<?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 Merl Syrex Chandana on Medium]]></title>
        <description><![CDATA[Stories by Merl Syrex Chandana on Medium]]></description>
        <link>https://medium.com/@Merl_Syrex?source=rss-356bb67a2844------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*_w1C5XagdimEVBtfqein6A.jpeg</url>
            <title>Stories by Merl Syrex Chandana on Medium</title>
            <link>https://medium.com/@Merl_Syrex?source=rss-356bb67a2844------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 01:46:50 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@Merl_Syrex/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[Week 1: Notes from the AI Hivemind]]></title>
            <link>https://medium.com/@Merl_Syrex/week-1-notes-from-the-ai-hivemind-cfb78764b456?source=rss-356bb67a2844------2</link>
            <guid isPermaLink="false">https://medium.com/p/cfb78764b456</guid>
            <dc:creator><![CDATA[Merl Syrex Chandana]]></dc:creator>
            <pubDate>Wed, 15 Apr 2026 18:10:15 GMT</pubDate>
            <atom:updated>2026-04-15T18:21:02.553Z</atom:updated>
            <content:encoded><![CDATA[<h4>On ambition, delusion, &amp; techno-optimism</h4><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/1*j5MrGbmu52BHQqp3vFeT7A.png" /><figcaption>It’s going to take a whole of society approach to see this one through</figcaption></figure><p>It is all Dean Ball this week. All of them are great.</p><ol><li>𝙈𝙚𝙖𝙨𝙪𝙧𝙚 𝙐𝙥, 𝘿𝙚𝙖𝙣 𝙒. 𝘽𝙖𝙡𝙡</li></ol><p>Link: <a href="https://www.hyperdimensional.co/p/measure-up-re-posted?fbclid=IwZXh0bgNhZW0CMTAAYnJpZBExeWRQenQ0VVBSa1pweDllTXNydGMGYXBwX2lkEDIyMjAzOTE3ODgyMDA4OTIAAR666UCRHdbBjLFaaMra9x-0yQILfn88uGgINzPOkL1qaMS-kEvsbvfg3SSqaQ_aem_GUTWvCawtrPGSe9qL1kcqw">https://www.hyperdimensional.co/p/measure-up-re-posted</a></p><p>This is a slightly philosophical essay on how new technology enables creations of greater complexity, scale and ambition. I felt it was also an ode to Hayek’s Knowledge Problem. Beethovan’s Vienna in the late 18th &amp; early 19th century, had hyper-local interactions between the piano manufacturers, composers and opera audiences that led to innovations in piano making.</p><p>AI is the modern day piano. But 100x or 1000x. In many different directions</p><p>2. 𝙏𝙝𝙚 𝙈𝙤𝙫𝙞𝙣𝙜 𝙖𝙣𝙙 𝙩𝙝𝙚 𝙎𝙩𝙞𝙡𝙡, 𝘿𝙚𝙖𝙣 𝙒. 𝘽𝙖𝙡𝙡</p><p>Link: <a href="https://www.hyperdimensional.co/p/the-moving-and-the-still?fbclid=IwZXh0bgNhZW0CMTAAYnJpZBExeWRQenQ0VVBSa1pweDllTXNydGMGYXBwX2lkEDIyMjAzOTE3ODgyMDA4OTIAAR4ZXa1IbCXBIftpCzaoEhxd7y39pkzF9FiUEHaxIkXw3A8yaRQigIg3zB2tCg_aem_gFnAr8I-apVq--rBDmN0mg">https://www.hyperdimensional.co/p/the-moving-and-the-still</a></p><p>It is a thought-provoking essay on why certain denialist, head-in-the-sand approaches to frontier AI developments may backfire for developing and middle-tier countries. The point is not that such countries should chase every fad. It is that, while they do need to make calculated bets, some now-familiar Global South talking points (like those below) can harden into bad policy if they become a way of looking away.</p><blockquote>→ “The winner of the AI race will be the people, organizations, and countries that diffuse small AI models and other sub-frontier AI capabilities the fastest.”</blockquote><blockquote>→ “Small models with low compute intensity are catching up rapidly to the largest frontier models.”</blockquote><blockquote>→ “Frontier AI advances are beginning to plateau.”</blockquote><p>There is a real balance to be struck here. One mistake is to deny the moving. Another is to become so captivated by the moving that you neglect the still. Countries like ours do need to pick our battles. But we should pick them with our eyes open.</p><p>3. 𝙏𝙝𝙚 𝙁𝙪𝙩𝙪𝙧𝙚 𝘼𝙣𝙙 𝙞𝙩𝙨 𝙁𝙧𝙞𝙚𝙣𝙙𝙨, 𝘿𝙚𝙖𝙣 𝙒. 𝘽𝙖𝙡𝙡 —</p><p>Link: <a href="https://l.facebook.com/l.php?u=https%3A%2F%2Fwww.hyperdimensional.co%2Fp%2Fthe-future-and-its-friends%3Ffbclid%3DIwZXh0bgNhZW0CMTAAYnJpZBExeWRQenQ0VVBSa1pweDllTXNydGMGYXBwX2lkEDIyMjAzOTE3ODgyMDA4OTIAAR6ucaG9l43T91PDZHi2whzEbQvcZCyKsnOdhqA4L_laGn7GohtUFiWatOHdhg_aem_vulrj2u-n9jBU9j6coq8qQ&amp;h=AT6lY8MaNl-IVgLSxvNRP3T9AkWazZkirqPBBAPN5EoEeBPlOT2ezcLQeIU8mreaQChfRDizPcz4i_gkk9c5VcCYbDP0amVqx2Qd0t46sxWBnroZlPYgFr2Y_LsARFyk_fgmMg76lDWFZ-w1_gJw6up2_pQwvOa-bTSPWA&amp;__tn__=-UK*F">https://www.hyperdimensional.co/p/the-future-and-its-friends</a></p><p>This is a techno-optimist essay. Not from the money-dangling VC kind. But a philsophical, almost-first principled inquiry into the kind of changes, institutions and thinking we need as a species to see this transition (of AI) through.</p><blockquote>“The future I hope will be hard-won. It will require more than just reasonable compromises on AI legislation-much more. It will require more than merely pushing back on ill-conceived regulations or cutting red tape. These things are small slivers — shadows, really — of what I have in mind.”</blockquote><blockquote>“Instead, the future I hope for will have to be advocated for with the utmost zeal and fierce belief. It will demand cajoling, coaxing, persuading, and not a small amount of fighting. Most importantly, though, it will require imagination: a willingness to invent the assembly line and the factory, to cast the old and start fresh.”</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cfb78764b456" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Notes from the AI Hive Mind]]></title>
            <link>https://medium.com/@Merl_Syrex/notes-from-the-ai-hive-mind-dca3dd692708?source=rss-356bb67a2844------2</link>
            <guid isPermaLink="false">https://medium.com/p/dca3dd692708</guid>
            <category><![CDATA[sensemaking]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[reading]]></category>
            <category><![CDATA[society]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Merl Syrex Chandana]]></dc:creator>
            <pubDate>Wed, 15 Apr 2026 18:05:37 GMT</pubDate>
            <atom:updated>2026-04-15T18:05:37.912Z</atom:updated>
            <content:encoded><![CDATA[<h4>Reading, thinking and scribbling about THE technolgical transformation of our time</h4><p>I have decided to make a more disciplined and organized effort around reading on what I think of as the societal implications of AI, and to write periodically (hopefully weekly) about what I am reading.</p><p>Most of this is to give room for myself to think more carefully about AI. But I also hope that, over time, it becomes useful to others trying to make sense of the field without getting lost in its noise.</p><p>The notes will be light: there won’t be impassioned defenses or refutations. Just observations and prompts to get thinking. I am staying lean to stay consistent.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*j5MrGbmu52BHQqp3vFeT7A.png" /><figcaption>It’s going to take a whole of society approach to see this one through</figcaption></figure><p>A few working heuristics on what/how I am going to read:</p><p>1. Read everything written by a small number of people worth following closely. I have started, for now, with Dean W. Ball (primary staff drafter of America’s AI Action Plan). His writings are here: <a href="https://www.deanball.com/writing-media?fbclid=IwZXh0bgNhZW0CMTAAYnJpZBExeWRQenQ0VVBSa1pweDllTXNydGMGYXBwX2lkEDIyMjAzOTE3ODgyMDA4OTIAAR6KwxT91nRbHMET1OIePvdfpifAW9AntNlCT2Rgs5S8umlsjHqpyDMhVOGCjA_aem_JXL64D4OtBUCLgmGKmbSOQ">https://www.deanball.com/writing-media</a></p><p>2. Read a handful of important and/or high-quality pieces each week. These might be newspaper articles, classic academic papers, essays, or transcripts of speeches and talks. I want to keep this deliberately flexible, because good signals can come from a variety of places.</p><p>3. Read fiction from earlier periods of technological and industrial change. I have a plan for this and will share more on that soon.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dca3dd692708" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[“Troubled” by Rob Henderson]]></title>
            <link>https://medium.com/@Merl_Syrex/troubled-by-rob-henderson-82b13d2d6daa?source=rss-356bb67a2844------2</link>
            <guid isPermaLink="false">https://medium.com/p/82b13d2d6daa</guid>
            <dc:creator><![CDATA[Merl Syrex Chandana]]></dc:creator>
            <pubDate>Sat, 08 Nov 2025 14:39:52 GMT</pubDate>
            <atom:updated>2025-11-08T16:15:12.318Z</atom:updated>
            <content:encoded><![CDATA[<h4>Some notes and reflections</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cHjrVz4jx9K86QIZ6Dn8vA.jpeg" /></figure><p>I can’t quite remember what led me to pick up “Troubled” by Rob Henderson, but I was hooked within the first few pages. I found myself stealing pockets of time (even within a very busy week at work) to keep reading.</p><p>1. I both could and couldn’t relate to Henderson’s story. Much of his childhood was nothing like mine — neglect from both parents, a mother battling addiction, years spent in multiple foster homes before turning ten, exposure to substances that numbed pain, and a constant tendency to engage in destructive behaviour, all underpinned by pain and purposelessness. None of that was part of my experience. Yet, some elements resonated deeply: the transformative power of having someone genuinely believe in you, inability to relate to some experiences of my own childhood friends, and the formative influence of early reading. The book made me reflect on my own childhood and helped me piece together explanations for some of the tendencies I still notice in myself today.</p><p>2. The implications of the book on education are profound. As a researcher thinking about how technology can improve education, the book also reminded me that educating the next generation isn’t just a pedagogical challenge — and certainly not just a technological one. Student engagement, which scholars now recognize as one of the strongest drivers of learning, ultimately stems from the value children place on learning itself.</p><p>What struck me most, though, was Henderson’s reflection on emotional security and its primacy over upward mobility. It floored me:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1001/1*_TARd5-9l2wrNtrAd9WM_g.jpeg" /></figure><p>3. It has a distilled treatise on “luxury beliefs”. I was already familiar with Rob Henderson’s concept of luxury beliefs — ideas and opinions that confer status on the upper class at little cost, while often imposing real costs on the lower classes. But Troubled expands on the concept in rich, personal detail. Through a series of vivid anecdotes, Henderson brings academic clarity and emotional weight to his argument. I also found his examination of “class” — not just in terms of income, but as expressed through vocabulary, ideas, and beliefs — particularly fascinating. The chapter on luxury beliefs alone makes the book worth reading, in my view.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/999/1*JbL1tIQz-qdS2qS-d3EisA.jpeg" /></figure><p>And finally, the book is written in remarkably readable prose — clear (one could even say it is “too raw” at times), engaging, and punctuated with sharp insights. It also helps that it’s under 250 pages; even with a busy schedule, it’s the kind of book you could easily finish within a week.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=82b13d2d6daa" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Blank Page, The Shitty Draft and The Machine]]></title>
            <link>https://medium.com/@Merl_Syrex/the-blank-page-the-shitty-draft-and-the-machine-7fc2bd860e42?source=rss-356bb67a2844------2</link>
            <guid isPermaLink="false">https://medium.com/p/7fc2bd860e42</guid>
            <dc:creator><![CDATA[Merl Syrex Chandana]]></dc:creator>
            <pubDate>Sun, 26 Oct 2025 10:17:15 GMT</pubDate>
            <atom:updated>2025-10-26T10:26:02.860Z</atom:updated>
            <content:encoded><![CDATA[<p>Thoughts on AI, judgement, and creative inertia</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*C4lKWLYYj5naAVLCl1d4ZA.jpeg" /><figcaption>Created Using ChatGPT 5</figcaption></figure><p>Ever since people first put words to page (stone, parchment, whatever), the hardest part of writing has, arguably, always been the beginning. The blank page feels daunting. Anne Lamott, in her book Bird by Bird, offered a deceptively simple (even offensive, one might argue) solution: give yourself permission to write a “shitty first draft.” Start messy, start wrong — just start. The act of putting words down, no matter how bad, breaks the paralysis. What Lamott did for writers, AI now does for anyone facing creative inertia. It gives you a collaborator who never stares back in silence. You don’t have to suspend your judgment, but it can dissolve the fear that stops you from exercising it.</p><p>But that gift only works if you’ve done the “busy work” before — if you’ve built things line by line, debugged, revised, and fought with details until they succumbed to the force of your skill and doggedness. People who’ve done that know how the machinery of good work actually runs. They know that a beautifully phrased answer can still be dead wrong. They have the judgement to completely discard a seemingly convincing argument. For them, AI is a tool to accelerate and provoke — a way to move from zero to something. Even when that something is completely discarded eventually, they have gained momentum. And momentum is all you need.</p><p>For those who’ve never learned to wrestle with the grind, the machine can become a mirror that flatters. It nods and bounces your brilliance back to you — never challenging, always agreeable. Psychologists call this kind of shared delusion a folie à deux, a “madness of two.” In this case, the partners are human and algorithm, spiraling together in perfect, confident error. Those who’ve fought their way through the work before recognize the danger. They know how easily fluency can be mistaken for truth (they’ve probably encountered a few snake-oil salesmen in their time) — and how essential it is to test what sounds right against what is right</p><p>Here’s Anne’s book, “Bird by Bird” — <a href="https://www.amazon.com/Bird-Some-Instructions-Writing-Life/dp/0385480016">https://www.amazon.com/Bird-Some-Instructions-Writing-Life/dp/0385480016</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7fc2bd860e42" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[An (A)I-dentity Crisis]]></title>
            <link>https://medium.com/@Merl_Syrex/an-a-i-dentity-crisis-8e279864a60e?source=rss-356bb67a2844------2</link>
            <guid isPermaLink="false">https://medium.com/p/8e279864a60e</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[generative-ai-use-cases]]></category>
            <category><![CDATA[ai-in-government]]></category>
            <category><![CDATA[responsible-ai]]></category>
            <category><![CDATA[ai-ethics]]></category>
            <dc:creator><![CDATA[Merl Syrex Chandana]]></dc:creator>
            <pubDate>Fri, 06 Jun 2025 05:48:08 GMT</pubDate>
            <atom:updated>2025-06-06T08:37:30.181Z</atom:updated>
            <content:encoded><![CDATA[<h4>The perils — and rare uses — of generative AI in criminal investigations</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/771/0*rxK7Bbt3VX6JpIdF" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/630/0*SFHJ3T7Vrdv7Vw0q.jpg" /><figcaption>Image Credits: <a href="https://x.com/SriLankaTweet">@SriLankaTweet</a> &amp; <a href="https://www.dailymirror.lk/breaking-news/CID-releases-AI-generated-images-of-wanted-suspects/108-310428">The Daily Mirror</a></figcaption></figure><p>A few days ago, a lawyer friend WhatsApped me the <a href="https://www.dailymirror.lk/breaking-news/CID-releases-AI-generated-images-of-wanted-suspects/108-310428">now viral story</a> about Sri Lanka’s CID using an AI-generated photorealistic image of a suspect. He asked what I thought. I’d seen the pictures on social media but hadn’t looked into it. So I replied, “I have loose opinions on this — the fact that it’s AI-generated isn’t inherently good or bad.” I still stand by that response — and I’d recommend it as a general rule for AI. But I could have done better. My reply was basically just “it depends.” I should have asked more questions, thought through the details, and told him what it actually depends on.</p><p>So here’s the response I should have given him. In fact, he’s already seen this.</p><p>If you want to skip to the conclusion, it’s this: Given what we know about the technology, its limitations, and the realities of the institutional context in which it’s being used, generating suspect images from sketches using AI is a bad idea.</p><p><strong><em>Disclaimer: </em></strong><em>This isn’t original reporting on the facts of the matter. It’s commentary on one use case of generative AI being used in the justice process — and a look at the dangers and possibilities that come with it. According to the news article, the images were created using AI, based on visual data of the suspects from the incident, with help from the Crimes Record Division. But beyond that, the exact steps — how the sketch was turned into a digital image using generative AI — aren’t clear.</em></p><p>If you are still here, let’s think this through. First, I get it. It’s easy to think, “What if we gave the rough sketch to ChatGPT and asked it to create a lifelike image of the suspect? It looks more real, so it must be more useful in catching the bad guy.”</p><p>But what’s wrong with that logic? Before we answer, let’s take a moment to understand — in simple terms — what’s actually happening when AI turns a sketch into a photorealistic face.</p><h4><strong>How can a pencil sketch can be turned into a photorealistic face using generative AI?</strong></h4><p>Suppose you have a pencil sketch, drawn from a witness’s or victim’s memory. That’s important — this detail is at the heart of many arguments that follow. Now, you feed that sketch into an AI tool to “fill in the blanks” and produce a realistic-looking face. What actually happens under the hood?</p><p>The <strong>AI doesn’t see the sketch</strong> like a human does. It breaks it down into patterns — lines, shading, facial structure. It detects the shape of the eyes, the tilt of the head, the curve of the jaw. Think of it like asking a highly imaginative artist, trained in photorealistic drawing, to recreate a lifelike face based on a rough sketch. This artist has seen millions of faces before.</p><p>In this analogy, the sketch and any text prompt you provide are like the instructions you’d give that artist. And the artist is the generative AI model — usually a diffusion model or a GAN. These models are trained on millions of real human faces scraped from the internet. They learn what eyes, noses, hair, and skin typically look like. They also learn what features tend to appear together — say, wider noses with certain ethnicities, or the kinds of wrinkles that come with age.</p><p>So when you give the model a sketch, it doesn’t reconstruct the real person. <strong>It generates a plausible face — one that could match the sketch, based on patterns it has seen before</strong>. The result is probabilistic: one of many realistic-looking faces that could have emerged from that input.</p><p><strong>Here’s the kicker</strong>: these models are sensitive. Even a small change — a stray pencil stroke, a slightly reworded prompt — can lead to a very different face. Here are the outputs I got when I use the pencil sketch from above and made only tiny changes to the input prompt.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DbiycLQlLee-mEhGsTr1gQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*InX21XBxFdrepmns2nT1Vg.png" /><figcaption>Source: ChatGPT-generated from minor variations of a base text prompt on the same image sketch as above</figcaption></figure><p>Do they look like the same person to you? You can probably already see how that could become a problem.</p><p>Let’s get specific.</p><h4><strong>So why is this problematic?</strong></h4><p>The biggest problem is simple: the AI-generated image might look real but be completely wrong. It could resemble an innocent person, who then becomes the focus of the investigation while the real suspect goes free. That innocent person may face mistreatment or even physical harassment — something all too familiar in the Sri Lankan context. Meanwhile, chasing the wrong lead wastes time and pulls attention away from real evidence. That’s the danger — a fake face leads to the mistreatment of an innocent person, and real justice gets delayed.</p><p>And the risks go further</p><ul><li>Because the image <em>looks</em> like a real photo, it can be mistaken for real evidence — even by trained investigators or eyewitnesses. It carries a <strong>false scientific certainty</strong>, as if it were a verified fact rather than a digital guess. In a sketch, it is very clear that it is an imperfect piece of information.</li><li>Worse, these images are hard to explain or trace. There’s <strong>no clear way to tell which parts came from the sketch and which parts were invented</strong>. That makes it nearly impossible to defend the process in court or justify it in public.</li><li>If an AI-generated image is later proven wrong, <strong>it damages trust</strong>. Not just in police work — but in the justice system, and in the government’s ability to use AI responsibly. It could undermine public support for broader digital transformation efforts.</li></ul><p>Then there’s the <strong>deeper issue: bias</strong>. If you prompt a generative AI model with something like “Generate a photorealistic face from this sketch for suspect identification,” things get trickier.</p><p>Models like DALL·E, Midjourney, Stable Diffusion — even ChatGPT with image tools — are trained on millions of image-caption pairs from the internet. That training teaches them patterns, including:</p><ul><li>What kinds of faces are labeled “suspect”</li><li>What kinds of lighting, expressions, or clothing are tied to words like “mugshot,” “criminal,” or “wanted poster”</li><li>Which features often appear in news images about crime or suspect identification</li></ul><p>So even if your sketch is neutral, your prompt might not be. If the model has learned to associate certain racial or ethnic features with the idea of a “suspect,” it may skew the output in harmful ways. It might darken skin tone. Sharpen the face to look more “hardened.” Add stereotypical cues — a frown, a beard, a hoodie or cap.</p><h4>Bias creeps in even without the keywords</h4><p>Some commercial generative models (like DALL·E or Midjourney) will block or restrict outputs when the prompt includes sensitive phrases like “suspect,” “mugshot,” “wanted,” etc. These are prompt-level safeguards, designed to prevent misuse. But, even if you avoid using words like “suspect” or “criminal” in your prompt, AI tools can still produce biased results — just from the way the sketch looks. Many forensic sketches share a similar style: front-facing pose, blank background, sharp lines. The AI may have learned to link that look with suspects during training. So even with a neutral prompt, the AI might still create faces that fit common “suspect” stereotypes.</p><p>Why does this happen? These AI models are trained on huge datasets that connect images, sketches, and words based on patterns. If a lot of that training data includes mugshots or police sketches, the model learns to associate certain styles with crime. So when it sees a sketch that looks like a suspect sketch — even if you never say it is — the model fills in the blanks based on what it has seen before. The result? Bias shows up not because someone told the AI to be biased, but because it learned biased patterns on its own — and those patterns are built into how it responds.</p><p>This isn’t because the AI is malicious. It’s because it learned from biased data. And it’s doing what it was trained to do: reflect patterns in what it’s seen — even when those patterns are wrong or unjust.</p><h4><strong>But you might ask “I see what you mean, but isn’t it still better than the sketch”?</strong></h4><p>At first, this logic seems sound. In fact, when discussing public-sector AI, we often use a principle like this: <em>“AI should be judged not by perfection, but by whether it’s better than the best available alternative.”</em> At first glance, that might apply here — but it doesn’t.</p><p>The reason is simple: we don’t have a photograph of the suspect. Technically speaking, <strong>we lack “ground truth”.</strong> The victim’s or witness’s endorsement of the AI-generated face as “better” is highly subjective and influenced by two critical biases:</p><p><strong>First</strong>, there’s <strong>fluency bias</strong> — the tendency to judge something as true or accurate just because it’s easy to process or looks polished. AI-generated faces are usually clear, symmetrical, and lifelike. They’re cognitively “fluent,” which our brains misinterpret as evidence of truth. In crime investigations, this bias is especially dangerous. Even if the generated image only vaguely resembles the real suspect, its realism lends it undeserved credibility. The image doesn’t just look believable — it <em>feels</em> right. But fluency isn’t evidence.</p><p><strong>Second</strong>, consider the <strong>cognitive vulnerability in states of trauma</strong>. Victims and witnesses often recall events through fear, trauma, or anxiety — states that distort memory. Sketches based on such memories are imprecise, shaped by stress, suggestion, or mistakes. When AI “enhances” these sketches, it doesn’t correct the errors — it magnifies and refines them, turning hazy memories into crisp, believable faces. These polished images feel real, increasing confidence in their accuracy, even if they’re fundamentally wrong. This creates a dangerous feedback loop: flawed memories, rendered with precision, are mistaken for truth.</p><p>We can’t rely on the principle of “better than the best available alternative” here. That principle only applies if we have a reliable, independent way to measure “better” — ideally against observable facts or validated outcomes. In this case, we simply don’t.</p><h4><strong>So are there no circumstances under which use of Generative AI is justified for criminal suspect identification?</strong></h4><p>Not exactly. There <em>are</em> rare, high-stakes situations where using AI to enhance a suspect sketch might be justified — even valuable.</p><p>Here are the conditions where it might be worth considering:</p><ol><li><strong>Human-AI collaboration to improve sketch accuracy<br></strong> Witness memory is flawed. But when guided carefully — through an iterative process with a trained forensic artist and investigator — AI can help reconstruct faces more vividly than hand-drawn sketches alone. When done carefully, here’s why the human-led, AI-assisted sketching can work:</li></ol><ul><li><strong>Human-led sketching is collaborative</strong>: A trained forensic artist — trained in facial anatomy, perception, and psychology, not just drawing — works closely with a witness, using careful interview techniques to draw out memory without leading or distorting it.</li><li><strong>The process is slow and deliberate</strong>: Features are adjusted step by step, with attention to trauma, memory decay, and the risk of suggestion — making the sketch a careful translation of uncertain memory.</li><li><strong>The artist plays a mediating role</strong>: They clarify ambiguous details and adapt the image based on ongoing feedback, while investigators are present, but typically stay uninvolved to avoid biasing the result.</li><li><strong>The output is transparent and open to scrutiny</strong>: While not a perfect likeness, a hand-drawn sketch reflects a documented, human-centered process that can be explained and questioned.</li></ul><p>Relying on Generative AI alone bypasses these human checks: it fills in gaps by guessing from patterns in unrelated data, not by engaging with the witness or verifying memory. The result may look real, but the process is opaque: There’s no way to tell what’s based on actual memory versus what the model invented — and no point of intervention before realism turns into fiction.</p><p><strong>2. Time-sensitive, high-risk scenarios where the cost of inaction is unacceptable<br></strong> If the stakes are life-and-death — say, a missing child, an active shooter, or a terrorist threat — and no photo of the suspect exists, a generative AI-enhanced image might help. Used with clear disclaimers and strong safeguards, it can trigger faster public tips than a vague pencil sketch.</p><p><strong>3. Controlled, internal use<br></strong> If the AI-generated image stays within the investigation — not shared with the public — and is only used to narrow down leads or identify patterns, the risk of wrongful identification drops significantly.</p><p><strong>4. Age progression from real photos (conditionally acceptable)<br></strong> When starting with verified, real photographs of a suspect, AI can be used to estimate how the person might look years later. This use is more grounded — it’s not fabricating a face from memory. But even here, the output must be clearly labeled as speculative, never used to imply guilt, and always paired with proper checks.</p><h4><strong>So what are the necessary transparency and due process safeguards?</strong></h4><p>Even in exceptional cases, strong safeguards are essential. Used carelessly, AI tools can cause real harm — especially when wielded by institutions that haven’t earned public trust.</p><p>At a minimum, these conditions must be met:</p><ul><li>Full disclosure that AI was used — the image should carry labels like <em>“artist’s rendering, machine-augmented.”</em></li><li>AI images should <em>never</em> be treated as proof of guilt — only as tools to support further inquiry.</li><li>Courts, oversight bodies, and investigators must be <em>fully informed</em> of how the image was generated.</li><li><em>Human oversight and witness involvement</em> must be retained at every stage.</li><li>No public naming, shaming, or arrest <em>based solely</em> on AI-generated images.</li><li>If someone is identified, their identity must be verified with other solid evidence <em>before</em> any action is taken.</li><li>There must be a <em>way to appeal or correct errors </em>— including public statements if someone was wrongly implicated.</li><li>In extreme cases, <em>support should be offered</em> to people falsely identified — psychological, legal, reputational.</li><li>Most importantly, AI-generated composites <strong><em>should be reserved for truly exceptional cases</em></strong> — where lives are at stake or public interest is overwhelming.</li></ul><h4><strong>What if the suspect turns out resembling one of the AI-generated images in this case?</strong></h4><p>Even if the real suspect ends up resembling one of the AI-generated faces, that doesn’t prove the method works. It only means the model happened to guess a plausible face — not that it can reliably reveal hidden truths.</p><p>The resemblance is coincidence, not evidence. Generative models produce statistically likely outputs based on patterns in their training data. Getting close once doesn’t make the process reliable — it’s not very different from a broken clock being right twice a day.</p><p>And none of the risks go away. Another version might have looked like an innocent person. That could have led to misidentification, reputational harm, or even wrongful arrest. One lucky hit doesn’t prove the system is safe — it just hides the danger and we can’t abandon oversight.</p><h4><strong>What does this say about the general use of AI in Government?</strong></h4><p>AI adoption in the public sector should never be just “can AI do this?”. Public sector AI doesn’t work in a vacuum. It’s used in systems — like policing, welfare, or healthcare — where mistakes can seriously harm people, especially those with the least power to speak up. Unlike private services, people can’t just walk away. So when AI goes wrong, it doesn’t just hurt individuals — it weakens trust in public institutions. With something like AI-generated suspect images, even one wrong face can lead to false suspicion, delayed justice, and a loss of faith in both the justice system and the government’s use of technology.</p><p>These risks are even sharper in developing countries. Public institutions often don’t have the people, tools, or systems they need to manage AI safely. They may be working with messy data, depend heavily on outside companies, and have limited understanding of how AI works. And public trust is often already shaky — all of which makes failure more likely. A single high-profile failure — like an AI-generated face that misidentifies someone — can sour public opinion not just on that tool, but on digital government more broadly. Innovation stalls. Legitimacy suffers.</p><p>That’s why public sector AI must begin with the problem, not the tool. Public institutions must ask: What are we trying to solve? Is AI the right fit? What are the alternatives? Public value must be defined upfront — clear, measurable, and grounded in citizen needs. As the suspect image case shows, tools that seem helpful can cause real harm if deployed without care.</p><p><em>Thanks to Helani Galpaya, Anish Fonseka, Kasun Amarasinghe, and Vishmila Fernando for reading drafts of this.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8e279864a60e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[My relationship with “whom”]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/writers-blokke/my-relationship-with-whom-dc579a09cdfe?source=rss-356bb67a2844------2"><img src="https://cdn-images-1.medium.com/max/2600/0*ziq0rqOpF2vYCMmE" width="5472"></a></p><p class="medium-feed-snippet">A peculiar affair!</p><p class="medium-feed-link"><a href="https://medium.com/writers-blokke/my-relationship-with-whom-dc579a09cdfe?source=rss-356bb67a2844------2">Continue reading on Writers’ Blokke »</a></p></div>]]></description>
            <link>https://medium.com/writers-blokke/my-relationship-with-whom-dc579a09cdfe?source=rss-356bb67a2844------2</link>
            <guid isPermaLink="false">https://medium.com/p/dc579a09cdfe</guid>
            <category><![CDATA[words]]></category>
            <category><![CDATA[language]]></category>
            <category><![CDATA[writing]]></category>
            <category><![CDATA[love]]></category>
            <category><![CDATA[hate]]></category>
            <dc:creator><![CDATA[Merl Syrex Chandana]]></dc:creator>
            <pubDate>Thu, 26 Nov 2020 16:39:13 GMT</pubDate>
            <atom:updated>2020-11-26T16:39:13.525Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Don’t Do These 3 Things If You Want to Think More Critically]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/writers-blokke/dont-do-these-3-things-if-you-want-to-think-more-critically-f6cac84ed9cd?source=rss-356bb67a2844------2"><img src="https://cdn-images-1.medium.com/max/2600/0*5lWWB0sESr_MVIJy" width="4608"></a></p><p class="medium-feed-snippet">Lessons from an early 20th-century guide on how to avoid the most common barriers to critical thinking.</p><p class="medium-feed-link"><a href="https://medium.com/writers-blokke/dont-do-these-3-things-if-you-want-to-think-more-critically-f6cac84ed9cd?source=rss-356bb67a2844------2">Continue reading on Writers’ Blokke »</a></p></div>]]></description>
            <link>https://medium.com/writers-blokke/dont-do-these-3-things-if-you-want-to-think-more-critically-f6cac84ed9cd?source=rss-356bb67a2844------2</link>
            <guid isPermaLink="false">https://medium.com/p/f6cac84ed9cd</guid>
            <category><![CDATA[self]]></category>
            <category><![CDATA[life]]></category>
            <category><![CDATA[critical-thinking]]></category>
            <category><![CDATA[personal-growth]]></category>
            <category><![CDATA[reasoning]]></category>
            <dc:creator><![CDATA[Merl Syrex Chandana]]></dc:creator>
            <pubDate>Fri, 04 Sep 2020 13:03:30 GMT</pubDate>
            <atom:updated>2020-09-04T13:03:30.840Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[When Truths Collide]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@Merl_Syrex/when-truths-collide-ba0c70f9c167?source=rss-356bb67a2844------2"><img src="https://cdn-images-1.medium.com/max/2600/1*HTF1-n3TBeEJKafLF9oAqw.jpeg" width="4589"></a></p><p class="medium-feed-snippet">&#x201C;True or False? Well, it&#x2019;s not that simple.&#x201D; writes Hector Macdonald, in his fascinating book &#x201C;Truth: 13 Reasons to Question Everything&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/@Merl_Syrex/when-truths-collide-ba0c70f9c167?source=rss-356bb67a2844------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@Merl_Syrex/when-truths-collide-ba0c70f9c167?source=rss-356bb67a2844------2</link>
            <guid isPermaLink="false">https://medium.com/p/ba0c70f9c167</guid>
            <dc:creator><![CDATA[Merl Syrex Chandana]]></dc:creator>
            <pubDate>Mon, 17 Aug 2020 21:16:04 GMT</pubDate>
            <atom:updated>2020-08-17T21:16:04.825Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[A Personal Love Affair (with the most decorated trio in men’s tennis)]]></title>
            <link>https://medium.com/@Merl_Syrex/a-personal-love-affair-with-the-most-decorated-trio-in-mens-tennis-e28b5338e232?source=rss-356bb67a2844------2</link>
            <guid isPermaLink="false">https://medium.com/p/e28b5338e232</guid>
            <category><![CDATA[lovestory]]></category>
            <category><![CDATA[tennis]]></category>
            <category><![CDATA[sports]]></category>
            <category><![CDATA[musings]]></category>
            <dc:creator><![CDATA[Merl Syrex Chandana]]></dc:creator>
            <pubDate>Mon, 28 Jan 2019 01:31:00 GMT</pubDate>
            <atom:updated>2019-01-28T01:31:00.717Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/910/1*lmKBoV_sGdAiKibWYOn78w.jpeg" /><figcaption>Image Source: Wimbledon</figcaption></figure><p>First came Roger Federer. It was love at first sight. No two ways about it. (Frankly, how could it not be?). The presence was majestic, the shots were divine, the mannerism was impeccable and the gracefulness of court movements would have (and still does) put many a feline to shame. Is he the greatest? I would say yes, but that is a debate. But what is not, is the fact that there has never been and probably will be, a player with such effortless, consummate artistry in his game. A player who reminds people of great musicians and painters more than his contemporaries when they think of him. And now in the twilight years of his career, during his graceful descent from the crest, you appreciate him even more. His game is not perfect: it has never been, he is not the most powerful or the most athletic: but he has enough tricks in his bag to still make jaws drop. Be it his lovely wife, the two absolutely adorable pairs of twins or that unfathomable ageing curve, you are left wondering “HOW. ON. EARTH.?”</p><p>Then it was the Djoker. Novak Djokovic. And that nickname does full justice to his off-court personality. He’s always got a prank or two up his sleeve and is probably the last person to take himself seriously on the men’s tour. Be it fooling around with Sharapova or imitating Serena, he’s always prime-time. But what that nickname absolutely, positively belies, is his on-court, game-time persona. He is laser-focused; his gaze would probably pierce through diamond. He is pumped; his expressions are intense and uninhibited. But strangely, he is also stoic; you probably wouldn’t find a player less ruffled by a loss point, a game or even a couple of sets (Just ask Roger Federer about the 2011 US Open). He is the master of comebacks and keeping it cool under pressure. He is insanely fit; 5-hour marathons of tennis matches are his bread and butter. And he is, without a shadow of a doubt, the best defensive player the game has ever seen. (And in my eyes the most complete player of all time as well). And what is absolutely mind-boggling is how well he has risen from the lowest point of his career, a break of nearly two years from professional tennis. This is the stuff of fairy tales and Disney movies,</p><p>And finally, perhaps somewhat surprisingly, comes Rafa. Rafael Nadal. I didn’t like him much at first. Afterall, he was the kryptonite of my first love. The destroyer of dreams who denied Roger a sixth straight title at Wimbledon in 2008. And it probably didn’t help that he has that weird resting-bitch-face on court. But as I matured in years, so did my love and appreciation towards Rafa. He has grown on me tremendously. The Spanish machismo, that steely, unwavering grit and that historically great clay-court dominance. They proved, as they have for countless others, too powerful a pull to ignore. I was hooked. And Rafa is also a gentleman in every sense of the word. He is unassuming in his speech, respectful towards his opponents and referees and conducts himself admirably on and off the court. While his game has Spanish written all over it (wearing down one’s opponents into submission and pouncing with the kill), that too has proven to be its own art form. Perhaps not the most flamboyant. But definitely art. The type of art which takes sweat, effort and hard work.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e28b5338e232" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Keep the Molecules Dancing — From Linear to Circular]]></title>
            <link>https://medium.com/@Merl_Syrex/keep-the-molecules-dancing-from-linear-to-circular-9a8e1a8c0c05?source=rss-356bb67a2844------2</link>
            <guid isPermaLink="false">https://medium.com/p/9a8e1a8c0c05</guid>
            <category><![CDATA[circulareconomy]]></category>
            <category><![CDATA[upcycling]]></category>
            <category><![CDATA[process-engineering]]></category>
            <category><![CDATA[sustainability]]></category>
            <category><![CDATA[responsible-business]]></category>
            <dc:creator><![CDATA[Merl Syrex Chandana]]></dc:creator>
            <pubDate>Sat, 07 Oct 2017 04:04:59 GMT</pubDate>
            <atom:updated>2017-10-07T04:12:01.275Z</atom:updated>
            <content:encoded><![CDATA[<p>Originally developed by Andrew Simms from the UK think tank <em>New Economics Foundation</em>, and annually published by <em>Global Footprint Network</em>,<strong> </strong>the <strong>Earth Overshoot Day</strong> calculates the calendar date on which humanity’s <a href="https://en.wikipedia.org/wiki/Resource_consumption">resource consumption</a> for the year, exceeds <a href="https://en.wikipedia.org/wiki/Earth">Earth</a>’s capacity to regenerate those resources that year. And this year, we hit it by the 2nd of August.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/624/1*o96WFBTZ4Rx3qS84ed8IsA.png" /></figure><p>With the exponential growth in population and our incompatible patterns of consumerism that accompany it, it really has been no brainer that we use more ecological resources and services than nature can regenerate, through overfishing, overmining, overharvesting forests, and emitting more carbon dioxide into the atmosphere than forests can sequester. We as humanity are in a century of fragility and uncertainty at a level not experienced before. And our problems with the scarcity of resources will only increase when 3 billion more people enter into what is known as the “middle class” in the next 30 years. This means more cars, more smart-phones, more food and more <strong>everything</strong>.</p><p>But what if we looked around and tried to learn a thing or two from the natural world? Living systems have been around for a few billion years, and will continue to be there for a few billion more. In the living world, there’s no land fill. Instead, materials flow. One species’ waste is another’s food. Energy is provided by the sun. Things grow and die and nutrients return to the soil safely. And it works. But as humans, we have adopted a linear approach. We take, we make and we dispose. A new phone comes out, so we ditch the old one. Our car becomes old-fashioned, we opt for another. Each time we do this, we are eating into the earth’s limited supply of resources.</p><p>The idea of replicating the earth’s system of a closed-loop world, where materials and resources are recycled endlessly and waste and pollution don’t exist, is at the heart of the concept of a <strong>Circular Economy. </strong>While there’s no universally adopted definition, the circular economy is really about “keeping the molecules dancing”. In such a system, products are made primarily from benign and non-toxic ingredients — “nutrients” which return to the soil or water safely. And whenever toxic ingredients are used they are continuously cycled back to productive use and are kept out of waste streams. And, of course, as much of this as possible, is powered by renewable energy.</p><p>But what is a chemical &amp; process engineer’s role in all of this? Well, in a circular economy, thermochemical processes become extremely important chemical processes. These are key processes that provide us with energy and the major base chemicals. In a circular economy, processes need to be designed to keep products and materials at their highest value for the longest possible time. Waste and energy usage must be minimized and kept within the economy for reuse. Chemical kinetic models are extremely powerful and valuable for this purpose. Chemical and Process Engineering at its heart, is about converting raw materials and energy into a variety of products. Therefore, from the choice of raw materials and energy sources, to designing and optimizing process steps, chemical engineering has the potential (perhaps more than any other profession) to adopt the principles of circularity.</p><p>A true “circular economy” requires <a href="https://www.weforum.org/pages/circular-economy-videos">seismic shifts</a> in the way we view the ownership of our products, organizations view their business models and interact with each other and the way the government regulates and incentivizes businesses for the greater good. Of course, the challenges are different both in nature and the degree of difficulty for different organizations. Companies producing technical products like engines, computers or consumer electronics have far greater capacity to move towards “closed loops” than say, a FMCG company or a clothing company. But if you look around, you won’t run short of inspiration. In 2015, <strong>adidas</strong> unveiled the first fruit of its collaboration with ocean conservation group <strong>Parley for the Oceans</strong>: a <a href="http://www.sustainablebrands.com/news_and_views/products_design/sustainable_brands/adidas_parley_oceans_unveil_first_footwear_concept">running shoe</a> made from up-cycled ocean plastic waste. And a <a href="http://www.thesundayleader.lk/2016/02/21/house-of-lonali-ethical-design/">Sri Lankan designer</a> has recently been making waves with her brand <strong>House of Lonali, </strong>by taking unwanted textiles, trims and clothing to transform them into something magical.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NPH9CvfmtO-FLtgj4v2OJQ.png" /><figcaption>In 2015, adidas, Parley for the Oceans Unveiled First Footwear Made from Upcycled Ocean Waste</figcaption></figure><p>Time Magazine, in its 3rd of April cover of 2006, pleaded humanity to “Be Worried. Be Very Worried” about climate change. Those words ring alarmingly true today, not just for climate change, but for humanity’s use of the earth’s finite supply of resources. Chemical engineers and all those involved in the design and operation of manufacturing processes need to fundamentally rethink what it means to be circular and sustainable.</p><p>The time for unit operations and thermodynamics to save our planet earth is here. We just might not survive without them.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9a8e1a8c0c05" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>