<?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 Greg Dale on Medium]]></title>
        <description><![CDATA[Stories by Greg Dale on Medium]]></description>
        <link>https://medium.com/@gregdale?source=rss-811146aabc07------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*tKNpC7CDWDEI-YBn-KcqIA.jpeg</url>
            <title>Stories by Greg Dale on Medium</title>
            <link>https://medium.com/@gregdale?source=rss-811146aabc07------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 10:53:34 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@gregdale/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[Facebook’s Political Ad Changes: Don’t Let the Good be the Enemy of the Great]]></title>
            <link>https://medium.com/techforcampaigns/facebooks-political-ad-changes-don-t-let-the-good-be-the-enemy-of-the-great-5b58820109f8?source=rss-811146aabc07------2</link>
            <guid isPermaLink="false">https://medium.com/p/5b58820109f8</guid>
            <category><![CDATA[politics]]></category>
            <category><![CDATA[advertising]]></category>
            <dc:creator><![CDATA[Greg Dale]]></dc:creator>
            <pubDate>Thu, 09 Jan 2020 21:26:14 GMT</pubDate>
            <atom:updated>2020-01-09T21:30:54.473Z</atom:updated>
            <content:encoded><![CDATA[<h3>Facebook’s Political Ad Changes: Don’t Let the Perfect be the Enemy of the Good</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7U8LG7VT8EIqgzADFRrHoQ.png" /><figcaption>Unsplash photo by NeONBRAND</figcaption></figure><p><em>January 9, 2020</em></p><p>Facebook’s political ad announcement today is, while by no means perfect, a good step forward and certainly avoids the <a href="https://techcrunch.com/2019/11/08/banning-digital-political-ads-gives-extremists-a-distinct-advantage/">dangerous consequences we publicly warned against</a> when Twitter and Google rashly changed their policies late last year. The announced changes increase transparency around disclosure and user experience, but represent a continuation of the status quo on targeting — <strong>and that’s a good thing</strong>.</p><p>Further restrictions to targeting options, like Google implemented, throw the baby out with the bathwater, hurting new candidates, those who don’t have reality-star-social-media level followings or bottomless checkbooks. The dollars would still flow — but to other, darker, places with limited transparency, as numerous other outlets and experts have noted.</p><p>From <a href="https://www.cnbc.com/2019/12/08/google-policy-change-has-political-advertisers-looking-elsewhere.html">CNBC</a>:</p><blockquote>Google’s new limitations on political advertisements won’t stop digital strategists from buying on their properties, but it will likely drive more dollars toward platforms with less ad transparency to the general public.</blockquote><p>From Alex Stamos, Facebook’s former Chief Security Officer:</p><h3>Squawk Box on Twitter</h3><p>People have over pivoted on the Democratic side pushing against this. I actually think in the end limiting the power of political ads is going to hurt Donald Trump&#39;s challenger,&quot; says @alexstamos on implications of $TWTR $FB $GOOGL policies on political ads ahead of #2020 https://t.co/Vtj3JJSymk</p><p>From the DCCC, DNC, and DSCC:</p><h3>Cole Leiter on Twitter</h3><p>Voters should be empowered with true and accurate information when they go to vote in 2020, but decisions like @Google&#39;s to restrict political advertising fail to help achieve that goal. Today, the @dccc is standing up to big tech companies&#39; reckless policy changes.</p><p>Misinformation and factual accuracy is the real issue. Microtargeting may make it more difficult to spot lies, but it’s a red herring — the amount of discussion and attention paid to this far outstrips its actual effect. As <a href="https://about.fb.com/news/2020/01/political-ads/">Facebook’s Rob Leathern noted today</a>:</p><p><em>“our data actually indicates over 85% of spend by US presidential candidates on Facebook is for ad campaigns targeted to audiences estimated to be greater than 250,000.”</em></p><p>Misinformation in advertising and free organic posts, posted by malign foreign actors but also by the average Jane. Facebook has taken steps to limit or label ‘fake news’, ban deep fakes, but can and should do more about fact-checking political content. It’s certainly a thorny technical and policy problem, but one that’s necessary to incredibly necessary to address, and TFC’s cofounder, <a href="https://techcrunch.com/2019/11/08/banning-digital-political-ads-gives-extremists-a-distinct-advantage/">Jessica Alter, shared some ideas in TechCrunch in November</a>, including adding a “nutrition label” to ads for disclosing their funding, targeting, etc. and warning users if ads have not been checked.</p><p>The consequence of idiosyncratic and inconsistent policy decisions across major digital communication platforms — Facebook, Google, Twitter, TikTok and off-platform ad exchanges — makes for, at best, a confusing landscape for all participants. Voters now confront a confusing landscape on how to assess and consume political content online. Thoughtful, holistic, and universal legislation is needed to address the obligations of platforms to limit misinformation and provide our civic discourse with a consistent and easily-negotiated roadmap for how political information will be communicated in the central medium of our age. In other words, this is the job of the government: not the platforms. So far, they have not lived up to that responsibility.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5b58820109f8" width="1" height="1" alt=""><hr><p><a href="https://medium.com/techforcampaigns/facebooks-political-ad-changes-don-t-let-the-good-be-the-enemy-of-the-great-5b58820109f8">Facebook’s Political Ad Changes: Don’t Let the Good be the Enemy of the Great</a> was originally published in <a href="https://medium.com/techforcampaigns">Tech for Campaigns</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[2020 Initial Priority State Targets]]></title>
            <link>https://medium.com/techforcampaigns/2020-initial-priority-state-targets-f4caee3e6aff?source=rss-811146aabc07------2</link>
            <guid isPermaLink="false">https://medium.com/p/f4caee3e6aff</guid>
            <category><![CDATA[politics]]></category>
            <dc:creator><![CDATA[Greg Dale]]></dc:creator>
            <pubDate>Sun, 22 Sep 2019 17:19:53 GMT</pubDate>
            <atom:updated>2019-09-22T17:19:53.454Z</atom:updated>
            <content:encoded><![CDATA[<h3>Tech for Campaigns’ 2020 Initial Priority State Targets</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WuYGCnQ_f69uUn0UuY8nIw.png" /></figure><p>There are <strong>7,383 state legislative seats</strong> in the United States. State Houses and State Senates govern the policy that affects our daily lives — on healthcare, guns, education, voting rights, women’s right to choose and much more. In most states, the legislatures draw the lines for theU.S. House districts, setting the stage for control of that chamber <strong>through 2032</strong>. For these reasons, Tech for Campaigns focuses its efforts on <strong>flipping state legislatures blue</strong>.</p><p>With these critical issues on the line, a coming redistricting process in 2022, and a dizzying array of thousands of elections next year, focus is a strategic imperative.</p><p>As of August 2019, TFC has <strong>completed 338 projects</strong> for <strong>189 campaigns</strong>, including a major and ongoing deployment to Virginia in 2019, where the State House and State Senate are both two seats away from flipping control this November.</p><p>We’re ready to continue scaling our work, and super excited to announce our 2020 state legislative priority list.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*osz19qv93sdkfDjg2a48vw.png" /></figure><h3>How we choose</h3><p>For federal and statewide races, there are no shortage of district-level predictions available from outlets like FiveThirtyEight — but for state legislatures, <strong>there is no public list</strong>. As in 2018, to guide our efforts, we needed to establish our own point of view and framework — and data was core to doing it.</p><p>Beginning in 2017, a team of TFC data scientists gathered data from a variety of government and nonprofit sources, cleaned it and used this information -from demographics, past election results, and incumbency — to predict Democratic and Republican results in thousands of state legislative districts across the country.</p><p>This project provides Tech for Campaigns with a single baseline to evaluate and prioritize, not just individual races, but entire state legislative chambers where we can help change electoral outcomes. This is the first step. We then talk with the state parties, caucuses and the like and begin to layer in qualitative assessments. These insights powered TFC strategy in 2018, 2019, and are continuing to do so for 2020.</p><p><strong>A huge thank you to the many TFC volunteers that have contributed and are currently contributing in our Build The List data science and engineering effort </strong>— this process would not have been possible without your hard work. (And a<strong> big shout-out to the open source </strong><a href="http://www.openelections.net/"><strong>OpenElections project</strong></a><strong> </strong>for continuing to push to structure and normalize the incredibly messy and disparate precinct-level election results found in all 50 states.)</p><h3>Let’s drive real change in 2020</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*szN0cjBTBjRdCn4n.png" /></figure><p>From 2009 to 2017, as shown above, Democrats lost a huge number of state government bodies.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*J-Qz5fHNysFfHEiMTzZECw.jpeg" /></figure><p>In 2018, Democrats partially rolled back those losses, gaining control of a net 13 chambers across the country. (We’re hoping to add two more from Virginia in November 2019!). For 2020, there’s so much work left to do to finish the job. We’re hungry, excited, and ready to go!</p><p>If you’re considering becoming a <strong>TFC volunteer</strong>, <a href="https://www.techforcampaigns.org/volunteers">sign up now</a> to be the first to hear as projects start up.</p><p>If you are a <a href="https://www.techforcampaigns.org/campaigns"><strong>candidate or on a campaign</strong></a> and would like to chat about becoming part of TFC’s efforts in these states, please let us know.</p><p>If you’re interested in talking about one of the states on the list or adding a state, <a href="/service/mailto:info@techforcampaigns.org"><strong>please reach out</strong></a><strong>.</strong></p><p><em>(Cross-posted from </em><a href="https://www.techforcampaigns.org/2020-priority-state-targets"><em>techforcampaigns.org</em></a><em>)</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f4caee3e6aff" width="1" height="1" alt=""><hr><p><a href="https://medium.com/techforcampaigns/2020-initial-priority-state-targets-f4caee3e6aff">2020 Initial Priority State Targets</a> was originally published in <a href="https://medium.com/techforcampaigns">Tech for Campaigns</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How 380 State Legislative Seats Flipped Blue]]></title>
            <link>https://medium.com/techforcampaigns/how-380-state-legislative-seats-flipped-blue-d823c9c75269?source=rss-811146aabc07------2</link>
            <guid isPermaLink="false">https://medium.com/p/d823c9c75269</guid>
            <category><![CDATA[tech]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[government]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[politics]]></category>
            <dc:creator><![CDATA[Greg Dale]]></dc:creator>
            <pubDate>Fri, 16 Nov 2018 21:02:58 GMT</pubDate>
            <atom:updated>2018-11-16T23:00:20.592Z</atom:updated>
            <content:encoded><![CDATA[<h3>5 Outperforming State Democrats — and the Forces Behind Their Blue Wave</h3><p><em>Co-written with </em><strong><em>Jim Sorenson</em></strong><em>, TFC volunteer data scientist, and the rest of the Build the List team</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xOnZuEfK4F-9OgRpslJkTA.jpeg" /></figure><p>In the 2018 elections, 380 state legislative seats flipped from red to blue as Democratic turnout surged and educated districts abandoned the GOP. While many eyes were on federal and statewide races, <a href="https://medium.com/techforcampaigns/7-reasons-you-should-pay-closer-attention-to-state-politics-94a087f2f56c">for reasons we have previously outlined</a>, the state legislative results were just as important. These flips produced eight new Democratic majorities in state legislatures — crucial for passing legislation that can have real impacts on our daily lives.</p><p><a href="http://www.techforcampaigns.org/">Tech for Campaigns</a> (TFC) concentrates on winning seats in state legislatures; however, with little public opinion polling available for these races, it can be difficult to prioritize our efforts. To help fix this problem, <a href="https://medium.com/techforcampaigns/how-we-used-data-to-select-our-priority-states-for-the-2018-elections-b97e72cda6">we built an election prediction model</a> — using demographics and past election results, generally from 2010–2016 — to provide a baseline for prioritizing state elections. While 2018’s votes are still being counted in Arizona, Florida, and other locales, we obtained 21 different State House election results, covering 1,635 races, to take a first look at how our model did and what we learned. While we knew that the long-term data underlying our model would be unlikely to fully capture the “blue wave” — a distinctly 2018 phenomenon — we were still astounded by how much State House Democrats outperformed our expectations. A few examples:</p><ul><li>In <strong>Iowa</strong> and <strong>Texas</strong>, Democrat vote share exceeded TFC’s expectations by more than 7.5%. In <strong>New Mexico</strong>, <strong>Pennsylvania</strong>, and <strong>Minnesota</strong>, it surpassed it it by more than 6%.</li><li>Even in <strong>Florida</strong>, where the statewide results are currently disappointing Democrats, Democratic vote share was 4.7% higher than expectations.</li></ul><p>The stories of individual candidates’ victories help illustrate the magnitude of this Blue Wave.</p><h3>Outperforming State Democrats Around the Country</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fc9ZnlnNQmhuexvXUihPcg.png" /></figure><p>As we reviewed 2018’s election results, these five winning state Democrats stood out versus expectations while flipping their districts from their Republican opponents:</p><ul><li><strong>Florida House District 47, Anna Eskamani</strong> — community activist and TFC candidate Anna Eskamani won her seat in 2018 with 57.3% of the vote, outpacing 2016’s Democratic result by ten points in this Orlando district!</li><li><strong>Kansas House District 45, Mike Amyx</strong> — in this district to the west and southwest of Lawrence, KS, Mike, a local barbershop owner and former Lawrence mayor, <strong>won with 69.9% of the vote</strong>. In 2016, retiring Republican Representative Thomas Sloan won with a 54.6% share of the vote. Mike’s 2018 results moved the needle by<strong> 24.5%</strong> — and he plans to work on adequately funding public education.</li><li><strong>New Mexico House District 68, Karen Bash</strong> — in northwest Albuquerque, Karen defeated incumbent Republican Rep. Monica Youngblood<strong> 58% to 42%</strong>, after Rep. Youngblood had run completely unopposed in 2014 and 2016 — typically a sign that Democrats don’t think they can win. (In a widely publicized September incident, Rep. Youngblood was convicted of <a href="http://www.santafenewmexican.com/news/local_news/judge-convicts-rep-monica-youngblood-of-aggravated-dwi/article_c598dabe-cdc7-5da8-a403-7823ed11f979.html">aggravated DWI</a>.) Karen plans to work on supporting public schools, improving public safety, and renewing the local economy.</li><li><strong>Iowa House District 55, Kayla Koether</strong> — in 2016, Republican Rep. Michael Bergan won this seat with 57.2% of the vote; in 2018, 29-year old Kayla, running on a progressive platform, fought Bergan to a race that is currently uncalled but has Kayla behind by only nine votes!</li><li><strong>Texas House District 114, John Turner </strong>— attorney and TFC candidate John Turner won this seat in 2018 with 55.6% of the vote; in 2016, the Republican earned 55.7% of the vote, an almost ten point swing in this North Dallas District. John is also an advocate of public education, having once sued the State of Texas on behalf of 88 public school districts to seek greater public school funding.</li></ul><h3>What drove this outperformance?</h3><p>The five candidates above are just a few of the hundreds of inspirational Democratic candidates who flipped their districts. Telling the full story of the work done by thousands of Democratic candidates on the campaign trail this year involves relating our model’s knowledge of past electoral results and demographics to 2018’s results.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/789/0*CiO2JTNTh0hmUugk" /></figure><p>In short, the share of votes garnered by Democrats in 2018 blew away our model’s expectations. In the chart above, every dot above the line represents a district where Democrats outperformed. Even in districts where Democrats ended up far short, they surged — reflective of a new political environment in 2018 where Democrats made an effort to show up everywhere. They ran in previously uncontested and gerrymandered districts, even when the odds looked steep, providing voters with a real choice for the first time in years.</p><h4><strong>Education</strong></h4><p>The chief driver of exceeded expectations was <strong>an increase in Democratic affinity among districts with higher education levels</strong>, which we measure as the percentage of a district’s population with a bachelor’s degree or higher. In 2018, the higher the percent of residents with bachelor’s degrees, the more Democrats outperformed. (<a href="https://www.vox.com/policy-and-politics/2018/11/12/18083014/2018-election-results-turnout">Catalist saw a similar trend in their voter file data (via Vox)</a>. The strength of this effect surprised our model quite a bit — in general we saw that the more people in a district had a bachelor’s degree, the more Democratic candidates over-performed relative to the model’s expectations. Districts in which at least 15% of the population had a bachelor’s degree over-performed our expectations by <strong>6.3%</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/526/0*60_Rk8knk0QjH03O" /></figure><h4><strong>High Voter Turnout</strong></h4><p>A second large driver of results was <strong>increased turnout. </strong>As <a href="https://fivethirtyeight.com/features/the-2018-midterms-in-4-charts/">FiveThirtyEight noted</a>, midterm voter turnout in 2018 was “unprecedented” at 49%, or 22.5% higher than the average 40% voter turnout during midterms. That turnout, often driven by competitive races for Senator or Governor, boosted Democratic candidates at the state legislative level. For example, in Texas, although Beto O’Rourke’s high-enthusiasm campaign for the U.S. Senate ultimately came up short, it helped propel Democrats in the Texas House of Representatives<strong> to pick up a dozen seats, the most since 1964.</strong></p><p>In our races, these additional voters were brimming with Democratic enthusiasm: each additional 1% of voter turnout added an <strong>additional 0.1%</strong> of Democratic performance.</p><p>The 49% Midterm turnout, compared to the typical 40%, alone would drive nearly a full percentage point of difference between expectations and 2018’s wave results. Add in other factors, and we can see where our model was too “conservative.”</p><h3>2018: a strong step forward but much left to do</h3><p>It’s not easy to predict whether the surge in turnout seen in 2018 will continue through 2020, a Presidential election year, where turnout averages are much higher. But the strong uptick for Democrats in more educated districts are a hopeful sign that the strong Democratic positions in these state elections is not just a momentary surge. For our model, 2018 now provides a rich set of data that helps reflect not only trends before 2016, but will help improve our accuracy for elections in 2019 and 2020.</p><p>Flipping 380 legislative seats is a monumental achievement, but still just progress along the way to recovering the more than 900 that Democrats lost between 2008–2016 — especially as we approach the critical 2020 redistricting cycle. Back to work.</p><p><em>Thanks to Susan Payne for design support on this article. Thanks to the current List Project team — Andrew Cron, Taylor Cunnington, Kelly Wiseman, Eddy Liu, Alastair Mackintosh — and long-time contributors Dan Egnor, Amy Skerry-Ryan, Louis Eisenberg, and Matt DeLand. Special thanks to Carl Klarner for his feedback.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d823c9c75269" width="1" height="1" alt=""><hr><p><a href="https://medium.com/techforcampaigns/how-380-state-legislative-seats-flipped-blue-d823c9c75269">How 380 State Legislative Seats Flipped Blue</a> was originally published in <a href="https://medium.com/techforcampaigns">Tech for Campaigns</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How Russian Trolls Won American Hearts and Minds]]></title>
            <link>https://medium.com/techforcampaigns/how-russian-trolls-won-american-hearts-and-minds-30037e1e13b7?source=rss-811146aabc07------2</link>
            <guid isPermaLink="false">https://medium.com/p/30037e1e13b7</guid>
            <category><![CDATA[tech]]></category>
            <category><![CDATA[facebook]]></category>
            <category><![CDATA[advertising]]></category>
            <category><![CDATA[politics]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Greg Dale]]></dc:creator>
            <pubDate>Mon, 21 May 2018 14:01:01 GMT</pubDate>
            <atom:updated>2018-05-21T14:01:01.331Z</atom:updated>
            <content:encoded><![CDATA[<h4>The techniques were not sophisticated, but the messages were on target.</h4><p><em>By Anela Chan and Greg Dale, Tech for Campaigns</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BDSvJbZcj3yWiE75NUEWww.png" /></figure><p>Mixing the likes of Laurel and Yanny with malicious intent, Russian trolls in 2015–17 produced thousands of pieces of divisive clickbait — and Americans consumed it voraciously.</p><p>On May 10, 2018, Democrats on the House Intelligence Committee released 3,500 Facebook and Instagram ads posted between June 2015 and July 2017 by the Internet Research Agency (IRA), a Saint Petersburg-based “troll farm” known for its efforts to influence public opinion through social media. <a href="http://www.techforcampaigns.org/">Tech for Campaigns</a> is focused on the intersection of digital tools and politics, and so we took a deep interest in this data. Joining our own efforts together with a dataset from <a href="http://datafordemocracy.org/">Data for Democracy</a> volunteers, we found that while the IRA’s technological efforts were far from sophisticated, American Facebook users ate up their content.</p><p>Our analysis confirms that the Russian strategy was not simply to advocate Trump or insult Hillary. As the February Mueller indictment of the IRA said, the strategy was to “sow discord,” targeting different American racial, regional, religious and sexual identities. Despite extraordinarily basic targeting and and crude graphic design, the IRA‘s work was very successful, delivering more than <strong>40 million ad views</strong> and achieving <strong>a click through rate almost 10x higher than Facebook’s averages</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*D-skRAbLlygy3FGEcQPDqA.png" /></figure><h3>Advocating for Trump or Hillary was not the focus</h3><p>As the chart below indicates, only a small portion of the ads mention Trump or Hillary. As the 2016 election drew near, the words “Trump” and “Hillary” began to appear more often. Looking at text likely understates the prominence of the election, but does portray that obvious messaging around the presidential election remained a small part of the overall effort.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*STsaOSYpZHxF39NkEqKYzA.png" /></figure><h3>Instead, the IRA took aim at identity politics</h3><p>The IRA’s modus operandi was targeting interest groups: pairing divisive messaging with Facebook’s interest advertising tools, sending users to Facebook Pages that the IRA had created, where ideological echo chambers had been established.</p><blockquote>“They sought to harness Americans’ very real frustrations and anger over sensitive political matters in order to influence American thinking, voting and behavior.” — Rep. Adam Schiff (D-CA)</blockquote><p>They targeted African Americans, conservatives, Latinos, Muslims, LGBTQ groups, and even Texas Secessionists, sending them to pages like “Brown Power”, “LGBT United”, and “Born Liberal”.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9YolffyScmo5ucf3CLnImA.png" /></figure><p>Although most of the activity went towards building these Facebook groups, several ads promoted actual physical events, including <a href="https://www.washingtonpost.com/news/the-switch/wp/2018/05/10/here-are-the-3400-facebook-ads-purchased-by-russias-online-trolls-during-the-2016-election/">a rally that was simultaneously pro-Beyonce and anti-Beyonce</a>, and self-defense classes for <a href="http://money.cnn.com/2017/10/18/media/black-fist-russia-self-defense-classes/index.html">African Americans</a>.</p><h3>The ads concentrated on African Americans and Conservatives</h3><p>The House released more than 3,500 ads. To produce an accounting of this large set, we used a combination of statistical techniques and manual review. The Russians placed <strong>85% of their spend</strong> towards African Americans and Conservatives, mostly across the entire United States, but with concentrations in Texas and “Black Lives Matter cities” like Ferguson and Baltimore. Here’s how their efforts broke down by targeted audience and geographies:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YDVetc48uAont-uVESWoyw.png" /></figure><h3>Nurturing their audience with a smile</h3><p>While other analyses have focused on polarization through outrage, the IRA’s trolls actually used a mix of angry, uplifting, funny, and sarcastic messaging. Commercial advertisers are well-versed in growing and retaining audiences on Facebook through diverse and “snackable” content, and the IRA was no different. Using sentiment analysis, we found the top-viewed and most-clicked ads were largely not angry in tone:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Kcqs7vJrM1hfIuazQ3d1fA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*R5OwTh-2Q6Oh9n96IX7IhA.png" /><figcaption>Two positive-sentiment messages from “Black Matters” and one negative from “Stop A.I.”</figcaption></figure><h3>The IRA’s digital prowess: crude but effective</h3><p>Relative to commercial standards for Facebook advertisers, the IRA’s methods were not sophisticated. They targeted their ads to very broad geographic areas, like the entire United States, with the majority of ads using Facebook’s interest targeting system, which targets ads according to users’ Likes and other expressed preferences. Interest targeting is considered a fairly crude targeting method. In fact, about <strong>6% of ad spend was</strong> <strong>completely untargeted</strong>, leaving these highly specific, polarizing ads to be seen by random US adults, a technique that would be malpractice for professional digital marketers.</p><p>As told by this dataset, the IRA was not investing in sophisticated targeting. They were not targeting specific lists of voters or only swing states. They didn’t need to; the content they produced, their relentless iteration, and the reach of Facebook’s platform enabled their ads to have an extreme amount of engagement.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5I2_ZqT73IhkQopun3eXuA.png" /></figure><p>Efforts at engaging Muslims were notably less successful — a likely reflection of content that didn’t resonate with its audience as well as others did.</p><h3>Strong messaging and relentless iteration paid off</h3><p>That the IRA’s project, paid for in Rubles, was aimed at distributing content to create strife in the United States is not new; what is now clear, though, is that American Facebook users responded incredibly well to it. A click is just a click; it does not equate to a changed mind or a manipulated election, but the high rates of engagement with this polarizing content suggests that it successfully registered with its viewers. Whether by bad intent or good luck, the IRA’s propaganda successfully tapped into a strong vein of divisive identity politics, and Facebook’s reach and algorithmic bias towards engagement enabled it to spread and resonate amongst more than millions of Facebook users.</p><p><em>Special thanks to Ethan Chen and Alex Lindsay, as well as Data For Democracy volunteers </em><a href="https://twitter.com/scottcame"><em>Scott Came</em></a><em> and </em><a href="https://twitter.com/jeffykao"><em>Jeff Kao</em></a><em> for their work parsing and clustering this dataset.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=30037e1e13b7" width="1" height="1" alt=""><hr><p><a href="https://medium.com/techforcampaigns/how-russian-trolls-won-american-hearts-and-minds-30037e1e13b7">How Russian Trolls Won American Hearts and Minds</a> was originally published in <a href="https://medium.com/techforcampaigns">Tech for Campaigns</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Cambridge Analytica is a Distraction]]></title>
            <link>https://medium.com/techforcampaigns/cambridge-analytica-is-a-distraction-f2f23bcc6e9?source=rss-811146aabc07------2</link>
            <guid isPermaLink="false">https://medium.com/p/f2f23bcc6e9</guid>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[facebook]]></category>
            <category><![CDATA[politics]]></category>
            <category><![CDATA[advertising]]></category>
            <dc:creator><![CDATA[Greg Dale]]></dc:creator>
            <pubDate>Tue, 10 Apr 2018 15:27:54 GMT</pubDate>
            <atom:updated>2018-04-10T16:31:10.112Z</atom:updated>
            <content:encoded><![CDATA[<h4>We need to protect our privacy. But that alone won’t return sanity to our politics.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lRgqusexBvavvDfQIDTGdA.png" /></figure><p>The 2016 presidential election was an explosive reinforcement of the dark potential of online influence. Cambridge Analytica, with its purloined Facebook data and skill at mythical self-promotion, has arisen as a perfect boogeyman for these gloomy times.</p><p>To be clear, the scandal exposed real privacy issues that need to be dealt with — urgently — but it shouldn’t distract the broader Democratic campaign community from the continuing need to uplevel its digital efforts. As a society, we need to reckon with the abuse of personal digital data and how to best stop it. Separately, campaigners should not take the scandal to mean that a strong digital effort requires shadowy tactics and data theft; instead, it needs a strong and effective use of existing tools and the serious devotion of time and money. Cambridge Analytica, which ultimately sold digital advertising in a shady way, is a distraction from this vital narrative.</p><p>Americans spend <strong>5.6 hours/day online</strong>. IPG Mediabrands, a large media buying agency, <a href="https://www.recode.net/2017/12/4/16733460/2017-digital-ad-spend-advertising-beat-tv">expects global advertisers in 2018</a> to spend <strong>$54 billion more</strong> on digital ad channels than TV. However, Democratic campaigns still are spending <strong>65 to 70 percent</strong> of their budget on <strong>direct mail and TV</strong> and just <strong>five to ten percent</strong> <strong>on digital</strong>. This incredible mismatch is still pervasive in 2018 as I talk to campaigns up and down the ballot and all around the country. It took commercial advertisers a decade to catch up to where their customers actually spend their time; as the world gets faster paced and the political stakes grow, Democratic political advertisers do not have the luxury of waiting that long. The good news is, catching up is a mindset change, not a technology change, and a look at the details of Facebook’s ad targeting capabilities and the Cambridge Analytica scandal illustrates how.</p><h4><strong>The evolution of Facebook’s ad platform showcases how limited demographic data is for targeting</strong></h4><blockquote><em>“Converting Facebook data into money is harder than it sounds, mostly because the vast bulk of your user data is worthless.” — </em><a href="https://www.theguardian.com/technology/2017/may/02/facebook-executive-advertising-data-comment"><em>Antonio Garcia Martinez</em></a><em>, former Facebook Ads product manager</em></blockquote><p>In 2012, Facebook had not yet figured out how to make serious money. Although Facebook knew who you were and what you had Liked, users just weren’t clicking on ads enough. With a looming IPO in May of that year, revenue was not where it needed to be.</p><p>That year, Facebook’s Ads team found their way to the solution — and as Garcia Martinez alludes to above, it was one that ignored the majority of the data you gave Facebook.</p><p><strong>Custom Audiences</strong>, announced in September 2012, let advertisers find their valued customers online. Facebook would say “give me 1,000 of your best customers”; an advertiser would hand over the list; and Facebook would say, “we’ll find these people — trust us.” Advertisers, although they did not know precisely who saw what ad, saw fantastic results out of his process, and so kept supplying the lists and buying the ads.</p><p>The special sauce was simply making sure the ads went to the specific people the advertisers cared about. Facebook didn’t need to know who bought what product, clicked on what link, or whose photos you had just scrolled through; just that it was showing ads to the right people. In 2013, Facebook announced “lookalike audiences,” which allowed advertisers to find new customers who “looked like” their old ones through an opaque black box of user data and targeting algorithms. These new products worked. Paired with a fortuitous shift to mobile device usage, click-through rates increased, advertiser results increased, and Facebook’s revenue shot up.</p><h4><strong>Selling your data is against Facebook’s interests</strong></h4><p>While these features are often construed as “selling your data to advertisers,” if they really did, advertisers take those online identities and instantly go around Facebook, advertising to people wherever they could be found, and Facebook would lose billions of dollars of revenue.</p><p>The Custom Audience and lookalike black boxes that match offline personas to online identities became black gold, and Facebook protected them however they could. In 2014, Facebook nearly bankrupted two ad measurement companies, HasOffers (now known as <a href="https://www.tune.com/">TUNE</a>) and Kontagent (now shuttered), for “<a href="https://venturebeat.com/2014/02/12/facebook-drops-mobile-ad-partners-hasoffers-and-kontagent-for-holding-on-to-user-data-too-long/">holding onto user data too long</a>.” In 2015, Facebook terminated the once officially sanctioned API that Cambridge Analytica exploited, but as we now know, only weakly enforced its closure.</p><h4><strong>Navigating the Gulf of Creepiness</strong></h4><p>There’s a longstanding gulf between the inherent creepiness of unknown companies knowing who your friends are and the real-life utility of that data.</p><p>Facebook users accept an implicit bargain: they will be shown targeted advertising in exchange for the free utility they find in the service. The social contract is finely drawn; when the data is possessed by another company, trust is violated. Although Facebook is catching all of the recent flak, companies do exist to explicitly sell your personal data. For instance, the <a href="http://www.bluekai.com/registry/">Oracle Data Cloud Registry</a>, formerly known as BlueKai, collects credit card receipts, income data, and store purchases to sell to marketers in search of people, say, ready for a new flavor of Doritos.</p><p>With politics, the social contract has always been different. The Obama 2008 operation was lauded for “microtargeting.” Can you imagine how creepy it would be if Doritos texted you, called you, and had staff come and knock on your door on account of what chips you had bought years prior? Although political campaigns operate under a different set of expectations, the Cambridge scandal reinforces the absolute need for those campaigns to be very conscious of preserving voter privacy.</p><h4><strong>Do Democrats need their own Cambridge Analytica?</strong></h4><p>Despite <a href="http://www.businessinsider.com/cambridge-analytica-boasts-won-trump-election-facebook-data-2018-3">their self-promotional efforts</a>, Cambridge Analytica did not elect Trump, and their efficacy as advertisers — even with the stolen data — seems limited. Ted Cruz, Cambridge’s marquee client, <a href="https://www.theverge.com/2018/3/20/17138854/cambridge-analytica-facebook-data-trump-campaign-psychographic-microtargeting">stopped using them after a 2016 primary in South Carolina</a>, and while the Trump campaign spent almost $6 million with Cambridge, that was just 13% of what the Trump campaign spent directly on Facebook.</p><p>By 2016, Facebook’s in-house tools — custom audiences and lookalikes — were perfectly capable of putting persuasive ads in front of the right people. <a href="http://adage.com/article/digital/trump-campaign-facebook-facebook-agrees/312974/">The Trump campaign outspent the Hillary campaign on Facebook, $44 million to $28 million, between June and November of 2016</a>. Spend does not equal persuasion, but it’s an important data point.</p><p>Democrats do not need their own Cambridge Analytica. Instead, they need to radically increase investment in the basics of a digital presence, up and down the ballot. A typical Democratic campaign plans to spend between <strong>five and ten percent</strong> of their campaign budget on digital advertising, with TV and direct mail taking up the larger share. <strong>These guidelines are severely out of date; digital ad spending for campaigns should be 30 percent of a campaign’s overall budget</strong>.</p><p>Facebook, for its part, is adapting to the new environment, releasing new ad disclosure features and verifying true ownership of pages. Facing possible regulation like the <strong>Honest Ads Act</strong>, Facebook is <a href="https://www.reuters.com/article/us-facebook-privacy/facebook-backs-political-ad-bill-sets-limits-on-issue-ads-idUSKCN1HD2JH">proactively endorsing it</a>, adopting its requirements ahead of schedule. At the same time, in the European Union, the new <strong>General Data Protection Regulation</strong> (GDPR) is enforcing the protection of personal data protection with new regulations and large fines. These new regulations and features will put the screws on online targeters, but not affect what Democrats need to do.</p><p>The net effect of these scandals will be some increased protection of your online identity — and that’s a great thing. Data holes will be plugged, Facebook will rightfully suffer in the public eye, and Cambridge Analytica will fade into the sunset. But do not confuse the defeat of the Cambridge boogeyman for a digital “job well done.” Being competitive online does not require data theft and shady targeting, but being serious and proactive about meeting voters where they are and spending serious time and money online.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f2f23bcc6e9" width="1" height="1" alt=""><hr><p><a href="https://medium.com/techforcampaigns/cambridge-analytica-is-a-distraction-f2f23bcc6e9">Cambridge Analytica is a Distraction</a> was originally published in <a href="https://medium.com/techforcampaigns">Tech for Campaigns</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How we used data to select our Priority states for the 2018 elections]]></title>
            <link>https://medium.com/techforcampaigns/how-we-used-data-to-select-our-priority-states-for-the-2018-elections-b97e72cda6?source=rss-811146aabc07------2</link>
            <guid isPermaLink="false">https://medium.com/p/b97e72cda6</guid>
            <category><![CDATA[democracy]]></category>
            <category><![CDATA[politics]]></category>
            <category><![CDATA[tech]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[elections]]></category>
            <dc:creator><![CDATA[Greg Dale]]></dc:creator>
            <pubDate>Wed, 28 Mar 2018 05:19:43 GMT</pubDate>
            <atom:updated>2018-03-28T05:19:43.104Z</atom:updated>
            <content:encoded><![CDATA[<h4>Bringing startup scale to progressive and centrist politics</h4><p>Last week, we published Tech for Campaigns’ <a href="https://medium.com/techforcampaigns/announcing-tfcs-2018-state-priorities-list-5b978abe6768">2018 Priorities list for State Legislatures</a>, detailing the fourteen states we are focused on for state legislative races. This week, we wanted to zoom out — and in — on how TFC established and is executing on its 2018 strategy. (Plus, I’ll be hosting a <a href="https://www.facebook.com/events/2074529379492016/">Facebook Live</a> on April 4th at 5PM to chat more about it — check it out!)</p><p>In 2018, TFC is aiming to allocate its 70% of its volunteer efforts to state legislature races and 30% to federal and statewide races (U.S. House, U.S. Senate, Governor, Lt. Governor, etc.) These categories include thousands of campaigns, so sorting through them is no trivial task!</p><p>For federal and statewide races, between outlets like <a href="https://fivethirtyeight.com/politics/">FiveThirtyEight</a> and <a href="http://cookpolitical.com/">The Cook Political Report</a>, there is no shortage of predictions and assessments. For state legislatures, though, there is no canonical public accounting of different races nationwide. Given our focus on helping hundreds of Democrats across multiple states in 2018, we needed to establish our own point of view and organizational framework — and data helped us do it.</p><h3>How we did it</h3><p>Being techies, we turned to data and predictive modeling to help us out. Our Build The List volunteer project gathered data from a variety of formats across government and commercial/nonprofit sources, normalized it, and used features like demographics, past election results, and incumbency to predict Democratic and Republican results in each district for 2018. The results provide Tech for Campaigns with a single baseline to evaluate and prioritize races across states and begin to layer in qualitative assessments.</p><p>Two main constraints on the list were finding routes to influence the 2020 redistricting process on way to flipping the U.S. House and TFC’s volunteer model. These drove us to de-prioritize state legislatures with only a single U.S. Congressional seat, state legislatures with multi-member election methods, state legislatures that have minimal influence over redistricting, and other factors.</p><h3>The results</h3><p>From this process, we emerged with <a href="https://medium.com/techforcampaigns/announcing-tfcs-2018-state-priorities-list-5b978abe6768">fourteen 2018 Priority states</a> for state legislative races, covering four key objectives:</p><ul><li><strong>Priority Flip</strong>: states where our modeling indicated that Democrats have a significant shot of retaking at least one state chamber in 2018;</li><li><strong>Potential Flip</strong>: states where our modeling indicated that favorable trends are in place to help make significant gains in the 2018 elections on the way towards a flip in 2020;</li><li><strong>Defends</strong>: states where a chamber has a thin Democratic majority; and</li><li><strong>Break Supermajorities</strong>: states where the Republican party controls a supermajority in State Senates, giving them veto-proof control over budgets and legislation without the natural consensus achieved by needing some Democratic votes.</li></ul><p>The predictive modeling process was key to informing the likelihoods of success of each of these four objectives, driving the selection of eight out of ten of our non-exploratory 2018 states.</p><h3>Prioritizing individual races</h3><p>With our initial 2018 State Prioritization complete, our focus will turn to our work with state parties and caucuses and prioritizing individual races. The modeling process is again key; it provides an excellent baseline to layer on qualitative assessments to determine the 10–15 races in each state where TFC volunteers would be best positioned to help.</p><p>For non-prioritized races, we will be offering tools and materials any progressive campaign can use to upgrade their digital presence.</p><h3>Going forward</h3><p>Tech for Campaigns was founded just over one year ago, in February 2017. Since then, we’ve completed 30+ projects and helped eight seats in the Virginia House of Delegates.</p><p>The modeling, inputs, and framework established in this process provide Tech for Campaigns with a repeatable organizational framework and baseline understanding of progressive political chances nationwide. Our prioritization process will be the center of our efforts for years to come as we build the digital arm for progressive politics in the United States.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b97e72cda6" width="1" height="1" alt=""><hr><p><a href="https://medium.com/techforcampaigns/how-we-used-data-to-select-our-priority-states-for-the-2018-elections-b97e72cda6">How we used data to select our Priority states for the 2018 elections</a> was originally published in <a href="https://medium.com/techforcampaigns">Tech for Campaigns</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Announcing TFC’s 2018 State Priorities List]]></title>
            <link>https://medium.com/techforcampaigns/announcing-tfcs-2018-state-priorities-list-5b978abe6768?source=rss-811146aabc07------2</link>
            <guid isPermaLink="false">https://medium.com/p/5b978abe6768</guid>
            <category><![CDATA[activism]]></category>
            <category><![CDATA[tech]]></category>
            <category><![CDATA[politics]]></category>
            <category><![CDATA[elections]]></category>
            <dc:creator><![CDATA[Greg Dale]]></dc:creator>
            <pubDate>Tue, 20 Mar 2018 17:46:31 GMT</pubDate>
            <atom:updated>2018-03-20T17:46:31.328Z</atom:updated>
            <content:encoded><![CDATA[<h4>Focusing digital volunteer efforts for maximum impact</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FgMnFaHJBFuxkc9ZBB8a_w.png" /></figure><p>Although a political news junkie for as long as I can remember, I had been satisfied as a bystander. I hosted a community event or two for Obama 2008 and donated some to Hillary 2016, obsessively refreshed Nate Silver and Politico like many, but doing more seemed like someone else’s job. In November 2016, like for many others, things changed. Within months, I had joined <a href="http://www.techforcampaigns.org/">Tech for Campaigns</a> as a volunteer and was running ads across Newport News and Williamsburg, Virginia for the successful re-election of <a href="http://www.mullinforvirginia.com/">Delegate Mike Mullin</a>. This week, I am thrilled to be joining TFC full-time as Director of Campaign Relations to help scale our digital volunteer projects more than 10x in 2018. I wanted to share a bit about the approach we are taking this year towards maximizing our impact in this year’s midterm elections.</p><p>Last year, Tech for Campaigns partnered up with the Virginia House Democratic Caucus to execute 35 projects across twelve separate campaigns for the House of Delegates and several projects for the Caucus itself. We helped to secure wins in nine of the races and flip eight seats in the biggest Democratic pickup in Virginia since 1899. We’re not done!</p><p><strong>In 2018, Tech for Campaigns will be continuing this effort, pairing up with key states, like we did in Virginia in 2017, to execute projects that range from paid social advertising to texting GOTV strategies to fundraising data analysis.</strong></p><p>TFC will be working on campaigns at all levels of the ballot, including federal and statewide races, but we will be placing special emphasis on state legislatures. In many states, state legislatures:</p><ul><li>Make most of the laws that govern Americans’ daily lives;</li><li>Are within just a few seats of flipping control of a chamber or breaking a Republican supermajority; and</li><li>Are ground zero for the demographic change that may help flip control ahead of the key 2020 redistricting process (phenomenally important for long-term Democratic control of the U.S. House!)</li></ul><p>Despite how important they are, state legislature races often fly below the radar and are the most in need of help from digital-savvy volunteers. For TFC, concentrating deeply within a state legislature has proven itself to be an effective strategy, so we’re back at it again.</p><p><strong>Tech For Campaigns is thrilled to announce our 2018 State Priorities List:</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*GhTnTdJC-s26jWXX." /></figure><p>In the coming weeks, we’ll be diving deep into our selection methodology and the political landscapes in some of our states. We’ve prioritized states based on our internal data team’s models and additional strategic decisions we’ve made to invest in important states. While there are fourteen total states, four remain in an exploratory stage for us.</p><p>A HUGE thank you to the many TFC volunteers participating in our Build The List data science and engineering effort — this process would not have been possible without your hard work.</p><p>If you are a candidate or represent one and would like to chat about becoming part of TFC’s efforts in these states, <a href="https://www.techforcampaigns.org/for-campaigns">please let us know</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5b978abe6768" width="1" height="1" alt=""><hr><p><a href="https://medium.com/techforcampaigns/announcing-tfcs-2018-state-priorities-list-5b978abe6768">Announcing TFC’s 2018 State Priorities List</a> was originally published in <a href="https://medium.com/techforcampaigns">Tech for Campaigns</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Succeed by designing machine learning products that fail]]></title>
            <link>https://gregdale.medium.com/succeed-by-designing-machine-learning-products-that-fail-9d433f0496e5?source=rss-811146aabc07------2</link>
            <guid isPermaLink="false">https://medium.com/p/9d433f0496e5</guid>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[user-experience]]></category>
            <category><![CDATA[design]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Greg Dale]]></dc:creator>
            <pubDate>Tue, 23 Jan 2018 06:09:48 GMT</pubDate>
            <atom:updated>2018-02-15T14:21:33.442Z</atom:updated>
            <content:encoded><![CDATA[<h3>Key design traits for successful machine learning products</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tbNJ85FWFPLHIsNlXLCgVQ.png" /><figcaption>My own Google Photos. Amazing, but those aren’t Bikes or Cars.</figcaption></figure><p>Computers are dumb. They’ve gotten faster throughout the decades, though, and programmers have increasingly stacked up larger amounts of simple instructions to make computers do ‘smart’ things. Computers, for their part, are happy to execute these instructions over and over again. Since the 1980s, when white collar workers began to use Lotus 1–2–3 to automate spreadsheets, software has subtly trained its users to expect predictable behavior. One interface may be more confusing than the next, but all users implicitly know that when one hits Play, the video should play. If it does not, something is wrong.</p><p>Now, in the era of machine learning, computers are a new kind of strict instructions: to teach themselves how to accomplish tasks. Drawing on computers’ power to do millions of instructions at once, these applications unlock amazing possibilities — like machine translation or comparing your selfie to historical paintings — that were once impossible to pull off via a programmer’s explicit instructions.</p><p>Computers, though, have not achieved sudden sentience. Machine learning systems are never 100% correct. <strong>Product design with machine learning, has a huge difference from previous software design paradigms:</strong></p><blockquote>Error is not an exception; it is a constant.</blockquote><p>Successful machine learning software must be designed from first principles to harness the interplay between rapid computational judgments and accurate human insights. Error handling can no longer be done with a red banner or a dialog box. It’s now a fundamental design requirement, but the good news is, there are emerging best practices and useful frameworks to think through the problem.</p><h3>Icarus.ai: binary products fly close to the sun</h3><p>Alexa is incredibly complex: it understands phonemes, extracts meaning and content from language, and interfaces with thousands of sub-services to answer queries.</p><p>That said, Alexa frequently misunderstands my wife. After Alexa says she doesn’t know how to “turn the lights awful,” my wife is innately compelled to repeat her question with a tone of voice that implies Alexa is actually a four-year-old refusing to eat her green beans.</p><p>Alexa gets one thing wrong and the user is instantly frustrated. Alexa isn’t an outlier; there’s an <a href="https://www.reddit.com/r/SiriFail/">entire SiriFail subreddit</a>. The traditional call-and-response “AI assistant” model has only one shot to get the answer right. Even for tech mega-companies like Amazon and Apple, it leads to frequently frustrated users — who expect software to behave logically.</p><p>Google Photos is incredibly impressive: in March 2015, Google released its FaceNet paper that outlined the methodology used for facial recognition. On a widely recognized benchmark, <strong>FaceNet was</strong> <strong>99.63% accurate</strong>. Just a few months later, in July, a scandal erupted when Google’s facial recognition models recognized two African-Americans as “gorillas.” The algorithm at fault may or may not have been FaceNet, but it’s clear that if one user becomes part of that 0.37% of error, the accuracy experienced by most of the population becomes quickly irrelevant.</p><h3>Graham Neubig on Twitter</h3><p>What an analogy! &quot;A 95% fully automatic high-quality translation system produces text analogous to a jar of cookies where 5% are poisoned. Such a cookie jar is useless without a professional analysis to localize the poisoned ones.&quot; (Carbonell&amp;amp;Tomita 1985) https://t.co/INeM2fwh7F</p><p>These apps use ML for <strong>binary outcomes</strong>. They are right or wrong: there is no gray area with text-to-speech, facial recognition, or many types of natural language understanding. The accuracy of the model controls the success of the product, and some disappointment is probabilistically inevitable. On the flip side, when these applications succeed, they open up new possibilities in software and society.</p><h3>jack on Twitter</h3><p>🤔</p><h3>Gradual ML products silently move us forward</h3><p>Alexa, Siri, and facial recognition apps are visible; they delight and disappoint us on a daily basis. However, there are numerous AI/ML products that we use just as often that don’t drive the same frustrations that the other systems do.</p><blockquote>8 million shows, still nothing to watch…</blockquote><p>When I sit down to watch something on Netflix, a melange of ML algorithms are guiding my selections. <a href="http://www.wired.co.uk/article/how-do-netflixs-algorithms-work-machine-learning-helps-to-predict-what-viewers-will-like">Over 80% of Netflix users’ viewing choices are driven by Netflix’s ML recommendation system</a>. Google and Amazon return search results; Spotify’s Discover Weekly finds new music; all systems enhanced by AI techniques.</p><p>These apps make <strong>gradual predictions</strong>—the accuracy of the model contributes to, but does not control, the success of the product. Users have lower expectations and allow these systems the latitude to improve over time. Whether your new favorite song showed up first or tenth in the recommendation list is immaterial; whether Lyft found you the absolute best driver or just the third-best is completely unclear; your expectations are managed.</p><h3>Mitigating risks in ML product design</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Mc3pncKu-6NybUwafoM3tA.png" /></figure><p>Building successful machine learning products involves navigating a complex soup of user expectations, workflow requirements, and design of fallbacks and suggestions. To reduce the risk while preserving the delight of binary outcome products, a number of risk mitigation techniques observed in gradual outcomes come in handy.</p><h4>Think through the product’s confusion matrix</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/487/1*0H3JiH7wXqdAmIyait16lw.png" /><figcaption>A confusion matrix.</figcaption></figure><p>Classification models, which try to predict whether something belongs to a certain group or not are oftentimes used in <strong>binary outcomes</strong>. (Is this photo of Barack Obama? Is this email spam? Is this blog post wonderful?) In building this type of model, data scientists use a <a href="https://en.wikipedia.org/wiki/Confusion_matrix"><strong>confusion matrix</strong></a> to monitor the four quantitative outcomes of the model: <strong>true positives, false positives, true negatives, and false negatives</strong>. Classification models encompass these four outcomes on a regular basis; error is inevitable.</p><p>An <a href="https://medium.com/google-design/human-centered-machine-learning-a770d10562cd">insightful article from two Google designers, Josh Lovejoy and Jess Holbrook, published last July</a> offers a framework for thinking through ML-driven experiences. A key insight is the use of a confusion matrix for product design: knowing that your users <strong>will</strong> <strong>experience all four outcomes</strong> means you need to <strong>design for all four modes</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/225/1*WNu8bghtr_nS5tBV3K4Y6w.png" /><figcaption>Hotdog!</figcaption></figure><p>In the classic “Not Hotdog” use case, the accuracy of the hotdog classifier has no consequence. For Google Photos, seeking to identify pictures of animals, <strong>true positives</strong> and <strong>true negatives</strong> that correctly identify animals are table stakes. In this context, incorrectly identifying a gorilla as a human can be thought of as a <strong>false negative</strong>, with little consequence. Identifying a human as a gorilla, though, as a <strong>false positive</strong>, is deeply offensive and creates a scandal. By thinking through the confusion matrix during the design phase, you’ll ensure users have a positive experience, no matter the prediction.</p><h4>Incorporate gradual attributes into binary outcomes</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/580/1*AdAvIwS9D0hDStib8R6_jg.png" /></figure><p>Google Inbox’s automatic responses is a powerful case study, neatly illustrating three UX tactics that lead to success:</p><ul><li><strong>Sets of predictions:</strong> Inbox provides three suggested replies, greatly lessening the accuracy requirements. If the user selects one of the three, the click can be recorded as positive feedback for the model. If the user ignores the set, all three predictions can be recorded as unhelpful.</li><li><strong>Human-in-the-loop</strong>: when the user selects an auto-response, it fills the reply window, with the cursor activated for easy customization and sending of the reply.</li><li><strong>Escape hatch:</strong> if the predicted responses are off the mark, the user ignores them and writes the email as before, with no added annoyance.</li></ul><p>This flow scales well beyond email to flows that require even higher accuracy. When radiologists are tasked with identifying malignant cells under a microscope, they start from scratch by scanning the photo with their eyes. Imagine a 70% accurate ML algorithm also scans the image, and draws ten outlines around what it thinks are bad cells. On average, three of the algorithm’s outlines will be wrong and seven will be right. With the right interface, the radiologist is well-positioned to rapidly reject the three bad outlines, approve the seven correct ones, and add new ones, generating a large boost in productivity.</p><p>With machine learning, the radiologist speeds through their work. Without the radiologist, 30% of cancer cells would go undetected — a massive failure.</p><h3>Conclusion</h3><p>Plenty of newsprint has been used to describe the time and effort that goes into training quality machine learning models. Less has been spent on the equally challenging task of designing applications for those models that delight users.</p><p>The incorporation of error not as an after-thought but as the primary consideration is a fundamentally different dynamic for experience design than before. As software designers and architects, we can navigate this complex arena by making informed choices about the role ML should play in solving a task for a user and offering that user complete control around the experience. If ML can help, it should; if it can’t, it needs to get out of the way, quickly. Only your user can tell the difference: don’t forget to help them out.</p><p><strong>Like this article? Give it a clap to help others find it.</strong></p><p><em>Comments, corrections? Would love to hear them at gregdale00 at gmail dot com.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9d433f0496e5" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Your chatbot needs a personality]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://gregdale.medium.com/your-chatbot-needs-a-name-b8f92f337386?source=rss-811146aabc07------2"><img src="https://cdn-images-1.medium.com/max/1600/1*Ftkx_YbEAyxadHj-pZTU5g.png" width="1600"></a></p><p class="medium-feed-snippet">Technology&#x2019;s gotten better, but history shows that personality is just as important.</p><p class="medium-feed-link"><a href="https://gregdale.medium.com/your-chatbot-needs-a-name-b8f92f337386?source=rss-811146aabc07------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://gregdale.medium.com/your-chatbot-needs-a-name-b8f92f337386?source=rss-811146aabc07------2</link>
            <guid isPermaLink="false">https://medium.com/p/b8f92f337386</guid>
            <category><![CDATA[psychology]]></category>
            <category><![CDATA[chatbots]]></category>
            <category><![CDATA[business]]></category>
            <category><![CDATA[tech]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Greg Dale]]></dc:creator>
            <pubDate>Tue, 28 Nov 2017 21:06:32 GMT</pubDate>
            <atom:updated>2021-01-03T22:27:40.841Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Snap’s ad business is in serious trouble]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://gregdale.medium.com/snaps-ad-business-is-in-serious-trouble-e8c431201de6?source=rss-811146aabc07------2"><img src="https://cdn-images-1.medium.com/max/2600/1*ORFH6BZi62FjNmAbZfBFJw.jpeg" width="4082"></a></p><p class="medium-feed-snippet">While it still has some tricks up its sleeve, there&#x2019;s no immediate route to re-ignite revenue growth.</p><p class="medium-feed-link"><a href="https://gregdale.medium.com/snaps-ad-business-is-in-serious-trouble-e8c431201de6?source=rss-811146aabc07------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://gregdale.medium.com/snaps-ad-business-is-in-serious-trouble-e8c431201de6?source=rss-811146aabc07------2</link>
            <guid isPermaLink="false">https://medium.com/p/e8c431201de6</guid>
            <category><![CDATA[investing]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[business]]></category>
            <category><![CDATA[snapchat]]></category>
            <category><![CDATA[advertising]]></category>
            <dc:creator><![CDATA[Greg Dale]]></dc:creator>
            <pubDate>Tue, 28 Nov 2017 01:49:42 GMT</pubDate>
            <atom:updated>2018-01-23T15:44:16.047Z</atom:updated>
        </item>
    </channel>
</rss>