<?xml version="1.0" encoding="UTF-8"?>
<rss  xmlns:atom="http://www.w3.org/2005/Atom" 
      xmlns:media="http://search.yahoo.com/mrss/" 
      xmlns:content="http://purl.org/rss/1.0/modules/content/" 
      xmlns:dc="http://purl.org/dc/elements/1.1/" 
      version="2.0">
<channel>
<title>Mete Akcaoglu</title>
<link>https://meteakcaoglu.com/blog.html</link>
<atom:link href="https://meteakcaoglu.com/blog.xml" rel="self" type="application/rss+xml"/>
<description></description>
<generator>quarto-1.8.27</generator>
<lastBuildDate>Wed, 18 Nov 2026 00:00:00 GMT</lastBuildDate>
<item>
  <title>A Modest Proposal and a Modest Conversation with AI</title>
  <link>https://meteakcaoglu.com/posts/2026-11-18-modest-proposal/</link>
  <description><![CDATA[ 





<p><img src="https://meteakcaoglu.com/images/marks/posts/modest-proposal.svg" class="post-mark img-fluid" alt=""></p>
<p>On the 15-minute drive to my Savannah office, I talked with Claude about jobs, something I had been thinking a lot about recently. I thought it would be an interesting blog post, so below is the conversation. I edited my own lines lightly (mostly the ums and false starts) and left Claude’s as they were, except for one line that answered a transcription error.</p>
<p>Also, a side note, and evolution in my thinking: In May, I wrote <a href="../../posts/2026-05-19-radiology-lesson/index.html">The Radiology Lesson</a>, where I argued that AI takes over the lower-order tasks of a job and leaves its purpose, so teaching will persist. I still think the tasks-and-purpose argument holds for a single job. This conversation is about something larger: the whole chain that produces knowledge workers is exposed at once, from the coder to the professor teaching coding to the schools preparing kids for college, and the new “workers” never buy what they make. On that question, I think I’m leaning pessimistic.</p>
<div class="chat">
<div class="msg me">
<p><span class="msg-who">Mete</span> People compare the AI movement to the Industrial Revolution: it disrupted jobs, and so on. One difference, of course, is how long that change took, and that’s an obvious one. But the other argument is that the AI revolution will erase jobs and not necessarily add new ones, as the Industrial Revolution did. Most of the jobs we do today exist, or there are more of them, because of the Industrial Revolution. I have a really hard time seeing it go that way this time, because the jobs that are going to be replaced, like a coder, and a professor who teaches people coding, or the schools that are preparing kids for college, might all be impacted. That’s nowhere near the effect the Industrial Revolution had, which was the opposite.</p>
</div>
<div class="msg ai">
<p><span class="msg-who">Claude</span> Yeah, I think you’ve put your finger on the real asymmetry. The Industrial Revolution automated muscle, and that raised the value of cognition. It’s literally why mass schooling exists: factories and offices needed literate, numerate people. AI goes after cognition itself, so the whole chain that produces knowledge workers is exposed at once, the coder, the person teaching coding, and the school preparing kids for that path.</p>
<p>The optimists’ best reply is that we’ve always been bad at imagining the new jobs. Nobody in 1850 could have pictured a UX designer. And when something gets cheap, demand for it often grows instead of shrinking. Cheaper code might mean far more software, not fewer people building it.</p>
<p>The weak spot in that argument is that every past wave left humans a refuge, some kind of work machines couldn’t do. If AI keeps closing in on general cognition, it’s not obvious where the next refuge is, apart from work where being human is the point itself: care, trust, physical presence, accountability.</p>
<p>So for education, maybe the question shifts from what jobs we’re preparing kids for to what’s worth knowing when expertise is cheap. Where do you lean on that?</p>
</div>
<div class="msg me">
<p><span class="msg-who">Mete</span> I hear what the optimists are saying, but going along the line of argument I made before, and what you raised about cognition, I think being pessimistic is more accurate, because I don’t see what could be generated as a new human job.</p>
</div>
<div class="msg ai">
<p><span class="msg-who">Claude</span> That’s a fair position. The optimists’ argument depends on jobs we can’t name yet, and that’s hard to test until it’s too late. One thing that might soften the pessimism a little: even if there aren’t new jobs, there may still be new roles, like supervising and judging AI output. Of course, that only works if people still build the expertise to do the judging, which runs straight into your point about the education pipeline.</p>
</div>
<div class="msg me">
<p><span class="msg-who">Mete</span> Yep. I don’t know.</p>
</div>
<div class="msg ai">
<p><span class="msg-who">Claude</span> Yeah, honestly, nobody does. It’s one of those questions where uncertainty is the right answer for now. Want to keep chewing on it, or switch to something else?</p>
</div>
<div class="msg me">
<p><span class="msg-who">Mete</span> Another way to add to this point: we are, in effect, adding an infinite number (not infinite, but you know what I mean) of PhD-level intelligences to our existing 8 billion people.</p>
</div>
<div class="msg ai">
<p><span class="msg-who">Claude</span> That’s a powerful way to frame it. It’s basically a massive labour supply shock, like the global workforce suddenly gaining millions of tireless experts who work for the cost of electricity. And unlike immigration or population growth, these new “workers” don’t consume anything. They don’t buy houses, eat at restaurants, or create demand that would generate jobs for other people.</p>
<p>That’s the part that breaks the usual economic story. Historically, more workers also meant more consumers, so the economy grew around them. Here you get the supply side without the demand side, so the gains flow to whoever owns the compute.</p>
</div>
<div class="msg me">
<p><span class="msg-who">Mete</span> Right. And the problem is, these free workers (I think that’s the point you raised as well) don’t need anything other than electricity. Obviously the data centers, the water, and all those environmental impacts are there. But compared to a human worker, they don’t consume the products we create as a byproduct of these activities, so that’s going to be disruptive.</p>
</div>
<div class="msg ai">
<p><span class="msg-who">Claude</span> Exactly, and it leads to a real paradox. If AI replaces enough paid human work, firms lose their customers too, because unemployed people can’t buy what the AI produces. Henry Ford supposedly understood this when he paid workers enough to afford his cars.</p>
<p>So the system can’t really run on the old logic of earning wages and then spending them. Something has to get the money back to people, whether that’s universal basic income, wider ownership of AI capital, or taxing compute. In the end the disruption may turn out to be less a technology problem than a distribution problem.</p>
</div>
<div class="msg me">
<p><span class="msg-who">Mete</span> What do you see coming out of this?</p>
</div>
<div class="msg ai">
<p><span class="msg-who">Claude</span> Honestly, I see two plausible paths, and the technology alone won’t decide between them.</p>
<p>In the hopeful one, productivity gains get shared, through policy or broad ownership. Work shrinks but meaning doesn’t, and people put more time into care, community, learning, and craft, things that matter partly because a human does them.</p>
<p>In the darker one, gains concentrate among those who own the compute, the middle class hollows out, and we get a lot of abundance with a lot of precarity sitting next to it.</p>
<p>My sense is that the transition is the dangerous part. Even if things settle somewhere good, the next 10 to 20 years could be rough, because institutions like schools, labour law, and tax systems move much more slowly than the technology. That’s where people like you in education really matter: you’re deciding now what expertise means for a generation that will live through it.</p>
</div>
<div class="msg me">
<p><span class="msg-who">Mete</span> It also reminds me of, I think it’s called <em>A Modest Proposal</em>. I read it in English 101, I think.</p>
</div>
<div class="msg ai">
<p><span class="msg-who">Claude</span> Ah, <em>A Modest Proposal</em>, Jonathan Swift’s 1729 satire. Yes, that fits darkly well. Swift was skewering people who looked at the Irish poor as an economic surplus to be managed rather than human beings, and he took that logic to its monstrous end: suggesting they sell their babies as food to the rich.</p>
<p>The parallel is the worry that if humans are no longer needed as workers, a cold economic logic starts treating them as a cost. Was that the connection you were drawing?</p>
</div>
<div class="msg me">
<p><span class="msg-who">Mete</span> Yeah, absolutely. I think he was even suggesting eating the…</p>
</div>
<div class="msg ai">
<p><span class="msg-who">Claude</span> Yes, eating the children was the whole shock of it. He laid it out in calm, rational, almost cheerful economic language, and that’s what makes it so chilling. The horror is in how reasonable it sounds once people are treated purely as numbers. That’s a good warning for AI debates too, since a lot of the talk about displaced workers already slips into that detached, spreadsheet tone.</p>
</div>
<div class="msg me">
<p><span class="msg-who">Mete</span> I remember there was a really iconic sentence, or passage. What was it?</p>
</div>
<div class="msg ai">
<p><span class="msg-who">Claude</span> The most famous passage is probably this one:</p>
<blockquote class="blockquote">
<p>“I have been assured by a very knowing American of my acquaintance in London, that a young healthy child well nursed is at a year old a most delicious, nourishing, and wholesome food, whether stewed, roasted, baked, or boiled; and I make no doubt that it will equally serve in a fricassee, or a ragout.”</p>
</blockquote>
<p>That cookbook tone is what makes it so devastating.</p>
</div>
<div class="msg me">
<p><span class="msg-who">Mete</span> Yeah, absolutely brilliant. Anyway, I just remember that.</p>
</div>
<div class="msg ai">
<p><span class="msg-who">Claude</span> It really is brilliant, and it’s stuck with people for nearly 300 years. Funny how an English 101 reading comes back when it suddenly fits. Anything else on your mind?</p>
</div>
<div class="msg me">
<p><span class="msg-who">Mete</span> Nope, that’s it. Thank you.</p>
</div>
<div class="msg ai">
<p><span class="msg-who">Claude</span> Thanks for the good conversation, Mete. Take care, goodbye!</p>
</div>
</div>



 ]]></description>
  <category>AI in Education</category>
  <guid>https://meteakcaoglu.com/posts/2026-11-18-modest-proposal/</guid>
  <pubDate>Wed, 18 Nov 2026 00:00:00 GMT</pubDate>
  <media:content url="https://meteakcaoglu.com/images/marks/posts/modest-proposal.svg" medium="image" type="image/svg+xml"/>
</item>
<item>
  <title>GreatSchools Ratings in Georgia</title>
  <link>https://meteakcaoglu.com/posts/2026-11-13-greatschools/</link>
  <description><![CDATA[ 





<p><img src="https://meteakcaoglu.com/images/marks/posts/greatschools.svg" class="post-mark img-fluid" alt=""></p>
<p>Many families choose a school by choosing a house. Economists call this residential sorting, often summed up as families voting with their feet: households move to the communities whose public services fit what they want and can afford (Tiebout, 1956). Schools are a large part of that choice for parents, and families pay measurably more for homes inside the attendance zones of schools they see as better (Black, 1999).</p>
<p>Many families now decide which schools are better by looking at GreatSchools. Its 1–10 rating appears next to the price on major home listing sites, and the site had more than 43 million visits in 2018 (Barnum &amp; LaMarr LeMee, 2019). A parent can see the number for a school long before visiting it, talking to other parents, or reading the state’s report card.</p>
<section id="earlier-work-on-greatschools" class="level2">
<h2 class="anchored" data-anchor-id="earlier-work-on-greatschools">Earlier Work on GreatSchools</h2>
<p>Researchers and journalists have asked what the rating captures. Private websites now do much of the public rating of US schools (Valant, 2017), and a Chalkbeat analysis found that GreatSchools ratings steer families toward whiter and more affluent schools (Barnum &amp; LaMarr LeMee, 2019). Two studies tested the ratings directly. In the first, researchers used the staged rollout of GreatSchools ratings across the country between 2006 and 2015 and found that once ratings appeared, home values, incomes, and racial composition across communities pulled further apart (Hasan &amp; Kumar, 2024). In the second, researchers compared GreatSchools components with lottery-based estimates of school quality in Denver and New York City middle schools and found that the test score component reflected who enrolls in a school more than what the school adds (Angrist et al., 2024).</p>
<p>The earlier studies covered a few cities, single components of the rating, or what the ratings do to neighborhoods. I wanted to see the summary rating, the number a parent sees, next to the state’s own scores for every school in one state.</p>
</section>
<section id="the-dataset" class="level2">
<h2 class="anchored" data-anchor-id="the-dataset">The Dataset</h2>
<p>In spring 2026, I used Claude Code to write a set of Python scrapers that collected GreatSchools ratings for Georgia’s public schools, starting from the federal list of 2,293 schools. I filled most of the gaps with retry passes and manual checks. I then joined four layers of data by school and ZIP code: the federal school records (with each school’s locale and its share of students on free or reduced-price lunch), the GreatSchools ratings, Georgia’s 2024 CCRPI scores, and the Zillow home value index. The final dataset holds 2,020 schools with both a GreatSchools rating and a CCRPI score, so I could read each rating against the school’s performance and its neighborhood at once.</p>
</section>
<section id="first-results" class="level2">
<h2 class="anchored" data-anchor-id="first-results">First Results</h2>
<div class="quarto-figure quarto-figure-center">
<figure class="figure">
<p><img src="https://meteakcaoglu.com/images/marks/greatschools.svg" class="post-mark left img-fluid figure-img" alt="A scatter of grey dots rising from lower left to upper right, with two orange dots high on the rating axis and a dashed horizontal line."></p>
<figcaption>An illustration, not the study data. Orange dots are schools rated high by GreatSchools that score below the state median.</figcaption>
</figure>
</div>
<p>I ran the analysis in R, with scripts I wrote using Claude Code. The results show that GreatSchools ratings and CCRPI scores only partly agree. The ratings follow the state’s test score measure moderately and its growth and achievement gap measures weakly, so a school rated high by GreatSchools can sit below the state median on its CCRPI scores, and a school rated low can sit above it. The full results are in a paper I am preparing for submission.</p>
</section>
<section id="a-dashboard-for-families" class="level2">
<h2 class="anchored" data-anchor-id="a-dashboard-for-families">A Dashboard for Families</h2>
<p>I built the dashboard in R Shiny with Claude Code and published it so that parents, teachers, and school leaders can check a school for themselves. You search for a school by name or ZIP code and see its GreatSchools rating beside its CCRPI scores, its student poverty rate, and local home values. You can also narrow the list by grade level and by locale (city, suburb, town, or rural).</p>
<div class="artifact-box">
<p><span class="artifact-kicker">Interactive dashboard · R Shiny</span></p>
<p><strong>Georgia School Ratings</strong></p>
<p>Look up any Georgia public school by name or ZIP code and see its GreatSchools rating next to its CCRPI scores, student poverty rate, and local home values.</p>
<p><a href="https://meteakca.shinyapps.io/georgia-school-ratings/" class="artifact-cta" target="_blank">Open the dashboard →</a></p>
</div>
<p>I would like to hear from parents and educators in Georgia who look up their school, especially if the comparison surprised them.</p>
<p><strong>References</strong></p>
<p>Angrist, J. D., Hull, P., Pathak, P. A., &amp; Walters, C. R. (2024). Race and the mismeasure of school quality. <em>American Economic Review: Insights, 6</em>(1), 20–37. <a href="https://doi.org/10.1257/aeri.20220292" class="uri">https://doi.org/10.1257/aeri.20220292</a></p>
<p>Barnum, M., &amp; LaMarr LeMee, G. (2019, December 5). How GreatSchools steers you toward whiter, more affluent schools. <em>Chalkbeat</em>. <a href="https://www.chalkbeat.org/2019/12/5/21121858/looking-for-a-home-you-ve-seen-greatschools-ratings-here-s-how-they-nudge-families-toward-schools-wi/" class="uri">https://www.chalkbeat.org/2019/12/5/21121858/looking-for-a-home-you-ve-seen-greatschools-ratings-here-s-how-they-nudge-families-toward-schools-wi/</a></p>
<p>Black, S. E. (1999). Do better schools matter? Parental valuation of elementary education. <em>Quarterly Journal of Economics, 114</em>(2), 577–599. <a href="https://doi.org/10.1162/003355399556070" class="uri">https://doi.org/10.1162/003355399556070</a></p>
<p>Hasan, S., &amp; Kumar, A. (2024). Who captures the value from organizational ratings? Evidence from public schools. <em>Strategy Science, 9</em>(3), 248–266. <a href="https://doi.org/10.1287/stsc.2023.0113" class="uri">https://doi.org/10.1287/stsc.2023.0113</a></p>
<p>Tiebout, C. M. (1956). A pure theory of local expenditures. <em>Journal of Political Economy, 64</em>(5), 416–424. <a href="https://doi.org/10.1086/257839" class="uri">https://doi.org/10.1086/257839</a></p>
<p>Valant, J. (2017, November 6). Who is evaluating US schools? <em>Brookings Institution</em>. <a href="https://www.brookings.edu/blog/brown-center-chalkboard/2017/11/06/who-is-evaluating-us-schools/" class="uri">https://www.brookings.edu/blog/brown-center-chalkboard/2017/11/06/who-is-evaluating-us-schools/</a></p>


</section>

 ]]></description>
  <category>School Ratings</category>
  <category>Computational Social Science</category>
  <guid>https://meteakcaoglu.com/posts/2026-11-13-greatschools/</guid>
  <pubDate>Fri, 13 Nov 2026 00:00:00 GMT</pubDate>
  <media:content url="https://meteakcaoglu.com/images/marks/posts/greatschools.svg" medium="image" type="image/svg+xml"/>
</item>
<item>
  <title>Learning-Focused Design for the LMS</title>
  <link>https://meteakcaoglu.com/posts/2026-10-29-course-timeline/</link>
  <description><![CDATA[ 





<div class="callout callout-style-simple callout-note">
<div class="callout-body d-flex">
<div class="callout-icon-container">
<i class="callout-icon"></i>
</div>
<div class="callout-body-container">
<p><strong>Update, November 12, 2026.</strong> I am now building this layout as a learning-focused LMS for any course. A course with a project shows its steps on the timeline, and a course without one shows its modules. The feed now has posts, comments, and a group chat, and a to-do list above the timeline shows what is late and what is due this week. <a href="https://claude.ai/artifact/HBCHKNbKZE4PSZ3t3EdvnS" target="_blank">Open the new prototype</a> to try the student and instructor views on invented data, in four visual designs I am still comparing. The essay below is Version 1, as deposited on Zenodo.</p>
</div>
</div>
</div>
<p><img src="https://meteakcaoglu.com/images/marks/posts/course-timeline.svg" class="post-mark img-fluid" alt=""></p>
<p>Every learning management system I know is a content management system, unfortunately (Brown et al., 2015). Instructors upload modules, files, discussion boards, and a gradebook, and the course takes the shape of those folders.</p>
<p>A learner needs to see two things in a course: their work over time and the people doing it with them. Students piece the first together from module pages and due dates, because the folders hide the shape of the project. For the second, the LMS offers little. Social presence grows when the design supports it, through small groups and through spaces students already use, such as Facebook groups (Lee &amp; Akcaoglu, 2017). Discussion boards crept in to fill the social gap. They did a poor job of reflecting the buzz of a classroom, and lately most students’ posts read as AI-generated.</p>
<p>I imagine a learning-focused layout with two parts. The work sits in the middle as a timeline of the course project, and each step opens into what to do, where to submit it, how much it counts, and how AI may be used at that step. The people sit on the side in an activity feed, where students share finds, questions, moments when AI helped or misled them, and updates on their work. Finishing a step on the timeline posts an update to the feed, so the work and the people stay connected.</p>
<p>I built the timeline for my Instructional Design course (FRIT 7231), and it runs as a page inside our LMS. For now, the timeline and the feed will be part of the existing course, a layer on top of the content system. The feed is designed and still to be built.</p>
<div class="quarto-figure quarto-figure-center">
<figure class="figure">
<p><img src="https://meteakcaoglu.com/posts/2026-10-29-course-timeline/timeline-screenshot.png" class="img-fluid figure-img" alt="A Gantt-style course timeline. Each assignment is a row across the 15-week term, with a red line marking today. One row, Part III Task analysis, is open and shows its due date, where to submit it, its weight, what to do, and the rule for AI use: only after the expert session."></p>
<figcaption>The project timeline for FRIT 7231, viewed as of Week 6, with Part III (task analysis) open.</figcaption>
</figure>
</div>
<div class="artifact-box">
<p><span class="artifact-kicker">Prototypes · HTML</span></p>
<p><strong>Learning-Focused LMS</strong></p>
<p>Try the student and instructor views of the new LMS on invented data, or open the original FRIT 7231 project timeline.</p>
<p><a href="https://claude.ai/artifact/HBCHKNbKZE4PSZ3t3EdvnS" class="artifact-cta" target="_blank">Open the LMS prototype →</a> <a href="https://claude.ai/artifact/8jXYsPCujKgBVyVsPinQqj" class="artifact-cta" target="_blank">Open the timeline →</a></p>
</div>
<p>If you teach online and are trying something similar, I would like to hear how it goes.</p>
<p><strong>Cite this post:</strong> Akcaoglu, M. (2026). <em>Learning-Focused Design for the LMS</em> (Version 1). Zenodo. <a href="https://doi.org/10.5281/zenodo.23773997" class="uri">https://doi.org/10.5281/zenodo.23773997</a></p>
<p><strong>References</strong></p>
<p>Brown, M., Dehoney, J., &amp; Millichap, N. (2015). <em>The next generation digital learning environment: A report on research</em> [ELI Paper]. EDUCAUSE. <a href="https://library.educause.edu/resources/2015/4/the-next-generation-digital-learning-environment-a-report-on-research" class="uri">https://library.educause.edu/resources/2015/4/the-next-generation-digital-learning-environment-a-report-on-research</a></p>
<p>Lee, E., &amp; Akcaoglu, M. (2017). Sociability of online learning environments. In J. M. Spector, B. B. Lockee, &amp; M. D. Childress (Eds.), <em>Learning, design, and technology</em>. Springer. <a href="https://doi.org/10.1007/978-3-319-17727-4_39-2" class="uri">https://doi.org/10.1007/978-3-319-17727-4_39-2</a></p>



 ]]></description>
  <category>Teaching</category>
  <category>Online Learning</category>
  <guid>https://meteakcaoglu.com/posts/2026-10-29-course-timeline/</guid>
  <pubDate>Thu, 29 Oct 2026 00:00:00 GMT</pubDate>
  <media:content url="https://meteakcaoglu.com/images/marks/posts/course-timeline.svg" medium="image" type="image/svg+xml"/>
</item>
<item>
  <title>The Offloading Ladder</title>
  <link>https://meteakcaoglu.com/posts/2026-10-28-offloading-ladder/</link>
  <description><![CDATA[ 





<p>This is my first preprint, and my first conceptual piece.</p>
<p>As we see AI taking over student work (and of course professional work), I have been thinking about a good way to think about its integration. Going back to theories in education (and recent studies on AI and learning), I drafted this piece and am sharing it as a preprint. It grew out of two earlier posts here, <a href="../../posts/2026-05-19-radiology-lesson/index.html">The Radiology Lesson</a> and <a href="../../posts/2026-05-19-beyond-the-panic/index.html">Beyond the Panic</a>.</p>
<p>The big idea is in the figure below. Tech-human partnership has always existed, but with AI, the cognitive partnership can be dangerous as well as helpful. I suggest that offloading happens on a ladder, and a task’s order depends on the person: it is high order while they still have to figure out what to do, and low order once they know how. Once a task becomes low order for a person, they can offload that part to technology, until they reach the core, which depends on the person and the job.</p>
<div class="quarto-figure quarto-figure-center">
<figure class="figure">
<p><img src="https://meteakcaoglu.com/posts/2026-10-28-offloading-ladder/figure-1-offloading-ladder.png" class="img-fluid figure-img" alt="Diagram of the Offloading Ladder: as experience with a task grows, more of it becomes low order and can be offloaded to AI, up to the core of the work, which stays with the person."></p>
<figcaption>The Offloading Ladder (Figure 1 in the preprint).</figcaption>
</figure>
</div>
<p>For learners, classrooms and courses need to determine the core and keep it off AI. AI companies need a better way to protect human learning.</p>
<div class="artifact-box">
<p><span class="artifact-kicker">Preprint · EdArXiv</span></p>
<p><strong>The Offloading Ladder</strong></p>
<p>Akcaoglu, M. (2026). <em>The Offloading Ladder: How Experience Changes What Learners and Practitioners Should Hand to Artificial Intelligence.</em> <a href="https://doi.org/10.35542/osf.io/ys2fu_v1" class="uri">https://doi.org/10.35542/osf.io/ys2fu_v1</a></p>
<p><a href="https://doi.org/10.35542/osf.io/ys2fu_v1" class="artifact-cta" target="_blank">Read the preprint →</a></p>
</div>
<hr>
<p><em>How I used AI.</em> Since the piece is about AI, let’s talk about it too. As I make the case in the paper, AI usage also ranges from low order to high order, and I believe I kept the high-order work for myself. I started from my initial idea and outline. I used Undermind to search the literature and Claude Code to gather notes from my readings (my highlights in Zotero). Over many sessions, I built the paper step by step, creating each paragraph and argument and using Claude Code to shape paragraphs from my arguments and reading notes, and I edited and rewrote each draft along the way. Claude Code also checked references and accuracy. I used Claude Design to draw the figure. Going back to the ladder, I kept the core (the idea and editorial ownership), and Claude Code helped with tasks that were low order for me, like pulling my highlights into my outline.</p>



 ]]></description>
  <category>AI in Education</category>
  <guid>https://meteakcaoglu.com/posts/2026-10-28-offloading-ladder/</guid>
  <pubDate>Wed, 28 Oct 2026 00:00:00 GMT</pubDate>
  <media:content url="https://meteakcaoglu.com/posts/2026-10-28-offloading-ladder/figure-1-offloading-ladder.png" medium="image" type="image/png" height="103" width="144"/>
</item>
<item>
  <title>Analyzing Classroom Images Locally with R and LLMs</title>
  <link>https://meteakcaoglu.com/posts/2026-05-26-local-llms-images/</link>
  <description><![CDATA[ 





<p><img src="https://meteakcaoglu.com/images/marks/posts/local-llms-images.svg" class="post-mark img-fluid" alt=""></p>
<p>We are almost done with our new book <a href="https://data-edu.github.io/computational-social-science-cookbook/">Computational Social Science Cookbook</a>, a collaborative open-access book aimed at education researchers working with, or looking to expand their research toolkits with, computational methods. Among many other things, I led the chapter on using local (or cloud) LLMs to conduct methodological image analyses. The chapter covers something I think is underused in the field: analyzing classroom images with vision-language models in R.</p>
<p>The chapter is practical and self-contained, like the rest of the book, which lets researchers follow the steps and repeat the analysis on their own datasets. If you have R and Ollama installed, you can follow along and run the code. The method is worth a brief argument, because most researchers who haven’t tried image analysis skip it for lack of awareness of how easy it can be and how rich the results can be. This point sits at the heart of the book: “things” beyond what we traditionally consider can be data, and these open new research questions. The barriers are mostly technical, and privacy is the largest concern, one that local models now largely solve.</p>
<p>Images are hard to analyze at scale, and that difficulty has kept them out of most research designs. Manual coding is slow, expensive, and hard to replicate. School photos in educational contexts contain students specifically, and sending them to a commercial API raises real privacy and IRB concerns that most researchers are unwilling to navigate.</p>
<p>This method has direct research applications. A researcher can record classroom interactions and use the resulting visuals to answer questions about instructional format, spatial arrangement, and student engagement, the example used in the book. The example focuses on student engagement, but the method extends to any research question that visual data can answer. For the chapter’s worked example, we used classroom photos pulled from Wikimedia Commons. I ran the workflow across three questions: instructional format, group composition, and student engagement. The results were accurate and easy to verify, and the workflow processed the images far faster than manual coding would allow. The model has limits, and careful qualitative analysis remains necessary alongside it. The workflow is a useful starting point, and a way to get more out of visual data than researchers have in the past. The book also has a full section on using LLMs for qualitative data analysis directly, with a responsible use framework built around correctness, transparency, and reproducibility. More on that <a href="https://data-edu.github.io/computational-social-science-cookbook/section-3-Intro.html">here</a>.</p>
<p>This analysis was possible through the <a href="https://github.com/frankiethull/kuzco"><code>{kuzco}</code> package</a>, which wraps Ollama’s local model inference and exposes four functions: classification, object recognition, sentiment estimation, and a free-form custom prompt. About a dozen lines of code takes a folder of classroom photos to structured, analyzable output, credit to <a href="https://bsky.app/profile/frankiethull.bsky.social">Frank Hull</a>, the package’s author, for making this so accessible.</p>
<p>Visual data has been treated as supplementary in educational research mostly because analyzing it at scale was impractical. Local vision-language models now make it practical. For researchers working with classroom images, observational photos, or any kind of visual data, education or not, the chapter is a working starting point in R. The book is open access, so you can read it online for free, and the code is all available on GitHub. I hope this encourages more researchers to consider visual data in their work, and to use local LLMs to analyze it in a way that is ethical, practical, and insightful.</p>
<div class="artifact-box">
<p><span class="artifact-kicker">Open-access book · R</span></p>
<p><strong>Computational Analysis of Educational Data</strong></p>
<p>Wang, W., Akcaoglu, M., Rosenberg, J. M., &amp; Kellogg, S. (in press). <em>Computational Analysis of Educational Data: A Field Guide Using R.</em> Chapman &amp; Hall/CRC.</p>
<p><a href="https://data-edu.github.io/computational-social-science-cookbook/" class="artifact-cta" target="_blank">Read the book →</a> <a href="https://data-edu.github.io/computational-social-science-cookbook/section-3-Intro.html" class="artifact-cta" target="_blank">LLMs for qualitative analysis →</a> <a href="https://www.routledge.com/Computational-Analysis-of-Educational-Data-A-Field-Guide-Using-R/Wang-Akcaoglu-Rosenberg-Kellogg/p/book/9781032436760" class="artifact-cta" target="_blank">Publisher page →</a></p>
</div>



 ]]></description>
  <category>Research Methods</category>
  <category>AI in Education</category>
  <category>Computational Social Science</category>
  <guid>https://meteakcaoglu.com/posts/2026-05-26-local-llms-images/</guid>
  <pubDate>Tue, 26 May 2026 00:00:00 GMT</pubDate>
  <media:content url="https://meteakcaoglu.com/images/marks/posts/local-llms-images.svg" medium="image" type="image/svg+xml"/>
</item>
<item>
  <title>The Radiology Lesson</title>
  <link>https://meteakcaoglu.com/posts/2026-05-19-radiology-lesson/</link>
  <description><![CDATA[ 





<p><img src="https://meteakcaoglu.com/images/marks/posts/radiology-lesson.svg" class="post-mark img-fluid" alt=""></p>
<p>I drive to Statesboro for work. It is an hour-long drive, enough to listen to podcasts but also to reflect on the changes in education, in higher ed and in K-12 (especially as a parent: “what will my son do in the future?” comes to mind a lot).</p>
<p>I was listening to Guy Raz’s <em>How I Built This</em> podcast, his interview with Jensen Huang. The conversation turned to jobs and AI. Huang used radiology, not teaching, as his example, and it stuck with me.</p>
<p>Experts declared that radiology would be the first profession AI eliminated. The logic made sense at the time: early AI was built for computer vision, computer vision could read scans better than any human, so radiologists were done. This prediction turned out to be right and wrong. Every radiologist now uses AI to read scans, faster and with fewer errors, which lets them see more patients. AI did not replace radiologists; it made them more productive. The job changed and persisted.</p>
<p>Huang pointed to the distinction between a job’s <em>purpose</em> and its <em>tasks</em>. The distinction seems obvious in hindsight, yet I had not framed it this way before. In the radiology example, reading scans is a lower-order task: routine, pattern-based, the kind of work AI handles well. Diagnosing disease is the purpose, the higher-order task, and it requires clinical judgment, context, and information no model has. AI took over the lower-order task, and human radiologists (in the loop) took on more of the higher-order work. Faster scanning let hospitals order more scans and see more patients, which increased demand for radiologists.</p>
<p>Huang located the harm in the prediction itself. When experts declared radiology obsolete, young people who might have entered the field chose something else instead, and the field ended up short on radiologists. The narrative did what the technology could not. A shrinking pool of radiologists could also give AI tools a larger share of the field than they would have earned otherwise, turning the narrative into a self-fulfilling prophecy.</p>
<p><em>(A grain of salt here: Huang is the CEO of Nvidia. He has a clear interest in AI optimism. But I think the point stands.)</em></p>
<p>I believe the same thing applies to teaching.</p>
<p>GenAI will take over the lower-order tasks of teaching: writing lessons, building assessments, giving feedback. The purpose of teaching is learning, and the work that serves it is building experiences, guiding students through complexities, and asking the question that shifts perspective and builds motivation to learn. Offloading the production tasks should push higher-order work to the center.</p>
<p>Education will have abundant AI tools, and, hopefully, teachers with more time to spend on that higher-order work.</p>
<p>AI will change teaching, as every technology integration does. The profession will persist, and it will demand more of the skills that are hardest to replace. The question is whether we say that clearly enough to keep people coming in.</p>
<hr>
<p><em>Update (October 2026): I developed the lower-order/higher-order argument into a paper, <a href="../../posts/2026-10-28-offloading-ladder/index.html">The Offloading Ladder</a> (<a href="https://doi.org/10.35542/osf.io/ys2fu_v1">preprint</a>).</em></p>
<p><em>Further reading: Hamilton, Rosenberg &amp; Akcaoglu (2016) examined the SAMR model as a framework for thinking about levels of technology integration in teaching, in <a href="https://doi.org/10.1007/s11528-016-0091-y">TechTrends, 60(5)</a>. Koehler, Mishra, Akcaoglu &amp; Rosenberg (2013) addressed TPACK (the knowledge teachers need to integrate technology effectively) in Bharati &amp; Mishra (Eds.), ICT Integrated Teacher Education Models.</em></p>



 ]]></description>
  <category>AI in Education</category>
  <guid>https://meteakcaoglu.com/posts/2026-05-19-radiology-lesson/</guid>
  <pubDate>Tue, 19 May 2026 00:00:00 GMT</pubDate>
  <media:content url="https://meteakcaoglu.com/images/marks/posts/radiology-lesson.svg" medium="image" type="image/svg+xml"/>
</item>
<item>
  <title>Beyond the Panic</title>
  <link>https://meteakcaoglu.com/posts/2026-05-19-beyond-the-panic/</link>
  <description><![CDATA[ 





<p><img src="https://meteakcaoglu.com/images/marks/posts/beyond-the-panic.svg" class="post-mark img-fluid" alt=""></p>
<p>Educators, rightfully so, are concerned about the impact of GenAI on their profession, myself included. The fear is that it will make it too easy for students to outsource the work of learning, and that what gets lost in the process is the thinking itself.</p>
<p>I understand the fear. I also think it may be mostly misplaced, or at least productive if we let it be.</p>
<p><em>This piece grew out of a keynote I gave at <a href="https://icits2024.kastamonu.edu.tr/index.php/en/">ICITS 2024</a>, Kastamonu University. I later developed its argument into a paper, <a href="../../posts/2026-10-28-offloading-ladder/index.html">The Offloading Ladder</a> (<a href="https://doi.org/10.35542/osf.io/ys2fu_v1">preprint</a>).</em></p>
<p>GenAI is genuinely disruptive. The disruption also reveals something we should have seen all along: many of our standard practices were never well-aligned with how people actually learn, the 250-student lecture hall among them. GenAI made that gap impossible to ignore.</p>
<p>The book summary illustrates the drift. For decades, we assigned it as a proxy for reading comprehension, critical thinking, and synthesis. What we were actually assessing was a product, a piece of text. The learning was supposed to happen in the process of writing it: the struggle to clarify your own thinking, to hold competing ideas in tension, to find the thread. Writing is thinking.</p>
<p>A student can now generate that product in seconds. Many do so because the assignment rewarded the deliverable and left the process invisible and ungraded.</p>
<p>The book summary was already broken. GenAI simply exposed that we were grading the wrong thing.</p>
<p>Learning is an active, constructive process. You learn to solve problems by solving them. You learn to write by writing. The struggle <em>is</em> the learning: “the conditions that produce the most errors during acquisition are often the very conditions that produce the most learning” (<a href="https://stats.andrewheiss.com/pinky-lychee/">Heiss, 2026</a>). LLMs, by design, eliminate that friction.</p>
<p>Learning is also social. It is students in a room, pushing back on each other’s ideas, sitting with confusion until something clicks. And it requires a <em>why</em>, not just grades, but the intrinsic experience of mastery and the personal relevance of what you are creating.</p>
<p>AI can produce a final product. It cannot replicate the cognitive or emotional experience of getting there.</p>
<p>Technology has three cognitive effects: you can think <em>with</em> it (partnering with a tool to accomplish a task), think <em>of</em> it (the cognitive residue left behind after the tool is gone), or think <em>through</em> it (the tool fundamentally reorganizes how you understand something) (Salomon &amp; Perkins, 2005).</p>
<p>For a veteran teacher who has designed thousands of lessons, using GenAI to brainstorm amplifies existing expertise. That is thinking <em>with</em>. A novice teacher or a student still developing those skills faces a different risk: if the tool does the cognitive work before it is internalized, the result is what researchers have called cognitive debt, the illusion of understanding without the substance. That is the dark side of thinking <em>of</em>.</p>
<p>The <a href="../../posts/2026-05-19-radiology-lesson/index.html">lower-order versus higher-order distinction</a> does the real work here. AI is excellent at offloading lower-order production tasks. The danger is letting it offload the higher-order thinking that was supposed to be the point.</p>
<p>Our <em>practices</em> have to change. Assignments can no longer treat the product as the point. Design has to assume AI use and make the process matter more than the deliverable; detection is a losing strategy (Piaget, 1970; Vygotsky, 1978; Lave &amp; Wenger, 1991).</p>
<p>Our <em>theories</em> remain sound, and new ones will keep emerging as theorizing continues. GenAI makes active learning, constructivism, situated cognition, and transparent pedagogy more urgent, not less. Students need an explicit <em>why</em> for the struggle we ask of them. Tasks need to make the doing the learning, authentic enough that delegating it to a machine would feel like giving away the interesting part.</p>
<p>I have seen this work in my own game design classes. Students building their own games put in the effort because they are intrinsically motivated to play and improve their own creations. The task itself makes “cheating” beside the point.</p>
<p>GenAI forces us to ask whether our educational practices actually reflect what we know about how people learn. In most cases, the honest answer is that they reflect it poorly.</p>
<p>The relationship between a good teacher and a curious student remains the foundation of education, and GenAI raises its stakes. Our job is to design learning experiences where students would want to keep the thinking for themselves.</p>
<p>A teacher who offloads lower-order tasks will hopefully have more time and energy to focus on that relationship. A student who has internalized the process of learning will be more motivated to engage with it. I believe the disruption, if we take it seriously, pushes us in that direction.</p>



 ]]></description>
  <category>AI in Education</category>
  <category>Teaching</category>
  <guid>https://meteakcaoglu.com/posts/2026-05-19-beyond-the-panic/</guid>
  <pubDate>Tue, 19 May 2026 00:00:00 GMT</pubDate>
  <media:content url="https://meteakcaoglu.com/images/marks/posts/beyond-the-panic.svg" medium="image" type="image/svg+xml"/>
</item>
</channel>
</rss>
