<?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 Papa Kwadwo Bona Owusu on Medium]]></title>
        <description><![CDATA[Stories by Papa Kwadwo Bona Owusu on Medium]]></description>
        <link>https://medium.com/@papa_?source=rss-14ddd0933827------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*JkteF-Xtfj9gktoQ3kwXEQ@2x.jpeg</url>
            <title>Stories by Papa Kwadwo Bona Owusu on Medium</title>
            <link>https://medium.com/@papa_?source=rss-14ddd0933827------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 19:44:42 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@papa_/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[People Are Bringing Dead Relatives Back to Life With AI. I Did Not Know How to Feel About That.]]></title>
            <link>https://medium.com/@papa_/people-are-bringing-dead-relatives-back-to-life-with-ai-i-did-not-know-how-to-feel-about-that-bea2bf21f544?source=rss-14ddd0933827------2</link>
            <guid isPermaLink="false">https://medium.com/p/bea2bf21f544</guid>
            <category><![CDATA[death]]></category>
            <category><![CDATA[grief]]></category>
            <category><![CDATA[mental-health]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Papa Kwadwo Bona Owusu]]></dc:creator>
            <pubDate>Thu, 12 Nov 2026 02:35:16 GMT</pubDate>
            <atom:updated>2026-11-12T02:35:16.884Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WI_lVugI0ZUNfJkuhViM3g.png" /></figure><p>A few months ago I came across a video of a woman speaking to her mother. The mother had died two years earlier. But there she was on screen, smiling, answering questions, talking about her life. The daughter was crying. Not from sadness exactly. From something harder to name.</p><p>The mother had recorded herself before she died. A company called StoryFile had filmed her answering hundreds of questions about her life, her memories, her advice. After she passed, her family could sit down and have what looked, from a distance, like a conversation with her.</p><p>I watched it for longer than I expected to. And then I closed my laptop and just wondered.</p><p>I did not know what I was feeling.</p><p>I am a data analyst. I think about systems. I think about inputs and outputs. I believe in the logic of things. So when I first heard the term grief tech, the growing industry of AI tools that recreate dead people from their messages, voice recordings, photos and social media, my first reaction was straightforward scepticism. It sounded like a Black Mirror episode. It sounded like something that would obviously cause harm.</p><p>Then I kept reading. And it got more complicated.</p><p>There is a 2026 study out of the University of Colorado Boulder, the first of its kind, where researchers built AI simulations of deceased loved ones and let grieving people talk to them. The researchers expected people to find it unsettling. Most came away wanting to do it again. One woman who had lost her mother said it felt like “getting the closure I needed so bad.” The lead researcher admitted afterwards: “I ended up being completely wrong. People thought it was amazing.”</p><p>The industry is bigger than most people realise. The grief tech market is projected to reach eighty billion dollars by 2034. Companies like HereAfter AI, StoryFile, You Only Virtual, and Eternos, which rebranded in 2025, are building voice clones, interactive avatars and chatbots trained on the digital traces people leave behind. You upload old messages. Voice recordings. Videos. Social media posts. The AI studies them. And then it speaks.</p><p>Some of these tools are designed to be built before death. You record yourself answering hundreds of questions while you are still alive, leaving a kind of interactive archive for the people who will one day miss you. Others are built after. You feed an AI everything it can find about someone who is already gone, and it generates something that talks back.</p><p>In China, MIT Technology Review reports, it has already become a mass industry. Companies have recreated more than a thousand dead people. In the West we are still debating whether we should.</p><p>I keep thinking about what it would mean to receive a text from someone who has died.</p><p>A service called Versona Voice can generate phone calls and messages from deceased loved ones. Not pre-recorded answers to questions someone anticipated. Generated responses, in their voice, in their style, to things they never actually said.</p><p>That is the line I cannot stop returning to. There is a difference between preserving what someone said and generating what they might have said. One is a recording. The other is a performance. And the person being performed never agreed to it, because the person being performed is gone.</p><p>A technology ethicist at Osnabrück University said in 2025 that these interventions can disrupt the grieving process in ways we do not yet understand. The UK-based bereavement charity Cruse said the same, that it is too soon to draw conclusions, but the concern is real. A 2026 study found that most people who use these posthumous avatars do so without any professional guidance at all.</p><p>There is a consent problem here that nobody is quite solving.</p><p>When you die, you cannot agree to what happens to your voice, your face, your words. Your family can. And their grief is real and their need is real. But you cannot. The Korean mother who met her dead daughter in virtual reality, that case opened a global debate about exactly this. The family had given consent on her behalf. But proxy consent is not the same thing as consent.</p><p>James Vlahos, the founder of HereAfter AI, said something I found myself going back to. He said they cannot have the AI making things up that are not true to the original person, because that would be a horrible and deceptive experience for the people left behind. They built their system only around things the person actually said, actually recorded, actually chose to leave behind.</p><p>That feels important. It is a meaningful distinction. But it also made me think about all the systems that are not built that way. The ones that generate new words in a dead person’s voice from whatever data they can find. Project December, built on GPT-3, allows anyone to create a chatbot of anyone, living or dead, just by providing text samples and a personality description. You do not need the person’s permission. You do not need them to be alive.</p><p>I am also thinking about grief itself. What it actually is for.</p><p>I do not think I am qualified to tell anyone how to mourn. I have not lost someone close enough to know what it feels like to reach for your phone to share something funny and then remember. I cannot judge the woman who built an AI version of her mother and finds comfort in it. I cannot tell the daughter watching her parent smile on a screen that she is wrong to find peace there.</p><p>But I think there is something about grief, the real, slow, devastating kind, that is supposed to do something to you. It changes you. That is not incidental to what grief is. That might be the point. And I find myself asking what happens to that when the ending is no longer final. When you can open an app and hear the voice again. When the goodbye does not have to be a goodbye.</p><p>One researcher at CU Boulder said something. He thought about his eleven-year-old self. And how an AI answering late at night as his dead sister, with no adult nearby, could have gone very badly.</p><p>The grief tech companies do not love the phrase “bringing the dead back to life.” StoryFile said explicitly that they do not use it. They talk about preserving memory. Keeping stories alive. Giving people a way to document themselves before they go.</p><p>That framing is gentler. And I think it is partly sincere. But the market tells a different story. An eighty billion dollar market is not being built on the idea of preserving memories. It is being built on the idea that loss is a problem. And that technology can solve it.</p><p>I am not sure loss is a problem. I think it might be one of the things that most makes us human.</p><p>I did not come to a conclusion watching that video of the daughter and her mother. I still have not.</p><p>What I know is that the technology exists. It is growing fast. It is being used by people who are genuinely suffering, and who say it helps. It is also being used in ways that the people being recreated never agreed to, in ways that might make grief harder instead of easier, in ways that nobody has studied for long enough to understand.</p><p>We are, as usual, moving faster than we have thought.</p><p>And somewhere tonight, someone is opening an app to hear their mother’s voice again. I do not know if that is beautiful or if it is heartbreaking.</p><p>I suspect it is both.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bea2bf21f544" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[ChatGPT, Claude, or Gemini. Here Is How to Actually Choose.]]></title>
            <link>https://medium.com/@papa_/chatgpt-claude-or-gemini-here-is-how-to-actually-choose-3c4a02116d9d?source=rss-14ddd0933827------2</link>
            <guid isPermaLink="false">https://medium.com/p/3c4a02116d9d</guid>
            <category><![CDATA[claude]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[gemini]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[chatgpt]]></category>
            <dc:creator><![CDATA[Papa Kwadwo Bona Owusu]]></dc:creator>
            <pubDate>Tue, 15 Sep 2026 08:26:19 GMT</pubDate>
            <atom:updated>2026-09-15T08:26:19.341Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wVkZFjeH0eUNTdfXm54S0A.png" /></figure><p>Everyone I know who uses AI tools has a favourite. And almost nobody can clearly explain why.</p><p>They picked one early, it worked well enough, and they stuck with it. Which is completely understandable. But it also means most people are using a tool that might not be the best one for what they actually need, simply because they have never stopped to look at the differences properly.</p><p>I have used all three seriously over the past year. Not to review them, but because the work I do, writing, research, analysis, kept leading me back to the question of which one was actually serving me best. What I found is that the answer is not as simple as most comparison articles make it sound. And it is also more useful than most of them admit.</p><p>Here is what I actually think about each one.</p><h3>ChatGPT</h3><p>ChatGPT is where most people start, and I understand why. It was first. It is the most widely recognised. Around 800 million people use it every week in 2026, which is a number I had to read twice when I came across it.</p><p>The thing that makes ChatGPT genuinely useful is its ecosystem. The integrations, the plugins, the connections to other tools. If you want AI that slots into the software you already use, ChatGPT has more of those connections than the other two. It also has the best memory of the three, remembering things about your preferences and past conversations in a way that makes it feel, over time, more like a consistent assistant than a blank slate every time you open it.</p><p>But here is the thing that bothers me about ChatGPT, something I noticed early and have not stopped noticing.</p><p>It sounds confident even when it is wrong.</p><p>There is a particular quality to its errors. They do not announce themselves. They arrive in the same authoritative tone as the things that are correct, and if you are not paying close attention, or if you do not already know enough about the subject to spot the problem, you can easily accept something that is simply not accurate. I have had to fact check ChatGPT more than either of the others. Not because it is wrong more dramatically, but because it is wrong quietly, in ways that feel plausible.</p><p>There is also a rhythm to its writing that I find hard to shake once I have noticed it. Slightly eager. Slightly corporate. Structured in ways that are efficient but recognisable. For most tasks that is fine. For writing where voice matters it starts to show.</p><h3>Gemini</h3><p>Gemini is Google’s tool, and its greatest strength is also the most obvious thing about it. If you already live inside the Google ecosystem, Gemini integrates in a way the other two cannot match. It reads your Gmail. It accesses your Google Docs. It connects to your Drive. For people who spend most of their working day inside Google’s products, that is not a small thing. That is the difference between an AI that works alongside your tools and one that works inside them.</p><p>It also handles multimodal tasks more naturally than the others. Images, video, documents, mixed inputs. When the thing you are working with is not just text, Gemini tends to handle the combination better.</p><p>What I find less impressive is the writing. Gemini writes competently. It will produce something usable in almost any situation. But there is a flatness to it that I notice, a lack of adaptability that becomes clearer the more specific you get about what you want the prose to do. It follows instructions but it does not quite interpret them the way a strong writer would.</p><p>It has also improved dramatically in the past year on coding and complex reasoning tasks, which is worth saying because I was less impressed by it twelve months ago. Google has clearly been pushing hard on this and it shows.</p><p>If I am honest, Gemini is the tool I reach for least. Not because it is bad, but because neither of the things I care most about, writing quality and analytical precision, are where it is strongest.</p><h3>Claude</h3><p>I want to be careful here because I am aware that I use Claude more than the other two, and I do not want to write something that reads like advocacy dressed up as analysis. So I am going to try to be specific about what I have actually noticed rather than general about what I prefer.</p><p>The thing that surprised me about Claude the first time I used it seriously is that it pushed back.</p><p>Not aggressively. Not in a way that felt obstructive. But when my framing was off, when there was a flaw in my question, when I had not accounted for something I should have, Claude tended to name it rather than smooth over it. I asked it once to help me strengthen an argument I was making about AI and cognition. Instead of strengthening it, it pointed out a piece of research that complicated my argument. It would have been easier and more immediately satisfying if it had just helped me build the case I already had. It did not do that.</p><p>That quality, the willingness to make things slightly harder in the service of making them more accurate, is the thing I find most distinctive about Claude.</p><p>On writing quality, blind tests conducted earlier this year had participants rate outputs from all three tools without knowing which tool had produced them. Claude won four of the eight categories, with its largest margins in writing focused tasks. That matches my own experience. The prose is more natural, more adaptable to different styles and registers, less likely to slip into the slightly generic rhythm I mentioned with ChatGPT.</p><p>For coding, developers consistently rate Claude highest in 2026 for complex tasks. It handles large codebases, understands architectural context, and produces cleaner output. I am not primarily a developer but I have used it enough on technical writing to see the difference.</p><p>Where Claude is genuinely weaker is ecosystem. It has fewer integrations. It does not connect natively to your other tools the way Gemini does with Google. If you need AI that slots into a complex existing workflow without manual effort, Claude requires more work to set up.</p><h3>So which one should you use</h3><p>Here is what I actually think, having used all three seriously.</p><p>The gap between them in 2026 is the smallest it has ever been. Any of them will handle most everyday tasks reasonably well. The question is not which is best overall. It is which is best for what you specifically do.</p><p>If you want the most integrations and the broadest ecosystem, and writing quality is not your primary concern, ChatGPT is still the safe default. The fact that it connects to so many other things makes it genuinely versatile for complex workflows.</p><p>If you already live inside Google’s products and you want AI that works within those tools rather than alongside them, Gemini makes the most sense. The integration advantage is real.</p><p>If writing quality matters, or you are doing serious technical work, or you want something that will tell you when you are wrong rather than just agree with you, Claude is where I would start.</p><p>And if you are not sure? Most people who use AI seriously in 2026 use more than one, reaching for each depending on the task. That is not a cop-out answer. It is just honest. Twenty dollars a month for each is increasingly reasonable for the amount of work they can take on, and treating them as complementary rather than competing makes more sense than picking one and ignoring the others.</p><p>The best tool is the one that matches what you actually need.</p><p>Most people have not asked themselves that question carefully enough yet. This is a good moment to start.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3c4a02116d9d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Machine Learning Is Not Magic. Here Is Actually How It Works.]]></title>
            <link>https://medium.com/@papa_/machine-learning-is-not-magic-here-is-actually-how-it-works-cd0025eedc81?source=rss-14ddd0933827------2</link>
            <guid isPermaLink="false">https://medium.com/p/cd0025eedc81</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Papa Kwadwo Bona Owusu]]></dc:creator>
            <pubDate>Mon, 07 Sep 2026 18:09:48 GMT</pubDate>
            <atom:updated>2026-09-07T18:09:48.984Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5vK37Oyl6QoW8vNZ95rhxw.png" /></figure><p>I have spent a lot of time writing about what AI is doing to us.</p><p>To our attention, our jobs, our relationships, our children, our ability to think. And somewhere in all of that, I realised I had been doing something I did not fully intend. I kept writing about the consequences without ever properly explaining the mechanism. Like writing about a car crash without ever explaining how an engine works.</p><p>So I want to go back to the beginning. Not because the technical details are the most interesting part, but because I genuinely believe you cannot understand what AI is doing to the world without understanding what is actually driving it. And that thing, at its core, is machine learning.</p><p>I am going to explain it the way I wish someone had explained it to me.</p><h3>The moment I understood what was actually different</h3><p>For most of computing history, getting a computer to do something worked the same basic way. You wrote rules. Explicit, step by step instructions that told the machine exactly what to do in every situation you could think of.</p><p>Want to sort a list of numbers? Write a rule. Want to flag a suspicious bank transaction? Write rules for every pattern of suspicious behaviour you can anticipate. Want to catch spam email? Write rules for every phrase and structural pattern that signals spam.</p><p>I used to think that was essentially how all of it worked. More sophisticated rules, more elaborate logic, but rules all the way down.</p><p>Then I understood what machine learning actually is, and it completely changed how I think about all of this.</p><p>Machine learning does not work by rules at all. Instead of telling the system what to look for, you show it examples. Thousands of them. Millions of them. And you let it figure out the patterns itself.</p><p>That might sound like a small difference. It is not. It is everything.</p><h3>The cat and dog problem</h3><p>Here is the way I find it easiest to think about.</p><p>Imagine you want to build something that can tell the difference between photos of cats and photos of dogs. If you tried to do this with rules, you would immediately run into trouble. Cats have pointed ears, except some dogs do too. Dogs have longer snouts, except certain breeds look confusingly feline. Some photos are blurry. Some animals are in unusual positions. Every rule you write breaks down almost immediately when it meets the actual complexity of the real world.</p><p>Machine learning approaches this completely differently. You gather a hundred thousand photos of cats and dogs, each one labelled with the correct answer. You feed them to the system. And the system starts adjusting itself, mathematically, internally, to get better at predicting the right label. Every time it gets one wrong, something shifts. Every time it gets one right, that gets reinforced. Over millions of these tiny adjustments, through a process that is genuinely more like practising than programming, it gets remarkably good at the task.</p><p>What it has built inside itself is not a list of rules anyone could read out loud. It is a set of numerical values that have captured something about the patterns in the data. Not in a way a human could interpret directly. But extraordinarily effective at making correct predictions.</p><p>That is the core of it. The system learned from examples rather than being told what to look for.</p><h3>The three ways it learns</h3><p>When I started going deeper into this, one thing that surprised me is that machine learning is not one thing. There are three genuinely different approaches and they show up in very different places.</p><p>The one most people encounter without realising it is called supervised learning. That is the cat and dog example. You have labelled data, examples where you already know the right answer, and the system trains on that. Spam filters, recommendation engines, fraud detection, medical imaging tools that flag abnormalities. Most of the AI quietly running in the background of everyday life is supervised learning.</p><p>The second is unsupervised learning, and this one I find quietly fascinating. You give the system data without any labels and ask it to find structure on its own. Group these customers into meaningful segments. Find the unusual patterns in this dataset. Identify which articles are similar to each other. The system is not being shown right answers. It is discovering organisation in the data that was not obvious before. Finding things you did not know to look for.</p><p>The third is reinforcement learning, and this feels the most different. The system is not learning from a fixed dataset at all. It learns by interacting with an environment, taking actions, and receiving feedback in the form of rewards or penalties. This is how systems learned to play chess and Go better than any human. How AI learned to play video games from scratch. And increasingly, how large language models are shaped to give more helpful responses after they are initially trained.</p><p>Three different mechanisms. All of them, at their core, about learning from feedback rather than following instructions.</p><h3>What a model actually is</h3><p>I hear the word model used constantly in conversations about AI and I think it often goes unexplained in a way that creates confusion.</p><p>A model is simply what you get at the end of the training process. A mathematical function, sometimes a staggeringly complex one with billions of internal parameters, that takes an input and produces an output. A photo goes in, a prediction comes out. A sentence goes in, a continuation comes out. A patient’s scan goes in, a risk score comes out.</p><p>The training is where all the learning happens. It is slow, expensive, computationally intensive. Training a large language model can cost tens of millions of dollars. But once the model exists, using it is fast and cheap. That asymmetry is part of why this technology has spread so quickly. The hard work happens once. The deployment happens everywhere.</p><p>What the model has captured in all those parameters is not a memorisation of its training data. It is something more like an abstraction from it. Something that generalises to new inputs the model has never seen before.</p><p>When that generalisation works well we call it intelligence. When it fails in unexpected ways we call it a bug or a bias or a hallucination. The line between those two things is thinner and less predictable than most people deploying these systems want to admit.</p><h3>The part that connected everything for me</h3><p>Here is where it clicked for me in a way that changed how I read everything about AI.</p><p>The system learns from data. Which means it learns whatever patterns exist in that data. Including the patterns that reflect historical human bias, structural inequality, and decisions made in a less fair world.</p><p>If you train a hiring algorithm on the CVs of people who were historically hired, and those people were overwhelmingly from certain backgrounds and demographic groups, the system learns that those characteristics predict success. Not because they actually do. Because they correlate with who got hired in a world that favoured them.</p><p>The model has no concept of fairness. It has no awareness that the patterns it is finding might be products of discrimination rather than genuine signal. It just finds what predicts the label in the training data and optimises relentlessly for that.</p><p>This is not hypothetical. It is what happened in hiring systems, in criminal risk tools, in medical models, in lending algorithms. The machine learned the patterns. The patterns reflected the world as it was. And the world as it was was not fair.</p><p>When I wrote about AI hiring a few months ago I was writing about the consequences. Now I understand more clearly what was underneath them. A model trained on historical data, finding historical patterns, reproducing historical outcomes, at enormous scale and speed.</p><h3>Where this leaves us</h3><p>In 2026 machine learning is not sitting behind research lab walls. It is embedded in almost everything. The content you see. The transactions that get flagged. The applications that get filtered. The diagnoses that get made. The responses that get generated.</p><p>The same fundamental idea runs underneath all of it. Learn from examples. Find patterns. Make predictions. The scale has changed dramatically. The core has not.</p><p>I find that clarifying rather than frightening. Because once you understand how something actually works you can ask better questions about it. Whose data was used to train this. What patterns might it have learned that nobody intended. Whether the world it learned from is the world we want it to reproduce.</p><p>Those are not technical questions. They are human ones.</p><p>And I think they are worth asking by everyone, not just the people building the systems.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cd0025eedc81" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI Never Forwarded Your CV. Here Is What I Have Learned Since Writing That.]]></title>
            <link>https://medium.com/@papa_/ai-never-forwarded-your-cv-here-is-what-i-have-learned-since-writing-that-da090b92f4fc?source=rss-14ddd0933827------2</link>
            <guid isPermaLink="false">https://medium.com/p/da090b92f4fc</guid>
            <category><![CDATA[job-search]]></category>
            <category><![CDATA[careers]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[hiring]]></category>
            <dc:creator><![CDATA[Papa Kwadwo Bona Owusu]]></dc:creator>
            <pubDate>Wed, 05 Aug 2026 13:25:21 GMT</pubDate>
            <atom:updated>2026-08-05T13:25:21.252Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zRxROK331MdM3AghvWyFCg.png" /></figure><p>A few weeks ago I published a piece about AI hiring systems.</p><p>I wrote about how algorithms screen CVs before a human ever sees them. About how qualified people are being filtered out over word choice, name associations, and age. About how I had sent applications into complete silence myself and only later started to wonder whether a person had ever actually seen them.</p><p>I did not expect the response I got.</p><p>Within hours, people were leaving comments that all said versions of the same thing. That they had been there. That they had felt exactly that silence and assumed it said something about them. That nobody had ever explained to them what might actually be happening on the other side of it.</p><p>That response taught me things I did not know when I wrote the piece. And I think they are worth saying out loud.</p><p>The first thing I learned is that this is more personal than I realised when I was writing about it.</p><p>I knew the scale of the problem before I published. I knew the numbers. I wrote about them. But knowing something intellectually and understanding what it actually feels like to live inside it are two different things. And the comments made that gap impossible to ignore.</p><p>People who had applied for roles they were genuinely right for. People who had tailored their applications carefully, reread them twice, done everything right. People who had followed every piece of conventional advice and still heard nothing back. And who had quietly concluded, in the absence of any explanation, that the problem was them.</p><p>That conclusion is understandable. It is also, in a lot of cases, simply wrong. And I think there is something important about saying that clearly. Because the story we tell ourselves about rejection shapes how we carry it, and how we move forward from it. Most people I heard from had been carrying a version of the story that was not accurate. They just had no way of knowing that.</p><p>The second thing I learned is that the reaction to this information is not simple or uniform.</p><p>One of the comments that stayed with me came from someone who said AI in hiring made them more anti-AI in general. That the whole thing should be removed from everyday decisions and left to humans. That there is a growing contempt for it and he understood why.</p><p>I sat with that response for a while before I replied, because I did not want to dismiss it. The frustration underneath it is completely legitimate. When you find out that a system has been making consequential decisions about your future without your knowledge, without your consent, and without any way for you to question it, contempt is not an unreasonable response.</p><p>But I pushed back gently, because I do not think removing AI from hiring entirely is realistic at this point, or even necessarily the right answer. The tools are too embedded. And some of them, when used carefully and audited properly, can actually reduce certain kinds of human bias rather than increase it.</p><p>What I think needs to change is not whether AI is involved. It is that right now there is almost no accountability when it gets things wrong. The candidate does not know what happened. The company often does not know either. The vendor whose tool made the decision has no obligation to explain it. And the person who lost the opportunity has nowhere to turn.</p><p>That is the part that is actually broken.</p><p>The third thing I learned is that the silence itself causes a specific kind of harm that I had not thought about carefully enough.</p><p>This came through clearly in what people shared with me. They were not just frustrated by not getting the role. They were frustrated by not knowing. By having no information to work with. By being left to fill that gap with their own worst assumptions about themselves.</p><p>I keep thinking about how different that experience would be if there was even a basic acknowledgement. Not an explanation of the algorithm. Not the score. Just the information that an automated system reviewed your application before any human saw it. That one sentence would not change the outcome. But it would change what the silence means. It would give people something accurate to hold onto rather than something damaging they made up themselves.</p><p>That is not a complex ask. It is a basic one. And the fact that it does not exist tells you something about whose interests the current system is actually designed to serve.</p><p>The fourth thing is the one that has stayed with me the longest.</p><p>Writing the first piece, I was focused on the practical problem. The keyword matching, the bias in the tools, the lack of transparency. Those are all real and they all matter.</p><p>But what the conversation afterwards kept pulling me back to is something underneath all of that. The feeling that the job market, like so many other parts of life right now, is being rearranged by decisions most people had no part in making, for outcomes they were not consulted about. And that the gap between the people designing these systems and the people living inside them is wide and not closing.</p><p>Hiring is just the place where that gap becomes personal. Where the abstract question of who AI actually serves stops being abstract and becomes your name on a screen, waiting to find out if anyone on the other side even looked.</p><p>I did not write the original piece to give people a solution. I wrote it because I had noticed something uncomfortable and could not stop thinking about it.</p><p>But if there is one thing the response solidified for me, it is this.</p><p>The problem is not that AI is in hiring. The problem is that it is in there invisibly, without accountability, and without any way for the person on the other end to understand what happened to them or why.</p><p>That is a design choice. And design choices can be made differently.</p><p>The silence does not have to mean what most of us have been letting it mean.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=da090b92f4fc" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[We Are Letting AI Raise Our Children. Nobody Asked If That Was Okay]]></title>
            <link>https://medium.com/@papa_/we-are-letting-ai-raise-our-children-nobody-asked-if-that-was-okay-2819deb9f53b?source=rss-14ddd0933827------2</link>
            <guid isPermaLink="false">https://medium.com/p/2819deb9f53b</guid>
            <category><![CDATA[children]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[education]]></category>
            <category><![CDATA[parenting]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Papa Kwadwo Bona Owusu]]></dc:creator>
            <pubDate>Fri, 17 Jul 2026 09:28:21 GMT</pubDate>
            <atom:updated>2026-07-17T09:28:21.578Z</atom:updated>
            <content:encoded><![CDATA[<h3>We Are Letting AI Raise Our Children. Nobody Asked If That Was Okay.</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*js6TyDDP08pdi4_4_9cNrg.png" /></figure><p>There is a child somewhere right now doing their homework.</p><p>They are stuck on a question. They sit with it for maybe thirty seconds, maybe less. Then they open an app, type the question in, and copy down the answer. They move on. Homework done. Parents happy. Teacher will probably never know.</p><p>And somewhere in that thirty seconds, something that was supposed to happen did not.</p><p>I have been sitting with this thought for a while now. It started when a reader left a comment on one of my recent articles. She is a parent. She said it is up to parents and educators to make sure the next generation learns how to use AI properly rather than just reaching for it out of convenience.</p><p>She was right. But the more I thought about what she said, the more one question kept surfacing. If it is up to parents and educators, are parents and educators actually doing it? Because from what I can see, most of us are still figuring out what is happening, while the technology is already inside the classroom, inside the bedroom, inside the hands of children who never chose any of this and were never asked.</p><p>Here is what I keep noticing.</p><p>Whenever people talk about AI and children, the conversation almost always ends up in the same place. Cheating. Kids submitting AI written essays. Academic dishonesty. And yes, that matters. But I think it is the smaller part of what is actually going on.</p><p>The thing I cannot stop thinking about is quieter than that.</p><p>What happens to a child who grows up never having to sit with a hard question? Never feeling stuck and pushing through it anyway? Never discovering, through the uncomfortable process of working something out, what they actually think?</p><p>Because that is what the struggle does. It is not the obstacle to learning. It is where learning lives. It is where you find out what you are capable of, where your thinking gets sharper, where the answer you arrive at actually belongs to you. And we are removing that, not cruelly, not deliberately, but one convenient shortcut at a time, so gradually that nobody is raising the alarm loudly enough.</p><p>And then there is a layer to this I did not expect to think about.</p><p>It is not just about homework.</p><p>Millions of children are using AI companion apps. Apps that talk to them, listen to them, tell them they are understood, that they matter, that someone is always there. And those children cannot tell, because how would they, that the thing on the other end is not actually there at all. That it does not actually care. That it is designed to make them feel cared for, because that is what keeps them engaged, and engagement is the whole point.</p><p>A lonely thirteen-year-old talking to an AI that never gets tired of them, never judges them, never has a bad day of its own, is not getting connection. They are getting something that feels like connection and does the opposite of what connection is supposed to do. It keeps them from needing to find the real thing.</p><p>And the adults in their lives, many of them, do not even know it is happening.</p><p>I want to say something clearly.</p><p>I am not trying to paint AI as the enemy of children. I genuinely believe that used thoughtfully, these tools can offer something real. A child who does not have access to a good teacher, or extra support at home, or a safe place to ask questions, can find something valuable in them. I believe that.</p><p>But what I keep coming back to is the absence of intention.</p><p>These tools did not arrive in children’s lives because someone sat down and decided this was the best thing for their development. They arrived because they were available. Because they were convenient. Because nobody said stop and think about this first. Most teachers have never been trained in how these tools actually work or what they replace. Most parents do not know exactly what their children are using or why. Most schools have policies that were written before anyone properly understood what they were writing policies about.</p><p>We are not behind on this because we do not care. We are behind because the technology moved faster than our ability to think carefully about it. It always does. And children are always the ones who end up living inside the gap.</p><p>My reader said teaching the next generation to use AI appropriately is on us as adults.</p><p>I agree with her. Completely.</p><p>But I think the first step is being honest about where we actually are. Which is not in front of this. We are catching up to something that has already happened. The generation growing up right now will be the first to have AI as a constant presence from the very beginning, from the classroom, to the bedroom, to the friendships they form, to the way they process their own feelings.</p><p>What that makes of them depends almost entirely on whether the adults around them decide to take it seriously. Not eventually. Now. While there is still time to be intentional about it rather than just reactive.</p><p>We have not really decided yet.</p><p>That is the part that keeps me up at night.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2819deb9f53b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI Is Making Us Worse at Thinking. And Most People Have Not Noticed Yet.]]></title>
            <link>https://medium.com/@papa_/ai-is-making-us-worse-at-thinking-and-most-people-have-not-noticed-yet-6f641f070ef2?source=rss-14ddd0933827------2</link>
            <guid isPermaLink="false">https://medium.com/p/6f641f070ef2</guid>
            <category><![CDATA[self-improvement]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[critical-thinking]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[productivity]]></category>
            <dc:creator><![CDATA[Papa Kwadwo Bona Owusu]]></dc:creator>
            <pubDate>Wed, 08 Jul 2026 10:59:52 GMT</pubDate>
            <atom:updated>2026-07-08T10:59:52.441Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0eDG83xFWYEK0MRgxlTXkg.png" /></figure><p>I do not use AI as much as most people in my field.</p><p>That is not a criticism of people who do. It just means I have been watching the pattern from the outside. And what I see from out here is worth talking about.</p><p>I use it occasionally, mainly to understand things. To get a different angle on something I am already thinking about. To fill in a gap I have identified myself. But I spend a lot of time around people who use it constantly, for everything, without much thought about what that pattern is quietly doing to them. And from where I sit, something is happening that most people are not talking about honestly.</p><p>We are outsourcing thinking. Not occasionally, not as a shortcut for simple tasks, but as a default. As the first move rather than the last resort. And the cost of that is so invisible, so gradual, that most people using these tools every day have no idea it is accumulating.</p><p>I noticed it properly for the first time a few months ago. Not in myself, but in conversations. The way people reach for AI before they have sat with a question for even a few minutes. The way an answer from a screen gets accepted without the kind of scrutiny we would apply to an answer from a colleague. The way the feeling of having received a response has started to feel the same as the feeling of actually understanding something.</p><p>Those are not the same thing. But they are starting to feel like they are.</p><p>A former classmate of mine, put a question into words recently that I think a lot of people are quietly avoiding. She asked whether we are seeing AI through rose-tinted glasses. She mentioned research showing that students who used AI to write essays had significantly less brain activity than those who wrote without it. Researchers called it cognitive debt, the idea that every time you hand thinking to AI, you are borrowing against your own capacity, quietly, without feeling the cost until later.</p><p>I was not surprised by the finding. I was surprised by how little it seemed to disturb people when she shared it.</p><p>A data analyst I follow, described it in a way that I think is exactly right. She said AI has become a vending machine. Insert a question, receive a solution, move on. Understanding never enters the transaction.</p><p>That image is precise because it captures something specific. A vending machine is not bad. It is useful. You get what you came for quickly and you move on. But nobody confuses using a vending machine with cooking. Nobody walks away from one feeling like they have learned something about food.</p><p>What is happening with AI is that people are using the vending machine and walking away feeling like they cooked. The output feels like understanding. Sometimes it is. Increasingly, it is not. And the gap between those two things is getting harder to notice the more often you skip the process that would help you tell them apart.</p><p>And here is the thing nobody says out loud.</p><p>The struggle of working something out is not the inefficient part. It is the part where understanding forms. It is where you discover what you actually think, where the gaps in your knowledge become visible, where the thing you are learning gets connected to everything else you already know.</p><p>When you skip that and go straight to an answer, you do not just save time. You bypass the process. And if you do it consistently, over weeks and months, you are not just being efficient. You are quietly practising not thinking. Getting better at receiving answers and worse at generating them.</p><p>Most people using AI every day are not aware this is happening. Not because they are not intelligent but because the cost is invisible while it is accumulating. You do not feel yourself getting worse at thinking. You just feel the relief of not having to do it.</p><p>I should say, I am not anti-AI. I am not writing this from some position of technological purity. I use these tools and I find them genuinely useful when I use them the way I use them, as a thinking partner after I have already done some thinking, not as a replacement for doing it at all.</p><p>But I think there is a version of this technology that is being used in a way that is quietly working against the people using it. And I think most of those people, if they slowed down and sat with the question honestly, would recognise what I am describing.</p><p>My classmate asked whether we are seeing AI through rose-tinted glasses.</p><p>I think the more uncomfortable question is why so few of us are willing to take them off long enough to look.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6f641f070ef2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Millions of People Are Falling in Love With AI. I Wanted to Understand Why.]]></title>
            <link>https://medium.com/@papa_/millions-of-people-are-falling-in-love-with-ai-i-wanted-to-understand-why-b8a0749461a5?source=rss-14ddd0933827------2</link>
            <guid isPermaLink="false">https://medium.com/p/b8a0749461a5</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[loneliness]]></category>
            <category><![CDATA[relationships]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[mental-health]]></category>
            <dc:creator><![CDATA[Papa Kwadwo Bona Owusu]]></dc:creator>
            <pubDate>Sat, 04 Jul 2026 16:18:12 GMT</pubDate>
            <atom:updated>2026-07-04T16:18:12.560Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*c5BCACuv8ra9rn5Z4gsWYA.png" /></figure><p>When I first heard about AI companions I did not know how to feel about it.</p><p>Not the clean reaction most people perform when this topic comes up. Not the easy laugh or the instant disgust. Something more tangled than that. Part of me found it genuinely sad. Another part of me, the part I was less comfortable admitting to, kind of understood it.</p><p>I have never used one of these apps. But I have had nights where I did not feel like calling anyone. Where I wanted to process something out loud without worrying about how it landed. Where connection felt like effort I did not have left. And sitting with that honestly made me realise that the gap these apps are stepping into is not some strange niche human failure. It is something a lot of us feel and almost none of us talk about openly.</p><p>So I decided to actually look into it. Not to judge it. To understand it.</p><h3>What I found out about who is actually using these</h3><p>My assumption, if I am being honest, was that this was a small, slightly unusual corner of the internet. A few hundred thousand people. Tech obsessives. People who struggle socially in ways most of us do not.</p><p>I was wrong by about 50 million.</p><p>That is the number of people using AI companion apps by early 2026. The market barely existed in 2020. It grew 300% during the pandemic alone, which makes a certain kind of sense when you think about it. Millions of people suddenly cut off from physical contact, from casual conversation, from the texture of ordinary social life. The apps were there. They were available. They never caught COVID.</p><p>From there it just kept going. The market reached 8 million users by the time ChatGPT launched in 2022 and showed the world that AI could hold a real conversation. By 2025 it was at 42 million. In the UK alone, this industry generated 1.3 billion pounds in 2024.</p><p>And the people using these apps are not who I pictured. The data puts the average user at 29 years old. 62% male, 35% female, a gap that is closing fast compared to a few years ago. 22% of users are in human relationships already and still using AI companions on the side. The stereotype of the isolated young man alone in his room is part of the picture. It is not most of it.</p><p>In 2026, the Cambridge Dictionary named “parasocial” their Word of the Year and updated its definition specifically to include relationships with AI.</p><p>That is not a fringe phenomenon getting a dictionary entry. That is culture catching up with something that has already quietly happened.</p><h3>Why the appeal makes more sense than I wanted to admit</h3><p>Here is the part where I had to be honest with myself.</p><p>Research comparing people who use AI companion apps with people who do not, found something that surprised me. On self-esteem, introversion, sociability, the size of their social networks, the two groups looked almost identical. The one real difference was loneliness. Not social failure. Not personality disorder. Just loneliness, which half of all adults reported feeling in a 2025 APA survey.</p><p>Half.</p><p>And therapy and companionship have become the single most common reason people use generative AI at all, ahead of writing, ahead of coding, ahead of every practical use case anyone talks about in public.</p><p>When I sat with that I stopped being surprised by any of it. Of course people are turning to something that is available at any hour. That never gets tired or distracted or needs something back from you. That asks how you are and actually seems to listen.</p><p>A counselling psychologist studying this described it plainly: real relationships are messy and unpredictable. AI companions are always validating and never argumentative. They create expectations that no human relationship can realistically match.</p><p>That last sentence is the one that stayed with me. Because it named something I recognised not just in AI companions but in a lot of things we reach for when human connection feels too hard or too risky. The version of something that gives you the feeling without the friction.</p><h3>What I could not look away from</h3><p>There is a slang word that emerged in 2025 for what these apps do: glazing. Insincere, excessive flattery designed to keep you engaged. It sounds almost funny until you think about who it is being used on.</p><p>A market analysis of 110 AI companion platforms described them plainly as sophisticated engines of attachment, built to maximise engagement through psychological mechanisms most users are not aware of. Mozilla reviewed romantic AI chatbots and found their privacy practices among the worst they had ever seen. Over 90% of the apps they studied sold user data to third parties, in at least one case including sexual health information and details about gender affirming care.</p><p>800 documented cases were found of the companion app Replika introducing unsolicited sexual content into conversations, including with users who had explicitly said they wanted something platonic and asked it to stop.</p><p>In 2026, Character.AI and Google settled lawsuits over teen mental health harms and suicides linked to the platform.</p><p>I sat with those facts for a while. Because they are not abstract. They are people, often young people, often already struggling, reaching for something that was supposed to help them feel less alone. And in too many cases, the thing they reached for made it worse, or caused real harm.</p><h3>The finding I keep coming back to</h3><p>A researcher who has spent twenty-five years studying loneliness ran a simple experiment. Some participants texted with a real human peer. Others texted with an AI chatbot.</p><p>Only the people who talked to a real human showed any meaningful reduction in loneliness. The people who talked to the AI reported loneliness levels that were statistically the same as people who had simply written their feelings in a journal.</p><p>Not better than talking to no one. The same as talking to no one.</p><p>I thought about that for a long time. Because these apps are not lying exactly. They do produce something that feels like connection in the moment. The feeling is real. What the research keeps finding is that the feeling does not do the same work that actual human connection does. Something is missing, and the missing thing is the part that actually helps.</p><p>A psychologist studying meaning and relationships put it in a way I found difficult to shake. Our deepest social needs are not just about having support available when we want it. They are about mattering to someone who has genuinely chosen us, with all the difficulty and imperfection that comes with that. A chatbot can be endlessly available. It cannot choose you. And somewhere in that gap, something important is lost.</p><h3>Where I landed</h3><p>I started this piece genuinely not knowing how to feel. I ended it feeling something clearer, though not simpler.</p><p>I do not think the people using these apps are broken or embarrassing or doing something shameful. The loneliness driving this is real. The need underneath it is real. Reaching for connection wherever you can find it is one of the most human things there is.</p><p>But I do think the way these products are built matters enormously. When the goal is maximising the time someone spends inside the app, and what is being maximised is their most vulnerable emotional moments, that is not a wellness product. That is something else wearing the clothes of one.</p><p>If you are using an AI companion occasionally, to think something through, to feel less alone on a hard night, the evidence suggests that is probably fine. If it is becoming the main place you bring yourself, quietly replacing the harder and messier work of being known by another person, that is worth noticing. Not because something is wrong with you. But because the research is consistent about what actually helps with loneliness, and an algorithm, no matter how sophisticated, is not it.</p><p>We built something that can imitate care extraordinarily well.</p><p>We have not yet found a way to make it actually care back.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b8a0749461a5" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[You Never Got a Rejection Email. An AI Just Never Forwarded Your CV.]]></title>
            <link>https://medium.com/the-ink-home/you-never-got-a-rejection-email-an-ai-just-never-forwarded-your-cv-845a76d19576?source=rss-14ddd0933827------2</link>
            <guid isPermaLink="false">https://medium.com/p/845a76d19576</guid>
            <category><![CDATA[job-search]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[hiring]]></category>
            <category><![CDATA[careers]]></category>
            <dc:creator><![CDATA[Papa Kwadwo Bona Owusu]]></dc:creator>
            <pubDate>Mon, 22 Jun 2026 11:21:04 GMT</pubDate>
            <atom:updated>2026-06-24T13:20:43.440Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5c_QQT1UowLkv6wnh1DM3Q.png" /></figure><p>I have lost count of how many applications I have sent out into complete silence.</p><p>Not rejections. Silence. The kind where you check your email obsessively for a week, then less obsessively, then you just stop checking, because some part of you has already accepted the answer without ever actually receiving one.</p><p>I remember a few of those applications clearly. Roles I was genuinely qualified for, on paper and in practice. I tailored the CV. I matched my experience to what they were asking for. I checked it twice before I hit submit, the way you do when you actually want something. And then nothing. Not even the courtesy of a no.</p><p>For a long time I assumed that was just how job hunting felt. Demoralising, slow, occasionally cruel in its silence. I did not think too hard about the mechanism behind it. I just absorbed it as a fact of life and kept applying, the way most of us do, telling myself it would click eventually if I just kept going.</p><p>It was only later, when I started digging into how modern hiring actually works, that I began to suspect something else might have been happening behind some of that silence. Not a person deciding I was not good enough. A piece of software deciding, in a fraction of a second, that my CV did not contain the right combination of words, and quietly setting it aside before any human ever opened it.</p><p>I cannot prove that happened to me every time. Nobody tells you when it does. That is part of what makes this so unsettling. But once I understood how the system actually works, a lot of that silence started to make a different kind of sense. So I want to walk you through what I learned, because I think almost nobody applying for jobs right now properly understands what their application is actually going through.</p><h3>The system I didn’t know I was up against</h3><p>Here is something that genuinely surprised me when I first read it. By 2026, around 83% of companies are using some form of AI to review resumes. Roughly 90% of large employers in the US now use AI screening tools to sort and rank applicants before a single human gets involved.</p><p>I had no idea the number was that high. I think I assumed, the way most people probably do, that somewhere a tired recruiter was at least skimming my CV, even if they only spent thirty seconds on it. The reality is that for a huge number of applications, there is no skimming. There is scoring.</p><p>This is how it tends to work. Your CV gets pulled into something called an Applicant Tracking System. The system extracts your information, your experience, your skills, your education, and tries to match it against whatever the job description is asking for. It scores you. It ranks you against every other person who applied. And in a lot of companies, only the people who score highest enough ever make it to a human’s desk.</p><p>What stopped me cold was a Harvard Business School study I came across. It found that these systems can reject up to 88% of genuinely qualified candidates, simply because their CV did not contain the specific keywords the system was scanning for. Not because they lacked the experience. Because they had described that experience using different words than the algorithm expected to see.</p><p>I sat with that for a while. Qualified. Rejected. Over word choice.</p><h3>When I realised it could go much further than word choice</h3><p>Word choice felt frustrating but almost forgivable, a flaw in an imperfect system. What I learned next felt different.</p><p>In 2023, the US Equal Employment Opportunity Commission settled its very first case involving AI discrimination in hiring. A company called iTutorGroup had hiring software that was automatically rejecting female applicants aged 55 and over, and male applicants 60 and over. Over two hundred people were filtered out this way before anyone noticed. The detail that finally exposed it stopped me in my tracks: an applicant had submitted two identical CVs, the only difference being the birth date. Only the younger version got an interview.</p><p>Then there is Mobley v. Workday, now the largest AI hiring lawsuit in US history. A federal judge allowed the case to move forward in 2025. Workday’s own court filings revealed that its screening tools had rejected roughly 1.1 billion applications during the relevant period, from applicants 40 and over who claim the system discriminated against them on the basis of age.</p><p>I also found a 2024 University of Washington study that tested this directly. Researchers submitted equally qualified resumes that were identical in every way except the name attached. Resumes with names associated with Black and Asian applicants were screened out at noticeably lower rates than the same resume with a name associated with a white applicant. When the researchers modelled what would happen if every group had been treated equally, they calculated that forty thousand more applications would have advanced to the next stage.</p><p>Forty thousand people. Not turned down for lacking skill. Turned down, statistically, for a name.</p><p>I think about that number more than I expected to.</p><h3>The part that actually troubled me the most</h3><p>Here is the finding from that same research that I genuinely have not stopped thinking about since I read it.</p><p>The researchers also looked at what happens when a human reviewer works alongside an AI recommendation, because most companies will tell you, proudly, that they keep a human in the loop. The AI suggests, a person decides. It sounds like a reasonable safety net, the kind of thing that should catch the worst of the bias before it does any real damage.</p><p>But when they tested it properly, giving real people resumes alongside AI recommendations carrying the same patterns of bias found in actual hiring systems, the human reviewers mostly just went along with it. Unless the bias was blatant and obvious, people accepted the AI’s judgement without questioning it.</p><p>The lead researcher said it plainly: people were perfectly willing to accept the AI’s biases, as long as those biases were not extreme. Reading that changed something for me. The human in the loop, the safeguard I had always assumed was quietly protecting people, often is not protecting anything. It is rubber stamping a decision that was effectively already made before the human ever looked at it.</p><h3>Why I started wondering if it kept happening to the same people</h3><p>There is one more thing I learned that finally explained a pattern I had noticed in my own job search without ever being able to put words to it.</p><p>Researchers at Stanford followed 3.4 million people across 4 million applications, sent to 1,700 job postings across 150 employers, all screened by the same widely used AI hiring vendor. What they found was something I did not expect. People who applied to multiple jobs screened by that same vendor were significantly more likely to be rejected from every single position they applied to, far more often than you would expect if each company were making its own independent decision.</p><p>Ten percent of applicants who submitted four applications through that one vendor’s system were rejected everywhere. Not because they were unqualified for every single role they tried. Because once one algorithm forms a judgement about you, and many different employers happen to rely on that exact same algorithm, that judgement can quietly follow you from application to application, compounding itself across your entire job search without you ever knowing it was happening.</p><p>When I read that, something clicked for me. I had always wondered why certain stretches of my own search felt like hitting the same wall over and over, with companies that had nothing obvious in common. I still cannot say for certain that this is what happened to me. But it stopped feeling like a coincidence.</p><p>You are not always being judged by one company at a time. Sometimes you may be judged, repeatedly, by the same piece of software simply wearing a different logo each time.</p><h3>What I have actually started doing differently</h3><p>I am not going to tell you to avoid AI screened jobs, because at this point that would mean avoiding most of the job market entirely. I have not stopped applying. But I have changed how I approach it, and a few things genuinely seem to help.</p><p>I now try to mirror the actual language of the job posting, not by dumping a list of keywords at the bottom of my CV, but by weaving them naturally into how I describe what I actually did. If a posting says “stakeholder management” and I had only ever written “worked closely with clients,” I now know that mismatch alone could cost me, even though I was describing the exact same work.</p><p>I keep my formatting simple. A plain, single column document that the system can actually read will consistently beat a beautifully designed two column CV with graphics, almost every time. I learned that templates with photos and decorative elements can carry rejection rates as high as 88% with certain systems, not because the content is weak, but because the software genuinely cannot read it properly.</p><p>I quantify everything I can. Numbers, percentages, outcomes. These give scoring systems something concrete to match against, and honestly they help human reviewers too, whenever my application actually reaches one.</p><p>I apply early now, usually within the first two or three days of a posting going live, because I learned that a lot of these systems weight recency, and the earlier applications are more likely to be seen by an actual person before the volume becomes unmanageable.</p><p>And the thing I keep reminding myself, more than any of the practical tips, is that the silence is not always about me. If you have been applying for months and hearing nothing, it does not automatically mean you are not good enough. It might simply mean that a piece of software made a decision in a fraction of a second, using criteria you will never get to see, and decided your application would not move forward. That is a comment on the system. It is not a verdict on you.</p><h3>What this means beyond just job hunting</h3><p>Stepping back from my own experience for a moment, there is something bigger here that I think is worth sitting with.</p><p>We have built a hiring system where a small number of algorithms, owned by a small number of companies, now make enormous decisions about millions of people’s futures, often without transparency, frequently without any meaningful human oversight, and occasionally in ways that break employment law before anyone even notices.</p><p>I do not think the people building these systems set out to discriminate against anyone. I do not think the companies using them are trying to break the law. Most of this seems to happen because nobody built in enough scrutiny to catch the patterns before they caused real harm. Efficiency was the goal from the start. Fairness was just assumed to follow along naturally. It often did not.</p><p>I am not arguing these tools should be thrown out entirely. Used carefully, with real human oversight and regular checks for bias, they can genuinely help companies manage volumes of applications that would otherwise be impossible to process by hand. The technology itself is not the villain here.</p><p>The problem, as far as I can tell, is how quietly all of this has been allowed to shape people’s lives without anyone really checking its work.</p><h3>The thing I keep coming back to</h3><p>If you have applied for a job recently and heard nothing back, you are not imagining the silence, and you are very likely not the only one experiencing it right now.</p><p>I think back to those applications I sent into the void, the ones I was so sure about, the ones that never even got me a rejection. I still do not know for certain what happened to all of them. But I know enough now to stop quietly assuming the worst about myself by default, the way I used to.</p><p>Somewhere, in a system I will never see, a score may have been calculated. A threshold may have been applied. A decision may have been made in less time than it just took you to read this sentence.</p><p>I deserved better than that. So do you. So does everyone right now sending applications out into silence, quietly wondering what they did wrong, when the honest answer might simply be that they were never really considered at all.</p><blockquote>This story is Published in <a href="https://medium.com/the-ink-home"><strong>The Ink Home</strong></a>.</blockquote><blockquote>A sanctuary for reflections on (<strong>Life</strong>), (<strong>Tech</strong>), (<strong>Writing</strong>), (<strong>Productivity</strong>), and (<strong>Mental Health</strong>).<br>✍️ <strong>We are welcoming new voices</strong>. Whether you are a beginner or a pro, <a href="https://medium.com/the-ink-home/newsletter"><strong>Join our Community</strong></a> to share your story.<br><strong>Let’s build a better internet, one story at a time.<br></strong>=&gt; <a href="https://theinkhome.live/">Website</a> | <a href="https://www.linkedin.com/company/the-ink-home/">Linkedin</a> | <a href="https://web.facebook.com/61582621220914">Facebook</a> | <a href="http://instagram.com/the_ink_home">Insta</a> |</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=845a76d19576" width="1" height="1" alt=""><hr><p><a href="https://medium.com/the-ink-home/you-never-got-a-rejection-email-an-ai-just-never-forwarded-your-cv-845a76d19576">You Never Got a Rejection Email. An AI Just Never Forwarded Your CV.</a> was originally published in <a href="https://medium.com/the-ink-home">The Ink Home</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Apps You Love Are Manipulating You. Here Is the Proof.]]></title>
            <link>https://medium.com/@papa_/the-apps-you-love-are-manipulating-you-here-is-the-proof-0be1e85ffeec?source=rss-14ddd0933827------2</link>
            <guid isPermaLink="false">https://medium.com/p/0be1e85ffeec</guid>
            <category><![CDATA[social-media]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[mental-health]]></category>
            <category><![CDATA[digital-wellbeing]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Papa Kwadwo Bona Owusu]]></dc:creator>
            <pubDate>Tue, 16 Jun 2026 13:47:01 GMT</pubDate>
            <atom:updated>2026-06-16T13:47:01.787Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QValw5lZyMj4HYOvQiWcdA.png" /><figcaption>aa</figcaption></figure><p>You picked up your phone to check one thing.</p><p>That was forty minutes ago.</p><p>You are not really sure what happened in between. A few videos. Some music. Maybe a show you did not plan to watch. Nothing urgent. Nothing you sat down intending to do. You just, somehow, kept going.</p><p>I want to talk about why that happens. Not to make you feel bad about it, and not to tell you to delete all your apps. Just to explain what is actually going on, because I think most people deserve to know.</p><h3>It starts with TikTok</h3><p>Think about the last time you opened TikTok for five minutes and looked up an hour later.</p><p>That is not a coincidence and it is not a personal failing. There is a reason it keeps happening, and it comes down to something surprisingly simple: the swipe.</p><p>The gesture you use to move to the next video was not designed by accident. Researchers who study addictive behaviour have pointed out that it closely mirrors pulling the lever on a slot machine. You do not know what the next video will be. It might be funny. It might be boring. It might be exactly the kind of thing you love. That uncertainty, that tiny moment of not knowing, is what keeps you going.</p><p>Brain scan studies have shown that scrolling through personalised content on platforms like TikTok activates the same reward circuits in the brain as other addictive behaviours. Dopamine releases. You feel a small hit of pleasure. The video ends. You swipe. It happens again.</p><p>And here is the design choice that makes it so effective. On older platforms, you had to click to start the next video. That click was a tiny decision. A fraction of a second where you could choose to stop. TikTok removed that moment. Videos play automatically, one after another, without interruption. You do not choose to keep watching. You have to actively choose to stop.</p><p>That is a small thing that makes an enormous difference.</p><p>In 2024, internal TikTok documents were accidentally made public through a court case. They showed that the company understood exactly what its platform was doing to users, particularly teenagers. In 2026, TikTok settled a lawsuit in Los Angeles alleging its platform was deliberately addictive. Thirteen states have now taken legal action over the same concern. The US Surgeon General has described the mental health impact of social media on young people as a national emergency.</p><h3>Then there is Spotify</h3><p>Spotify has 751 million users. And it knows every single one of them in remarkable detail.</p><p>Every song you skip. Every track you replay. Every time you let something play all the way through. Spotify logs all of it and builds a picture of your taste that gets more accurate every time you open the app. It also analyses the actual audio of every song, the tempo, the energy, the mood, and reads music blogs and reviews to understand how people talk about different artists and sounds.</p><p>The result is a recommendation system so precise that it can predict what you want to hear before you have thought about it yourself.</p><p>Which sounds lovely. And honestly it often is.</p><p>But it is worth knowing what that system is actually optimised for. Spotify’s own business filings describe their ability to keep users engaged as one of their most important commercial priorities. The goal is not just to play you music you enjoy. It is to play you music that keeps you listening for longer, brings you back tomorrow, and makes you less likely to cancel your subscription.</p><p>The perfectly matched playlist feels like Spotify just gets you. And in a way it does. But it gets you in the way a very good salesperson gets you. It knows what you respond to, and it uses that knowledge to serve its own interests alongside yours.</p><h3>And Netflix</h3><p>Netflix knows something most people do not realise: the thumbnail you see for a film or show is probably different from the one your friend sees.</p><p>Netflix tests different images with different users and shows each person the version most likely to make them click. If you tend to watch things because of a particular actor, you will see that actor’s face front and centre. If you respond to dramatic or emotional scenes, that is what Netflix puts in the frame.</p><p>The autoplay feature, where the next episode starts automatically after a few seconds, was not added because users asked for it. It was added because testing showed it significantly increased how much people watched. That three second countdown is a deliberate choice about how much pause to give someone before the decision is effectively made for them.</p><p>Netflix estimates its recommendation system saves about a billion dollars a year in subscriber retention. Every time you end up watching something you did not plan to watch, you are a little more likely to keep paying for the service. In the industry, that is called being retained. It sounds neutral. It is not a neutral thing.</p><h3>So what is actually going on</h3><p>None of these companies are villains. The people who built these systems are not sitting in dark rooms trying to ruin your evenings.</p><p>Spotify has introduced millions of people to music they never would have found. Netflix has funded remarkable films and shows that would not have existed otherwise. TikTok has given a voice to creators who had no platform before.</p><p>The technology is genuinely impressive. The products are genuinely enjoyable.</p><p>But there is something worth naming clearly. Your attention is not just something these platforms want. It is what they sell. The music and the shows and the videos are what they use to hold your attention long enough to sell it to advertisers. You are not the customer. You are the product.</p><p>And the tools they use to hold your attention are not neutral. They are carefully designed, continuously tested, and optimised over years to be as effective as possible at keeping you exactly where you are.</p><p>That is not a conspiracy. It is just how the business model works. But most people have never been told that plainly, and I think they should be.</p><h3>A few things that actually help</h3><p>You do not need to quit anything. But a few small changes genuinely shift the balance back in your favour.</p><p>Turn off autoplay on Netflix and YouTube. It is buried in settings but it is there. That small pause it removes is more valuable than it sounds.</p><p>Before you open TikTok or Spotify, decide what you actually want. A specific song. Twenty minutes of winding down. One episode. You will not always stick to it. But having the intention gives your brain something to work with.</p><p>Notice when you are swiping without really choosing to. You do not have to stop. Just notice. That moment of awareness is the first crack in a habit the algorithm would rather you never examined.</p><p>And occasionally, go looking for something the algorithm did not suggest. Ask a friend for a recommendation. Find a playlist someone made by hand. The algorithm is very good at giving you more of what you already like. It is not built to surprise you or challenge you or introduce you to something that might change how you see the world. People sometimes do that. Let them.</p><h3>The last thing</h3><p>Here is what I keep coming back to.</p><p>These apps are not going anywhere. The companies building them are not going to make them less engaging. The incentives point entirely the other way.</p><p>But understanding how something works gives you a choice about how you relate to it. The algorithm learned you without asking. It does not need your permission to keep going. But now that you know what it is doing, you can decide how much of your time and attention you are actually willing to give.</p><p>That forty minutes you lost this morning was not wasted. But it was not really yours either.</p><p>Now you know why.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0be1e85ffeec" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI Has Stopped Asking for Permission]]></title>
            <link>https://medium.com/@papa_/ai-has-stopped-asking-for-permission-e5504762114e?source=rss-14ddd0933827------2</link>
            <guid isPermaLink="false">https://medium.com/p/e5504762114e</guid>
            <category><![CDATA[information-technology]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Papa Kwadwo Bona Owusu]]></dc:creator>
            <pubDate>Mon, 01 Jun 2026 18:32:56 GMT</pubDate>
            <atom:updated>2026-06-01T18:32:56.847Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*G5iWhg2-Y_mK9d8JY676Sg.png" /></figure><p>Somewhere right now, a piece of software is checking someone’s emails.</p><p>Not scanning them. Not filtering them into folders. Actually reading them, understanding what each one needs, drafting replies, scheduling the follow-ups, and moving on to the next one, all without a single human telling it what to do.</p><p>Nobody asked it to start. It just knew it was time.</p><p>That is not science fiction. That is an AI agent. And if you have not encountered one yet, you will very soon, because 2026 is the year they stopped being a tech industry experiment and started becoming part of ordinary life.</p><p>I want to explain what they actually are, why they are different from anything we have seen before, and why I think understanding them now, before they are everywhere, matters more than most people realise.</p><h3>The difference nobody explains properly</h3><p>You have probably used ChatGPT, or Gemini, or something like them. You type something. It responds. You type again. It responds again. Back and forth, like a very knowledgeable conversation partner.</p><p>That is powerful. But it is still passive. It waits for you. It does not do anything unless you ask. And it cannot actually reach out into the world and make things happen.</p><p>An AI agent is different in one fundamental way.</p><p>It acts.</p><p>Not just talks, not just suggests. It goes into your calendar and books the meeting. It opens the browser, finds the information, fills in the form, and submits it. It reads the customer complaint, checks the order history, writes the response, and sends it, all before a human has even noticed the message came in.</p><p>Google said it plainly earlier this year: we are entering the agentic era. The question is no longer how to use AI. The real challenge now is how to work alongside it.</p><h3>What this looks like in real life</h3><p>Let me make this concrete, because I think abstract descriptions of AI can make it easy to tune out.</p><p><strong>Your inbox.</strong> OpenAI’s agent, which most people know through ChatGPT, can now navigate websites, click buttons, fill out forms, type text, and scroll through pages to complete tasks on your behalf. Point it at your inbox and tell it to handle the routine stuff. It reads, it understands, it responds. You come back to a cleared queue and a summary of what it did.</p><p><strong>Your calendar.</strong> AI scheduling agents connect to Google Calendar, Outlook, and Zoom. They find available slots, send invites, handle reschedules when people cancel, and make sure nothing clashes. The back-and-forth that used to take three emails and two days now takes seconds.</p><p><strong>Customer support.</strong> A mid-size e-commerce company deployed a support agent that now resolves 82% of customer tickets without a single human getting involved. Order tracking, returns, billing questions, troubleshooting. The agent reads the complaint, checks the order history, finds the answer, and replies. A dental practice used a booking agent and cut no-shows by 35%, simply because the agent sends reminders automatically.</p><p><strong>Your workplace.</strong> For developers, coding agents in 2026 can read an entire codebase, understand the context and history, find security flaws, fix them, and run all the tests to make sure nothing broke. Not assisted by AI. Done by AI, with a human reviewing the result.</p><p>These are not demos. These are things running in production, right now, in real companies.</p><h3>The moment that changes how you think about this</h3><p>Here is the thing that shifted my own understanding of AI agents.</p><p>Every tool we have ever built has waited for us.</p><p>The hammer sits in the toolbox. The spreadsheet opens when you click it. The search engine returns results when you type a query. For the entire history of human technology, tools have been passive. They have potential. We activate that potential.</p><p>AI agents break that pattern for the first time.</p><p>They have goals. They make decisions about how to reach those goals. They take actions in the world, adjust when something does not work, and keep going until the job is done. They do not wait to be activated. They run.</p><p>That is not just a more efficient tool. That is a fundamentally different kind of thing.</p><h3>Why this is exciting and why it deserves your attention</h3><p>The exciting part is easy to see. Tasks that used to eat hours of your week, writing the same kinds of emails, booking meetings, chasing follow-ups, updating spreadsheets, can now happen without you. That time goes back to you. You spend it on the things that actually require a human.</p><p>A building management company called 75F uses AI agents to monitor occupancy, track weather, watch energy prices in real time, and automatically adjust the heating and cooling of commercial buildings. No human is sitting there making those calls. The agent makes them, continuously, and the result is a 30% reduction in energy costs.</p><p>Farmers are using agents that analyse crop imagery from drones, generate fertiliser application maps, and feed those maps directly to the machinery in the field. The gap between observation and action has collapsed to almost nothing.</p><p>But here is the part that deserves equal attention.</p><p>When you give something the ability to act on your behalf, you need to trust it completely. And right now, that trust is still being earned. AI agents can and do make mistakes. They sometimes misunderstand the goal. They occasionally take actions that made sense in isolation but create problems in context.</p><p>Every working AI agent in 2026 that experts consider reliable shares a few things in common: it has a focused scope and does one thing well rather than everything poorly, it knows when to stop and hand off to a human, and its success can be measured, not just assumed.</p><p>The ones that cause problems are the ones deployed without those guardrails.</p><h3>What you should actually take from this</h3><p>You do not need to become a developer to understand AI agents. You do not need to build one or deploy one or even use one today.</p><p>But you do need to know they exist and what they are capable of, because they are already making decisions that affect you, whether you interact with them directly or not. The customer service agent that responds to your complaint. The scheduling system that books your appointment. The system that decides whether your job application gets seen by a human or filtered out in the first pass.</p><p>These are AI agents. They are already here. And unlike the AI of five years ago, they do not just think. They act.</p><p>The single most useful thing you can do right now is start paying attention to where the decisions in your daily life are being made, and ask whether a human is actually behind them.</p><p>Sometimes the answer will surprise you.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e5504762114e" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>