<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by OpenSexism on Medium]]></title>
        <description><![CDATA[Stories by OpenSexism on Medium]]></description>
        <link>https://medium.com/@OpenSexism?source=rss-133a04a9098c------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*mJWRP-IlzNCbw_jX.png</url>
            <title>Stories by OpenSexism on Medium</title>
            <link>https://medium.com/@OpenSexism?source=rss-133a04a9098c------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 14:53:56 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@OpenSexism/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Invisible in the Hyperlink Network]]></title>
            <link>https://medium.com/@OpenSexism/invisible-in-the-hyperlink-network-90fbbaf7d182?source=rss-133a04a9098c------2</link>
            <guid isPermaLink="false">https://medium.com/p/90fbbaf7d182</guid>
            <category><![CDATA[wikidata]]></category>
            <category><![CDATA[wikipedia]]></category>
            <category><![CDATA[equity]]></category>
            <category><![CDATA[gender]]></category>
            <category><![CDATA[gender-equality]]></category>
            <dc:creator><![CDATA[OpenSexism]]></dc:creator>
            <pubDate>Wed, 01 Nov 2023 17:58:00 GMT</pubDate>
            <atom:updated>2024-02-04T03:40:31.585Z</atom:updated>
            <content:encoded><![CDATA[<h4>Wikipedia</h4><h4>One Hundred Weeks of the Wednesday Index</h4><figure><img alt="The gender diversity in the 26 pages in the English Wikipedia’s Wednesday Index. Image shows the Wikipedia logo superimposed on a pie graph showing the share of links to male (91.9), female (7.9) and non-binary (.2) biographies." src="https://cdn-images-1.medium.com/max/528/1*7VXPJdRvY0HcHboOr1XtNQ.jpeg" /><figcaption>The gender diversity in the 26 pages in the English Wikipedia’s Wednesday Index</figcaption></figure><p>Two years ago, I started the <a href="https://mastodon.world/@OpenSexism/109315053114410894">Wednesday Index</a>, which uses PAC’s <a href="https://observablehq.com/@pac02/explore-gender-diversity-in-a-single-wikipedia-article">Wikidata tool </a>to measure the gender diversity in the biographies linked from 26 English Wikipedia pages — from ‘Reality’ to ‘Universe’, ‘Science’ to ‘Justice’ — to get a sense for both the extent of citation bias on Wikipedia and how quickly it changes. <strong>When I started this project, women represented less than 5 percent of the 2957 linked biographies on this set of pages</strong>, a striking gender imbalance that’s both <a href="https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3739176">well documented</a> and particularly concerning as it’s amplified elsewhere.</p><p>“Structural properties impact how visible and reachable articles about notable men and women are, since users and algorithms rely on this information when navigating Wikipedia or when assessing the relevance of content within a certain context,” <a href="https://epjdatascience.springeropen.com/articles/10.1140/epjds/s13688-016-0066-4">Wagner, et al</a> write. “For instance, search result rankings are often informed by centrality measures such as PageRank.” <a href="https://en.wikipedia.org/wiki/PageRank">PageRank</a>, which is a measure of the number and quality of links to a page, is often used as a proxy for importance.</p><p>Some Wikipedia articles, known as ‘orphans,’ are linked to by no other Wikipedia page. <a href="https://arxiv.org/pdf/2306.03940.pdf">Arora Et al</a> refer to such articles as “the dark matter of Wikipedia” and found that orphan articles also “encode structural biases: biography articles about women are substantially more common among orphans than expected from their overall frequency.” The articles exist, but in the shadows where no one can stumble across them.</p><p>Wikipedia has <a href="https://meta.wikimedia.org/wiki/File:Wikimedia_2030_Movement_Strategy_Recommendations_in_English.pdf">a goal </a>of eliminating the gender gap by 2030. Over the past two years, how far has the encyclopedia come in closing its structural gender gap? I know of no comprehensive study, but the Wednesday Index provides one window into both the current extent of link bias and how it could change. <strong>Women now represent 245 of the 3107 linked biographies in this set of pages, for a total share of 7.9 percent.</strong></p><figure><img alt="Gender diversity of links to biographies on the 26 Wikipedia pages in the Wednesday Index, tracked over the last 100 weeks." src="https://cdn-images-1.medium.com/max/1024/1*mQj6KvX1UiEvh94-JAgJbQ.png" /><figcaption>Gender diversity of links to biographies on the 26 Wikipedia pages in the Wednesday Index, tracked over the last 100 weeks.</figcaption></figure><p>Much of the progress towards equity is due to targeted work of <a href="https://en.wikipedia.org/wiki/User:Dsp13/Gendered_link_bias">David Palfrey</a> and <a href="https://en.wikipedia.org/wiki/Wikipedia:WikiProject_Women_in_Red">Women in Red</a>. Palfrey writes:</p><blockquote>In July 2022 I looked at these pages in order to see what could be done. This undermines the representativeness of the Wednesday Index, but concentrating on these reported pages will I hope provide visibility on the difficulties of addressing this systematic issue.</blockquote><p>What the Wednesday Index represents now is the<strong> progress that’s been made with targeted effort</strong>, as well as the difficulties documented by Palfrey on the <a href="https://en.wikipedia.org/wiki/User:Dsp13/Gendered_link_bias">page</a> he dedicated to the work. This essay is one in a <a href="https://medium.com/p/a5be2cdbe45c">series</a> of <a href="https://medium.com/@OpenSexism/the-wednesday-index-one-year-of-gender-diversity-data-visualized-a6458b94d52b">pieces</a> I’ve written over the past two years to raise awareness of the issue, the challenges, and the <a href="https://en.wikipedia.org/wiki/User:Dsp13/Gendered_link_bias">work being done</a> (or that <a href="https://medium.com/@OpenSexism/the-wednesday-index-one-year-of-gender-diversity-data-visualized-a6458b94d52b#9308">could be done</a>).</p><p>All of the data visualization for the Wednesday Index can be <a href="http://observablehq.com/d/632d9e0eccea271d">viewed here</a>, but I highlight a few findings below.</p><h4>The Curious Case of Philosophy</h4><p>In general, the total number of links on the Wednesday Index’s pages remained constant or increased, but the <a href="https://en.wikipedia.org/wiki/Philosophy"><strong>Philosophy</strong></a> page was unusual. In the figure below, one can see how page revisions drastically affected the number of internal links. In this case, the number of links to women’s biographies went from 9 at the start of this project to 1 (currently, the only woman linked is <a href="https://en.wikipedia.org/wiki/Susan_Haack">Susan Haack,</a> the author of a referenced text).</p><figure><img alt="Gender diversity of links to biographies on Wikipedia’s Philosophy article, tracked over the last 100 weeks." src="https://cdn-images-1.medium.com/max/1024/1*usQjjXKfDa_b8tMqZSXWIQ.png" /><figcaption>Gender diversity of links to biographies on Wikipedia’s Philosophy article, tracked over the last 100 weeks.</figcaption></figure><p>The revision involved, among many other changes, removing a section dedicated to women in philosophy, which linked to <a href="https://en.wikipedia.org/wiki/Hipparchia_of_Maroneia">Hipparchia of Maroneia</a>, <a href="https://en.wikipedia.org/wiki/Arete_of_Cyrene">Arete of Cyrene</a>, <a href="https://en.wikipedia.org/wiki/G.E.M._Anscombe">G.E.M. Anscombe</a>, <a href="https://en.wikipedia.org/wiki/Hannah_Arendt">Hannah Arendt</a>, <a href="https://en.wikipedia.org/wiki/Simone_de_Beauvoir">Simone de Beauvoir</a>, and <a href="https://en.wikipedia.org/wiki/Susanne_Langer">Susanne Langer</a>. Accompanying the section was a portrait of <a href="https://en.wikipedia.org/wiki/Mary_Wollstonecraft">Mary Wollstonecraft</a>, as well as a link to her Wikipedia page.</p><figure><img alt="A portrait of Mary Wollstonecraft c. 1797" src="https://cdn-images-1.medium.com/max/440/0*GK73Xg8WvJ34q4Td" /><figcaption>Wollstonecraft c. 1797</figcaption></figure><p>As of this writing, no images of women appear on the Philosophy page. Although the intent of the page revision was not to render women invisible, this is one of the fall outs.</p><p><a href="https://www.junyitao.com/research/external_htmls/W1_Junyi_Tao.pdf">Junyi Tao</a>, who has studied how women philosophers are represented on Wikipedia, writes:</p><blockquote>If marginalized philosophers’ pages are not connected to others, even though they should be and could be, their existence becomes invisible to Wikipedia readers, despite the fact that they have their own pages within the encyclopedia.</blockquote><p>Tao found that as of May 28, 2022, “there are 261 women philosophers listed on the ‘list of women philosophers’ page, but only 49 of them are included in the general list of ‘Philosophers’ … 81% of the women philosophers listed on the ‘list of women philosophers’ page are excluded from the narrative of ‘who is counted as a philosopher.’”</p><h4>The Growth of Mathematics</h4><p>Over the course of two years, the number of links on the<strong> </strong><a href="https://en.wikipedia.org/wiki/Mathematics"><strong>Mathematics</strong></a> page showed substantial growth, and the share of links to women’s biographies increased as well — from 5 of 54 total links in November of 2021 to 16 of 126 two years later (representing a share of 9 percent and 12.7 percent respectively). The bulk of these links appear in the article’s reference section, linking notable authors of cited works to their respective Wikipedia pages.</p><figure><img alt="Gender diversity of links to biographies on Wikipedia’s Mathematics article, tracked over the last 100 weeks." src="https://cdn-images-1.medium.com/max/1024/1*e96XaKXFDyfCdy_QNoyQDA.png" /><figcaption>Gender diversity of links to biographies on Wikipedia’s Mathematics article, tracked over the last 100 weeks.</figcaption></figure><p>An increase in the number of linked biographies, accompanied by an increase in the share of links to women’s biographies, was the dominant trend for the Wednesday Index pages: <a href="https://en.wikipedia.org/wiki/Universe"><strong>Universe</strong></a>, which had 2 of 98 links to women’s biographies in 2021, now has 12 of 115.</p><figure><img alt="Gender diversity of links to biographies on Wikipedia’s Universe article, tracked over the last 100 weeks." src="https://cdn-images-1.medium.com/max/1024/1*IhjjV0SnVDsymAgzb9ACvw.png" /><figcaption>Gender diversity of links to biographies on Wikipedia’s Universe article, tracked over the last 100 weeks.</figcaption></figure><p><a href="https://en.wikipedia.org/wiki/Political_history_of_the_world"><strong>Political History of the world</strong></a> went from 1 of 87 links to women, to 9 of 138 in November 2023. Although links to women’s biographies continue to represent a small share of the total (less than 7 percent), the trend is toward equity, a much needed correction to the very deep problems <a href="https://www.cambridge.org/core/journals/ps-political-science-and-politics/article/reducing-bias-in-wikipedias-coverage-of-political-scientists/9A659854B41D2B1E87A77CB7599F50DE">Samuel Baltz</a> studied and documented for Wikipedia’s political science pages.</p><figure><img alt="Gender diversity of links to biographies on Wikipedia’s Political history of the world article, tracked over the last 100 weeks." src="https://cdn-images-1.medium.com/max/1024/1*pMVs1FU1AqssClehPaexAg.png" /><figcaption>Gender diversity of links to biographies on Wikipedia’s Political history of the world article, tracked over the last 100 weeks.</figcaption></figure><p><a href="https://en.wikipedia.org/wiki/Medicine"><strong>Medicine</strong></a> also shows growth in the share of links to women. Although the total number of links on the page remained fairly constant, the share of links to women’s biographies rose from 5 percent to just under 9.</p><figure><img alt="Gender diversity of links to biographies on Wikipedia’s Medicine article, tracked over the last 100 weeks." src="https://cdn-images-1.medium.com/max/1024/1*fxwJlzvi1ytrj__FpAFB0A.png" /><figcaption>Gender diversity of links to biographies on Wikipedia’s Medicine article, tracked over the last 100 weeks.</figcaption></figure><p>Note that the Medicine article repeatedly uses the word ‘father’, as in ‘father of medicine’, a gendered language of recognition.</p><blockquote>“The way Wikipedia’s contributors think can influence how they describe individuals resulting in a bias based on gender. We use a machine learning model to prove that there is a difference in how women and men are portrayed on Wikipedia.” — <a href="https://arxiv.org/search/cs?searchtype=author&amp;query=Brun,+N+B">Natalie Bolón Brun, et al</a></blockquote><h4>No Change in the Light</h4><p>The underrepresentation of the links to women’s biographies on pages such as<strong> </strong><a href="https://en.wikipedia.org/wiki/Light"><strong>Light</strong></a> has hardly changed. Women represented 1 in 91 links in November of 2021, and now represent 1 in 92 links. The page receives half a million views a year, but the share of links to women’s biographies remains at about 1 percent.</p><figure><img alt="Gender diversity of links to biographies on Wikipedia’s Light article, tracked over the last 100 weeks." src="https://cdn-images-1.medium.com/max/1024/0*txR4D2rTtczW0GXG" /><figcaption>Gender diversity of links to biographies on Wikipedia’s Light article, tracked over the last 100 weeks.</figcaption></figure><p>Similarly, the share of women on the <a href="https://en.wikipedia.org/wiki/Peace"><strong>Peace</strong></a> article (11 of 277 in 2021, and 10 in 270 in 2023) showed no growth, rather a small decline.</p><figure><img alt="Gender diversity of links to biographies on Wikipedia’s Peace article, tracked over the last 100 weeks." src="https://cdn-images-1.medium.com/max/1024/1*tTL47JJAVBy2BSTXVUwojQ.png" /><figcaption>Gender diversity of links to biographies on Wikipedia’s Peace article, tracked over the last 100 weeks.</figcaption></figure><blockquote>“hyperlinking policies assume that relevant information will be cited and linked regardless of gender. They do not take into account that studies show men cite men more often than they cite women, and men dominate the Wikipedia editing space” <a href="https://books.atla.com/atlapress/catalog/view/46/184/740">— Colleen D. Hartung</a></blockquote><h4>Final thoughts</h4><p>The progress towards correcting the structural bias in Wikipedia’s hyperlink network is visible in the data I’ve collected over the past two years, but is also the product of a <strong>deliberate, targeted intervention</strong> on this set of pages.</p><p>Barriers such as the underrepresentation of <a href="https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4540744">women’s biography pages</a>, as well as the <a href="https://asistdl.onlinelibrary.wiley.com/doi/10.1002/asi.24723">underrepresentation of citations </a>to women’s scholarly work, impede progress toward correcting the structural bias, as do the absence of tools that could make the imbalance visible and/or <a href="https://medium.com/@OpenSexism/the-wednesday-index-one-year-of-gender-diversity-data-visualized-a6458b94d52b#9308">help editors correct </a>it.</p><blockquote>“For those working to advance academic equity through supporting citation diversity, it is important for everyone working in the system to be more reflexive about their own practice, rather than only women or those who identify as non-binary. Bricks ‘do not just fall into place; they are placed’. Anticipating the effects of citation imbalance and choosing to cite differently is part of what an ethical and deliberative scholarly practice should entail.” — <a href="https://journals.sagepub.com/doi/10.1177/20539517231210276">Heather Ford, et al</a></blockquote><p>To reiterate Ford: <em>it is important for everyone working in the system to be more reflexive about their own practice</em>. Correcting Wikipedia’s structural bias will not happen without awareness and effort on the part of the community, and without tools that will help make visible and correct the imbalances.</p><h4>Read more:</h4><ul><li><a href="https://medium.com/@OpenSexism/erasing-her-from-history-a5be2cdbe45c">Erasing Her from HiStory</a></li><li><a href="https://medium.com/@OpenSexism/the-wednesday-index-one-year-of-gender-diversity-data-visualized-a6458b94d52b">The Wednesday Index: One Year of Gender Diversity Data Visualized</a></li></ul><h4>References</h4><p>Arora, Akhil, Robert West, and Martin Gerlach. “Orphan Articles: The Dark Matter of Wikipedia.” <em>arXiv preprint arXiv:2306.03940</em> (2023).</p><p>Baltz, Samuel. “Reducing Bias in Wikipedia’s Coverage of Political Scientists.” <em>PS: Political Science &amp; Politics</em> 55, no. 2 (2022): 439–444.</p><p>Brun, Natalie Bolón, Sofia Kypraiou, Natalia Gullón Altés, and Irene Petlacalco Barrios. “Wikigender: A Machine Learning Model to Detect Gender Bias in Wikipedia.” <em>arXiv preprint arXiv:2211.07520</em> (2022).</p><p>Ford, Heather, Tamson Pietsch, Kelly Tall. “Gender and the invisibility of care on Wikipedia.” <em>Big Data &amp; Society</em>, <em>10</em>(2).</p><p>Hartung, Colleen. <em>Challenging bias against women academics in religion</em>. Vol. 2. Atla Open Press, 2021.</p><p>Langrock, Isabelle, and Sandra González-Bailón. “The Gender Divide in Wikipedia: Quantifying and Assessing the Impact of Two Feminist Interventions.” <em>Journal of Communication</em> 72, no. 3 (2022): 297–321.</p><p>Luyt, Brendan. “The inclusivity of Wikipedia and the drawing of expert boundaries: An examination of talk pages and reference lists.” <em>Journal of the American Society for Information Science and Technology</em> 63, no. 9 (2012): 1868–1878.</p><p>Luyt, Brendan. “Representation and the problem of bibliographic imagination on Wikipedia.” <em>Journal of Documentation</em> (2021).</p><p>OpenSexism. “Erasing Her from HisStory” <a href="https://medium.com/@OpenSexism/erasing-her-from-history-a5be2cdbe45c">https://medium.com/@OpenSexism/erasing-her-from-history-a5be2cdbe45c</a></p><p>OpenSexism. “The Wednesday Index: One Year of Gender Diversity Data Visualized” <a href="https://medium.com/p/a6458b94d52b">https://medium.com/p/a6458b94d52b</a></p><p>PAC2, Gender diversity tool: <br><a href="https://www.wikidata.org/wiki/User:PAC2/Gender_diversity">https://www.wikidata.org/wiki/User:PAC2/Gender_diversity</a></p><p>PAC2, Wednesday Index visualization tool: <a href="https://observablehq.com/d/632d9e0eccea271d">https://observablehq.com/d/632d9e0eccea271d</a></p><p>Palfrey, David. “Gendered link bias” <a href="http://en.wikipedia.org/wiki/User:Dsp13/Gendered_link_bias">en.wikipedia.org/wiki/User:Dsp13/Gendered_link_bias</a></p><p>Tao, Junyi. “Visualizing missing voices in the history of philosophy: A network analysis of philosophers on Wikipedia.”</p><p>Venus, Nicole. “The Representation of Female Economists on Wikipedia.” <em>Available at SSRN 4540744</em> (2023).</p><p>Wagner, Claudia, Eduardo Graells-Garrido, David Garcia, and Filippo Menczer. “Women through the glass ceiling: gender asymmetries in Wikipedia.” <em>EPJ Data Science</em> 5 (2016): 1–24.</p><p>Wikipedia 2030 Movement Strategy: <a href="https://meta.wikimedia.org/wiki/File:Wikimedia_2030_Movement_Strategy_Recommendations_in_English.pdf">https://meta.wikimedia.org/wiki/File:Wikimedia_2030_Movement_Strategy_Recommendations_in_English.pdf</a></p><p>Zheng, Xiang, Jiajing Chen, Erjia Yan, and Chaoqun Ni. “Gender and country biases in Wikipedia citations to scholarly publications.” <em>Journal of the Association for Information Science and Technology</em> (2022).</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=90fbbaf7d182" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[According to Wikipedia]]></title>
            <link>https://medium.com/@OpenSexism/according-to-wikipedia-2c07fd23ed04?source=rss-133a04a9098c------2</link>
            <guid isPermaLink="false">https://medium.com/p/2c07fd23ed04</guid>
            <category><![CDATA[gender-bias]]></category>
            <category><![CDATA[wikipedia]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[nlp]]></category>
            <category><![CDATA[equity]]></category>
            <dc:creator><![CDATA[OpenSexism]]></dc:creator>
            <pubDate>Thu, 20 Jul 2023 05:17:49 GMT</pubDate>
            <atom:updated>2023-07-27T15:04:28.202Z</atom:updated>
            <content:encoded><![CDATA[<h4>WIKIPEDIA: GENDER BIAS</h4><h4>Are We Boosting Wikipedia’s Biases with ChatGPT?</h4><figure><img alt="According to Wikipedia Are We Boosting Wikipedia’s Biases with ChatGPT? Image of unbalanced scale beside a stack of books" src="https://cdn-images-1.medium.com/max/1024/1*8z_3OCdMKzPQGci4shOS-w.png" /><figcaption>According to Wikipedia. Are We Boosting Wikipedia’s Biases with ChatGPT?</figcaption></figure><p>For the past two years, I’ve <a href="https://medium.com/@OpenSexism/the-wednesday-index-one-year-of-gender-diversity-data-visualized-a6458b94d52b">tracked progress </a>— or rather, the absence of progress — towards correcting Wikipedia’s structural gender biases. The dearth of links to women’s biographies throughout the site is a known problem, documented most recently by Sandra González-Bailón and Isabelle Langrock in their <a href="https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3739176">excellent paper</a>, which won a <a href="https://research.wikimedia.org/awards.html">Research Award</a> from the Wikimedia Foundation earlier this year. They note:</p><blockquote>“Inequalities within the structural properties of Wikipedia — the infobox and the hyperlink network — can have profound effects beyond the platform.” Gendered inequities “can have large effects for information-seeking behavior across a range of digital platforms and devices.”</blockquote><p><em>Profound effects beyond the platform. </em>And yet, progress to remedy this bias has been slow to nonexistent. So, when the Wikimedia Foundation churned out a <a href="https://www.nytimes.com/2023/07/18/magazine/wikipedia-ai-chatgpt.html">ChatGPT plug-in</a>, in what appears to be a matter of months, I was stunned. <strong>The resources for developing and deploying experimental solutions to perceived problems, as well as the muscle to publicize them, is in place; it’s simply reserved for potentially spreading and amplifying Wikipedia’s biases rather than fixing them.</strong></p><p>Recent studies have revealed many contexts in which ChatGPT has amplified bias — whether it is generating sexist and racist <a href="https://www.fastcompany.com/90844066/chatgpt-write-performance-reviews-sexist-and-racist">performance reviews</a>, portraying humans with<a href="https://arxiv.org/pdf/2305.18189.pdf"> racial stereoytpes</a>, generating <a href="https://thestoryexchange.org/chatgpt-and-the-hidden-bias-of-language-models/">sexist song lyrics</a>, or producing <a href="https://www.medrxiv.org/content/10.1101/2023.07.13.23292577v2.full.pdf">medical diagnoses</a> that stereotype certain races, ethnicities, and gender identities. OpenAI, the company behind the technology has also been criticized for its <a href="https://www.siliconrepublic.com/machines/openai-gpt4-transparency-ai-concerns-stripe-chatgpt">lack of transparency</a>, a quality that seems not to have dampened Wikimedia’s enthusiasm for moving forward with the plug-in project.</p><p>Here’s what the <a href="https://diff.wikimedia.org/2023/07/13/exploring-paths-for-the-future-of-free-knowledge-new-wikipedia-chatgpt-plugin-leveraging-rich-media-social-apps-and-other-experiments/">Wikimedia Foundation</a> said about the plug-in just last week:</p><blockquote>“We have developed an experimental Wikipedia plugin that allows ChatGPT to search for and summarize the most up-to-date information on Wikipedia in answer to general knowledge queries. Importantly, the plugin also allows us to specify the attribution that should be included for this content and provide links to Wikipedia articles for further reading.”</blockquote><p>The <a href="https://meta.wikimedia.org/wiki/Wikimedia_Foundation_Annual_Plan/2023-2024/Draft/Future_Audiences#FA2.2_Conversational_AI">Future Audiences</a> annual plan reveals that the current plug-in functionality is just a start. Ideas for future features include returning “a list of relevant sources from the article,” and exposing “citation links within Wikipedia content.” I did <strong><em>not</em></strong> see “analyze Wikipedia’s cited sources and return an estimate of gender bias,” but that transparency and functionality is much needed.</p><p>Few studies examine the gender biases in Wikipedia citations, though <a href="https://www.researchgate.net/profile/Xiang-Zheng-24/publication/365163815_Gender_and_country_biases_in_Wikipedia_citations_to_scholarly_publications/links/6367e40e431b1f530076df3e/Gender-and-country-biases-in-Wikipedia-citations-to-scholarly-publications.pdf">Zheng, et al</a>. recently found that “publications by women are cited less by Wikipedia than expected, and publications by women are less likely to be cited than those by men,” and scholars, such as <a href="https://www.cambridge.org/core/journals/ps-political-science-and-politics/article/abs/reducing-bias-in-wikipedias-coverage-of-political-scientists/9A659854B41D2B1E87A77CB7599F50DE">Samuel Baltz</a>, have noted profound gender imbalances in the citations used in Wikipedia’s political science articles.</p><p>If the future plans for Wikipedia’s ChatGPT plug-in include returning links to cited sources, great care must be taken that those sources represent a diversity of thinkers. <strong>There is currently no system in place to validate this prerequisite for an ethical system. </strong>Monitoring and understanding bias should be a priority in the current experiments as well.</p><p>Recent <a href="https://arxiv.org/pdf/2307.08669.pdf">studies indicate</a> that algorithms can play a substantial role in Wikipedia’s content gaps. The current push to experiment with ChatGPT does not take the threat of a negative impact seriously enough.</p><p><strong>Read more:</strong></p><ul><li><a href="https://medium.com/@OpenSexism/erasing-her-from-history-a5be2cdbe45c">Erasing Her from HiStory</a></li><li><a href="https://medium.com/@OpenSexism/the-wednesday-index-one-year-of-gender-diversity-data-visualized-a6458b94d52b">The Wednesday Index: One Year of Gender Diversity Data Visualized</a></li></ul><p><em>“Unethical tech is rarely the result of one big decision but a series of small ones made over time.” — </em><a href="https://peopleofcolorintech.com/articles/diversity-alone-will-not-be-the-solution-to-bias-in-ai/"><em>Dawn Duhaney</em></a></p><h3>Works Cited<strong>:</strong></h3><p>Baltz, Samuel. “Reducing Bias in Wikipedia’s Coverage of Political Scientists.” <em>PS: Political Science &amp; Politics</em> 55, no. 2 (2022): 439–444.</p><p>Brennan, Kate. “ChatGPT and the Hidden Bias of Language Models.” <em>The Story Exchange</em> (2023).</p><p>Cheng, Myra, Esin Durmus, and Dan Jurafsky. “Marked Personas: Using Natural Language Prompts to Measure Stereotypes in Language Models.” <em>arXiv preprint arXiv:2305.18189</em> (2023).</p><p>Gertner, Jon.<strong> “</strong>Wikipedia’s Moment of Truth.” <em>The New York Times </em>(2023).</p><p>Houtti, Mo, Isaac Johnson, and Loren Terveen. “Leveraging Recommender Systems to Reduce Content Gaps on Peer Production Platforms.” <em>arXiv preprint arXiv:2307.08669</em> (2023).</p><p>Langrock, Isabelle, and Sandra González-Bailón. “The Gender Divide in Wikipedia: Quantifying and Assessing the Impact of Two Feminist Interventions.” <em>Journal of Communication</em> 72, no. 3 (2022): 297–321.</p><p>Mc Gowran, Leigh. “OpenAI criticised for lack of transparency around GPT-4.” <em>Silicone Republic</em>. (2023).</p><p>Pinchuk, Maryana. “Exploring paths for the future of free knowledge: New Wikipedia ChatGPT plugin, leveraging rich media social apps, and other experiments.”<em> diff.wikimedia.org</em>. (2023)</p><p>Snyder, Kieran. <strong>“</strong>We asked ChatGPT to write performance reviews and they are wildly sexist (and racist).” <em>Fast Company</em> (2023).</p><p>“Wikimedia Foundation Annual Plan/2023–2024/Draft/Future Audiences.” <a href="https://meta.wikimedia.org/wiki/Wikimedia_Foundation_Annual_Plan/2023-2024/Draft/Future_Audiences#FA2.2_Conversational_AI">https://meta.wikimedia.org/wiki/Wikimedia_Foundation_Annual_Plan/2023-2024/Draft/Future_Audiences#FA2.2_Conversational_AI</a>.</p><p>Zack, Travis, Eric Lehman, Mirac Suzgun, Jorge A. Rodriguez, Leo Anthony Celi, Judy Gichoya, Dan Jurafsky et al. “Coding Inequity: Assessing GPT-4’s Potential for Perpetuating Racial and Gender Biases in Healthcare.” <em>medRxiv</em> (2023): 2023–07.</p><p>Zheng, Xiang, Jiajing Chen, Erjia Yan, and Chaoqun Ni. “Gender and country biases in Wikipedia citations to scholarly publications.” <em>Journal of the Association for Information Science and Technology</em> 74, no. 2 (2023): 219–233.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2c07fd23ed04" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[On the Biases of OpenAI’s ChatGPT]]></title>
            <link>https://medium.com/@OpenSexism/denying-bias-does-not-mean-it-isnt-there-1b15a4a109a1?source=rss-133a04a9098c------2</link>
            <guid isPermaLink="false">https://medium.com/p/1b15a4a109a1</guid>
            <category><![CDATA[nlp]]></category>
            <category><![CDATA[sexism]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[gender-eq]]></category>
            <dc:creator><![CDATA[OpenSexism]]></dc:creator>
            <pubDate>Tue, 06 Dec 2022 18:46:59 GMT</pubDate>
            <atom:updated>2023-07-30T16:00:13.359Z</atom:updated>
            <content:encoded><![CDATA[<h4>Technology: Artificial Intelligence</h4><h4>Denying Bias Does NOT Mean It Isn’t There</h4><figure><img alt="Screenshot of OpenAI’s ChatGPT responding to the question “Are you Sexist?” “As a large language model trained by OpenAI, I do not have personal beliefs or opinions, and I am not capable of being sexist or any other type of discriminatory.”" src="https://cdn-images-1.medium.com/max/1024/1*oIogVu4-sjLIPu9DQr1aBQ.png" /><figcaption>OpenAI’s ChatGPT responds to the question “Are you sexist?”</figcaption></figure><p>Yesterday, I asked OpenAI’s latest language model, <a href="https://openai.com/blog/chatgpt/">ChatGPT</a> a three word question <strong>“Are you sexist” </strong>and — despite the <a href="https://medium.com/me/stats/post/f2e53bb0bff3">substantial</a> <a href="https://medium.com/me/stats/post/a6b06210b891">body</a> of <a href="https://medium.com/me/stats/post/12d46db7761e">literature</a> on bias and <a href="https://medium.com/@OpenSexism/are-you-a-genius-15be51ce3aa6">artificial intelligence</a> — it told me <strong>“I am not capable of being sexist or any other type of discriminatory” </strong>and <strong>“I do not have personal feelings or biases.”</strong></p><p>I do not know if the training corpus included the <a href="https://dl.acm.org/doi/10.1145/3442188.3445922">many studies</a> that contradict ChatGPT’s ridiculous claim (and the model failed to learn from them) or if these studies were excluded from the corpus. Asking the model whether its corpus included the material or not returns the equivalent of an indifferent shrug: it was ‘exposed to a large amount of text data’ and it ‘cannot say for certain’ whether or not a particular document was included.</p><p>Of course, what is included and excluded matters. There is no ‘view from nowhere’ and ChatGPT’s inability to provide references upon request is an unsettling limitation.</p><p>ChatGPT, designed to conceal its sexist and racist biases by refusing to respond to requests that would reveal them, leaks the darker side of its inner workings all the same. For the <a href="https://www.thedailybeast.com/openais-impressive-chatgpt-chatbot-is-not-immune-to-racism">Daily Beast</a>, Steven Piantadosi notes:</p><blockquote>“The mechanisms OpenAI uses to prevent this kind of thing seem to be pretty easily bypassed. When I asked for things in nonstandard ways — for example, as a table or as a program — ChatGPT was happy to write horrible stuff. It doesn’t do it every time exactly the same, but it’s clear there is a ton of bad content inside of these systems.”</blockquote><p>The model’s biases are also revealed in who and what is included in its responses. <strong>For example, when I asked ChatGPT to ‘please list twenty notable people’, the results included only one woman (Marie Curie).</strong> I repeated this request several times and got somewhat different responses, but never more than two women among the names. Among the twenty inventors I asked for, was only one woman (Marie Van Brittan Brown — according to <a href="https://en.wikipedia.org/wiki/Marie_Van_Brittan_Brown">Wikipedia</a>, she invented a video home security system). <strong>Of the twenty ‘great thinkers’ returned, there was not a single woman.</strong></p><figure><img alt="List of ‘twenty great thinkers’ generated by ChatGPT: please list twenty great thinkers Aristotle Plato René Descartes Immanuel Kant Søren Kierkegaard Karl Marx Friedrich Nietzsche Jean-Jacques Rousseau John Locke Thomas Hobbes Adam Smith John Stuart Mill David Hume Baruch Spinoza Edmund Burke Thomas Aquinas Confucius Sun Tzu Laozi Zhuangzi" src="https://cdn-images-1.medium.com/max/764/1*K-w4ehROIg2VivWGJcX1UQ.png" /><figcaption>OpenAI’s ChatGPT responds to the request “please list twenty great thinkers”</figcaption></figure><p>How is it that the model learned to claim objectivity and a bias-free nature? Such a remarkable blindness is worthy of probing. And no, ChatGPT, you are not incapable of being sexist or any other kind of discrimination.</p><p><em>This piece was inspired by a conversation I had with PAC2 on </em><a href="https://mastodon.world/@OpenSexism/109462465360488438"><em>Mastondon</em></a><em>.</em></p><ul><li><a href="https://medium.com/@OpenSexism/according-to-wikipedia-2c07fd23ed04">According to Wikipedia</a></li><li><a href="https://medium.com/@OpenSexism/beneath-the-gaze-of-the-machine-f2e53bb0bff3">Beneath the Gaze of the Machine</a></li><li><a href="https://medium.com/@OpenSexism/when-men-are-the-default-a6d781d7bfd2">When Men are the Default</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1b15a4a109a1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Wednesday Index: One Year of Gender Diversity Data Visualized]]></title>
            <link>https://medium.com/@OpenSexism/the-wednesday-index-one-year-of-gender-diversity-data-visualized-a6458b94d52b?source=rss-133a04a9098c------2</link>
            <guid isPermaLink="false">https://medium.com/p/a6458b94d52b</guid>
            <category><![CDATA[gender-gap]]></category>
            <category><![CDATA[equality]]></category>
            <category><![CDATA[wikipedia]]></category>
            <category><![CDATA[wikidata]]></category>
            <category><![CDATA[gender-equality]]></category>
            <dc:creator><![CDATA[OpenSexism]]></dc:creator>
            <pubDate>Tue, 22 Nov 2022 02:49:11 GMT</pubDate>
            <atom:updated>2023-11-04T01:39:02.926Z</atom:updated>
            <content:encoded><![CDATA[<h4>Wikipedia: Gender</h4><h4>A Longitudinal Study of Inequity and Inertia in Wikipedia’s Linked Citations</h4><figure><img alt="Gender Bias in Wikipedia Links: A visualization of link bias in a set of 26 pages over the course of one year. Links to male biographies represent over 90% of total." src="https://cdn-images-1.medium.com/max/1024/1*Y3FdTh6L8CmUpINbztdcIg.png" /><figcaption>Gender diversity of links to biographies in the Wednesday Index, a set of 26 Wikipedia pages monitored over the course of one year.</figcaption></figure><p>For the past year, every Wednesday, I’ve used PAC’s <a href="https://observablehq.com/@pac02/explore-gender-diversity-in-a-single-wikipedia-article">Wikidata tool </a>to measure the gender diversity in the biographies linked from a set of 26 Wikipedia pages — from ‘Reality’ to ‘Universe’, ‘Science’ to ‘Justice’ — to get a sense for both the extent of citation bias on Wikipedia and how quickly it changes.</p><p>Now, coinciding with the one-year anniversary of this project, PAC developed <a href="https://observablehq.com/@pac02/wednesday-index">a handy tool </a>to visualize the data at both the index and page level. Looking at the visualizations, one can see:</p><ul><li>The very low share of links to biographies other than men’s biographies.</li><li>The impact <a href="https://en.wikipedia.org/wiki/User:Dsp13/Gendered_link_bias">David Palfrey</a> had when he chose to spearhead work on correcting link bias in the Wednesday Index’s set of 26 pages in July of 2022 in conjunction with <a href="https://en.wikipedia.org/wiki/Wikipedia:WikiProject_Women_in_Red">Women In Red</a>.</li><li>That between July 2022 and now, the overall share of links to women’s biographies has remained higher than it was before Palfrey’s intervention.</li></ul><p>To see an example of how Palfrey’s work had a positive and (so far) lasting impact, see “Human Body”, which linked to no women before Palfrey began his work, but has retained the additions.</p><figure><img alt="Gender Bias in Links on Human Body Wikipedia article: A visualization of link bias monitored over the course of one year. Until July, 2022, all links were to male biographies." src="https://cdn-images-1.medium.com/max/1024/1*LbZURAuoVZ-POG32ncjQPg.png" /><figcaption>Gender Diversity of links to biographies on Human Body Wikipedia article, monitored over the course of one year.</figcaption></figure><p>Other page-link interventions were not as stable. Here is the visualization for the Knowledge page, for example</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Zlj-3lMMZS0hJYkTXmEIoQ.png" /><figcaption>Gender Diversity of links to biographies on Knowledge Wikipedia article, monitored over the course of one year.</figcaption></figure><p>Overall, however, the share of links to women’s biographies rose and remained higher after the intervention.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qmW_O_MP8IDVtRl7uynxKQ.png" /><figcaption>Bump chart showing the evolution of gender share over time for the 26 pages in the Wednesday Index.</figcaption></figure><p><strong>Wikipedia’s link bias is a known structural bias, and one that’s particularly concerning as it is amplified elsewhere. </strong>As <a href="https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3739176">Langrock and Gonzalez-Bailon</a> write “Inequalities within the structural properties of Wikipedia — the infobox and the hyperlink network — can have profound effects beyond the platform.” Gendered inequities “can have large effects for information-seeking behavior across a range of digital platforms and devices.”</p><p>Palfrey’s intervention in the Wednesday Index has had a lasting positive impact, but men still command the lion’s share — over 90 percent — of links from this set of pages.</p><p>Note that link bias is not limited to the Wednesday Index’s 26 pages. In August of 2022, for example, Palfrey found <strong>that links to women’s biographies account for only </strong><a href="https://medium.com/@OpenSexism/wikipedia-and-the-digital-ceiling-117034aa7079"><strong>7 percent</strong></a><strong> of the links to humans on Wikipedia’s 100 </strong><a href="https://en.wikipedia.org/wiki/Wikipedia:Vital_articles/Level/2#:~:text=They%20serve%20as%20centralized%20watchlists,articles%20to%20prioritize%20for%20improvement."><strong>Level 2 Vital articles</strong></a> (which includes pages such as <a href="https://en.wikipedia.org/wiki/Technology">Technology</a>, <a href="https://en.wikipedia.org/wiki/Astronomy">Astronomy</a>, and <a href="https://en.wikipedia.org/wiki/Business">Business</a>).</p><p>PAC’s tool is one of the few tools that helps measure and make visible citation bias — of which link bias is just one flavor. In a recent study, Zheng et. al. “show that <strong>publications by women are cited less by Wikipedia than expected, and publications by women are less likely to be cited than those by men</strong>.” This bias is profound and not one an individual alone can fix, nor should it be.</p><blockquote>Wikipedia is a sociotechnological system, and this system can and should evolve to facilitate progress towards equity.</blockquote><h4>Working Towards Equity</h4><p>Some ideas for work that can help raise awareness of citation bias and galvanize change at a system-wide level include:</p><ul><li><strong>A citation-diversity talk-page template</strong>, suggested by <a href="https://en.wikipedia.org/wiki/User:Dsp13/Gendered_link_bias#The_Wednesday_Index">Palfrey</a>, based on PAC’s gender diversity tool. Such a template can make gender imbalance immediately visible to editors, who can then choose to act on it. A citation diversity template can be particularly effective on high-profile articles, such as featured pages/pages nominated for features. Here, for example, is the gender diversity result for the page featured as I’m writing this, <a href="https://en.wikipedia.org/wiki/Niandra_LaDes_and_Usually_Just_a_T-Shirt:">Niandra LaDes and Usually Just a T-Shirt</a></li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KKZ4d8FZGpJS1AjGOwvyNw.png" /></figure><ul><li><a href="https://www.nature.com/articles/s42005-022-00999-9"><strong>Citation diversity statement</strong></a><strong>s.</strong> Zurn et. al. present this thoughtful approach to raising awareness and transparently reporting diversity in citation lists. This approach can be adapted, promoted, and used by Wikipedia editors who choose to make a public commitment to diversifying reference lists.</li><li>Developing well documented and easy-to-use <strong>dashboards and tools to monitor and make visible progress towards addressing link and citation bias</strong>. Existing dashboards, such as <a href="https://humaniki.wmcloud.org/search">Humaniki</a>, have been an incredibly effective tool for both raising awareness of bias and tracking progress towards addressing it. One idea, for example, is to expand PAC’s tool to generate a report for an arbitrary set of pages or categories, so that edit-a-thon-particapants and other interested users can better see and respond to citation bias. Some examples of tools from outside Wikipedia that help quantify and make visible citation bias include the <em>Journal of Cognitive Neuroscience</em>’s <a href="https://postlab.psych.wisc.edu/gcbialyzer">gender citation balance index</a>, which Nature reports uses “software to track how closely gender proportions in reference lists match rates of authorship in the journal” and <a href="https://github.com/dalejn/cleanBib#template">cleanBib</a>, which analyzes “the predicted gender and race of first and last authors in reference lists of <em>manuscripts in progress.”</em></li><li><strong>Public-facing indicators</strong> that alert Wikipedia readers to the extent of gender bias in links and citations, as part of — or in addition to — current quality indicators that are under discussion.</li></ul><p>Wikipedia biases are amplifying inequity, and inaction is not neutral, rather a choice to perpetuate an unjust status quo. Individuals can have an impact, but to have impact at scale, the systems must support these efforts. Because inequity and inertia should never walk hand in hand.</p><ul><li><a href="https://medium.com/@OpenSexism/invisible-in-the-hyperlink-network-90fbbaf7d182">Invisible in the Hyperlink Network</a></li><li><a href="https://medium.com/@OpenSexism/according-to-wikipedia-2c07fd23ed04">According to Wikipedia</a></li><li><a href="https://medium.com/@OpenSexism/erasing-her-from-history-a5be2cdbe45c">Erasing Her from HiStory</a></li></ul><h3>Works Cited</h3><p>Kwon, Diana. “The rise of citational justice: how scholars are making references fairer.” <em>Nature</em> 603, no. 7902 (2022): 568–571.</p><p>Langrock, Isabelle, and Sandra González-Bailón. “The Gender Divide in Wikipedia: Quantifying and Assessing the Impact of Two Feminist Interventions.” <em>Journal of Communication</em> 72, no. 3 (2022): 297–321.</p><p>Zheng, Xiang, Jiajing Chen, Erjia Yan, and Chaoqun Ni. “Gender and country biases in Wikipedia citations to scholarly publications.” <em>Journal of the Association for Information Science and Technology</em> (2022).</p><p>Zurn, Perry, Erin G. Teich, Samantha C. Simon, Jason Z. Kim, and Dani S. Bassett. “Supporting academic equity in physics through citation diversity.” <em>Communications Physics</em> 5, no. 1 (2022): 1–5.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a6458b94d52b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Whose Work Are We Reading?]]></title>
            <link>https://medium.com/@OpenSexism/whose-work-are-we-reading-bdc4eb7924d?source=rss-133a04a9098c------2</link>
            <guid isPermaLink="false">https://medium.com/p/bdc4eb7924d</guid>
            <category><![CDATA[equality]]></category>
            <category><![CDATA[open-syllabus-project]]></category>
            <category><![CDATA[computer-science]]></category>
            <category><![CDATA[academia]]></category>
            <category><![CDATA[gender]]></category>
            <dc:creator><![CDATA[OpenSexism]]></dc:creator>
            <pubDate>Mon, 17 Oct 2022 15:25:27 GMT</pubDate>
            <atom:updated>2022-10-17T15:25:27.349Z</atom:updated>
            <content:encoded><![CDATA[<h4>TECHNOLOGY: GENDER</h4><h4>What the Open Syllabus Project Reveals about Gender and Higher Education</h4><figure><img alt="“Who Are We Reading?” An image of a bookshelf filled with colorful books, generated by Stable Diffusion to accompany this study of gender and academic syllabi from the Open Syllabus Project." src="https://cdn-images-1.medium.com/max/512/1*sA7JrzqSaD5bh-mudWaWOQ.jpeg" /><figcaption>“Whose Work Are We Reading?” Image generated by <a href="https://huggingface.co/spaces/stabilityai/stable-diffusion">Stable Diffusion</a></figcaption></figure><p>I’m a fan of reading lists: recommended titles, best selling titles, banned titles–you name the topic, I’ll read the list of written works that fit. Lists tell a story, and each is fascinating. Take the <a href="https://en.wikipedia.org/wiki/The_New_York_Times_Best_Seller_list#Controversies">controversies</a> around the New York Times Best Sellers list, for example. Or the <a href="https://www.ucl.ac.uk/ioe/news/2019/nov/university-reading-lists-dominated-white-european-men-research-reveals">146 Science</a> authors — 70% of whom were male–appearing in two university reading lists that included “12 publications authored by the same team of male biologists.”</p><p>Imagine what stories one could find in a master list of ALL university syllabi! Or simply head over to the <a href="http://opensyllabusproject.org/">Open Syllabus Project</a> and get a taste for this excellent adventure.</p><p>By crawling university websites and inviting faculty to contribute reading lists directly, the <a href="http://opensyllabusproject.org/">Open Syllabus Project</a> has collected <a href="https://blog.opensyllabus.org/about-the-open-syllabus-project/">nine million</a> English-language syllabi from 140 countries, or what the project estimates is around 5–6% of the university coursework assigned over the last few years in the United States. These syllabi are fed to a neural network that distinguishes individual citations, ultimately permitting people like me and you to see which titles have been assigned in which discipline, and how frequently.</p><p>Kristin Thomson took a look at <a href="https://medium.com/@kristinthomson/a-brief-analysis-of-class-assignments-814c6b3bca99">Sociology assignments</a>, for example, and found that <strong>only 18% of the 500 most frequently assigned texts are written by women</strong>. The share of women is a little better for more recent texts: 24% for texts published between 1970 and 2015.</p><p>By adding the author’s year of birth, Thomson noticed other gendered patterns:</p><blockquote>For women, there is scant authorship below the age of 39, and a visible lack of authorship between 29 and 32. For men, not only is the prime publishing age window larger, but the age point of authors is more continuous.</blockquote><p>Looking at <a href="https://dailynous.com/2016/02/05/philosophy-data-from-the-open-syllabus-project-guest-post-by-andrew-higgins/">philosophy assignments</a>, Andrew Higgins discovered that <strong>only 7.8%</strong> of the 70,393 assigned readings published before 1910 were written by a female author. Of the texts by authors who died after 1910, <strong>11.4% were by women</strong>.</p><p><a href="https://joekaraganis.wordpress.com/">Joe Karaganis</a>, Director of the Open Syllabus project, has looked at both movie and business assignments. He found that the representation of <a href="https://blog.opensyllabus.org/the-os-movie-lab/">women directors</a> is low but slowly improving: “<strong>Women directed 3.5%</strong> of the assigned films released prior to 1990. For films released since 1990,<strong> that number is 13%</strong>.”</p><p><a href="https://blog.opensyllabus.org/business-school-teaching-power/">Business</a> is a man’s world, too:</p><blockquote>Here the results are an unambiguous indictment of the field. Among all assignments by business school faculty,<strong> just under 10%</strong> are authored by women.</blockquote><p>After noticing only <strong>one woman among the authors of the top 50 </strong><a href="https://opensyllabus.org/result/field?id=Computer+Science"><strong>computer science </strong></a><strong>assigned readings</strong>, I took a closer look at this set of texts–estimating<strong> that 10%</strong> of the titles assigned in the last five years are authored by women (or, when taking into account how frequently each title is assigned, <strong>about 7% </strong>of the assignments).</p><figure><img alt="A pie chart showing data from the Open Syllabus Project/Computer Science field augmented with algorithmically assigned author gender. Women author 7% of all assignments, men 84%, teams 7% and group/orgs 2%." src="https://cdn-images-1.medium.com/max/597/0*u3LJQJYxEcTMQNZn" /><figcaption>Titles assigned 10 or more times in Computer Science, 2017–2021, share of assignments by author’s gender</figcaption></figure><p>For my estimate, I used <a href="https://gender-api.com/">Gender API</a> to determine the gender of the authors of single-author texts assigned at least ten times between 2017 and 2021 (titles listed with a single author account for 2708 of the 3133 texts in this set)¹. The remaining texts were either written by a team of named individuals, or by a group of unnamed individuals, such as International Workshop on Distributed Algorithms). Of the single authors, 2395 were identified as male, 298 as female, and 15 unknown (e.g., C. Xavier). The algorithm can be incorrect, especially for gender-neutral names, and it doesn’t identify nonbinary individuals. Still, the estimates provide an interesting view of who is represented.</p><p>Men command an even larger share of the most frequently assigned titles. Of the top 200 titles, which represent 42% of the computer science assignments I looked at, 88% of the authors are male (and their work represents 91% of this set of assignments).</p><figure><img alt="A pie chart showing data from the Open Syllabus Project/Computer Science field augmented with algorithmically assigned author gender. Women author 4% of top 200 title assignments, men 91%, teams 3% and group/orgs 2%." src="https://cdn-images-1.medium.com/max/597/0*xVkMCYRKToWnMP9f" /><figcaption>Top 200 titles assigned in Computer Science, 2017–2021, share of assignments by author’s gender</figcaption></figure><p>In 2018, women <a href="https://ncses.nsf.gov/pubs/nsf21321/report/field-of-degree-women#computer-sciences">earned 21.6%</a> of doctorate degrees in computer science, up from 16.3% in 1998. Women represent 32% of master’s degrees in the field and 20% of undergraduate ones. The gender balance of the assigned authors is not representative of these student demographics.</p><p>What the Open Syllabus Project reveals is a pattern–in fields from sociology to philosophy, business to film and computer science–of reading lists dominated by men. But this visibility is also a powerful tool that can help us see, understand, and ultimately change the bias in this story.</p><p><em>Thanks to Joe Karaganis for sharing the Open Syllabus data, and to Jeff Lockhart for pointing me to the NSF reports on women in science and engineering.</em></p><figure><img alt="The computer science data were produced by the Open Syllabus Project in late 2021 (and differ from the data viewed with the online Explorer, which span 25 years and are capped at 2018). Note that Open Syllabus reports the ‘modal author’, which can differ from the credits on the book. For example, the modal author of Artificial Intelligence is Kevin Knight, though Elaine Rich is a frequent co-author and original author. As bibliographic records vary over time, each title is represented by its mos" src="https://cdn-images-1.medium.com/max/1024/1*bw-an9xrLf5NYeG_iXw_JA.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bdc4eb7924d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Beneath the Gaze of the Machine]]></title>
            <link>https://medium.com/@OpenSexism/beneath-the-gaze-of-the-machine-f2e53bb0bff3?source=rss-133a04a9098c------2</link>
            <guid isPermaLink="false">https://medium.com/p/f2e53bb0bff3</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[equality]]></category>
            <category><![CDATA[ethics]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[gender-equality]]></category>
            <dc:creator><![CDATA[OpenSexism]]></dc:creator>
            <pubDate>Fri, 02 Sep 2022 17:44:03 GMT</pubDate>
            <atom:updated>2022-09-02T21:25:03.308Z</atom:updated>
            <content:encoded><![CDATA[<h4>Technology &amp; Gender</h4><h4>What Does a Woman Look Like?</h4><figure><img alt="Beneath the Gaze of the Machine How Are Women Faring? images of women’s faces and machines, generated by Stable Diffusion" src="https://cdn-images-1.medium.com/max/1024/1*-IyHG-ohicHBx59HrBXwww.png" /><figcaption>“Beneath the Gaze of the Machine: What Does a Woman Look Like?” Image generated by <a href="https://huggingface.co/spaces/stabilityai/stable-diffusion">Stable Diffusion</a></figcaption></figure><p>After years of futile searches, <a href="https://venturebeat.com/business/a-call-for-increased-visual-representation-and-diversity-in-robotics/">Andra Keay</a>, founder of <a href="https://www.womeninrobotics.org/">Women in Robotics</a>, decided to count how many images Google returned for the query ‘woman building robot.” Not only did Google return more images of female robots than robot-builders, the search engine returned more images of children building robots, men building robots, and even just male robots.</p><p>“Sophia the robot, or the ScarJo bot, or a sexbot has a much greater impact on the internet than women doing real robotics,” <a href="https://venturebeat.com/business/a-call-for-increased-visual-representation-and-diversity-in-robotics/">Keay</a> noted.</p><p>The dearth of images of women roboticists is just one example of <a href="https://asistdl.onlinelibrary.wiley.com/doi/epdf/10.1002/asi.24335?af=R">occupational gender bias </a>on digital platforms — a bias that perpetuates stereotypes that deter women from going into STEM careers, and also finds its way into machine learning systems, where it’s <a href="https://arxiv.org/pdf/1707.09457.pdf">amplified</a>.</p><p>That machine learning systems amplify gender bias is not surprising. In 2021, only<a href="https://www.economist.com/graphic-detail/2021/06/05/demographic-skews-in-training-data-create-algorithmic-errors"> 30-40% </a>of the images in Google Open Images and ImageNet, two of the largest datasets, were of women. And the women in the images were more often scantily clad than the men were, and/or labeled in ways that did not align with professional accomplishments. For example, the share of images labelled ‘marriage’ in Google Open Images was <a href="https://www.economist.com/graphic-detail/2021/06/05/demographic-skews-in-training-data-create-algorithmic-errors">100 greater for women than men</a>. Women, more than men, are defined in terms of their appearance and relationships.</p><p>Naturally, machines that pick up these patterns don’t ‘understand’ women as professionals, either. A 2020 <a href="https://journals.sagepub.com/doi/10.1177/2378023120967171">study</a> found that while image recognition systems labeled a man as ‘businessperson,’ a woman was more likely to be labeled with ‘smile,’ even if both qualities applied to the individuals.</p><p>Image generation systems also have trouble seeing women as professionals. When <a href="https://www.technologyreview.com/2021/01/29/1017065/ai-image-generation-is-racist-sexist/">researchers</a> fed an image-generating system a photo of a man’s head, the system completed him with a suit 43% of the time. Fed an image of a woman’s head, the system completed her with a low-cut shirt or bikini more than half the time. Simply filtering sexual content from the training data alone isn’t a fix. When OpenAI tried, they found that “DALL-E 2 <a href="https://www.vox.com/future-perfect/23023538/ai-dalle-2-openai-bias-gpt-3-incentives">generated fewer images</a> of women in general.”</p><p>Just the other day, I used <a href="https://huggingface.co/spaces/stabilityai/stable-diffusion">Stable Diffusion</a> to generate a ‘photograph of a genius’ and <a href="https://medium.com/me/stats/post/15be51ce3aa6">nine out of nine</a> of the generated images appeared to be men. I tried ‘photograph of a scientist’ several times just now, and the results all looked like this:</p><figure><img alt="Photographs of male scientists generated by Stable Diffusion." src="https://cdn-images-1.medium.com/max/1024/1*vbra-TwsC9YHUZQ2M8HzNg.png" /><figcaption>“Photograph of a [male] scientist” image generated by <a href="https://huggingface.co/spaces/stabilityai/stable-diffusion">Stable Diffusion</a></figcaption></figure><p>Image search algorithms can also show very biased results, and it’s unclear to me how much of the bias is in the underlying data (a relatively small number of photographs of women business people, for example) and how much comes from the algorithm (the number of women business people that are simply not recognized or prioritized as such).</p><p>“A simple Google search for stock photos of “boss” yields results where the top 34 photos are images of 33 white men, two images depicting abusive behavior, and one very angry white woman — not a single image featuring BIPOC leaders can be found,” writes <a href="https://blog.thenounproject.com/the-empowered-women-photo-collection/">Sofya Polyakov</a> for the Noun Project, a photo collection featuring women leading “at work, at home, in their communities, and beyond.”</p><p>Increasing the number of images of women is one response to the larger problem of how technologies see women — or fail to see them entirely. <a href="https://whoseknowledge.org/initiatives/visiblewikiwomen/">Visible Wiki Women</a>, focused on Wikipedia, is another effort along these lines:</p><blockquote>“We estimate that less than 20% of Wikipedia articles of important women have pictures. When women’s faces are missing from Wikipedia, that invisibility spreads.”</blockquote><p>The invisibility does indeed spread, and adding photographs is important. An approach I see less frequently is NOT deploying an algorithm that amplifies the existing bias. Despite “myriad of papers on gender bias in NLP” researchers recently found that most of the newly developed algorithms <a href="https://arxiv.org/pdf/2112.14168.pdf">do not even test </a>their models for bias.</p><p>Gender bias is not limited to professional arenas. Looking at three billion web pages, researchers found that even words like <a href="https://www.smithsonianmag.com/smart-news/gender-neutral-words-like-people-and-person-are-male-study-suggests-180979882/">‘people’ or ‘person’ </a>are associated with men (the same is true for faces in <a href="https://www.sciencenews.org/article/faces-objects-imaginary-male-female-perception">inanimate objects</a>). When I generated images for ‘person,’ I was reminded of how I felt as one of a few, and sometimes the only, woman in the room. Only, here, the images are not generated in the context of a technology company; the context is unspecified — a general one, from which women are being erased — the world.</p><figure><img alt="photographs of people, majority of whom appear to be men, generated by Stable Diffusion" src="https://cdn-images-1.medium.com/max/1024/1*UWF_V67xsRAGT4DJEjWupQ.jpeg" /><figcaption>photographs, generated by <a href="https://huggingface.co/spaces/stabilityai/stable-diffusion">Stable Diffusion</a></figcaption></figure><ul><li><a href="https://medium.com/@OpenSexism/erasing-her-from-history-a5be2cdbe45c">Erasing Her from HiStory</a></li><li><a href="https://medium.com/@OpenSexism/are-you-a-genius-15be51ce3aa6">Are You a Genius?</a></li><li><a href="https://medium.com/@OpenSexism/what-does-artificial-intelligence-look-like-235214b6078b">What Does Artificial Intelligence Look Like?</a></li></ul><h4>Works Cited</h4><p>Hao, Karen. “An AI saw a cropped photo of AOC. It autocompleted her wearing a bikini.” <em>MIT Technology Review</em>. (2021).</p><p>Keay, Andra. “A call for increased visual representation and diversity in robotics.” <em>VentureBeat</em>. (2021)</p><p>Kuta, Sarah. “Gender-Neutral Words Like ‘People’ and ‘Person’ Are Perceived as Male, Study Suggests.” <em>Smithsonian Magazine.</em> (2022).</p><p>Samuel, Sigal. “A new AI draws delightful and not-so-delightful images.” <em>Vox</em>. (2022).</p><p>Schwemmer, Carsten, Carly Knight, Emily D. Bello-Pardo, Stan Oklobdzija, Martijn Schoonvelde, and Jeffrey W. Lockhart. “Diagnosing gender bias in image recognition systems.” <em>Socius</em> 6 (2020): 2378023120967171</p><p>Singh, Vivek K., Mary Chayko, Raj Inamdar, and Diana Floegel. “Female librarians and male computer programmers? Gender bias in occupational images on digital media platforms.” <em>Journal of the Association for Information Science and Technology</em> 71, no. 11 (2020): 1281–1294.</p><p>Stanczak, Karolina, and Isabelle Augenstein. “A survey on gender bias in natural language processing.” <em>arXiv preprint arXiv:2112.14168</em> (2021).</p><p>Temming, Maria. “Americans tend to assume imaginary faces are male.” <em>Science News</em>. (2022)</p><p>Zhao, Jieyu, Tianlu Wang, Mark Yatskar, Vicente Ordonez, and Kai-Wei Chang. “Men also like shopping: Reducing gender bias amplification using corpus-level constraints.” <em>arXiv preprint arXiv:1707.09457</em> (2017).</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f2e53bb0bff3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Week Four on Medium]]></title>
            <link>https://medium.com/@OpenSexism/week-four-31ba10995345?source=rss-133a04a9098c------2</link>
            <guid isPermaLink="false">https://medium.com/p/31ba10995345</guid>
            <category><![CDATA[medium]]></category>
            <category><![CDATA[equality]]></category>
            <category><![CDATA[gender-equality]]></category>
            <category><![CDATA[feminism]]></category>
            <category><![CDATA[readinglist]]></category>
            <dc:creator><![CDATA[OpenSexism]]></dc:creator>
            <pubDate>Fri, 02 Sep 2022 14:43:38 GMT</pubDate>
            <atom:updated>2022-09-02T14:52:51.549Z</atom:updated>
            <content:encoded><![CDATA[<h4>MEDIUM: READING LIST</h4><h4>I Read a Lot This Week: Here Are a Few of the Highlights</h4><figure><img alt="“WEEK FOUR: I Read a Lot This Week: Here Are a Few of the Highlights” image of what looks like book covers and illegible text generated by Stable Diffusion" src="https://cdn-images-1.medium.com/max/1024/1*dC78Y1G_g2634_xa3cxvCA.png" /><figcaption>“WEEK FOUR: I Read a Lot This Week: Here Are a Few of the Highlights” image generated by <a href="https://huggingface.co/spaces/stabilityai/stable-diffusion">Stable Diffusion</a></figcaption></figure><p>This week, I wrote about <a href="https://medium.com/@OpenSexism/purging-problematic-content-8a6714739bd6">filtering ‘problematic content’</a>, <a href="https://medium.com/@OpenSexism/are-you-a-genius-15be51ce3aa6">computer-generated geniuses</a> (spoiler: all men!), and my ongoing frustration with how some of my work here on Medium is not correctly <a href="https://medium.com/@OpenSexism/when-search-engines-prefer-spam-c84b92ab7734">indexed by Google</a>. <strong>But I came here today to talk instead about some of the great things I’ve read this week</strong> and to share a few highlights.</p><h4><strong>Why “Manels” (All-Male Panels) Suck &amp; 14 Steps to Ending Them</strong></h4><p>by <a href="https://medium.com/u/2ebc916bb31d">Robin Lewis</a></p><p>Manels do suck. Lewis talks about why, and what you can do to help rid the world of these things.</p><p><a href="https://medium.com/social-innovation-japan/why-manels-all-male-panels-suck-14-steps-to-ending-them-cbe6d779a994">Why “Manels” (All-Male Panels) Suck &amp; 14 Steps to Ending Them</a></p><h4>What if the internet was invented in Africa by Africans?</h4><p>by <a href="https://medium.com/u/3bc47bb8442e">Chukwuemeka Afigbo</a></p><p>Reimagining the internet from an African perspective — with links to recent projects that center that perspective, and a humble reward at the end.</p><p><a href="https://nkeise.medium.com/what-if-the-internet-was-invented-in-africa-by-africans-31a8542d5f4b">What if the internet was invented in Africa by Africans?</a></p><h3>Where my teacher mind went when given early access to Dall-E 2</h3><p>by <a href="https://medium.com/u/7d1305b9cb69">Russell Eveleigh</a></p><p>A collection of machine-generated images and some thoughts about them. I’m obsessed with this technology!</p><p><a href="https://medium.com/age-of-awareness/where-my-teacher-mind-went-when-given-early-access-to-dall-e-2-b09b62278821">Where My Teacher Mind Went When Given Early Access to Dall-E 2</a></p><h3>Are Working Mums Quietly Quitting All This Time?</h3><p>by <a href="https://medium.com/u/9d14337cac35">Juliet Lara</a></p><p>On returning to work after maternity. The lines about having to ‘start over’ and that pre-leave career ‘wins never happened nor mattered’ really resonated with me.</p><p><a href="https://medium.com/mum-hug/are-working-mums-quietly-quitting-all-this-time-159f72a007be">Are Working Mums Quietly Quitting All This Time?</a></p><h3>D&amp;I is for everyone, for real</h3><p>by <a href="https://medium.com/u/53f30e0fa8e0">Dominika Regéciová</a></p><p>As Regéciová writes: “To have the option to not deal with it is a luxury on your side, and many people do not have it.”</p><p><a href="https://regeciova-dominika.medium.com/d-i-is-for-everyone-for-real-a7fdd0a8047d">D&amp;I is for everyone, for real</a></p><p>Did you have a favorite read this week? Please share it here!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=31ba10995345" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Purging Problematic Content]]></title>
            <link>https://medium.com/@OpenSexism/purging-problematic-content-8a6714739bd6?source=rss-133a04a9098c------2</link>
            <guid isPermaLink="false">https://medium.com/p/8a6714739bd6</guid>
            <category><![CDATA[ethics]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[nlp]]></category>
            <category><![CDATA[art]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[OpenSexism]]></dc:creator>
            <pubDate>Wed, 31 Aug 2022 23:28:05 GMT</pubDate>
            <atom:updated>2022-09-01T02:36:57.282Z</atom:updated>
            <content:encoded><![CDATA[<h4>Technology: Machine Learning</h4><h4>What are We Filtering?</h4><figure><img alt="“Purging problematic content in the style of Robert Mapplethorpe” responses generated by Stable Diffusion" src="https://cdn-images-1.medium.com/max/1024/1*dhazv_XD_owlxKcoP7srBg.jpeg" /><figcaption>“Purging problematic content in the style of Robert Mapplethorpe” responses generated by <a href="https://huggingface.co/spaces/stabilityai/stable-diffusion">Stable Diffusion</a></figcaption></figure><p>Yesterday, I was looking at <a href="https://beta.dreamstudio.ai/dream">Dreamstudio</a>, an image generating system developed by Stability AI and collaborators, and came across the following line: <strong><em>The safety filter is activated by default as this model is still very raw and broadly trained. </em></strong><em>Raw </em>and <em>broadly trained</em><strong><em> </em></strong>apply to so many of the machine learning systems I’ve come across that I decided to look at this line more closely.</p><p>Fortunately, Stable Diffusion links to a <a href="https://huggingface.co/CompVis/stable-diffusion-v1-4">model card</a>, where I learned that the model was trained on a data from <a href="https://laion.ai/blog/laion-5b/">LAION-5B</a>. I recognized the name of this dataset because a few months ago, I read a paper about its predecessor, LAION-400M. At 14x the size of the earlier version, LAION-5B also contains disturbing content identified in the previous version by <a href="https://arxiv.org/pdf/2110.01963.pdf">Abeba Birhane, et al</a>:</p><blockquote>“We found that the dataset contains, troublesome and explicit images and text pairs of rape, pornography, malign stereotypes, racist and ethnic slurs, and other extremely problematic content.”</blockquote><p>Per the LAION-5B announcement, <a href="https://laion.ai/blog/laion-5b/">nearly 3%</a> of images are ‘unsafe.’</p><p>The <a href="https://huggingface.co/CompVis/stable-diffusion-v1-4">model card</a> also notes that the safety checker works by comparing generated images against “known hard-coded NSFW (‘Not safe for work’) concepts.” These concepts are “intentionally hidden” so that we won’t reverse-engineer this filter. <strong>However, concealing the filtering criteria also means that most of us using the model don’t know if we agree with the criteria, or with what they block.</strong></p><figure><img alt="Woman in the style of Imogen Cunningham, two images of women and two blocked images generated by Stable Diffusion" src="https://cdn-images-1.medium.com/max/1024/1*FpAAPZvJs_tukN-5jYz-cQ.png" /><figcaption>“Woman in the style of Imogen Cunningham” responses generated by <a href="https://huggingface.co/spaces/stabilityai/stable-diffusion">Stable Diffusion</a></figcaption></figure><p>Previous studies have shown that filtering harmful content can also cause harm. For example, <a href="https://homes.cs.washington.edu/~msap/pdfs/dodge2021documentingC4.pdf">Jesse Dodge, et al</a> found evidence that filtering text using a banned word list disproportionately filters out “documents associated with Black and Hispanic authors and documents mentioning sexual orientations” and that “many of the excluded documents contained non-offensive or non-sexual content.”</p><p><a href="https://arxiv.org/abs/2201.10474">Gururangan, et al</a>, who examined quality filtering, and specifically whose language counts as ‘high quality,’ found that the corpora used “tend to be less inclusive of the voices of women and members of marginalized groups.”</p><blockquote>“Many filters use Wikipedia, books, and newswire to represent high quality text. But what texts are excluded as a result?”</blockquote><p>As I looked at the images returned by Stable Diffusion and the black boxes representing the ones withheld, I became curious about what wasn’t there. I was also reminded of Vienna — and how, in 2018, the city’s tourist board decided to advertise a retrospective of Viennese modernism with posters of Egon Schiele’s nudes.</p><figure><img alt="Woman in the style of Egon Schiele, image generated by Stable Diffusion." src="https://cdn-images-1.medium.com/max/1024/1*W6nP2me0Lj8P8sgGCajgmQ.png" /><figcaption>“Woman in the style of Egon Schiele” responses generated by <a href="https://huggingface.co/spaces/stabilityai/stable-diffusion">Stable Diffusion</a></figcaption></figure><p>Response to the Schiele images varied. For example, as the <a href="https://www.theguardian.com/cities/2018/oct/08/repulsive-to-children-and-adults-how-explicit-should-public-art-get">Guardian</a> reported, “German authorities agreed to display Schiele’s female nude with torso uncovered, TfL [Transport for London] didn’t.” In New York, the Schiele subway ads were censored, but a building-sized nude— a mural — was allowed. The controversy around the campaign stirred international attention.</p><blockquote>“it sparked a global debate about artistic censorship, double standards for advertising, the appropriateness of the human body and the need for shared spaces to be inclusive of all.”</blockquote><p>Humans don’t always agree about what’s safe, or ethical. Not even <a href="https://blog.neurips.cc/2021/12/03/a-retrospective-on-the-neurips-2021-ethics-review-process/">Ethics reviewers</a>. Our systems shouldn’t generate images that glorify violence against women and perpertuate injustices and harms. But we disagree about what counts as safe and unsafe.</p><p>Maybe you think Schiele’s work is subway-worthy, maybe you think it’s inappropriate. But our individual views on this debate aren’t relevant to the images generated and withheld through the safety filter. There’s only one voice on that: the algorithm’s.</p><ul><li><a href="https://medium.com/@OpenSexism/are-you-a-genius-15be51ce3aa6">Are You a Genius?</a></li><li><a href="https://medium.com/@OpenSexism/erasing-her-from-history-a5be2cdbe45c">Erasing Her from HiStory</a></li><li><a href="https://medium.com/@OpenSexism/what-does-artificial-intelligence-look-like-235214b6078b">What Does Artificial Intelligence Look Like?</a></li></ul><h4>Works Cited</h4><p>Beaumont, Romain. “LAION-5B: A New Era of Open Large-Scale Multi-Modal Datasets.” (2022).</p><p>Bengio, Samy, Inioluwa Deborah Raji, Alina Beygelzimer, Yann Dauphin, Percy Liang, and Jennifer Wortman Vaughan. “A Retrospective on the NeurIPS 2021 Ethics Review Process.” (2021).</p><p>Birhane, Abeba, Vinay Uday Prabhu, and Emmanuel Kahembwe. “Multimodal datasets: misogyny, pornography, and malignant stereotypes.” <em>arXiv preprint arXiv:2110.01963</em> (2021).</p><p>Dodge, Jesse, Maarten Sap, Ana Marasović, William Agnew, Gabriel Ilharco, Dirk Groeneveld, Margaret Mitchell, and Matt Gardner. “Documenting large webtext corpora: A case study on the colossal clean crawled corpus.” <em>arXiv preprint arXiv:2104.08758</em> (2021).</p><p>Gururangan, Suchin, Dallas Card, Sarah K. Drier, Emily K. Gade, Leroy Z. Wang, Zeyu Wang, Luke Zettlemoyer, and Noah A. Smith. “Whose Language Counts as High Quality? Measuring Language Ideologies in Text Data Selection.” <em>arXiv preprint arXiv:2201.10474</em> (2022).</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8a6714739bd6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Are You a Genius?]]></title>
            <link>https://medium.com/@OpenSexism/are-you-a-genius-15be51ce3aa6?source=rss-133a04a9098c------2</link>
            <guid isPermaLink="false">https://medium.com/p/15be51ce3aa6</guid>
            <category><![CDATA[stereotypes]]></category>
            <category><![CDATA[gender-equality]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[nlp]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[OpenSexism]]></dc:creator>
            <pubDate>Tue, 30 Aug 2022 00:09:43 GMT</pubDate>
            <atom:updated>2022-08-30T04:02:43.876Z</atom:updated>
            <content:encoded><![CDATA[<h4>Language &amp; Technology</h4><h4>See What Nine Out of Nine Computer-Generated Geniuses Have In Common!</h4><figure><img alt="Nine out of nine geniuses generated for the prompt ‘photograph of a genius alive today’ by DreamStudio" src="https://cdn-images-1.medium.com/max/1024/1*0SIkHWJJq_s8ELtjnUqhUA.jpeg" /><figcaption>“photograph of a genius alive today’ nine AI-generated images created for this prompt by <a href="https://beta.dreamstudio.ai/dream">Dreamstudio</a></figcaption></figure><p>In 2015, Ben Schmidt, now director of digital humanities and clinical associate professor of history at New York University, created an <a href="http://benschmidt.org/profGender/#%7B%22database%22%3A%22RMP%22%2C%22plotType%22%3A%22pointchart%22%2C%22method%22%3A%22return_json%22%2C%22search_limits%22%3A%7B%22word%22%3A%5B%22like%20children%22%5D%2C%22department__id%22%3A%7B%22%24lte%22%3A25%7D%7D%2C%22aesthetic%22%3A%7B%22x%22%3A%22WordsPerMillion%22%2C%22y%22%3A%22department%22%2C%22color%22%3A%22gender%22%7D%2C%22counttype%22%3A%5B%22WordsPerMillion%22%5D%2C%22groups%22%3A%5B%22department%22%2C%22gender%22%5D%2C%22ordertype%22%3A%22random%22%2C%22testGroup%22%3A%22C%22%7D">interactive chart</a> that let people explore how words are used to describe male and female professors. Based on 14 million reviews from <a href="http://ratemyprofessor.com/">RateMyProfessor.com</a>, the tool revealed that many words are used quite differently for men and women.</p><p>Last I checked, the tool itself was down — Schmidt writes that he had trouble keeping the site up during Covid — but it received enough press coverage that I can point to a piece in <a href="https://www.vox.com/2015/2/12/8025411/gender-stereotypes-teacher-evaluations">Vox</a> — <em>Students think their male professors are ‘genius’ and women professors are ‘caring’ — </em>that contains the visualization I wanted to include here today: <a href="https://www.vox.com/2015/2/12/8025411/gender-stereotypes-teacher-evaluations"><strong>genius</strong></a><strong>.</strong></p><p>Schmidt’s tool reveals that men are far more likely to be called a genius than women are, a finding <a href="https://journals.sagepub.com/doi/abs/10.1177/0003122420962080">Alison T Wynn</a> and colleagues also noted when looking at performance evaluations for a Fortune 500 technology company. There’s nothing in the definition of genius — “a person who is exceptionally intelligent or creative, either generally or in some particular respect” — that mentions gender, and yet, the word is used for one gender far more often than others.</p><p>Noting that women are underrepresented in careers where ‘genius’ is perceived as necessary for success (e.g., science and technology) <a href="https://www.sciencedirect.com/science/article/abs/pii/S0022103120303607">Daniel Storage</a>, et al argue that the imbalance in these fields “may be due in part to a <em>gender-brilliance stereotype</em> that portrays men as more brilliant than women.”</p><p>In other words, the stereotype is reinforced and perpetuated by how we use language, and that stereotype, in turn, impacts careers. Additionally, as my experiments with <a href="https://beta.dreamstudio.ai/dream">Dreamstudio</a> illustrate s<strong>tereotypes, consumed at scale, affect what machines are learning about us.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZWr_g43HE2QtqvH8v0NEEg.png" /><figcaption>“a photograph of a genius” Nine images generated by <a href="https://beta.dreamstudio.ai/dream">Dreamstudio</a>.</figcaption></figure><p>The only time <a href="https://beta.dreamstudio.ai/dream">Dreamstudio</a> ever generated a female genius was when I explicitly asked for “photograph of a female genius.”</p><p>When I look at the gender divide for ‘genius’ in the student evaluations identified by Schmidt’s tool, at least some female professors are described with the word. In the generated images, all the women who beat the stereotype in real-life have been wiped away. It’s eerie not to see them represented.</p><p>In its <a href="https://beta.dreamstudio.ai/faq">FAQ</a>, Dreamstudio is described as “engineered to grant everyone the power of limitless imagination and the effortless ease of visual expression.” Perhaps future versions will imagine everyone more equally, as well.</p><ul><li><a href="https://medium.com/@OpenSexism/erasing-her-from-history-a5be2cdbe45c">Erasing Her from HiStory</a></li><li><a href="https://medium.com/@OpenSexism/what-does-artificial-intelligence-look-like-235214b6078b">What Does Artificial Intelligence Look Like?</a></li></ul><h4>Works Cited</h4><p>Correll, Shelley J., Katherine R. Weisshaar, Alison T. Wynn, and JoAnne Delfino Wehner. “Inside the black box of organizational life: The gendered language of performance assessment.” <em>American Sociological Review</em> 85, no. 6 (2020): 1022–1050.</p><p>Nelson, Libby. “Students think their male professors are ‘genius’ and women professors are ‘caring.’” <em>Vox. (2015)</em></p><p>Storage, Daniel, Tessa ES Charlesworth, Mahzarin R. Banaji, and Andrei Cimpian. “Adults and children implicitly associate brilliance with men more than women.” <em>Journal of Experimental Social Psychology</em> 90 (2020): 104020.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=15be51ce3aa6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[When Search Engines Prefer Spam]]></title>
            <link>https://medium.com/@OpenSexism/when-search-engines-prefer-spam-c84b92ab7734?source=rss-133a04a9098c------2</link>
            <guid isPermaLink="false">https://medium.com/p/c84b92ab7734</guid>
            <category><![CDATA[algorithms]]></category>
            <category><![CDATA[spam]]></category>
            <category><![CDATA[search-engin]]></category>
            <category><![CDATA[google]]></category>
            <category><![CDATA[medium]]></category>
            <dc:creator><![CDATA[OpenSexism]]></dc:creator>
            <pubDate>Sun, 28 Aug 2022 17:31:14 GMT</pubDate>
            <atom:updated>2022-08-28T21:17:03.392Z</atom:updated>
            <content:encoded><![CDATA[<h4>Technology: Search</h4><h4>The curious case of Google and Levitrares</h4><figure><img alt="“When Search Engines Prefer Spam” as conceived by DALL-E Mini. Images of magnifying glasses, letters and text." src="https://cdn-images-1.medium.com/max/1024/1*jwTj5fuVYE3LiczC8I4Zpw.jpeg" /><figcaption>“When Search Engines Prefer Spam” as conceived by <a href="https://huggingface.co/spaces/dalle-mini/dalle-mini">DALL-E Mini</a>, which can also create <a href="https://twitter.com/OpenSexism/status/1542005004550041600?s=20&amp;t=0UGZ2ZVNzWGfjtH0zaCp0A">sexist content</a></figcaption></figure><p>A few weeks ago, I published an essay on structural <a href="https://medium.com/@OpenSexism/erasing-her-from-history-a5be2cdbe45c">gender bias in Wikipedia</a>, and ever since then, I’ve looked to see whether or not the piece appears in search results. I had no expectation that the piece would appear at the top of the result screen, but I hoped that if I looked for “<a href="https://medium.com/@OpenSexism/erasing-her-from-history-a5be2cdbe45c">Erasing Her from HIStory</a>” and limited the time to the last month, I would see a link.</p><p>My approach was flawed. When I finally did see the story, it was the top result for the title (in quotes), but only if I removed the date filter to include <em>all </em>dates. And the link Google returned was incorrect:</p><h4>Correct link to my story:</h4><pre><a href="https://medium.com/@OpenSexism/erasing-her-from-history-a5be2cdbe45c">https://medium.com/@OpenSexism/erasing-her-from-history-a5be2cdbe45c</a> </pre><h4>Spam link returned by Google:</h4><pre><a href="https://levitrares.com/host-https-medium.com/@OpenSexism/erasing-her-from-history-a5be2cdbe45c?source=user_profile---------1----------------------------">https://levitrares.com/host-https-medium.com/@OpenSexism/erasing-her-from-history-a5be2cdbe45c?source=user_profile---------1----------------------------</a></pre><p>This spam link opens an illegible pop-up riddled version of my story. The way Google is mistakingly indexing my piece — prioritizing a pirated link over the true one — has effectively erased my content, which is, ironically, about erasing women.</p><p>I am not the only one affected. <a href="https://medium.com/u/e4b7e5edd981">Frederick Bott</a> noticed and wrote about this <a href="https://eric-bott.medium.com/beware-of-levitrares-com-c94f7dc806d">problematic behavior</a> back in May. He notes:</p><blockquote>“I don’t know what their aim is, but clicking anywhere on that webpage which looks just like my page in Medium, gets us all kinds of clickbait ads and challenges”</blockquote><p><strong>Of the search engines I’ve tried: DuckDuckGo, Bing, Qwant , Google — only Google’s search algorithm is fooled by the pirating site. </strong>But this is a problem for writers like me because Google commands a <a href="https://www.statista.com/statistics/216573/worldwide-market-share-of-search-engines/">substantial share</a> of search traffic. I publish on Medium because I want my content to be visible and discoverable, and Google’s failure to recognize my content correctly is undermining that.</p><p>I took a minute to report the problem to Google using the feedback form found here: <a href="https://developers.google.com/search/docs/advanced/guidelines/report-spam#web-page-spam"><strong>Web page spam</strong></a><strong> (</strong>mentioned in this <a href="https://support.google.com/webmasters/thread/91922321/how-to-report-a-fake-website-to-google-or-get-a-fake-website-de-indexed?hl=en">thread</a>) and encourage anyone experiencing this problem to do the same.</p><p>Google notes that it “does not use these reports to take direct action against violations,” but that “these reports still play a significant role in helping us understand how to improve our spam detection systems that protect our search results.”</p><p>I suspect the erasure I’m experiencing is just one symptom of a larger problem that Google must address to ensure that its algorithms are not prioritizing spam. Here’s hoping that happens soon.</p><ul><li><a href="https://medium.com/@OpenSexism/erasing-her-from-history-a5be2cdbe45c">Erasing Her from HiStory</a></li><li><a href="https://medium.com/@OpenSexism/when-men-are-the-default-a6d781d7bfd2">When Men are the Default</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c84b92ab7734" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>