<?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 J.D. Maddox on Medium]]></title>
        <description><![CDATA[Stories by J.D. Maddox on Medium]]></description>
        <link>https://medium.com/@jdmaddox?source=rss-c371dc19822a------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*IgN3MJyej-_8Zb86</url>
            <title>Stories by J.D. Maddox on Medium</title>
            <link>https://medium.com/@jdmaddox?source=rss-c371dc19822a------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 05:27:25 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@jdmaddox/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[I asked my grad students to use AI to write their exam papers, and…]]></title>
            <link>https://medium.com/@jdmaddox/i-asked-my-grad-students-to-use-ai-to-write-their-exam-papers-and-3d481dcaabaa?source=rss-c371dc19822a------2</link>
            <guid isPermaLink="false">https://medium.com/p/3d481dcaabaa</guid>
            <dc:creator><![CDATA[J.D. Maddox]]></dc:creator>
            <pubDate>Tue, 03 Oct 2023 23:23:52 GMT</pubDate>
            <atom:updated>2023-10-03T23:23:52.291Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/768/1*D41Jj9CkII7zHV3Zg6NTXA.jpeg" /><figcaption>An AI-generated image to accompany a paper about AI-generated text seemed appropriate. This image was generated using Stable Diffusion.</figcaption></figure><p>This semester I asked my <a href="https://mymasonportal.gmu.edu/bbcswebdav/orgs/AU_Schar_Web/course-syllabi/fall-2023/POGO/Maddox-POGO-750-012-FA-23.pdf">Disinformation and Policy Responses</a> grad students to write two short midterm exam papers — first, a traditional human-written paper, and second, a paper produced by an Artificial Intelligence (AI) text generator like ChatGPT. I asked the students to hand them in simultaneously and not to tell me which one was AI-generated so I could guess.</p><p>I thought this little experiment might offer lessons in how far text generators have come. The emergence of AI text generators may have <a href="https://cset.georgetown.edu/publication/ai-and-the-future-of-disinformation-campaigns-2/">profound implications</a> for the issue of disinformation, since these systems offer to automate and up-scale disinformation production in unpredictable ways. Also, I thought this little experiment might simply help other professors to spot fakes.</p><p><em>Here’s what happened.</em></p><p>I correctly identified all but one AI-written paper. I don’t think it’s because I’m particularly good at AI content detection; instead, the AI “tells” just seemed pretty obvious.</p><ul><li>The AI text generators frequently used lists of paragraph-length evidentiary examples instead of deep explanations of singular topics. These examples never went beyond first-order analysis to consider complex implications of unsettled disinformation issues or concepts.</li><li>AI text generators used fairly pedantic conclusions in the papers. They didn’t stray from basic mainstream thinking on a topic. If I had hopes of off-the-wall creative ideation, I was sorely let down by the robots.</li><li>The AI “authors” seemed entirely self-assured. The artificially-generated papers did not offer any doubts about the facts they presented, and did not offer questions about the limits of the authors’ own knowledge of a subject.</li><li>The AI text generator repeated its evidentiary examples in multiple papers on the same topic. After encountering the evidentiary examples in one paper, it became very easy to determine that the next, similar-sounding, paper was AI-generated. For example, multiple AI-generated papers made the same three suggestions for technological disinformation countermeasures.</li><li>One of the strongest “tells” of <em>authentic</em> content was <em>unsupported opinion.</em> If a writer made a biased statement without sourcing, it stood out as an original human thought. In contrast, AI-derived statements typically sounded like patent groupthink. Some critics of AI content generators deride them as “drunken uncles” because of their unpredictable reliance on illusory information, but this experiment showed the opposite — that the robot writers were regurgitating passable common concepts while some human writers didn’t care to offer proof of their sometimes passionate opinions. Maybe passion is what will continue to distinguish humans.</li><li>A few students reported that the AI text generators were inefficient. A few had to make multiple prompts to generate a sufficiently long paper, obscure subjects resulted in substantively thin AI text generation, and a few had to do some laborious editing to produce a cogent paper.</li><li>I failed to identify one AI-written paper because of my own biases. The student’s authentic paper contained some technical jargon that sounded “mechanistic” — as if a machine might’ve been repeating the jargon from an obscure technical source — and it led me to incorrectly believe that the other, flimsier, paper was the student’s. It turns out that the student author just did deep research… and maybe they write a bit like a robot (sorry not sorry, student).</li></ul><p>To sum up, this little experiment did not offer any frightening hints of imminent AI dominance, and certainly not a <a href="https://terminator.fandom.com/wiki/Skynet">Skynet</a> moment. Instead, it hinted at the insouciant defensive shrug of all the AI users to come: “stop worrying, old people, it’s just AI,” they might say. The relatively unsophisticated responses of the AI text generators simply did not support the idea of an all-knowing or strongly influential system, but instead offered a hallmark of emotional emptiness. But this is just one little experiment.</p><p><strong><em>If you’re interested, here are one student’s final exam papers (one written using AI and the other using a mere human brain — in no particular order). Can you guess which one was written using an AI text generator?</em></strong></p><ul><li><a href="https://drive.google.com/file/d/1xrk4-aIOfKUqNe3aQWV6w6-lCwiDkn5f/view?usp=sharing">DPR Pizzagate Student Midterm 1.pdf</a></li><li><a href="https://drive.google.com/file/d/1PqAvtpRJjOe9uKwKm7HUDwjja0nsiFPI/view?usp=sharing">DPR Pizzagate Student Midterm 2.pdf</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3d481dcaabaa" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[An incomplete map of US information assurance capabilities.]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@jdmaddox/an-incomplete-map-of-us-information-assurance-capabilities-c9ad691e69c4?source=rss-c371dc19822a------2"><img src="https://cdn-images-1.medium.com/max/2600/1*bviiWh6YrktKlP4NMfeH1A.png" width="5100"></a></p><p class="medium-feed-snippet">A fresh look at the broad opportunities for US information operations.</p><p class="medium-feed-link"><a href="https://medium.com/@jdmaddox/an-incomplete-map-of-us-information-assurance-capabilities-c9ad691e69c4?source=rss-c371dc19822a------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@jdmaddox/an-incomplete-map-of-us-information-assurance-capabilities-c9ad691e69c4?source=rss-c371dc19822a------2</link>
            <guid isPermaLink="false">https://medium.com/p/c9ad691e69c4</guid>
            <dc:creator><![CDATA[J.D. Maddox]]></dc:creator>
            <pubDate>Mon, 13 Jun 2022 12:32:19 GMT</pubDate>
            <atom:updated>2022-06-13T12:32:19.898Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[A reductive look at narrative dynamics contributing to the US Capitol violence on 6 January 2021.]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@jdmaddox/a-reductive-look-at-narrative-dynamics-contributing-to-the-us-capitol-violence-on-6-january-2021-4e095a33f03f?source=rss-c371dc19822a------2"><img src="https://cdn-images-1.medium.com/max/2600/1*8204Aj5PCUOl3aFRHAfyrQ.png" width="3300"></a></p><p class="medium-feed-snippet">What motivated J6 participants</p><p class="medium-feed-link"><a href="https://medium.com/@jdmaddox/a-reductive-look-at-narrative-dynamics-contributing-to-the-us-capitol-violence-on-6-january-2021-4e095a33f03f?source=rss-c371dc19822a------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@jdmaddox/a-reductive-look-at-narrative-dynamics-contributing-to-the-us-capitol-violence-on-6-january-2021-4e095a33f03f?source=rss-c371dc19822a------2</link>
            <guid isPermaLink="false">https://medium.com/p/4e095a33f03f</guid>
            <category><![CDATA[disinformation]]></category>
            <category><![CDATA[politics]]></category>
            <category><![CDATA[j6]]></category>
            <category><![CDATA[violence]]></category>
            <category><![CDATA[influence]]></category>
            <dc:creator><![CDATA[J.D. Maddox]]></dc:creator>
            <pubDate>Mon, 03 Jan 2022 01:18:27 GMT</pubDate>
            <atom:updated>2022-01-03T01:18:27.241Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Russia’s disinformation doctrine, rewritten by Artificial Intelligence]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@jdmaddox/russias-disinformation-doctrine-rewritten-by-artificial-intelligence-60edb27d8ac0?source=rss-c371dc19822a------2"><img src="https://cdn-images-1.medium.com/max/1280/1*xqvMj5rv8P6M_4aHu8SBxQ.jpeg" width="1280"></a></p><p class="medium-feed-snippet">The analyst who first published an English-language translation of the so-called &#x201C;Gerasimov Doctrine&#x201D;&#x200A;&#x2014;&#x200A;a speech about the new importance&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/@jdmaddox/russias-disinformation-doctrine-rewritten-by-artificial-intelligence-60edb27d8ac0?source=rss-c371dc19822a------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@jdmaddox/russias-disinformation-doctrine-rewritten-by-artificial-intelligence-60edb27d8ac0?source=rss-c371dc19822a------2</link>
            <guid isPermaLink="false">https://medium.com/p/60edb27d8ac0</guid>
            <category><![CDATA[propaganda]]></category>
            <category><![CDATA[future]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[disinformation]]></category>
            <category><![CDATA[security]]></category>
            <dc:creator><![CDATA[J.D. Maddox]]></dc:creator>
            <pubDate>Mon, 18 Nov 2019 15:54:08 GMT</pubDate>
            <atom:updated>2019-11-18T15:54:08.103Z</atom:updated>
        </item>
    </channel>
</rss>