:41 @am_anatiala sending hugs 2021-02-07 17:39:01 18 yrs?! I feel so old https://t.co/faI9o4GWdR 2021-02-07 13:09:24 @BeukelaerSophie beautiful 2021-02-07 10:07:09 a̵n̵t̵i̵ racist = a̵n̵t̵i̵ white racist = white https://t.co/2yBf1TP5gg 2021-02-07 09:57:23 RT @dfreelon: Cool, cool. So can my kids opt out of white history? https://t.co/rsSQ3epoXP 2021-02-07 09:47:04 @WiringTheBrain @tyrell_turing 12 2021-02-07 08:55:51 RT @AfricaFactsZone: Zambia. https://t.co/X21Vbz8lRc 2021-02-07 08:44:11 RT @DerilloEyob: When researching the origins of #chess, historians only look towards India, Persia, Arabia or China. Sänṭäräž ሰንጠረዥ a va… 2021-02-07 08:39:21 RT @ProSyn: The removal of the Tigray People's Liberation Front will bring peace to the region's people and enable closer cooperation withi… 2021-02-07 08:35:27 Where's 0 tho https://t.co/5naqtI3EcY 2021-02-06 22:12:44 RT @mer__edith: We can't count on Obama-era policies to fix BigTech. "This is why the union effort [at Google/in tech] is so positive, wh… 2021-02-06 21:46:36 While simultaneously reinforcing a 'publish or perish' system https://t.co/W79mCqfYRn 2021-02-06 17:57:23 @ShannonVallor sending much love 2021-02-06 17:00:37 @DhikusookaRodg1 yes 2021-02-06 16:59:57 RT @canalpeixebabel: Thread linda! É fácil se empolgar com modelos de linguagem como a GPT3, mas será que é interessante criar modelos mai… 2021-02-06 16:44:20 Happy Saturday (Flowers have become my simple form of self care) https://t.co/direRJPqe5 2021-02-06 14:19:15 I hope to be back with tweet-storms of books that #amreading in the coming weeks. In the meantime, enjoy this thread (the paper that got Google's Ethical AI co-lead fired). https://t.co/TkXIY0ovJ9 2021-02-06 14:10:49 End #amreading On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? By Emily M. Bender, Timnit Gebru, Angelina McMillan-Major, & 2021-02-06 14:07:54 What is also needed is scholarship on the benefits, harms, and risks of mimicking humans and thoughtful design of target tasks grounded in use cases sufficiently concrete to allow collaborative design with affected communities #StochasticParrots 2021-02-06 14:07:12 We call on NLP researchers to carefully weigh these risks while pursuing this research direction, consider whether the benefits outweigh the risks, and investigate dual use scenarios utilizing the many techniques (e.g. those from value sensitive design) that have been put forth. 2021-02-06 14:06:26 In this paper, we have invited readers to take a step back and ask: #StochasticParrots https://t.co/yjxQWsWss3 2021-02-06 14:05:02 We advocate for research that centers the people who stand to be adversely affected by the resulting technology, with a broad view on the possible ways that technology can affect people. #StochasticParrots 2021-02-06 14:04:10 Research and development of language technology, at once concerned with deeply human data (language) and creating systems which humans interact with in immediate and vivid ways, should be done with forethought and care. #StochasticParrots 2021-02-06 14:03:22 Just because a model might have many different applications doesn’t mean that its developers don’t need to consider stakeholders. #StochasticParrots https://t.co/hRpg4G25us 2021-02-06 14:01:47 Simply turning to massive dataset size as a strategy for being inclusive of diverse viewpoints is doomed to failure. #StochasticParrots https://t.co/7wjYdOS0E8 2021-02-06 13:59:56 In order to mitigate the risks that come with the creation of increasingly large LMs, we urge researchers to shift to a mindset of careful planning, along many dimensions, before starting to build either datasets or systems trained on datasets. #StochasticParrots 2021-02-06 13:59:16 Building training data out of publicly available documents doesn’t fully mitigate this risk: just because the PII was already available in the open on the Internet doesn’t mean there isn’t additional harm in collecting it and providing another avenue to its discovery. https://t.co/3QTRqojzdo 2021-02-06 13:57:54 Machine Translation (MT) systems can (and frequently do) produce output that is inaccurate yet both fluent and (again, seemingly) coherent in its own right to a consumer who either doesn’t see the source text or cannot understand the source text on their own. #StochasticParrots https://t.co/xIHmhiArwR 2021-02-06 13:56:11 A third category of risk involves bad actors taking advantage of the ability of large LMs to produce large quantities of seemingly coherent texts on specific topics on demand in cases where those deploying the LM have no investment in the truth of the generated text. https://t.co/wgsnfupH9I 2021-02-06 13:54:55 LMs producing text will reproduce and even amplify the biases in their input #StochasticParrots https://t.co/4PYeDKalHg 2021-02-06 13:48:09 As people in positions of privilege with respect to a society’s racism, misogyny, ableism, etc., tend to be overrepresented in training data for LMs, this training data thus includes encoded biases, many already recognized as harmful. #StochasticParrots https://t.co/HA3IK7gDFE 2021-02-06 13:46:00 Contrary to how it seems when we observe its output, LM is a system for haphazardly stitching together sequences of linguistic forms it observed in vast training data, according to probabilistic info about how they combine, but w/out any reference to meaning:a stochastic parrot 2001-01-01 01:01:01