<?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 Weidan Li on Medium]]></title>
        <description><![CDATA[Stories by Weidan Li on Medium]]></description>
        <link>https://medium.com/@weidanlee?source=rss-870e9ba832c8------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*9OI4u3ZmUzVz-oNictOgxg.png</url>
            <title>Stories by Weidan Li on Medium</title>
            <link>https://medium.com/@weidanlee?source=rss-870e9ba832c8------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 08:05:38 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@weidanlee/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[When Your Agent Has All the Data and Still Gets It Wrong: A Lesson from Hans-Georg Gadamer]]></title>
            <link>https://medium.com/@weidanlee/when-your-agent-has-all-the-data-and-still-gets-it-wrong-a-lesson-from-hans-georg-gadamer-2a452b8b39c6?source=rss-870e9ba832c8------2</link>
            <guid isPermaLink="false">https://medium.com/p/2a452b8b39c6</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ai-design]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[user-experience]]></category>
            <category><![CDATA[user-research]]></category>
            <dc:creator><![CDATA[Weidan Li]]></dc:creator>
            <pubDate>Wed, 29 Apr 2026 09:57:47 GMT</pubDate>
            <atom:updated>2026-04-29T09:57:47.550Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/968/0*cfSdLM92Bb-kYP7W.png" /><figcaption><a href="https://x.com/HannesKerber/status/1776722913359544324">Hans-Georg Gadamer in 1974 during his first visit as a guest professor at BostonCollege</a></figcaption></figure><p>There is a kind of AI agent failure that is easy to miss, because on the surface, it does not look like a failure.</p><p>The agent gives an answer. The answer is fluent. It uses relevant information. It draws from the right documents. Nothing sounds obviously wrong. And yet, something still feels off.</p><p>The response sits slightly beside the point. It is technically correct, but contextually wrong. It feels like talking to someone who heard the words you said, but missed what you actually meant.</p><p>The user then rephrases the question. The agent adjusts a little. The conversation continues, but it never quite connects.</p><p>Let’s make this more concrete.</p><p>Imagine a UX designer working on a healthcare app. She asks an internal AI agent:</p><p>“What are the best practices for designing onboarding flows?”</p><p>The agent retrieves all the right material. It pulls in Nielsen’s usability heuristics, the company’s design system guidelines, a few onboarding case studies, and maybe some internal research notes. Then it gives a solid summary of onboarding best practices: reduce friction, explain value early, keep steps short, use clear language, support progress visibility.</p><p>Nothing in that answer is wrong.</p><p>But it is not what the designer needed.</p><p>She has been designing products for eight years. She already knows the general principles. What she was really asking was something more specific and more situated:</p><p>“Do these usual onboarding principles still hold when the users are elderly patients with low digital confidence, using a medical interface for the first time?”</p><p>She did not need the conventions and principles repeated. She needed them questioned.</p><p>The agent answered the question that was typed, not the question that was meant.</p><p>This is the important part: the retrieval pipeline did not fail. The documents were relevant. The answer was accurate. The system worked as designed.</p><p>And yet the interaction still failed.</p><p>So the issue is somewhere else. Until we can name where it is, we will keep building agents that make the same mistake, just with better tools.</p><h3><strong>The hidden assumption in how we think about context</strong></h3><p>When people building AI agents talk about context, they often treat it as a resource.</p><p>More context means better output. So the work becomes: give the agent more context. Build better retrieval. Add memory. Expand the context window. Connect more tools. Pull in more documents.</p><p>To be clear, all of that matters. Better retrieval and stronger memory systems are useful. They are not the problem.</p><p>But underneath this way of thinking is a quiet assumption: if an agent has more relevant information, it will produce a more relevant response.</p><p>Context goes in. Quality comes out.</p><p>That sounds reasonable. But the earlier healthcare onboarding design example shows where this model breaks down. The agent had relevant information, but it still missed the point.</p><p>Why?</p><p>Because the problem was not simply whether the agent had enough context. The problem was whether the agent had made contact with the user’s context.</p><p>Those are not the same thing.</p><p>The agent treated the user’s question as a signal to match against documents. It treated the retrieved material as information to summarise. But it did not ask a deeper question:</p><p>“Who is this person, what do they already know, and what are they really trying to work out?”</p><p>That is the missing layer.</p><p>Supplying context to an agent is not the same as helping the agent understand the user’s world.</p><p>This is where Hans-Georg Gadamer, a twentieth-century German philosopher, becomes useful for AI design.</p><h3><strong>What Gadamer means by a horizon</strong></h3><p>Gadamer’s major work, <em>Truth and Method</em>, is about how understanding actually happens.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*uwcrdgHK5dBfyXMd" /></figure><p>His philosophical idea is powerful: understanding is never context-free.</p><p>Whenever we try to understand a text, a conversation, a person, or a situation, we do it from somewhere in our history. We do not arrive as blank minds. We bring our background, our experience, our language, our assumptions, our professional habits, and even the things we take for granted without noticing.</p><p>Gadamer calls this a “horizon.”</p><p>A horizon is everything you can see from where you stand.</p><p>Your horizon includes your role, your knowledge, your culture, your past experiences, your expectations, and the questions that feel natural to you. It shapes what you notice. It shapes what you ignore. It shapes what you think something means.</p><p>And importantly, Gadamer does not treat the horizon as a bad thing.</p><p>It is not simply bias (in fact, bias is not a negative word at all in Gadamer’s context). It is not something we can fully remove. It is the condition that makes understanding possible in the first place.</p><p>Without standing somewhere, you cannot interpret anything at all.</p><p>This matters for AI agents because users always come with a horizon.</p><p>The user is not just a prompt. They are a person with a role, a task, a history, a skill level, a business problem, and a situation they are trying to move forward.</p><p>And the agent has something like a horizon too. It has its training, its system instructions, its retrieved documents, its memory, its previous turns in the conversation, and the framing built into its design.</p><p>So every interaction between a user and an agent is not just a question meeting an answer.</p><p>It is one horizon meeting another.</p><p>The real design question is: what happens when those two horizons meet?</p><h3><strong>The core idea: horizon fusion</strong></h3><p>This is where Gadamer gives us a useful way to think about better agent design.</p><p>For Gadamer, understanding does not happen when one person simply absorbs another person’s view. It also does not happen when one side overwrites the other.</p><p>Understanding happens through what he calls “horizon fusion.”</p><p>Horizon fusion means two perspectives meet, interact, and produce a broader understanding than either side had alone.</p><p>For example, when you read an old text, you do not fully become the author. You also do not stay locked inside your own modern assumptions. If real understanding happens, something shifts. Your horizon expands. The text becomes meaningful in your world, but your world is also changed by the encounter.</p><p>Something new is created between the two horizons.</p><p>Now return to the UX designer and the onboarding experience design question.</p><p>The agent had its own horizon. It had the design guidelines, the heuristics, the case studies, and the common onboarding advice.</p><p>But it did not really meet the designer’s horizon.</p><p>It did not recognise that she was experienced. It did not recognise that the real challenge was a vulnerable user group. It did not recognise that she was not asking for general best practice, but for help judging whether those practices should change in a specific context.</p><p>So the agent’s horizon stayed separate from the user’s horizon.</p><p>There was no fusion.</p><p>That is why the answer felt accurate but unhelpful or uncreative. It was full of information, but it did not connect to the situation that gave the question its meaning.</p><p>This is not a retrieval failure.</p><p>It is a fusion failure.</p><p>And retrieval alone cannot fix it.</p><p>You can fill an agent’s context window with highly relevant documents and still get an answer that ignores the user. More data does not solve a fusion problem. But better interaction design between the user and the agent might.</p><h3><strong>What this means for agent design</strong></h3><p>If horizon fusion is the goal, then agent design needs to move beyond “how do we give the agent more context?” and toward “how do we help the agent meet the user’s context?”</p><p>That shift has three practical implications.</p><h4><strong>1. Surface the user’s horizon, not just their query</strong></h4><p>A query is only the visible tip of what the user means.</p><p>When someone asks a question, they bring a lot of hidden context with them: what they already know, what they are worried about, what they have tried, what decision they need to make, and what kind of answer would actually help.</p><p>Agents that only respond to the typed query will always work with a thin version of the user.</p><p>Better agents need ways to understand the fuller picture.</p><p>For example:</p><p>What is this user’s role?</p><p>How experienced are they?</p><p>What project are they working on?</p><p>What decision are they trying to make?</p><p>What kind of answer would be useful right now: a summary, a critique, a recommendation, a challenge, or a next step?</p><p>In the designer’s case, even a small amount of user context would have changed the answer. If the agent knew she was a senior designer working on healthcare accessibility, it would have been less likely to give a generic onboarding summary. It might have said:</p><p>“The standard onboarding principles still apply, but for this user group, the bigger risk is cognitive load and anxiety. So the question is not just how to make onboarding shorter, but how to build confidence and trust before asking users to act.”</p><p>That is a very different answer.</p><p>Memory systems, onboarding questions, user profiles, and smart clarification are not nice extras. They are ways of making the user’s horizon visible.</p><h4><strong>2. Treat retrieval as horizon-building, not fact-fetching</strong></h4><p>Retrieval is often treated as a way to find the answer.</p><p>But from a Gadamerian view, retrieval should do something deeper. It should help build the agent’s horizon so the agent can meet the user more intelligently.</p><p>That changes what “good retrieval” means.</p><p>Topical relevance is only the starting point. The retrieved document should not just match the keywords. It should help the agent understand the situation.</p><p>So we need to ask:</p><p>Is this document current enough for the user’s problem?</p><p>Does it reflect the user’s domain?</p><p>Does it match the level of depth the user needs?</p><p>Does it help explain why the question matters now?</p><p>Does it support the decision the user is trying to make?</p><p>In other words, retrieval should not just stock the agent with information. It should shape the agent’s point of view for this specific interaction.</p><p>That is a different standard from “find the most relevant chunks of information.”</p><p>It is closer to: “find the material that helps the agent understand what the user is really trying to work through.”</p><h4><strong>3. Design for iterative understanding</strong></h4><p>Gadamer also gives us another useful idea: understanding is not built in one pass, but through a hermeneutic circle.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*aTgoVJ4T4hXvrIi3.jpg" /></figure><p>We start with an initial interpretation. Then we encounter more information. Then we revise. Then we return with a better view. This back-and-forth is how understanding deepens.</p><p>Human conversation works like this all the time.</p><p>Someone asks a broad question. We answer tentatively. They clarify. We ask a better question. They reveal more of what they mean. The answer becomes sharper.</p><p>But many AI agents are still designed around a single prompt-response pattern. The user asks. The agent answers. The interaction is treated as a task to complete.</p><p>That makes sense for simple questions. But for complex work, especially design, strategy, research, or any work that involves critical decision-making moments, understanding often needs a few turns.</p><p>Clarification should not be treated as a failure mode. It should be part of the design.</p><p>A good agent should know when to slow down and ask:</p><p>“Are you looking for general best practice, or are you trying to adapt this for a specific user group?”</p><p>That one question could prevent a technically correct but useless answer.</p><h3><strong>The deeper shift</strong></h3><p>None of these practices are completely new.</p><p>Teams are already building memory systems. They are improving retrieval. They are designing multi-turn agents. They are experimenting with user profiles and clarification flows.</p><p>But often, these practices are treated as separate features.</p><p>Gadamer gives us a stronger reason for why they matter.</p><p>Memory is not just convenience. It is how the user’s horizon becomes visible over time.</p><p>Retrieval is not just document search. It is how the agent builds a better vantage point.</p><p>Clarification is not just a fallback when the agent is confused. It is how understanding is formed through interaction.</p><p>That reason matters.</p><p>When a feature becomes expensive, slow, or hard to maintain, teams will cut it unless they understand why it is fundamental.</p><p>If memory is seen as a nice-to-have, it can be removed. But if memory is understood as part of how the agent recognises the user’s world, it becomes much harder to dismiss.</p><p>Gadamer helps us reframe context. Context is not just technical input. It is part of the relationship between the user and the agent.</p><p>So the better question is not:</p><p>“Does the agent have enough context?”</p><p>The better question is:</p><p>“Has the agent deeply met the user’s horizon?”</p><p>Did it understand not only the words, but the world behind the words?</p><p>Did the interaction produce an answer that neither the user nor the agent could have reached alone?</p><p>That is a much higher bar.</p><p>But it is also the right one.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2a452b8b39c6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[When AI does the thinking in customer research, innovation suffers]]></title>
            <link>https://medium.com/design-bootcamp/when-ai-does-the-thinking-in-customer-research-innovation-suffers-338350c9946e?source=rss-870e9ba832c8------2</link>
            <guid isPermaLink="false">https://medium.com/p/338350c9946e</guid>
            <category><![CDATA[ux]]></category>
            <category><![CDATA[research]]></category>
            <category><![CDATA[user-experience]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[customer-experience]]></category>
            <dc:creator><![CDATA[Weidan Li]]></dc:creator>
            <pubDate>Wed, 18 Feb 2026 05:22:20 GMT</pubDate>
            <atom:updated>2026-02-19T11:48:16.337Z</atom:updated>
            <content:encoded><![CDATA[<p>There’s an explosion of training on AI tools: how to prompt ChatGPT for analysis, how to use AI for coding interviews, how to generate insights faster. These skills matter. But something crucial is disappearing from the curriculum: sensemaking.</p><p>Sensemaking is the ability to connect dots and understand the why behind patterns in serendipity. It’s different from pattern-finding, which AI does brilliantly. Sensemaking is about developing judgement (knowing which patterns matter, understanding what they mean, and imagining what customers might need that they can’t articulate yet).</p><p><strong>Here’s why this matters for innovation:</strong></p><p>AI excels at finding patterns in existing behaviour. It can tell you what customers currently do, what they say they want, what features they use. This drives incremental improvement, which is valuable.</p><p>But innovation requires understanding unmet and unarticulated needs (the things customers don’t know they want yet because they can’t imagine the solution).</p><p>No customer data would have predicted demand for the iPad. People didn’t know they wanted a device between a phone and laptop because that category didn’t exist. Apple’s innovation came from a deep understanding of how people wanted to interact with technology, not from analysing what they currently did.</p><p><strong>The problem with AI-reliant research:</strong></p><p>When AI does the pattern-finding and people just validate the outputs, we’re training people to check boxes rather than understand deeply. Through research democratisation, we’re creating a generation of customer research professionals who can efficiently analyse existing behaviour but struggle to imagine what comes next.</p><p>High engagement with tools and processes. Low engagement with actual meaning-making.</p><p>This creates a business risk. Companies become very good at incremental improvements (optimising current features) but lose the capability for breakthrough innovation (understanding unarticulated needs). A company’s competitors who retain deep sensemaking capabilities will out-innovate the company itself.</p><p><strong>What to do about it:</strong></p><p>We need to fundamentally rethink how we cultivate customer understanding. Sensemaking isn’t only an analytical skill. It’s a way of engaging with the world. Here are practices that go beyond the screen:</p><ul><li>Cliché but rarely done well: Observe customer behaviour in context, paying attention to what people do versus what they say.</li><li>Notice the unspoken evidence: hesitations, workarounds, body language, environmental factors.</li><li>Spend time in customers’ physical spaces to understand their lived experience beyond the digital interface.</li><li>Create space for insights to emerge through movement (take walks, exercise, let your mind wander and then come back). Step away from screens regularly to let patterns synthesise subconsciously.</li><li>Listen to music, read philosophies, pay attention to international politics, engage in creative activities that allow different kinds of thinking. Read widely beyond your domain to build associative thinking.</li><li>Look for analogies from other industries, other life experiences. Notice how insights from one area of life illuminate customer needs in another.</li><li>Regularly ask: “What did I miss?” not just “What did I find?” Regularly observe discourse: Not only “What was talked about by people”, but also “what was NOT talked about by people?”</li><li>Practise articulating why something matters, not just what it does.</li></ul><p>AI tools are incredibly valuable for productivity, but not for sensemaking. Sensemaking is a full-body, full-context, deeply human capability that requires practices far beyond prompt engineering.</p><p>Innovation doesn’t come from efficiently processing information. It comes from deeply making sense of human needs in ways that reveal what’s possible but not yet imagined.</p><p>Are we making space for that kind of understanding? Or are we accidentally optimising it away in pursuit of efficiency?</p><p><strong>What practices do you use to cultivate deep customer understanding beyond the data?</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=338350c9946e" width="1" height="1" alt=""><hr><p><a href="https://medium.com/design-bootcamp/when-ai-does-the-thinking-in-customer-research-innovation-suffers-338350c9946e">When AI does the thinking in customer research, innovation suffers</a> was originally published in <a href="https://medium.com/design-bootcamp">Bootcamp</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Cognitive Engagement in Human-AI Interaction]]></title>
            <link>https://medium.com/@weidanlee/cognitive-engagement-in-human-ai-interaction-72be44a9ff4a?source=rss-870e9ba832c8------2</link>
            <guid isPermaLink="false">https://medium.com/p/72be44a9ff4a</guid>
            <category><![CDATA[human-ai-collaboration]]></category>
            <category><![CDATA[ux]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[human-ai-interaction]]></category>
            <category><![CDATA[ux-research]]></category>
            <dc:creator><![CDATA[Weidan Li]]></dc:creator>
            <pubDate>Tue, 10 Feb 2026 09:37:07 GMT</pubDate>
            <atom:updated>2026-02-10T09:38:35.552Z</atom:updated>
            <content:encoded><![CDATA[<h3>Why AI Assistance Might Be Making Us Think Less</h3><p><strong><em>Note</em></strong><em>: This article is based on a </em><a href="https://www.youtube.com/watch?v=JybLJiJpKS0&amp;t=1833s"><em>lecture</em></a><em> by Krzysztof Gajos from Harvard University’s Intelligent Interactive Systems group, delivered at Stanford. The research findings, examples, and frameworks discussed are drawn from his work and that of his collaborators. Full credit for the original insights belongs to Professor Gajos and his research team.</em></p><h3>1. Introduction: The Paradox of AI Assistance</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*r09NkevproemuxAEIbK1ww.png" /></figure><p>Imagine this scenario: A clinician sits before her computer screen, reviewing a patient’s case. The AI system has already analysed the symptoms, medical history, and recent lab results. Within seconds, it recommends Treatment A with a confidence score of 87% and provides three bullet points explaining why. The clinician glances at the recommendation, nods, and clicks “approve.” The entire decision takes less than thirty seconds.</p><p>Is it efficient? Certainly. But did she actually think about the decision, or did she simply defer to the machine?</p><p>This scenario illustrates a troubling paradox at the heart of AI-assisted work. We build AI systems with the promise that they will augment human intelligence and make us more effective in making better decisions. Yet mounting evidence suggests that AI assistance might be reducing how deeply we think, subtly eroding the very cognitive engagement that builds expertise, enables learning, and grounds accountability.</p><p>Krzysztof Gajos, who leads the Intelligent Interactive Systems group at Harvard University, has dedicated recent years to studying what happens when human intelligence meets machine intelligence. In a lecture at Stanford, he revealed an uncomfortable truth:</p><p>“Decision makers presented with AI-generated decision recommendations and explanations rarely engage cognitively with the content of what the AI communicates.”</p><p>In other words, we’re not thinking about what the AI tells us — we’re simply accepting it.</p><p>His lecture examines a critical but often overlooked dimension of human-AI interaction: the design of AI assistance systems fundamentally shapes not just task outcomes, but the quality of human cognitive engagement.</p><p>Current approaches, optimised for efficiency and accuracy, may be optimising for precisely the wrong things — creating systems that make us faster but less thoughtful, more productive in the moment but less capable over time.</p><h3>2. Understanding Cognitive Engagement</h3><h4>What Cognitive Engagement Means</h4><p>In the context of human-AI interaction, <strong>cognitive engagement refers to the depth and quality of mental processing that occurs when people interact with AI systems.</strong></p><p>It exists on a spectrum:</p><p><strong>Shallow processing</strong> involves using superficial cues to make decisions without genuine understanding. A person might see that an AI has provided an explanation with many specific facts and conclude the AI is competent, without actually reading or evaluating those facts. They pattern-match, take shortcuts, and accept AI outputs without critical evaluation.</p><p><strong>Deep processing</strong> involves active reasoning, critical evaluation, and integration of AI-provided information with existing knowledge. A person engaged in deep processing questions the AI’s logic, considers alternative explanations, and synthesises the AI’s input with their own expertise.</p><p>Both modes have their place. The problem arises when AI systems consistently push users towards shallow processing in contexts where deep processing matters — which turns out to be most contexts where AI assistance is deployed.</p><h4>Why This Matters Beyond Task Performance with AI</h4><p>The common metrics we use to evaluate AI systems such as decision accuracy, time to completion, user satisfaction, often miss a question that is significantly important: <strong>whether people are actually thinking and learning.</strong></p><p><strong>Skill development and expertise acquisition depend fundamentally on cognitive engagement.</strong> When a junior clinician works through a difficult case, the struggle to reason through symptoms, consider differential diagnoses, and weigh treatment options builds clinical judgement. If an AI simply tells them the answer, that learning opportunity is gone.</p><p><strong>Incidental learning</strong> is the informal acquisition of knowledge through everyday work. This type of <strong>learning </strong>accounts for roughly half of workplace learning. But it only occurs when people cognitively engage with their tasks. If an AI simply does the work for a user, the user learns nothing. The efficiency gain today becomes a capability gap tomorrow.</p><h3>3. The Current State: How AI Reduces Cognitive Engagement</h3><h4>The Shortcut Problem</h4><p>Gajos and his colleagues conducted a study asking people to write captions for images using a predictive keyboard they had built and controlled. The results revealed that predictive text doesn’t just make typing faster — <strong>it fundamentally changes what people write</strong>.</p><p>First, text became more predictable. Users were more likely to choose words aligned with the language model’s predictions. The mechanism was simple substitution: a person thinking “the train is approaching the station” would see the suggestion “on” and write “the train is approaching on the station” instead — not because “on” was better, but because it was available.</p><p>More surprisingly, people using predictive text wrote shorter captions. The predictive system rarely suggested adjectives, adverbs, or embellishments. Someone wanting to write “An old train is approaching a quiet outdoor station” would find “outdoor” unpredicted but “station” predicted. So they’d skip the elaboration, producing simpler, shorter text.</p><p>When the researchers trained language models on either predominantly positive or predominantly negative restaurant reviews, then had people write reviews using those models, the model’s bias substantially affected how readers perceived the reviews — even though the writers had already committed to their star ratings before writing.</p><p>This is the shortcut problem: <strong>AI provides a path of less cognitive resistance, and users take it, often unconsciously.</strong> The result is content that’s more homogeneous, less thoughtful, and more reflective of the AI’s training data than the user’s actual intentions.</p><h4>The Explanation Illusion</h4><p>The field of explainable AI emerged from a reasonable premise: if we show people why an AI recommends something, they’ll understand the recommendation better and make better decisions.</p><p>Research on how people actually respond to AI explanations reveals a more troubling pattern.</p><p>Gajos’s work on AI-assisted decision-making shows that <strong>explanations often increase trust without increasing understanding</strong>. People see that the AI has provided an explanation — especially one filled with concrete facts and specific details — and conclude that the AI is competent. They then rely on the AI’s recommendation without actually engaging with the content of the explanation.</p><p>In one study examining how clinicians make treatment decisions with AI assistance, researchers found evidence of systematic overreliance. More concerningly, adding explanations didn’t improve the situation. People weren’t reading the explanations and thinking more carefully. They were seeing the explanations and thinking, “This system must know what it’s doing.”</p><p>When evaluating information from a source they don’t know well, people pay attention to superficial features: the number of concrete facts presented, the confidence of the presentation, the presence of supporting detail. This is similar to the way we typically design AI explanations using lots of specific facts, confident language, and detailed supporting information.It is perfectly designed to make people trust the AI without actually reading or thinking about what it says.</p><p>In summary, the assumption was:</p><p><strong>Explanations → Understanding → Better decisions.</strong></p><p>The reality is:</p><p><strong>Explanations → Perceived competence → Over reliance.</strong></p><h4>The Decision Recommendation Trap</h4><p>Consider the traditional workflow for complex decision-making:</p><p><strong>(1) Analyse the situation and gather information</strong></p><p><strong>(2) Reason through possible options and their implications</strong></p><p><strong>(3) Make a decision based on that reasoning</strong></p><p><strong>(4) Observe the outcome and learn from it</strong></p><p>Now consider the AI-assisted workflow:</p><p>(1) Receive AI recommendation</p><p>(2) Briefly review (or not)</p><p>(3) Accept or reject</p><p>(4) Move to the next case</p><p>Steps 1–3 from the traditional workflow (the reasoning process that builds expertise) have been removed. And because the AI is usually correct, users have little reason to engage more deeply.</p><p>Gajos’s research demonstrates the consequence: no incidental learning occurs with standard explainable AI designs. His team tested this by having people make nutritional decisions with AI assistance, then later testing whether they’d learned anything that would help them make similar decisions without the AI. People given decision recommendations and explanations learned essentially nothing. They performed no better than people who’d received no feedback at all. They’d made correct decisions with the AI’s help, but they hadn’t actually thought about those decisions in a way that built knowledge.</p><h4>Cognitive Offloading vs Cognitive Partnership</h4><p>These examples illustrate different points on a spectrum of AI assistance:</p><p><strong>1. Full cognitive offloading</strong>: The AI makes decisions and humans simply rubber-stamp them. Example: clicking “approve” on an AI-generated report without reading it.</p><p><strong>2. Decision support</strong>: The AI recommends and humans validate. This sounds better, but validation often means superficial engagement. Example: a clinician who glances at the AI’s treatment recommendation, sees it matches their initial intuition, and approves it.</p><p><strong>3. Information augmentation</strong>: The AI provides context and information; humans do the reasoning. Example: an AI that summarises a patient’s medical history, but leaves treatment decisions to the clinician.</p><p><strong>4. Collaborative reasoning</strong>: Both AI and human contribute to the thinking process, with genuine back-and-forth. Example: an AI that asks probing questions about a case, prompting the clinician to articulate their reasoning.</p><p>Current commercial AI systems fall predominantly into the first two categories — the ones requiring least cognitive engagement. We’ve systematically optimised for efficiency whilst ignoring engagement.</p><h3>4. Why Standard Interventions Don’t Work</h3><h3>4.1 The Cognitive Forcing Paradox</h3><p>Recognising that people over-rely on AI recommendations, researchers tried an intervention called cognitive forcing: make people form their own judgement first, then show them the AI’s recommendation, then let them revise if they want.</p><p>Gajos and his colleagues tested this approach. As expected, cognitive forcing reduced overreliance on AI. When the AI was wrong, people in the cognitive forcing condition made better decisions than those who saw the AI recommendation immediately.</p><p>But is it a big success? Not quite. The researchers then tested whether cognitive forcing actually increased cognitive engagement, still using incidental learning as their measure. The results were striking: <strong>people in the cognitive forcing condition showed no evidence of learning.</strong></p><p>The researchers hypothesise that cognitive forcing didn’t actually make people think more deeply. It just shifted what they over-relied on. Instead of over-relying on the AI, people over-relied on their initial snap judgement.</p><p>The lesson is that disrupting shortcuts doesn’t automatically create engagement. You can’t force people to think more deeply simply by rearranging <strong>when</strong> they see information.</p><h3>4.2 Why More Information Doesn’t Equal More Thinking</h3><p>A natural response might be: provide more detailed explanations, give people more information. Surely more information leads to better understanding?</p><p>The paradox is that additional information can actually reduce engagement. People have limited cognitive resources, and when presented with extensive explanations, t<strong>hey often satisfice — they take the path of least mental effort to achieve a particular goal</strong>. An explanation with twenty detailed points is more likely to be skimmed or ignored than an explanation with three key points.</p><h3>4.3 The Accountability Illusion</h3><p>Much discussion of AI decision-making emphasises keeping “humans in the loop.” The ethical reasoning is sound: humans should make final decisions and bear responsibility for consequential choices.</p><p>But Gajos’s research exposes the gap between nominal and actual accountability. Yes, a human technically makes the final decision when they click “approve” on an AI recommendation. But if they haven’t genuinely evaluated that recommendation, in what real sense are they accountable if they’ve deferred to the AI without critical engagement?</p><p>Current systems create an accountability illusion: the appearance of human oversight without the cognitive engagement that would make that oversight carry a meaning.</p><h3>5. Design Approaches That Enhance Cognitive Engagement</h3><h3>5.1 The Informative Explanation Approach</h3><p>Given these failures, Gajos and his team tried a radically different approach. They removed the decision recommendation entirely. Instead of telling people what to do and why, just give them information that helps them figure it out themselves.</p><p>In their nutrition study, instead of saying “Meal A has more carbohydrates” (a decision recommendation), the system said “Beans are a significant source of carbohydrates” (an informative explanation). This requires users to make a small cognitive leap: if beans are a significant source of carbohydrates, and Meal A contains beans whilst Meal B doesn’t, then Meal A probably has more carbohydrates.</p><p>The results were remarkable. People made decisions just as accurately as those who received direct recommendations. But unlike the recommendation group, the informative explanation group showed substantial evidence of learning. When tested later without AI assistance, they performed significantly better. They’d actually absorbed knowledge that transferred to new situations.</p><p>The key insight in this experiment is that making even a small cognitive leap is enough to trigger genuine engagement. When people have to connect the dots themselves, even when the dots are very close together, they think about what they’re doing.</p><h3>5.2 Evaluative AI Framework</h3><p>Miller (2023) has proposed a related approach called evaluative AI. Instead of recommending one option, the AI provides a balanced synthesis of reasons for and against different alternatives.</p><p>For example, in medical treatment selection:</p><p><strong>Traditional explainable AI</strong>: “Recommend Treatment A. Reasoning: Most effective for this condition, low side effect profile, covered by insurance.”</p><p><strong>Evaluative AI</strong>:</p><ul><li><strong>Treatment A</strong>: Most effective for this condition, low side effects, covered by insurance, but requires daily dosing</li><li><strong>Treatment B</strong>: Moderate effectiveness, very low side effects, once-weekly dosing, but higher cost</li><li><strong>Treatment C</strong>: Good effectiveness, may interact with patient’s current medications, daily dosing, fully covered</li></ul><p>We have seen the similar response structure from most of the LLMs today. The evaluative approach presents information rather than conclusions. It enables informed choice rather than guided choice. The human must weigh trade-offs and consider context.</p><h3>5.3 Question-Prompting Systems</h3><p>Research by Maes and colleagues explored another alternative. Instead of giving people information or recommendations, <strong>ask them questions</strong> that focus their attention on relevant aspects of the problem (Alqaraawi et al., 2020).</p><p>For a loan application decision, rather than “Deny this application — income insufficient, debt-to-income ratio 47%,” the AI might ask: “How does this applicant’s debt-to-income ratio compare to our threshold?” “Are there compensating factors in their credit history?”</p><p>The questions guide thinking without replacing it. Remarkably, this research found that people prompted with such questions performed better than either people alone or AI alone. They achieved genuine complementarity, the best practice of human-AI collaboration.</p><h3>5.4 Design Principles Summary</h3><p>In summary, how to design approaches that enhance cognitive engagement? Here are some common principles:</p><ol><li><strong>Remove decision shortcuts</strong>: Don’t tell users what to do. Provide information, context, or questions that help them figure it out themselves.</li><li><strong>Require small cognitive leaps</strong>: The gap between information and conclusion should be small enough to bridge easily, but large enough that bridging it requires thought.</li><li><strong>Support the reasoning process</strong>: Augment human thinking rather than replacing it.</li><li><strong>Enable exploration</strong>: Allow users to consider multiple options, examine trade-offs, and explore what-if scenarios.</li><li><strong>Preserve learning opportunities</strong>: Design AI systems so that using them builds knowledge and capability over time, not just completing tasks in the moment.</li></ol><h3>6. Measuring Cognitive Engagement</h3><h3>6.1 Beyond Task Performance</h3><p>Traditional metrics for evaluating AI systems — decision accuracy, time to completion, user satisfaction — are necessary but insufficient. A system could score well on all three whilst completely undermining cognitive engagement.</p><p>We need new metrics that capture whether people are actually thinking:</p><p><strong>Incidental learning measures</strong>: Test whether users can perform similar tasks without AI assistance after using the system.</p><p><strong>Knowledge retention</strong>: Assess what users understand and remember days or weeks after using the system.</p><p><strong>Quality of reasoning</strong>: Evaluate not just whether users reach correct conclusions, but whether they can explain their reasoning.</p><p><strong>Transfer to novel situations</strong>: Test whether users can apply what they’ve learned to new contexts.</p><h3>6.2 Incidental Learning as a Proxy</h3><p>Gajos’s use of incidental learning as a measure of cognitive engagement offers a practical methodology. The logic is straightforward: <strong>if people engage cognitively with AI-provided information, they should learn from that engagement.</strong></p><p>The measurement approach:</p><ol><li>Have users complete tasks with AI assistance</li><li>Later, test their performance on similar tasks without AI assistance</li><li>Compare to appropriate baselines</li></ol><p>If AI assistance produces learning comparable to expert feedback, users were engaging deeply. If it produces learning no better than no assistance at all, users were accepting recommendations without thought.</p><p>This approach has revealed that most current AI assistance produces minimal learning, which is exactly the evidence of shallow engagement. But informative explanation approaches show learning comparable to expert feedback. It implies the evidence that small design changes can fundamentally alter engagement levels.</p><h3>6.3 Process Measures</h3><p>Beyond outcome measures, we can also assess engagement through<strong> process observation</strong>:</p><ol><li><strong>Eye tracking</strong> reveals where users look and for how long. Do they read explanations or just glance at recommendations?</li><li><strong>Think-aloud protocols</strong> expose reasoning processes. Are users critically evaluating AI information or simply accepting it?</li><li><strong>Interaction patterns</strong> show how users engage with the interface. Do they explore alternatives or immediately accept the first recommendation?</li></ol><p>These process measures work alongside outcome measures to reveal how people actually think when using AI assistance.</p><h3>7. Conclusion: Towards Thoughtful AI Assistance</h3><p>Krzysztof Gajos’s research reveals an uncomfortable truth about current AI assistance: we’ve optimised for the wrong outcomes. We’ve built systems that make people faster and more accurate in the moment, but we’ve done so by reducing how deeply people think, undermining learning, and eroding expertise. This will be very risky to those who should benefit from the expertise, such as a patient, a student, or an athlete.</p><p>The path forward requires recognising cognitive engagement as a critical design objective alongside efficiency and accuracy. It requires measuring not just whether people complete tasks successfully, but whether they learn and become more capable. It requires design patterns that provide purposeful information rather than conclusions, that require small cognitive leaps, and that support reasoning processes rather than skipping them.</p><p>Most fundamentally, it requires rethinking what “helpful AI” means. Is AI most helpful when it gives us answers, or when it helps us think better? When it saves us time now, or when it builds our capability for the future?</p><p>Gajos’s work suggests that the most valuable AI assistance doesn’t minimise human cognitive effort. Rather, it optimises the quality and appropriateness of that effort. It makes us not just faster workers, but better thinkers. It builds expertise instead of replacing it.</p><p>The question isn’t whether AI will reshape human cognition. It already is. The question is whether we’ll design that reshaping intentionally, with full awareness of whether we are promoting or preventing cognitive engagement.</p><p><strong>Reference</strong></p><p>Alqaraawi, A., Schuessler, M., Weiß, P., Costanza, E., &amp; Berthouze, N. (2020). Evaluating saliency map explanations for convolutional neural networks: A user study. <em>Proceedings of the 25th International Conference on Intelligent User Interfaces</em>, 275–285.<a href="https://doi.org/10.1145/3377325.3377519"> https://doi.org/10.1145/3377325.3377519</a></p><p>Miller, T. (2023). Explainable AI is dead, long live explainable AI! Hypothesis-driven decision support. <em>Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency</em>, 333–342.<a href="https://doi.org/10.1145/3593013.3594001"> https://doi.org/10.1145/3593013.3594001</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=72be44a9ff4a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Toward Human-Centred AI Research: A Framework for Evolving UX Research in the Age of Artificial…]]></title>
            <link>https://medium.com/@weidanlee/toward-human-centred-ai-research-a-framework-for-evolving-ux-research-in-the-age-of-artificial-8a3e550f2999?source=rss-870e9ba832c8------2</link>
            <guid isPermaLink="false">https://medium.com/p/8a3e550f2999</guid>
            <category><![CDATA[human-centric-ai]]></category>
            <category><![CDATA[ux-research]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[research]]></category>
            <category><![CDATA[user-experience]]></category>
            <dc:creator><![CDATA[Weidan Li]]></dc:creator>
            <pubDate>Sun, 01 Feb 2026 23:30:46 GMT</pubDate>
            <atom:updated>2026-02-01T23:30:46.210Z</atom:updated>
            <content:encoded><![CDATA[<h3>Toward Human-Centred AI Research: A Framework for Evolving UX Research in the Age of Artificial Intelligence</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TCWgfi96Ejm3qyox4tmnzA.png" /></figure><h3>Abstract</h3><p>The rapid advancement of artificial intelligence technologies has created both opportunities and challenges for user experience research. However, the current relationship between UX research and AI remains underdeveloped, primarily characterised by instrumental applications of AI tools to enhance research efficiency. This article proposes a comprehensive framework for “human-centred AI research” that extends beyond tool-based efficiency gains. Drawing upon Frayling’s seminal taxonomy of design research — research into, for, and through design — this framework articulates three complementary dimensions: <strong>research into AI, research for AI, and research through AI.</strong> This tripartite model positions UX research at the intersection of traditional human-centred research and AI-focused inquiry, establishing a foundation for the discipline’s evolution in an AI-driven era.</p><h3>1. The Problem: An Underdeveloped Relationship</h3><p>The integration of artificial intelligence into user experience research currently suffers from a fundamental limitation: the relationship is predominantly conceived as unidirectional and instrumental. Most discourse surrounding AI and UX research centres on leveraging AI tools to augment research operations — automating transcription, accelerating analysis, or generating insights from synthetic users. Whilst these applications offer tangible efficiency enhancement, they represent an incomplete and potentially reductive vision of how UX research should evolve in response to AI technologies.</p><p>This narrow conceptualisation creates several critical gaps.</p><p>First, it fails to position UX research as a discipline capable of generating knowledge about AI systems themselves, thereby<strong> missing opportunities </strong>to contribute meaningfully to AI development and deployment.</p><p>Second, it neglects the substantial domain of human-AI interaction as a distinct research territory requiring <strong>specialised methodological approaches</strong>.</p><p>Third, and perhaps most consequentially, it offers <strong>no clear pathway for UX researchers</strong> to develop distinctive expertise that transcends operational efficiency — a limitation that constrains both the strategic value of the discipline and the career trajectories of its practitioners.</p><p>The current state stands in contrast to what human-centred AI research could become: <strong>a discipline situated at the convergence of AI research — with its focus on technical performance and engineering optimisation — and traditional human-centred research — with its emphasis on user experience, behaviour, and needs.</strong></p><p>Neither purely technical nor purely experiential, human-centred AI research occupies a unique position that demands its own theoretical frameworks, methodological approaches, and knowledge domains.</p><p>Without a comprehensive framework to guide this evolution, UX research risks becoming marginalised in AI development processes, relegated to either late-stage validation activities or operational support functions. Thus, the discipline requires a conceptual model that articulates multiple dimensions of engagement with AI — one that positions researchers as knowledge producers across technical, experiential, and operational domains.</p><h3>2. Frayling’s Framework: Research Into, For, and Through Design</h3><p>Christopher Frayling’s 1993 paper “<a href="https://researchonline.rca.ac.uk/384/9/frayling_research_in_art_and_design_1993_OCR.pdf">Research in Art and Design</a>” established a foundational taxonomy that has profoundly influenced how creative and design disciplines conceptualise research. Developed during a period when design struggled for recognition as a legitimate research domain within academia, Frayling’s framework distinguished three distinct modalities of research practice: research into design, research for design, and research through design.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*x8MdVKxoxx8ar4zuy8sDfA.png" /><figcaption>Christopher Frayling’s 1993 paper</figcaption></figure><p><strong>Research into design</strong> encompasses traditional scholarly inquiry where design serves as the object of study. This includes historical analysis, theoretical investigation, and empirical studies examining design processes, artefacts, and impacts. Such research generates propositional knowledge <strong>about</strong> design as a phenomenon.</p><p><strong>Research for design</strong> involves investigative work conducted to inform and enable design practice. This includes materials research, user studies, market analysis, and other inquiries undertaken to provide inputs <strong>for</strong> design decisions. The primary output is not necessarily generalised knowledge but rather insights that support specific design outcomes.</p><p><strong>Research through design</strong> represents the most distinctive contribution of Frayling’s taxonomy — the notion that the design process itself constitutes a legitimate mode of inquiry. <strong>Through</strong> iterative making, testing, and refinement, designers generate embodied knowledge that may not be fully articulable in propositional terms. The designed artefact becomes both the means and the record of research activity.</p><p>Frayling’s framework emerged from a specific historical imperative: establishing design as an intellectually rigorous discipline within research-intensive universities. By articulating multiple valid forms of research engagement, the taxonomy validated design’s epistemological contributions whilst acknowledging the discipline’s distinctive practices and knowledge forms.</p><p>The framework remains valuable because it recognises an important principle: any discipline can — and should — conduct research in multiple ways at once. The framework does not dictate one “correct” way for a field to engage with research. Instead, it validates different approaches that each serve distinct purposes and produce different forms of knowledge. This flexibility is especially useful for new disciplines trying to establish themselves, or for established disciplines — like UX research — that are grappling with transformative technologies like AI.</p><p>Applied to the current challenge facing UX research, Frayling’s framework implies that a mature relationship with AI <strong>cannot be unidimensional</strong>. Just as design research encompasses studying design, conducting research to support design practice, and using design as a research method, UX research must develop multiple modes of engagement with artificial intelligence. This recognition provides the conceptual foundation for the human-centred AI research framework proposed in this article.</p><h3>3. The Human-Centred AI Research Framework</h3><p>Building upon Frayling’s tripartite structure, this article proposes a framework comprising three complementary dimensions: research into AI, research for AI, and research through AI. Together, these dimensions define human-centred AI research as a distinct domain positioned at the intersection of technical AI research and traditional human-centred inquiry.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TCWgfi96Ejm3qyox4tmnzA.png" /><figcaption>The Human-Centred AI Research Framework</figcaption></figure><h3>3.1 Research Into AI: Understanding Technology and Trends</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Lmd7u01x5do_jDgunHoP6g.png" /></figure><p>Research into AI establishes UX researchers as producers of knowledge <strong>about</strong> artificial intelligence technologies themselves. This dimension positions researchers not merely as users or observers of AI but as scholars who generate systematic understanding of AI phenomena relevant to human experience and organisational deployment.</p><p>Typical research questions in this dimension include:</p><ul><li><em>How have large language models’ capabilities evolved from 2023 to 2025 in terms of multimodal understanding?</em></li><li><em>What are the current technical limitations of AI-generated images in terms of brand consistency and accuracy?</em></li><li><em>How do fine-tuning approaches (LoRA, full fine-tuning, prompt tuning) impact model behaviour and output quality for enterprise use cases?</em></li><li><em>How do different attention mechanisms and context window sizes affect AI models’ ability to maintain coherence in long-form interactions?</em></li></ul><p>Methodologically, research into AI draws upon approaches including systematic literature reviews, technology landscape analysis, comparative benchmarking of AI models, and longitudinal tracking of AI capabilities and limitations. Unlike purely technical AI research that directly contributes to algorithmic performance, model training, or engineering optimisation, research into AI maintains a <strong>human-centred lens</strong> — examining how AI technologies function, evolve, and differ in ways consequential for users, organisations, and society.</p><p>For example, research into AI might involve comparative analysis of large language models’ capabilities in specific domains relevant to an organisation’s users, investigation of how AI systems’ behaviours shift across versions or training approaches, or synthesis of emerging AI paradigms and their implications for interaction design.</p><p>This type of research generates knowledge that is simultaneously technically grounded AND human-centred — informed by understanding of AI systems whilst oriented towards implications for human experience and use.</p><p>The strategic value of “research into AI” lies in establishing UX researchers as credible voices on AI technology itself, not solely on user reactions to predetermined systems. This positions the discipline to contribute earlier in AI development and deployment processes, informing fundamental decisions about technology selection, customisation, and application rather than merely validating completed systems.</p><h3>3.2 Research For AI: Understanding Human-AI Relationships and Interactions</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1gfcvsiqrOvx-mEccb1Upw.png" /></figure><p>Research for AI encompasses empirical investigation of how humans <strong>experience, interact with, and make sense of AI products</strong>. This dimension represents perhaps the most intuitive extension of traditional UX research into the AI domain, yet it demands significant methodological evolution beyond conventional user research approaches.</p><p>Typical research questions in this dimension are very similar to conventional UX research questions, which mainly focus on<strong> users’ experiences</strong>, such as:</p><ul><li><em>What factors influence whether users trust AI-generated recommendations in healthcare contexts?</em></li><li><em>How does prolonged use of AI assistants affect users’ problem-solving skills and confidence?</em></li><li><em>What interaction patterns emerge when users collaborate with AI in creative tasks versus analytical tasks?</em></li><li><em>How do users calibrate their trust after experiencing AI errors or hallucinations?</em></li></ul><p>The research agenda within this dimension includes understanding users’ mental models of using AI systems, identifying usability challenges unique to AI-mediated interactions, examining trust formation and calibration, investigating how users detect and respond to AI errors or limitations, and analysing how AI integration affects task performance and user satisfaction.</p><p>“Research for AI” requires adapting established UX research techniques whilst developing new approaches suited to AI’s distinctive characteristics — including its probabilistic nature, opacity, explainability, and capacity for unexpected behaviours.</p><p>Critically, research for AI differs from conventional product research in its focus on the human-AI relationship as the primary unit of analysis. Rather than treating AI as simply another interface element, this type of research recognises AI systems as distinct interaction partners whose characteristics — learning capabilities, probabilistic behaviours, potential biases — fundamentally shape the nature of user experience.</p><p>The knowledge generated through research for AI directly informs the design of AI-powered products and experiences. By providing empirical evidence about how users interact with and experience AI systems, this type of research is critical because it helps AI product decisions grounded in actual human needs, capabilities, and contexts rather than purely technical considerations.</p><h3>3.3 Research Through AI: Enhancing Research Operations and Capabilities</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wUX5uYmApYNx3I7QZLTPdw.png" /></figure><p>“Research through AI” involves the strategic adoption of AI tools and techniques to augment research processes, enhance analytical or creative capabilities, and expand the scope or scale of research activities. This dimension acknowledges that AI technologies constitute powerful instruments for conducting research itself, whilst emphasising intentionality and criticality in their application.</p><p>Applications of research through AI span the research lifecycle: AI-assisted participant recruitment and screening, automated user interview transcription and summarisation, AI-augmented qualitative analysis, synthesis of large-scale unstructured quantitative data, pattern detection across diverse data sources, and generation of research artefacts such as personas, reports and journey maps.</p><p>Beyond operational efficiency, AI tools can enable research approaches previously impractical at scale — such as analysing thousands of customer support conversations or conducting multilingual research across geographic regions.</p><p>However, research through AI demands more than mere tool adoption. It requires developing critical capabilities of the teams who use the AI tools, including: understanding AI tools’ capabilities and limitations, evaluating output quality and potential biases, determining appropriate human oversight and validation processes, integrating AI-generated insights with human interpretation, and maintaining ethical standards when using AI in research contexts.</p><p>The strategic value of research through AI extends beyond time savings. When deployed thoughtfully, AI augmentation can enable organisations to increase coverage (reaching more users or use cases), improve consistency (reducing analyst variability in interpretation), enhance accessibility (supporting multilingual or multi-modal research), and elevate focus (allowing researchers to concentrate on higher-order synthesis rather than mechanical tasks).</p><p>Importantly, research through AI positions UX researchers as <strong>sophisticated users of AI technologies</strong> — developing practical expertise that both enhances their research efficiency and deepens their understanding of AI tools’ strengths and limitations from a user perspective.</p><h3>3.4 Integration and Synergy</h3><p>Whilst articulated as distinct dimensions, research into, for, and through AI function synergistically.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CSsmj0eBc6rCf_54qgTwKg.png" /></figure><ol><li>Conducting “research through AI” using various tools generates experiential knowledge about AI capabilities and limitations that informs “research into AI”.</li><li>Understanding AI technologies by “research into AI” provides context essential for designing rigorous “research for AI” examining human-AI interaction.</li><li>Findings from “research for AI” regarding how users experience AI systems inform more effective deployment of “research through AI tools”.</li></ol><p>This integration positions human-centred AI research as a cohesive domain rather than a collection of disconnected activities. Researchers operating across all three dimensions develop comprehensive expertise that spans technical understanding, experiential insight, and methodological capability — a combination that establishes UX research as essential to effective AI development and deployment.</p><h3>4. Building Human-Centred AI Research Capability</h3><p>Realising the potential of human-centred AI research requires deliberate capability development at individual, team, and organisational levels. This section articulates the vision, requisite competencies, and organisational conditions necessary for implementing the proposed framework.</p><h3>4.1 Vision: Positioning UX Research for the AI Era</h3><p>The vision underlying human-centred AI research positions the discipline as an essential bridge between technical AI development and successful human-AI integration.</p><p><strong>Rather than being consumers of AI technologies developed elsewhere or validators of completed systems, UX researchers become active participants in shaping how AI manifests in the experiences of products and services.</strong></p><p>This vision recognises several key premises.</p><p>First, effective research in AI products development requires deep understanding of both technical capabilities and human needs — a convergence that defines the human-centred AI research domain.</p><p>Second, the complexity and novelty of AI systems create knowledge gaps that cannot be adequately addressed by either technical AI research or traditional UX research alone.</p><p>Third, organisations that develop strong human-centred AI research capabilities will achieve more successful AI integration, better user outcomes, and sustainable competitive advantage.</p><p>For individual researchers, this vision offers an aspirational identity:</p><blockquote><strong>becoming specialists in the human dimensions of AI whilst developing sufficient technical literacy to engage credibly with AI development processes.</strong></blockquote><p>For the UX research discipline, it articulates a pathway towards greater strategic influence and intellectual leadership in an increasingly AI-mediated world.</p><h3>4.2 Individual Competencies: Knowledge and Skills</h3><p>Practising human-centred AI research effectively requires developing competencies across several domains, though individual researchers may emphasise different dimensions based on organisational context and personal interests.</p><p><strong>Technical AI literacy</strong> forms the foundation, encompassing understanding of how major AI paradigms (machine learning, deep learning, large language models) function at a conceptual level, familiarity with AI development processes and terminology, awareness of current AI capabilities and limitations across domains, and knowledge of common AI failure modes and biases. This does not require engineering expertise but demands sufficient understanding to engage meaningfully with technical stakeholders and evaluate AI systems critically.</p><p><strong>Human-AI interaction expertise</strong> represents a specialised application of UX research skills to AI-mediated contexts, including understanding distinctive characteristics of AI as an interaction partner, methods for evaluating AI-assisted or AI-driven experiences, frameworks for analysing human-AI collaboration, and approaches to studying emergent behaviours in AI systems. This expertise bridges traditional UX research and AI-specific considerations.</p><p><strong>AI-augmented research workflow</strong> encompasses practical skills in deploying AI tools for research purposes whilst maintaining quality and rigour, including selecting appropriate AI tools for research tasks, evaluating AI-generated outputs critically, designing effective human-AI workflows in research processes, and understanding ethical considerations in using AI for research.</p><p><strong>Synthesis and translation capabilities</strong> enable researchers to integrate insights across technical and experiential domains, translating between technical and user-centred perspectives, identifying implications of AI technical characteristics for user experience, and communicating findings effectively to diverse stakeholders including engineers, designers, and executives.</p><p>Organisations can support competency development through structured learning programmes, cross-functional collaboration opportunities, attendance at both AI-focused and UX research conferences, and creation of communities of practice where researchers share learnings about AI-related research.</p><h3>4.3 Organisational Embedding: Strategic Value and Integration</h3><p>For human-centred AI research to fulfil its potential, organisations must recognise and structure its strategic value beyond operational efficiency gains.</p><p><strong>Strategic positioning</strong> involves integrating human-centred AI research into AI governance frameworks. This strategic positioning enables research teams to influence AI decisions from the outset — from selecting and customising technologies, to shaping responsible AI practices, to creating competitive advantage through better AI integration that serves user needs.</p><p><strong>Organisational structures</strong> that enable effective human-centred AI research include cross-functional teams combining UX researchers with AI engineers and product managers, dedicated human-centred AI research roles or specialisations, clear pathways for research findings to influence AI development and adoption decisions, and mechanisms for research to contribute to organisational AI literacy and capability.</p><p><strong>Value articulation</strong> requires demonstrating concrete contributions across multiple dimensions: reducing AI implementation risks through early identification of usability or experience challenges, accelerating AI adoption by designing more intuitive and trustworthy AI-mediated experiences, informing build-versus-buy decisions through comparative AI system research, and contributing to responsible AI practices through systematic investigation of AI impacts on users.</p><p>Organisations that successfully embed human-centred AI research capabilities position themselves to navigate AI integration more effectively — <strong>moving beyond reactive responses to AI technologies towards proactive shaping of how AI enhances rather than complicates human experiences.</strong></p><h3>4.4 Evolution Pathway: From Current State to Human-Centred AI Research</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yZHKAsqkoCL3JKtiLY8luQ.png" /></figure><p>Transitioning from current UX research practices to comprehensive human-centred AI research capabilities occurs through progressive expansion rather than wholesale transformation.</p><p>Initial steps often emphasise <strong>research through AI</strong> — adopting AI tools for specific research tasks whilst developing critical evaluation capabilities. This builds practical AI experience whilst generating immediate value through efficiency improvement.</p><p>Parallel development of <strong>research for AI</strong> capabilities involves adapting and extending existing UX research methods to study AI-mediated interactions. This typically begins with evaluating specific AI features within products, then expands to designing comprehensive human-AI interactions across entire product experiences.</p><p><strong>Research into AI</strong> capabilities typically develop later, building upon experiential knowledge gained through research for and through AI. As researchers accumulate understanding of various AI systems and their characteristics, they become positioned to conduct more systematic technology analysis and trend research.</p><p>Organisations at different maturity levels may emphasise different dimensions initially, but comprehensive capability requires eventual development across all three areas. <strong>The framework itself provides a roadmap for assessing current capabilities and identifying development priorities.</strong></p><h3>Conclusion</h3><p>The framework for human-centred AI research proposed in this article offers a pathway for UX research to evolve beyond instrumentalist applications of AI tools towards a comprehensive engagement with artificial intelligence across multiple dimensions. By distinguishing research into AI, research for AI, and research through AI as complementary and mutually reinforcing activities, the framework positions UX research at a critical intersection: between technical AI development and human-centred inquiry.</p><p>This positioning addresses the inadequacies of current relationships between UX research and AI whilst establishing a foundation for the discipline’s continued relevance and influence. Rather than being diminished by AI technologies or relegated to operational support functions, evolving UX research into human-centred AI research becomes essential to supporting that AI integration serves users’ needs and values.</p><p>As AI technologies continue to evolve and permeate products and services, the need for rigorous, systematic, human-centred investigation of AI phenomena will only intensify.</p><p>Organisations and individuals that develop strong human-centred AI research capabilities can do more than adapt to an AI-mediated future — they can actively shape it so that artificial intelligence augments rather than undermines human flourishing.</p><p>Human-centred AI research represents NOT an abandonment of UX research’s foundational principles but their necessary evolution for an era in which human experience is increasingly mediated by, dependent upon, and shaped by artificial intelligence.</p><p>The framework articulated here provides both theoretical grounding and practical direction for this essential evolution.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8a3e550f2999" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[You wake up before your alarm, not because you feel fresh, but because your brain is already…]]></title>
            <link>https://medium.com/@weidanlee/you-wake-up-before-your-alarm-not-because-you-feel-fresh-but-because-your-brain-is-already-80946a9c2e25?source=rss-870e9ba832c8------2</link>
            <guid isPermaLink="false">https://medium.com/p/80946a9c2e25</guid>
            <category><![CDATA[corporate-culture]]></category>
            <category><![CDATA[philosophy]]></category>
            <category><![CDATA[burnout]]></category>
            <dc:creator><![CDATA[Weidan Li]]></dc:creator>
            <pubDate>Tue, 27 Jan 2026 00:19:28 GMT</pubDate>
            <atom:updated>2026-01-27T02:54:13.142Z</atom:updated>
            <content:encoded><![CDATA[<h3>Self-Surveillance and Burnout</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*YS6QrwMjZdr3gqh1.jpeg" /></figure><p>You wake up before your alarm, not because you feel fresh, but because your brain is already running a meeting without you. You lie still for a second and try to remember what day it is. It’s Wednesday. It’s usually the busiest day of the week. The answer comes with a list in your head: the report, the stakeholders meeting, the “quick sync” that never stays quick.</p><p>Thanks for reading Weidan’s Substack! Subscribe for free to receive new posts and support my work.</p><p>Your hand goes to your phone without you even thinking. You tell yourself you’ll check one thing, but minutes slip by. Soon your chest feels tight, like the day is already chasing you.</p><p>When you finally get to work, the building looks the same as always. Bright lights. Clean floors. The smell of coffee that almost helps. You tell yourself you can handle it. You’ve handled worse.</p><p>You sit down, open your laptop, and the screen fills with little demands that all sound polite but feel sharp. “Following up.” “Just checking.” “Can you share by EOD?” Nobody is yelling. That’s the strange part. Everything looks calm, so you wonder why you feel like you’re sinking.</p><p>The morning goes by in chopped pieces. You start one task, then stop to answer a message. You go back, then a meeting reminder pops up.</p><p>In the meeting, you listen to people trade updates that feel like they’re floating above the real work. You say a few words so you don’t look silent. You smile when someone makes a joke. Your body is there, but your attention keeps slipping away, like it’s tired of being pulled.</p><p>Lunch arrives, and you realise you haven’t stood up in hours.</p><p>You stepped into a restaurant near the office and took a seat at a table. You take a bite of food and taste almost nothing.</p><p>You’re halfway through lunch when your phone lights up with a Slack message. You tell yourself you’ll wait. But your thumb taps the screen anyway, and suddenly you’re back inside the work storm.</p><p>You think, <em>I just need one day off. One weekend that actually feels like a weekend.</em></p><p>But even when you do get time off, the work follows you in your head, like a song you can’t stop hearing.</p><p><strong>Burnout doesn’t always show up as a big crash.</strong> <strong>For you, it’s more like a slow fading.</strong></p><p>You still show up.</p><p>You still get things done.</p><p>People still say you’re reliable.</p><p>And yet you start noticing little changes that scare you. You forget small details. You feel annoyed at questions that used to be easy. You stare at a simple email, and nothing moves in your head — not because it’s hard, but because you feel empty.</p><p>At home, you sit down and try to relax, but your body doesn’t believe it’s safe. Your shoulders stay high. Your jaw stays tight. You keep thinking about what you missed in work, what you need to fix, what someone might ask you tomorrow. Even fun things feel far away, like they belong to another version of you. The hardest thought creeps in when you’re brushing your teeth or staring at the ceiling:</p><p><em>“Maybe I’m just not strong enough.”</em></p><p>But, my friend, <strong>burnout is not a weakness.</strong></p><p>It’s what happens when you live too long in “go” mode and never get to “stop.”</p><p>Your mind is built to handle stress in short waves. A hard week. A big deadline. A tense project that ends. Then you rest, and the wave passes.</p><p><strong>However, modern work often turns that wave into a whole ocean. </strong>The workday has no clear edge anymore. Messages can arrive anytime. New tasks appear the moment you finish the old ones. It’s like the tap never turns off, so you keep trying to empty the bathtub faster, and you blame yourself for getting tired.</p><p>There’s also a quiet rule in many offices that nobody says out loud:</p><p>You are supposed to be reachable.</p><p>You are supposed to be quick.</p><p>You are supposed to be “on.”</p><p>If you respond fast, you look helpful. If you slow down, you worry you’ll look lazy. If you say no, you worry you’ll be seen as difficult.</p><p>Over time, you start acting like you’re being watched, even if nobody is watching. You check your status. You measure your output. You keep proving you’re useful, because you think being useful is the same as being safe in your company.</p><p>Michel Foucault, a philosopher, spent his life studying this kind of invisible pressure.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*DKu0FBJ0wNA7KlIz.jpeg" /><figcaption>Michel Foucault</figcaption></figure><p>He wasn’t talking about power only as a boss shouting or a rulebook on the wall. He looked at how systems shape people until people start shaping themselves.</p><p>He pointed to a strange building idea: a circular prison called the <strong>Panopticon</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*USawlsR_yYmi0KFq.jpeg" /><figcaption>Jeremy Bentham’s Panopticon</figcaption></figure><p>Cells sit around the edge, and a guard tower sits in the center. The prisoners can’t see inside the tower, so they never know if someone is looking at them. After a while, they stop needing a guard to control them. They control themselves.</p><p>His idea can sound heavy, but it lands in a simple way: <strong>sometimes the strongest control is the one you learn to do to yourself.</strong> You don’t need someone to stand over you if you already feel guilty the moment you slow down. You don’t need someone to demand perfection if you’ve trained yourself to believe that mistakes make you less worthy.</p><p>When you see burnout through that lens, something important changes.</p><p><strong>You should stop thinking of it like a personal failure, and you should start treating it like a pattern you can interrupt.</strong></p><p>If work has trained you to always rush and stay “on,” then healing starts when you choose to slow down in small, real ways — without waiting for someone to tell you it’s okay, because nobody might tell you.</p><p>The first step is not a grand speech. It’s a quiet sentence you tell yourself that sounds plain and true:</p><p><em>I’m overloaded.</em></p><p>Not “I’m weak.”</p><p>Not “I’m failing.”</p><p>Just overloaded.</p><p>That one word matters because it removes the shame. Shame is like carrying an extra backpack on top of everything else, and you don’t need that weight. This is your life, and you, instead of shame, get to steer it. Shame doesn’t deserve your attention.</p><p>Once you name what’s happening, you can start making choices that protect your energy instead of spending it like it’s endless.</p><p><strong>Think of your energy like a battery that has to last the day.</strong> If you treat every email like an emergency, your battery drains by noon. If you treat every meeting like it deserves your full heart, you’ll have nothing left for the work that actually needs your brain.</p><p>So you begin the day by deciding what matters most. You pick a small number of real priorities, not because you don’t care about the rest, but because your attention is not infinite. The world will keep asking for more. You have to choose.</p><p><strong>Then you build boundaries</strong> that feel almost too small to count, because small boundaries are the ones you can keep.</p><p>Maybe you stop answering messages the moment they arrive and instead reply in blocks.</p><p>Maybe you protect one hour where you do one task without switching.</p><p>Maybe you stop agreeing to meetings that have no point and ask for a clear reason first.</p><p>At first, this will feel uncomfortable, like you’re breaking a secret rule, and that’s exactly the Self-Surveillance Foucault was pointing at. The discomfort does not mean you’re wrong. It means you’re changing a habit.</p><p><strong>You also learn a new kind of “yes.”</strong></p><p>Instead of saying yes like you’re opening a door to unlimited work, you shape it. You say yes with time, yes with trade-offs, yes with reality. “I can do that by Friday.” “If I take this on, what should I pause?” “I can help, but I’ll need support.” These sentences keep you kind without letting you become a sponge that soaks up everything.</p><p><strong>Rest</strong>, too, has to become real. Not the fake rest where you scroll until your eyes hurt, and then wonder why you still feel tired. Real rest is the kind that lowers the noise inside you. A slow walk. Read a book with no phone nearby. Sitting outside for ten minutes and letting your mind wander without feeding it new problems. It sounds small, almost silly, but small is the point. You are teaching your body a new lesson: it is allowed to turn off.</p><p><strong>If you want to work out, try boxing. I really recommend it. It’s the best kind of training I’ve ever done.</strong></p><p><strong>When you do intense brain work all day, your body still holds all that stress.</strong> Boxing helps you move it out. It sounds a little strange, but it works for a lot of people. After one hour, you feel lighter. Your energy comes back. The worry and anger and pressure don’t follow you home — you leave them on the heavy bag, or in the round with your sparring partner.</p><p>And if you can, you bring one clear and honest feeling into the workplace, said calmly to the right person. You don’t have to make it dramatic. You can talk like someone describing a schedule, not like someone asking to be rescued. “I’m at capacity. If we add this, something else will slip — can we choose what matters most?” When you speak like that, you stop being a silent container for impossible expectations. You become a person who helps the work stay sane.</p><p>Recovery won’t look like a movie scene where you suddenly feel perfect. It will look more ordinary than that.</p><p>You’ll notice that you can breathe again in the afternoon.</p><p>You’ll finish work and your head won’t keep spinning as hard.</p><p><strong>You’ll catch yourself paying attention to little things again</strong>, like a dog at a café leaning against its owner, and you’ll wonder, for no reason at all, how old it is.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*VH58i31lZFqLy-9U.png" /></figure><p>You’ll laugh and it will feel real. Not every day, but more days than before.</p><p>Burnout tries to convince you that you are a machine that is failing.</p><p><strong>But you are not a machine.</strong></p><p>You are a person who has been running too long without enough air, and you can start giving yourself air in ways that are simple and steady.</p><p>Thanks for reading Weidan’s Substack! Subscribe for free to receive new posts and support my work.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=80946a9c2e25" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Disappearance of Subjectivity in Human-Centred Design]]></title>
            <link>https://medium.com/@weidanlee/the-disappearance-of-subjectivity-in-human-centred-design-d7f59369ab4b?source=rss-870e9ba832c8------2</link>
            <guid isPermaLink="false">https://medium.com/p/d7f59369ab4b</guid>
            <category><![CDATA[human-centred-design]]></category>
            <category><![CDATA[research]]></category>
            <category><![CDATA[human-centric-design]]></category>
            <category><![CDATA[design]]></category>
            <dc:creator><![CDATA[Weidan Li]]></dc:creator>
            <pubDate>Tue, 22 Apr 2025 12:12:03 GMT</pubDate>
            <atom:updated>2025-04-22T12:12:03.702Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*avfrfwA1IK7GnHGav7EThQ.png" /></figure><p>At the end of <em>The Order of Things</em>, Michel Foucault said, “one can certainly wager that man would be erased, like a face drawn in sand at the edge of the sea.”</p><p>If we scrutinize human-centred design today, this sentiment deeply resonates. Despite the popularity of the term, human-centred design is largely a myth because true subjective experience cannot be fully grasped, especially within the purposeful, profit-driven context of business.</p><p>Human experience is vast and endlessly complex. It encompasses the full reality of an individual’s life — emotions, thoughts, culture, beliefs, worldviews, social connections, habits, cognition, self-awareness — everything that shapes how one engages with the world. Truly understanding another person in all their depth is nearly impossible in practice. That’s why conversations about human experience often rely on theoretical reflection — what we commonly call philosophy.</p><p>We might strive to empathize in design practice, but true empathy, in its absolute form, remains elusive. Understanding is inherently subjective and interpretive. It depends not only on the information provided but also on the interpreter’s biases, experiences, and assumptions.</p><p>Consider this: imagine I’m a user, and you are a person studying me on behalf of your company, even if I explicitly tell you how I think or feel, how do you know you are accurately interpreting my inner reality? How do you know my words match my genuine feelings or beliefs through my language? More importantly, how could you discern whether I even fully understand my own experiences?</p><p>To manage this complexity, commercial organizations simplify people into clearly defined categories, making their “experiences” easier to grasp. But this simplified representation doesn’t reflect true human experiences; it merely offers a convenient version. People are segmented into groups based on behaviours, attitudes, demographics, regions, professional backgrounds, family statuses, and similar criteria. These segments don’t naturally emerge from who people genuinely are; rather, they are shaped by how individuals interact with products or how effectively they contribute to business goals.</p><p>Above, I described how individual subjectivity is lost due to organizational needs. People are viewed through fragmented categories and collective groups, rather than as complete individuals. This reveals how experiences don’t naturally present themselves — they are reshaped through business segmentation. Now, zooming in on an individual within an invisible, extensive, and unavoidable network woven by thousands of commercial organizations, we see how a person is no longer viewed as a subjective being but as objective data.</p><p>Imagine a Saturday morning: someone goes for a run wearing Nike shoes like always, monitoring their workout with an Apple Watch. After exercising, they grab coffee from Starbucks, shower at home with a Dyson hairdryer, and later drive to lunch guided by Google Maps.</p><p>In this brief scenario, the person splits into multiple fragmented identities — an amateur athlete, a health-conscious individual, a habitual coffee drinker, a tech-savvy consumer, and an efficient navigator — each selectively identified and interpreted by different companies. None of these labels genuinely captures the fullness of who this person is; they represent only partial views. These labels are chosen specifically because they serve distinct commercial purposes.</p><p>Take Nike as an example: Suppose Nike identifies a market opportunity for 2025–2026, targeting frequent runners experiencing minor chronic injuries. Nike’s entire focus then shifts to understanding this individual solely as a runner — their stride, posture, foot impact, and injury patterns. Nike has no interest whatsoever in how this person drys their hair, drives their car, orders coffee, or navigates the city.</p><p>But stepping beyond this narrow business lens, is this individual merely one isolated identity (as a runner) to the exclusion of all others? Clearly not. Each identity coexists simultaneously, and none alone fully represents the complete human being.</p><p>This fragmentation is at the heart of exploitation. Each company picks and chooses the data points it needs, and accessible to the company, to drive growth, enhance profitability, or increase market share. Individuals are dissolved into metrics, becoming sources of conversion, adoption, usability, desirability, or engagement data. The personal and holistic experiences of individuals are lost. The experiences are transformed instead into fuel that powers business engines.</p><p>Internet companies demonstrate this clearly. Their fundamental business models revolve around measurable outcomes, such as conversions, active users, session durations, click-through rates, to name a few. Human experiences become abstracted and simplified into these indicators. Anything outside this narrow framework of measurable business success becomes invisible or irrelevant. Businesses do not aim to understand people fully; they aim to improve indicators tied directly to financial outcomes. The true depth of human experiences — implicit emotions, social bonds, cultural subtleties — falls outside this scope.</p><p>Now, let’s shift perspective from the “others” — customers and users — to ourselves. We, the designers, researchers, or business strategists, perceive ourselves as empathetic and thoughtful practitioners, actively reaching out to understand people’s preferences and behaviours. Yet, at the very same moment, we are those “others” too.</p><p>We use products like Uber, Netflix, Airbnb, and Google Maps in our daily lives. We too are distilled into data points, and we become fuel for other businesses’ growth funnels. We reflect and amplify the same indicators of success we track professionally — adoption rates, retention figures, satisfaction scores, and spending habits. This creates an ironic paradox. While we strive to understand users authentically in our professional roles, we simultaneously exist as fragmented data in someone else’s analysis. We become a part of the very system that fragments us. Ultimately, no one subjective experience exists, we all become “others”.</p><p>Companies may genuinely believe they are delving into authentic human experiences. However, what they’re capturing are not genuine human experiences but merely structured, categorized data viewed through the inevitable lens of profitability and growth. This lens is never removed because it fundamentally defines business priorities. Consequently, the data collected never aligns with true human experiences. It merely parallels the true experiences, providing an imagination of depth and empathy where none genuinely exists.</p><p>What, then, is the result of all this? The outcome is a widespread misconception: that human-centred design is based on human experiences. Instead, what we have is a carefully crafted illusion — an illusion maintained by businesses to make their products and services feel meaningful and human-centred. Yet, this human-centred narrative remains shallow, constructed solely from categorized insights driven by commercial goals.</p><p>The hard truth is that we cannot escape this dilemma. Our survival, success, and prosperity are inherently tied to business outcomes. We must continue wearing our business hats, using fragmented data to improve our products and drive profits. We participate in a system that simultaneously exploits everyone of us, transforming our lives and experiences into data, just as we do to others.</p><p>Ultimately, authentic human-centred design remains unattainable in the current business context. Human experience cannot and will not be fully understood when viewed through lenses shaped primarily by financial incentives and corporate interests. This doesn’t mean we abandon empathy or stop attempting to understand users deeply. It simply means we must acknowledge the inherent limitations and biases built into the practice.</p><p>My final words: honestly, it’s unrealistic to expect commercial organizations to genuinely prioritize humanity. Their survival depends on putting people at the center — not to understand them fully, but to extract commercial value. If human experience isn’t framed in ways that serve a profit-driven system, there’s no revenue; then there’s no bread and butter. Thankfully, I’m able to sit here writing this article with a roof over my head and food on the table because of that very system.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d7f59369ab4b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Some random tips on UX research]]></title>
            <link>https://medium.com/@weidanlee/some-random-tips-on-ux-research-e9af089a04e8?source=rss-870e9ba832c8------2</link>
            <guid isPermaLink="false">https://medium.com/p/e9af089a04e8</guid>
            <dc:creator><![CDATA[Weidan Li]]></dc:creator>
            <pubDate>Wed, 14 Dec 2022 02:41:21 GMT</pubDate>
            <atom:updated>2022-12-14T02:41:21.006Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*9qRKmfqW3eI_JZEC.jpeg" /></figure><blockquote><em>“Research is a craft… Research isn’t like a craft. It is a craft. If you know what people have to go through to become skilled carpenters or makers of clothes, you have some idea of what it takes to learn the skills for doing research. It takes practice, practice, and more practice.”</em></blockquote><blockquote><em>Research Methods in Anthropology by H. Russell Bernard</em></blockquote><p>Maybe your team doesn’t have a design researcher, so you have to do user interviews yourself; or the team has design researchers, but you are keen to do research on your own, either way, some of these tips might help.</p><h3>Write up research plan and interview guide</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*R4CJUnKMWr0P7DLH.jpeg" /></figure><p><strong>· Make each individual research question as concise as possible.</strong> If necessary, break it down to several questions. Clear and short sentences save audiences a great amount of time for reading. Compared to a complicated sentence structure, there will be a shorter journey between people start making sense of the sentence and catching the meanings of it. A great book for writing: <a href="https://www.amazon.com/Elements-Style-Fourth-William-Strunk/dp/020530902X/ref=pd_sbs_1/141-5763267-3406536?pd_rd_w=dRuro&amp;pf_rd_p=3676f086-9496-4fd7-8490-77cf7f43f846&amp;pf_rd_r=P95D14ZPS9H9QVHFWYVR&amp;pd_rd_r=2ae533df-f77c-49cf-898f-04218ad760ab&amp;pd_rd_wg=zR7Vd&amp;pd_rd_i=020530902X&amp;psc=1">The elements of style</a>. It saved my thesis writing.</p><p><strong>· Always check your interview questions against the research questions.</strong> Interview questions are the specific questions you ask participants in interview sessions. Research questions are the high level questions you’d find answers for to represent the goal of the research. Make sure the interview questions have reflected all the research questions.</p><p><strong>· Imagine yourself as the participant.</strong> Run through the script with your design a few times before the study. Make sure there are no leading questions such as using words that appear in the UI. For example, asking “How would you proceed …” while the button says PROCEED. That is a leading question. Thus, rephrase it.</p><h3>Conduct usability testing or interviews</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*uZ1pHXDob-FVyGCk.jpeg" /></figure><p><strong>· Don’t panic even if you are a social introvert</strong>. Treat the participant as someone who tells you stories, but will never show up again in your whole life. Focus on the stories, not the participant. Prepare some chit chat before meeting them — weather, food, pandemic lockdown, vaccination etc. I usually take the first 2 minutes of interviews to make both of us relaxed before starting the introduction.</p><p><strong>· Be nice but also keep a distance with participants — I mean emotionally, not 1.5m social distance.</strong> Assume that, on a scale of 1–10, 1 as “very cold” attitude (following and reading through a questionnaire like a robot and no interaction at all), and 10 as “very friendly” attitude (making very interactive conversations with humor and empathy to show respect), then in a usability testing, it’s appropriate to keep your attitude at 4–5. You just need to maintain the basic politeness (for example, some smile), but let the participants lead their conversation after you ask the question. Minimize your verbal response (such as “wow nice!”) and body language (such as frequent nodding).</p><p><strong>· Tasks and scenarios are key in usability.</strong> Read the tasks a few times to make sure they’ve covered research questions or designed features you’d validate. Every word in a task should be useful to inform participants. No one word is useless in tasks or scenarios. Make them as clear as possible. This might be a useful <a href="https://www.nngroup.com/articles/qual-usability-testing-study-guide/?utm_source=userweekly.com&amp;utm_medium=referral&amp;utm_campaign=userweekly.com">resource</a> from Nielsen Norman Group.</p><p><strong>· “Prioritization” is always a fun and useful activity.</strong> You could ask the participants to rank and prioritize items on a variety of topics, such as information on the website, tools they use, or people in their network. By asking why people prioritize that way, you could uncover the mental model, the value perception, and more importantly, the reasons behind.</p><h3>Conduct unmoderated user testing</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*QKENpTbmyZj9ZxNa.jpeg" /></figure><p><strong>· Include a screening question about the candidates’ experience with unmoderated testing</strong>, i.e. how often do they attend unmoderated testing. You don’t want to recruit many “professional testers”. They make your design look perfect and give you amazing improvement suggestions, but they aren’t the average users. You’d better recruit people with different unmoderated testing experiences to minimize bias.</p><p><strong>· Don’t send out the testing invitations to all the participants at the same time.</strong> Sometimes you don’t know where the design or questions could go wrong, even if you’ve run through the testing script several times with the team. Send the invitation to two or three participants first to trial your testing. If everything works fine, then send the invitations out to the rest.</p><p><strong>· Treat participants as kids</strong>. <strong>Keep reminding them about thinking aloud</strong>. I put <em>[Speak to your computer or microphone]</em> at the end of every question on <em>Lookback</em> because people could just click through the prototype without talking through their experience. If it’s a scrollable page, say it’s scrollable in the beginning. You can definitely replace the invalid participants, but it takes time.</p><p><strong>· One question at a time</strong>. If you ask “What is the most important … ? Why is it important?”, separate it into two questions. Participants often don’t realize the “why” question following the main question. If they missed it, you don’t have a second chance to ask because it’s unmoderated. Therefore, when setting up questions in remote research tools such as Lookback, make sure participants will only answer one question before clicking on the ‘complete task’ button to proceed to the next question.</p><h3>Data analysis (qualitative)</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*3yvIvHzVxQi_ZPWt.jpeg" /></figure><p><strong>· Pay more attention to the intensity of an attitude, instead of only the number of occurrences of an opinion.</strong> For example, if 8/10 participants said the design was good, but 2 expressed intense frustration, then it’s worth digging into the reasons behind the frustration. Watch out for tones and keywords in people’s narrative.</p><p><strong>· Don’t create tags consisting of only general keywords.</strong> Use brief phrases or sentences to show the <em>meaning</em> of the chunk of the narrative. For example, a participant talks about a frustrating pick-up experience in a supermarket because staff always take a long time to find their purchased product, don’t just tag <em>“bad pick-up experience”</em>. A better tag might be like <em>“frustrating — long wait when picking things up”</em>. A principle is, when you cover the transcripts, you can still tell what the person has expressed by reading through the tags, so that you don’t have to go back to reading the raw data (transcripts) over and over again.</p><p><strong>· When you are analysing data, the aha! moments come and go quickly.</strong> It’s very easy to forget important findings if you start writing up insights only after completing the time-consuming tagging process. You could write down your thoughts and insights while analyzing data — it doesn’t matter if they are messy at the moment. They help you grab the moments when you are in the zone and feel you’ve found something important.</p><h3>Write a report</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*1-dmKWqAChIuUv5A.jpeg" /></figure><p><strong>· Pick up the most comprehensible, interesting or punchy quotes from the transcript.</strong> A good quote from research participants tells a much more intriguing story than your description. If you use multiple quotes from different participants, take advantage of the diversity. You could either show different quotes that represent the same topic to strengthen your point of view (<em>for example, a few participants talked about the navigation problem using different styles of words</em>), or use different quotes with different focuses to reveal the multiple facets of a topic (<em>for example, some participants reckoned the navigation problem is caused by the website color scheme, but some others thought it was due to poor information structure</em>).</p><p><strong>· Same with writing up research questions, insights are better to be concise too.</strong> A short sentence with simple structure is punchier than a complicated sentence. Also, remove unnecessary adverbs that describe intensity or level. Significantly, tremendously, greatly, mildly… different audiences assign different meanings to these words.</p><p><strong>· When writing an insight, ask yourself — “so what?”.</strong> An insight is more than a statement of a phenomenon or fact. It elaborates relationships such as “cause and effect” or “behaviour and impact”. Don’t just show what people did — they did that, so what? What are the consequences of the behaviour? These connectivities need to be in an insight.</p><p><strong>· Some insights may show the most severe usability problems; if so, don’t put them in the very beginning of your report.</strong> It can easily hurt some stakeholders’ feelings because they didn’t expect the problems, or the problems disprove their assumptions and caused awkwardness. If you’d engage people in your presentation, put the most important and impactful insights in the middle, so that people would not be bored after the amazing first 5 to 6 minutes.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e9af089a04e8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why UX research democratization is an educational practice]]></title>
            <link>https://uxdesign.cc/user-research-democratization-is-an-educational-practice-2ea97c36a980?source=rss-870e9ba832c8------2</link>
            <guid isPermaLink="false">https://medium.com/p/2ea97c36a980</guid>
            <category><![CDATA[research-democratization]]></category>
            <category><![CDATA[mentoring]]></category>
            <category><![CDATA[user-reserach]]></category>
            <category><![CDATA[ux]]></category>
            <category><![CDATA[ux-research]]></category>
            <dc:creator><![CDATA[Weidan Li]]></dc:creator>
            <pubDate>Sun, 27 Nov 2022 10:19:07 GMT</pubDate>
            <atom:updated>2022-12-01T10:48:32.059Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>UX research democratization is an educational practice</strong></h3><h4>— and five learning theories that inspire research education</h4><figure><img alt="should be user research democratized? the debate has never stopped" src="https://cdn-images-1.medium.com/max/1024/1*UdY-0lqcJeqNUL3d0BhL5w.png" /></figure><p>Should be user research democratized?</p><p>The debate has never stopped.</p><p>Some opinions support it because it scales the value of research in an organization. Some are against it because it dilutes the profession. <a href="https://dovetailapp.com/outlier/jared-spool-democratization-research/">Jared Spool </a>thought both were wrong because they treated research as the end goals rather than the process of understanding problems.</p><p>We will not have a winner in this debate because when people are arguing over a trend, the trend is already happening. It will not cease and it is inevitable whether we are against it or not. Is user research an independent profession or just a tool? Maybe the debate is not important.</p><p>What’s important is the signal that there are more and more non-research roles such as product designers and product managers are keen to do user research themselves and learn from it.</p><p>I regard this a positive signal because it’s an opportunity to decentralize the authoritative power and share the right of generating user research knowledge among those who want to, without the limitation of roles.</p><p>If user research is not democratized in an organization, only the research team generates insights and manages the end-to-end user research activities.</p><p>Imagine this: I am a user researcher and you are not. I own every piece of knowledge from research and you don’t. At the end of the day, you have to listen to me telling you how I did the research.</p><p>Even if I bring you in as a stakeholder and engage you in interview sessions and data analysis, it was my strategy to get you informed to realize my research values, not to help you truly understand and practice research on your own. The user research process didn’t help you with it because I didn’t intend to. After all, I am a user researcher and you are not.</p><p>It leaves a question here: in a subject that relies on individual interpretation, how could we overcome the power dynamics in knowledge generation? In such a subject, if there is no one universal law to judge the research methods and knowledge to be right or wrong, how could we make sure the knowledge that one single institute has created is not biased within their own interpretation?</p><p>This is not a question I raised just right now, but a problem in the power-knowledge relationship that Foucault has pondered in his entire life (Rouse, 1994).</p><p>It is interesting that authors argue about the risks and benefits of democratizing research, and suggest strategies to facilitate the process, such as better tooling and more efficient research ops. I would rather focus on educating and coaching the learner in user research democratization to improve research quality.</p><p>Eventually we would like to see more user research done by the broader teams, but at the same time, the process of research democratization should help those who conduct research to advance their research mindsets and skills. We want to see the improvement of research quality that scales user research values in the organization.</p><h4>2. User research is a craft</h4><figure><img alt="user research is a craft" src="https://cdn-images-1.medium.com/max/1024/0*7RG22Sq5FD_-xVMR" /></figure><p>A craft is composed of technical skills and foundational mindsets. Skills in user research are so diverse that include interviewing skills, inductive skills, deductive skills, reading skills, listening skills, and so on.</p><p>A learner can get familiar with skills by doing research themselves. However, having the skills only won’t make a good user researcher. The learner needs the right mindsets to approach the research problems in a more appropriate way — I didn’t say the correct way. The user research mindset development is just as hard as skills. It takes years.</p><p>An analogy I’ve always been using is boxing. Maybe you can throw a sharp jab, a powerful straight, and a vicious body shot on the heavy bag too. Those are skills. The mindset is that instinct helps you analyze and predict your opponent as a whole moving target.</p><p>With the right mindset, your opponent is almost throwing the punches that you expect them to throw. Not because they really do so, but because you read them by feeling their movement and gestures so that you could react actively. The mindset of “reading your opponent” transforms into the skills of defending and counterpunching. A boxing coach is the one who corrects their boxers skills and nurtures their mindsets as an intimate outsider.</p><p>In user research democratization, the research coach should work similarly as a boxing coach. They give feedback to the learner, put them on the right track for user research, and provide training sessions for skill improvement, and so forth. The research coach doesn’t have to be alway bugging the learner, but they need to show up when the learner needs them for answers. The coach needs to have enough knowledge of the end-to-end design research process and when to use which methods. Also they need to be a patient instructor so that they could get along with the team over a long period.</p><p>The requirement list for a user research coach can go on and on, but it is also crucial for the coach to figure out how they might provide appropriate mentorship to help the learner achieve the goals.</p><p>There are five primary learning theories that research coaches can refer to. These theoretical perspectives were born to respond to the different stances among each other in history. But there are no right or wrong, worse or better perspectives. The research coach can even apply all the five learning theories in their work at the same time.</p><h4>3. Behaviorism learning theory and behavior reinfocement</h4><p>Learning is based on a system of routines that help students memorize the information in those routines. In order to put the information in students’ memory, it is important for the teachers to give direct feedback on the behaviors. If students did a good job, they should receive positive feedback and encouragement. The purpose is to reinforce the recognition of the correct behaviors in a rewarding manner. If the students did things incorrect, there should also be instant feedback to correct the behavior.</p><p>In user research, the coach needs to give feedback at the right timing to the learner. There are two scenarios that can be the best timing to give feedback. One is when the learner is planning the research activities yet not implementing. The other is when the person has done the activities and is reflecting how well they’ve done it.</p><p>The coach might want to avoid giving direct feedback while during research, such as between interview sessions, or after a survey has launched. These are not good timings because correcting the behavior while people are doing the tasks can lead to stress and frustration.</p><p>Repetition and reinforcement are important in behaviorism learning. The coach should regularly emphasize on the correct behaviors and highlight the incorrect ones for improvement. For example, if the person wrote a good interview question in the discussion guide, the coach should give positive feedback not only once, but at least a few times in the future to imply the correctness till the person deeply embeds the behavior into the research mindset.</p><h4>4. Cognitivism learning theory and learners’ initiative</h4><p>To respond to behaviorism, cognitivism was born in the 1950s. Behaviorism learning theory focuses on the external force (i.e. teachers’ feedback) to change behavior, in comparison, cognitivism learning process shows the importance of internal thinking processes. The learner is not considered a pre-designed object to be changed, but a human mind that can absorb cognitive materials, digest information and generate knowledge on their own.</p><p>Cognitive learning process emphasizes that the learner can take the initiative in understanding learning materials such as texts and instructions. The learner should not just repeat their behaviors for the best outcome of learning. More importantly, they need to make sense of the meaning behind the learning materials and reflect on what they’ve learned.</p><p>Cognitive learning theory tells us that the user research learners can learn actively and reflectively. Firstly, there should be some instructions on user research for the learners to read and watch by themselves. It could be texts about how to run user interviews, presentation recordings of how to analyze survey data, or video tutorials on writing up a good research plan. The key is to offer the learner the cognitive stimulus, which is the learning materials, and give them the time to digest.</p><p>Secondly, the research coach lets the learner run their own research under guidance, then reflect on what they have done and how well they think they’ve done. It is important to notice that, in the cognitive learning approach, the coach is better to give feedback after the learner shares their feelings and thoughts.</p><p>For example, in a 1 on 1 chat between the coach and the learner, the coach could let them talk about feelings on the recent unmoderated design testing, ask about the difficulties and barriers they had, and how they want to improve next time. Based on these cognitive reactions towards the specific research, the coach could give positive or negative feedback on certain aspects. Notice the timing of giving feedback is to ensure the coach’s perspectives will not pressure and lead the learner’s thoughts in the conversation.</p><h4>5. Constructivism learning theory and the former knowledge</h4><figure><img alt="Constructivism learning" src="https://cdn-images-1.medium.com/max/1024/0*dBkSIit9m2tKc-d-" /></figure><p>Constructivism is related to the psychologist Jean Piaget’s<a href="https://www.simplypsychology.org/piaget.html"> theory of cognitive development</a>. <a href="https://www.wgu.edu/blog/what-constructivism2005.html#close">Constructivism</a> holds the belief that a learner is not an empty vessel or passive recipient of information. They construct new skills based on previous experience and knowledge.</p><p>This learning concept emphasizes that there is a strong and constant connection between what the learner has already known and what they want to learn next, as well as how they will learn it. Current knowledge acquisition is always built on the learner’s older knowledge.</p><p>In user research coaching, constructivism implies the individualization of the learning experience. Since each learner comes from a particular background with unique professional knowledge such as UX design, product management, marketing, business strategy or visual design, it is crucial to know the learner’s current understanding of user research, what research methods they’ve been using most often, and what research skills they want to learn in the future.</p><p>By understanding this background, the research coach can make sense of how the learner constructs the new knowledge, and how they might obtain the future knowledge based on the current practice.</p><p>For example, if the learner is learning usability testing, and they are from a statistical background who designed hundreds of quantitative surveys, the coach may need to talk them through more fundamental mindsets of interview facilitation, which is a very different way of collecting data than quantitative survey, but also shares the commonality of validating hypothesis.</p><p>The research coach should also respect the learner and take care of them for new knowledge construction if they have limited experience. In-depth user interviewing might be a useful research skill, but it could be intimidating and stressful to the learner if the coach asks them to directly jump in the interview facilitation. Constructivism reminds the research coach that always bear in mind where the learner’s current skills are from, which level they are on, and what their biggest interests are in future learning.</p><h4>6. Humansim learning theory and self-actualization</h4><p>Humanistic learning approaches are based on Maslow’s theoretical foundation (Maslow, 1954). It holds the moral stance that humans should be in the center of scientific inquiry. Scientific activities should help the individual person pursue a self-defined and meaningful life.</p><p>A fundamental notion in humanism philosophy, based on Kant, is that there is no one universal moral and ethical truth for all humans, but ethics arise in the humanistic way via which people seek for their individual good (Dierksmeier, 2011).</p><p>In user research coaching, humanism theory shares a commonality with constructivism, which is the learner-centric approach. Humanistic learning approaches propose that the learning process is a pathway to reach the learner’s self-actualization. Therefore, coaching user research is about helping the individual learner become the person they want to be.</p><p>The center of learning research is not the user research profession, but the individual momentum of how the learner wants to use research in their work, or what type of research skills they think will benefit them the most for a better career path.</p><p>There might be the difference between high-quality and low-quality user research outcomes, but there is no one universal law defining what user research is and how it should be used for all professionals. The way it is leveraged has to be aligned with the overarching actualization of learners’ personal values.</p><p>To realize the humanistic value in user research education, the learner’s managers can play an important role. Whenever the learner chats with their managers, they could share thoughts about what their goals are on learning and practicing user research. For example, is it more about meeting OKR requirements for the year, or are they interested in transferring into a more user-focussed role in the future, or just some ad hoc curiosity? No matter what it is, none of these goals are unreasonable. They are all precious motivations of learning user research that managers and the coach should respect.</p><p>Managers are usually the people that most easily listen to the learner’s thoughts and feelings. Therefore the user research coach should keep a close relationship with managers to ask about the learner’s goals and motivations.</p><h4>7. Connectivism learning theory and social learning</h4><figure><img alt="Connectivism learning" src="https://cdn-images-1.medium.com/max/1024/0*1DdS_QvegzCKqd4S" /></figure><p>Connectivism theory was developed in the technological and digital era. Specifically, connectivism is a social learning approach. It claims that the learner obtains knowledge not in a single channel of resource, but within the networked social connections. It is only through these personal networks that the learner can acquire new viewpoints and diversity of opinion to learn to make critical decisions (Duke et al, 2013).</p><p>A key concept is that the learner has or should nurture the ability to see connections between different information sources and to maintain that connection to facilitate continual learning.</p><p>In an organizational environment, it is not difficult for user research education to adopt the connectivism learning theory because an organization itself is a network where the learner can obtain new knowledge.</p><p>I consider connectivism learning approach can be practiced via two ways for user research upskilling.</p><p>The first method is about the connection between the learner and their peers. If the learner is not very experienced with user research, they are suggested to pair up with peers in user research projects, which we call a research buddy. The research buddy usually has more research experience than the learner. There could be a few research buddies for one learner, and they can be either designers or researchers. The learner will shadow their research buddies in a few research sessions until they feel comfortable to run their own research.</p><p>It is very important for the research buddies to give positive or negative feedback at the right timing, just like a coach’s action under the behaviorism learning approach. What’s also crucial is that the research coach should set up meetings with research buddies to see how they’ve been helping the learner to make sure the research activities are on track and the learners are comfortable.</p><p>The second method is about the connection between the learner and the research coach. The research coach should not only rely on a few big workshops or lengthy research coaching sessions to urge the learner to digest all the knowledge, because it is better for the learner to connect the knowledge fragments they acquired from different resources, and digest the connection of knowledge over time.</p><p>The one-time coaching sessions can be useful for basic user research concepts, but the ad hoc chats are equally valuable when the learner needs to resolve their instant problems. The research coach should be open and happy to answer any questions that any learners come up with at any time. The timely ad hoc chat on research questions creates the most effective moments that the learner absorbs knowledge and gains new skills.</p><h4>8. Back to user research democratization</h4><p>I talked about the learning theories that can help the education for democratizing user research. One of my UX research friends once asked me - Have you ever thought, after years, if non-research people learn and practice research better and better, what do we do then? Are we going to lose our jobs?</p><p>I asked him, so what do you do now?</p><p>He said, I do service design now. I feel they are all interconnected.</p><p>Separation of design and research is what this capitalistic industry defined, not what people naturally generated to achieve our personal goals. As Max Weber believed,</p><blockquote>bureaucratic organization is an efficient way of managing laborers. It defined the meticulous division of labor, the responsibility, and the clear hierarchies and professions to maintain its efficiency.</blockquote><p>I regard the democratization of user research across non-research roles as an attempt to step out of the taken-for-granted scope that the bureaucratic organizational power has set.</p><p>In a sense, people can be driven by their desire to learn new knowledge beyond the professional role definitions. Since non-research roles could learn and practice user research, then researchers can also learn and practice other skills and step into new fields too.</p><p>At the end of the day, you study and understand human experience, based on which you create something new to help people’s lives. That’s what we call innovation.</p><h4>References</h4><p>Maslow, A. (1954). <em>Motivation and personality</em>. New York: Harper.</p><p>Duke, B., Harper, G., &amp; Johnston, M. (2013). Connectivism as a digital age learning theory. <em>The International HETL Review</em>, <em>2013</em>(Special Issue), 4–13.</p><p>Rouse, J. (1994). Power/knowledge. <em>The cambridge companion to Foucault</em>, <em>2</em>.</p><p>Dierksmeier, C. (2011). Kant’s humanist ethics. In <em>Humanistic ethics in the age of globality</em> (pp. 79–93). Palgrave Macmillan, London.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2ea97c36a980" width="1" height="1" alt=""><hr><p><a href="https://uxdesign.cc/user-research-democratization-is-an-educational-practice-2ea97c36a980">Why UX research democratization is an educational practice</a> was originally published in <a href="https://uxdesign.cc/">UX Collective</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The invisible relationship: How power misrepresents knowledge in UX research]]></title>
            <link>https://uxdesign.cc/the-invisible-relationship-how-power-misrepresents-knowledge-in-ux-research-d77cba21d20c?source=rss-870e9ba832c8------2</link>
            <guid isPermaLink="false">https://medium.com/p/d77cba21d20c</guid>
            <category><![CDATA[politics]]></category>
            <category><![CDATA[ux-research]]></category>
            <category><![CDATA[editor-picks]]></category>
            <category><![CDATA[ux]]></category>
            <category><![CDATA[design]]></category>
            <dc:creator><![CDATA[Weidan Li]]></dc:creator>
            <pubDate>Mon, 20 Jun 2022 06:33:16 GMT</pubDate>
            <atom:updated>2022-09-10T01:22:09.193Z</atom:updated>
            <content:encoded><![CDATA[<h4><strong>1. Power in political affairs</strong></h4><p>“Virus! Get the fuck outta here! Go back to Asia!” A middle-aged Caucasian man shouted at me from the street corner. I’m obviously an Asian even if I was wearing a face mask. It was the autumn of 2021 in Melbourne when I was walking to a beach for some fresh air, but the racial discrimination polluted the air.</p><p>I didn’t think I was a virus because a virus doesn’t speak English. So I spoke to him right away, “How’s it going, mate?” He must’ve realized I could speak proper English and that was not what a virus can do. So he turned back and walked away.</p><p>Covid-related racism once surged during the pandemic. The main reason for that was the coronavirus was first reported in Wuhan, China in 2020. Whether for political purposes or not, mass media had strengthened the impression that coronavirus came from China and was made in China. Thus, it was once named “Wuhan virus”. Just like another misnomer “Spanish flu”.</p><p>The Asian communities have strived for equality in many western countries by protests, such as “Stop Asian Hate”. Covid-related racism in the Asian community was largely triggered by the political and economic conflicts between western countries and China. The conflicts were intensified in recent years by the different forms of ideologies and governments.</p><figure><img alt="Covid-related Asian hatred occurred in some western counties" src="https://cdn-images-1.medium.com/max/1024/0*wTOykR3ClfCiinRy" /><figcaption>Image from <a href="https://www.businessinsider.com/trump-sued-calling-covid-19-the-china-virus-kung-flu-2021-5">businessinsider.com</a></figcaption></figure><p>Whether intentional or not, mass media has created and exacerbated such a strong misbelief because it has the power to communicate with the target audience. Power dynamics have always existed in multicultural societies. Some communities are in the center of the social system because the society was built upon the major communities’ values, whereas some are marginalized because of lacking the resources and channels to be presented to the entire society.</p><figure><img alt="The mainstream society and marginalized cohorts" src="https://cdn-images-1.medium.com/max/1024/0*8xYcI_GkkN_LlAWt" /><figcaption>Image from <a href="https://news.utexas.edu/2019/05/13/political-controversies-about-marginalized-groups-increase-bullying-in-youths/">news.utexas.edu</a></figcaption></figure><p>When mass media aligns with the government’s and major communities’ ideological beliefs and expectations, power relationships are created. Under the predominant ideologies and beliefs, the media communicates knowledge to society and shows audiences a biased reality. As a result, the audiences see the world in the way that the major communities want them to see it. The knowledge that audiences obtain is the knowledge that the predominant parties want them to absorb.</p><figure><img alt="Power relationships in political affairs" src="https://cdn-images-1.medium.com/max/1024/0*P-kAMwYjWqPhD75m" /></figure><p>However, media was not the power itself. Media was only a neutral instrument being utilized by the dominant parties, but the political knowledge that media generates can be strong enough to influence the entire society’s point of view.</p><h4><strong>2. Knowledge and the elusive power exercise: Philosophers’ points of view</strong></h4><p>In Foucault’s perspective, power and knowledge are inextricably linked. Knowledge, as he called <strong>“politics of truth”, is formed in a system of communications, records, accumulation and displacement which is in itself a form of power </strong>(Foucault, 2019).</p><p>Foucault (1977, p27) said,</p><blockquote>“we should admit . . . that power produces knowledge (and not simply by encouraging it because it serves power or by applying it because it is useful); that power and knowledge directly imply one another; that there is no power relation without the correlative constitution of a field of knowledge, nor any knowledge that does not presuppose and constitute at the same time power relations”.</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/0*cLPo7eoEr_N7rdh0" /><figcaption>Michel Foucault</figcaption></figure><p>The bridge between knowledge and power is discourse. Foucault regarded that discourses are dynamic:</p><blockquote>“I do not mean ‘Power’ as a group of institutions and mechanisms that ensure the subservience of the citizens of a given state. By power, I do not mean, either, a mode of subjugation, which, in contrast to violence, has the form of the rule. Finally, I do not have in mind a general system of domination exerted by one group over another, a system whose effects, through successive derivations, pervade the entire social body” (Foucault, 1990, p. 92).</blockquote><p>Foucauldian theorists considered that knowledge agents draw upon the media resources through which power is exercised (Giddens, 1986). People act in the social system where the resources that actors utilize are not equally distributed due to the exercise of power, thus it leads to deeper systems of domination (Gaventa, 2003).</p><p>The way that Giddens (1984, p15) described power echoes Foucault’s perspective on power, knowledge, and media in a social system,</p><blockquote>“Resources (focused via signification and legitimation) are structured properties of social systems, drawn upon and reproduced by knowledgeable agents in the course of interaction. … <strong>Resources are media</strong> <strong>through which power is exercised</strong>, as a routine element of the instantiation of conduct in social reproduction”.</blockquote><figure><img alt="Power practiced in the mass communication process" src="https://cdn-images-1.medium.com/max/1024/0*3lYOJVoJ5wPhQqPU" /></figure><p>Power is implicitly embedded and exercised in media activities because media produces knowledge, which could be news and reports. The opinion of “Coronavirus is from China, so Chinese is virus” is knowledge because most of the world believes that it is true. If we think about the knowledge production around the pandemic, drawing upon Giddens’ statement, the <em>resources</em> are the <strong>capability of making people believe what the truth is, whether the truth is a fact or not</strong>. The knowledgeable agents are governments and mainstream media who are able to use the resources to produce “true information” in the power relationships between media and audiences. It’s only that the audiences are not hundreds or thousands, but millions or billions.</p><p>It is worth noticing here that perhaps neither the governments nor mainstream media are evil that twisted the pandemic facts on purpose — I hope they didn’t. They broadcast what they believe to be true. If they believe a communist’s government impedes human civilization in the modern world, then it does. But the effects of power still exist because when there is a flow of knowledge between two parties in the <strong>using resources — producing knowledge relationship</strong>, power is created without awareness.</p><p>Philosopher Russell (2004) coined that individuals are influenced by power in three ways:</p><ul><li>First is direct physical power over the individuals’ bodies, which is someone being imprisoned or killed.</li><li>Second is the power influence through rewards and punishments as an inducement, such as giving bonuses to employees.</li><li>The third is power influences opinions, such as “propaganda in its broadest sense”.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*P0dxgsYLD_ZSqLPG" /><figcaption>Bertrand Russell</figcaption></figure><p>Among all the types of power, Russell believes that <strong>opinion</strong> is one of the strongest power in social affairs because opinion power is pervasive and intangible. Again, opinion is a form of knowledge as long as the opinion is believed to be true. Opinion wasn’t only defined among the apparent unequal roles, such as the king vs civilians, or capitalists vs proletariat, but it moves as an invisible force through persuasion.</p><p>Russel (2004, p. 110) said,</p><blockquote>“First, pure persuasion leading to the conversion of a minority; then force exerted to secure that the rest of the community shall be exposed to the right propaganda; and finally a genuine belief on the part of the great majority, which makes use of force again unnecessary”.</blockquote><p>I talked about power theories to show that in a society where people play their roles to make a living, power itself is not always vicious and negative because it does keep the status quo of the operation of the social structure.</p><p>But the problem that power brings can cause conflicts and biases. The tricky part is that, as emphasized, <strong>power is often not a straightforward production by some person or institute, but it filtrates the social system like a liquid and prevents people from understanding the facts.</strong></p><p>After decades, Foucault’s perspectives on power remain convincing regarding what we saw during the pandemic. It also triggers my thoughts on the power relationships in UX research.</p><p>There is no one dominant group in UX research that evidently controls the other. Organizations, UX researchers and research participants work in their own roles to enable the UX research system operating. However, <strong>the latent negative effects of power in legitimate UX research activities have misrepresented insights and covered the knowledge of facts</strong>.</p><h4><strong>3. Power relationships in UX research</strong></h4><p>Power could shift overtly or covertly in UX research. Overt power appears when participants choose not to tell some stories that could’ve been of great value to the research, or being dishonest on purpose. Covert power is hidden and not aware, for example, when the researcher judged the participant’s language or communication skills yet not realized the bias has skewed data analysis. Ignoring the power dynamics in UX research could lead to a perfunctory understanding of user experience.</p><p>I will talk about the covert power relationship among the UX researchers, customers and organizations.</p><p><strong>3.1 Language and interpretation: The power between researchers and participants</strong></p><p>A main power problem in researcher—participant relationships is language. Language conveys the meaning of words in the communication between researchers and research participants. It is one of the most important methods in data collection. Whether in qualitative research where researchers talk to participants, or in quantitative research where researchers write up survey questions for participants to read through, language is the fundamental way to deliver information. However, the meaning in language can be misinterpreted if researchers are not aware of it.</p><p>One of the most significant language powers is unintentionally judging research participants by how they communicate and behave in qualitative research sessions. The judgment exists in the consensus that a good research session equals good communication between the researcher and the participant. The communication experience ultimately relies on language.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/0*YWwsWHlNK2pEhUuu" /><figcaption>Image from pexels.com</figcaption></figure><p>I referred to language not to the intensity of tones or sentence structures, but how language and expression are fundamentally shaped in people’s original national cultures and ethnicities.</p><p>After being involved in many usability testing and user interviews, I found there is an<strong> implied typical character</strong> <strong>in UX researchers’ mind that depicts what a good research participant should look like</strong>. They’d better be able to elaborate themselves well with clear words. They’d better be honest and brave enough to express what they honestly thought about a topic or design. They’d better be, most of the time, native English speakers. They’d better have quick reactions so that the research will yield a great amount of narrative data in 60 minutes. This list goes on and on.</p><p>Several years ago I observed an interview session moderated by another researcher. The participant was a Japanese lady in her 30s. She was on a working holiday in Australia. She has been living in Sydney for two years and applying for a student visa at that time.</p><p>After the interview session, I had a debrief with other researchers. One of the researchers said,</p><blockquote><em>“Hmm… She was too quiet,</em> s<em>he was not as elaborate as the last one. I don’t feel I’ve got enough insights from her”</em>.</blockquote><p>The other researcher agreed. I was a bit sad.</p><p>Unfortunately, the participant’s performance did not meet the researcher’s expectation, in comparison to the earlier and more “adorable” participants, as if there was an expectation on participants’ performance.</p><p>However, what I observed was she had limited language skills and barriers to express her ideas. Japanese was her native language and her English speaking skills were a disadvantage among the rest of the interviewees.</p><p>The problem here was two-fold. First, the participant struggled to use accurate English words to support her viewpoints on the prototype. The words she picked were rather general and basic than specific. She mentioned one feature was <strong>good</strong> and the other was not <strong>useful</strong>. Lack of elaboration made her look like a quiet and uncooperative participant from the moderator’s perspective.</p><p>The second one is a deeper problem. The language gap revealed a significant issue in cultural differences. <strong>The behavioural pattern of generating and expressing thoughts in English cultures was different to people who were born and raised in most East Asian cultures because of different educational systems and social values</strong>.</p><p>In China, for example, when I was in primary, middle and high school, expressing ideas and speaking was far less important than inner thinking and reflection. In fact, I could’ve been rebuked by teachers if I expressed too much in a class, even if my talk was relevant to the class.</p><p>I talked less, I did more homework, and I was a good thinker than a talker, which made me a “good student”. Ironically, the unspoken standards of judging participants in UX research are literally the opposite of those I experienced in my student time.</p><p>The main data collection activities in UX research, especially via qualitative research, are human communication. But fundamentally, <strong>the way of thinking shapes the way of communicating</strong>. Power has occurred in the social interaction process if the agents hold different cultural and ideological values.</p><p>In the interview, the Japanese participant had a difficult time communicating with the researcher due to her English speaking level, but more importantly, because of the cultural difference, she was deemed a “quiet participant” in the western context whereas quietness has another definition in the East Asian cultures. <strong>Having a moderate motivation of expressing opinions is not equal to being quiet.</strong></p><p>The consequences led by this power relationship is that research findings that inform design decisions could be culturally identical. Under the power of researchers’ own cultural identities and world values, the researchers may not realize that they have <strong>tagged participants as a “good one or bad one” rather than a “different one” through a predominant cultural lens that the researchers own</strong>. Ultimately the knowledge produced is the skewed research insights from the predominant cultural values.</p><figure><img alt="power in researcher — participant relationship" src="https://cdn-images-1.medium.com/max/1024/0*1z-C5xXiGX6J9mjn" /></figure><p>When the majority of insights come from those well-performed participants, they are from similar participants that are regarded as good participants by researchers, and it will be a risk that user experience was understood from a one-dimensional cultural perspective that UX researchers can hardly get rid of.</p><p>But will companies only design products for the users who are similar to those well-performed, elaborate and extroverted participants during interviews, and those who have the same cultural identity as the UX researchers’?</p><p>Same to what Foucault thought, no one institute or person creates and implements power in the UX researcher-participant relationship. The UX researcher did not pressure any participant. The participants were not trying to cover any information on purpose during interviews. <strong>But the pervasive power stems from the subjectivity of cultural homogenization, and it led to the judgment on research participants and further bias in data collection.</strong></p><p><strong>3.2 Generate cross-cultural knowledge to mitigate the power of cultural homogenization</strong></p><p>Inspired by ethnographic methodology, cultural awareness is a basic but core quality. The awareness and sensitivity of cultural difference is a result of curiosity in fieldwork. Hasbrouck (2017) described that ethnographers keep seeking for the “why” in every social situation and trying to uncover the meaning of the objects or events that people take for granted. The momentum of such curiosity stems from an outsider&#39;s perspective.</p><p>It could be easier for ethnographers to gain the outsider lens if they delve into a cultural context that is different to their own than into an identical culture. Though it doesn’t mean that ethnographers can not study a culture where they were born. The British anthropologist Daniel Miller beautifully wrote about the ordinary people in South London and the material culture (Miller, 2008).</p><p>However, when anthropologists like Clifford Geertz and Claude Lévi-Strauss stepped into the exotic cultures in Bali and Brazil, they were astonishingly intrigued by the contrast between the modern and primitive societies. Then they started as an outsider — a modern person — to uncover the primitive social norms and structures to reflect on the current and future society.</p><p>Engagement in different cultures made a great positive impact on research findings. UX research practice can borrow the mindset from ethnography to mitigate the power in the researcher—participant relationship. <strong>That is to include participants from a diverse range of cultural and lingual backgrounds.</strong></p><p>As discussed, language is one of the mediums that embodies cultural, political and social values. Therefore, UX researchers should maintain the diversity of research participants’ national cultures because languages root in national cultures rather than only races. Two people with different races and ethnicities who grew up in one country could share the same culture due to the same language and values.</p><p>Maintaining cultural diversity is important to qualitative research because I found that <strong>the disparity of national cultures makes participants convey their attitudes towards the same topic in different manners</strong>. When talking through the experience of the same prototype, in one of my studies, the Australian and British participants are more likely to share their thoughts about the features they liked first and then the aspects they suggested us to improve. It sounded like</p><blockquote><strong>“I loved the purchase flow because it was soooo simple to use. If this product comparison page could be improved, it would be fantastic!”</strong></blockquote><p>However in the same study, the Malaysian Chinese participants would share the design drawbacks first before talking about the features they liked. The attitude was</p><blockquote><strong>“I would fix this product comparison page ASAP because it wasn’t a pleasant experience, otherwise I won’t use it again. I had no problem with the purchase flow though” </strong>.</blockquote><p>Obviously, language customs played a significant role here. The conclusion of both the participants</p><p><strong>1) The product comparison page design is not good enough</strong></p><p><strong>2) The purchase flow overall has been satisfactory</strong></p><p>But the language habits and straightforwardness of expressing dissatisfactions in western cultures and Asian cultures are different.</p><p>This may lead UX researchers to different interpretations — should the report highlight the comparison page problem as the Asian participants addressed, or emphasize the positive purchase flow experience as the western participants loved?</p><p>The team can certainly go for the simple way and recruit the culturally similar participants so that the behavioural patterns are much more straightforward — they’ll all love the same feature and hate the other based on the way of communication.</p><figure><img alt="cultural homogenization — all oranges are the the same" src="https://cdn-images-1.medium.com/max/800/0*ADom8vOOVaOQhq9h" /><figcaption>Image from pexels.com</figcaption></figure><p>But the findings will not stay close to the factual user experience under cultural homogenization because UX researchers will only be exposed to the similar way of saying, thinking and behaving that might be quite similar to their own.</p><p>In order to help researchers remain the outsider view, during user recruitment for UX research, they could add questions in the screener, such as:</p><ul><li>Is English your native language?</li><li>What languages do you use other than English?</li><li>Please specify your ethnicity.</li><li>Where were you born?</li><li>If applicable, please specify your religion.</li></ul><p>The advice here is that try to diversify the participants’ culture when inviting them to qualitative studies where language plays a key role.</p><p><strong>3.3 Market segmentation vs. cultural communities: The power between organizations and customers</strong></p><p>Organizations define customers as the individuals who constitute the market. In UX research, researchers recruit research participants to cover the most of the market segments. They also frame research questions and showcase research findings based on market segments, which are usually represented by customer archetypes.</p><p>For example, when I worked at a health insurance company, I recruited participants with consideration of the types of insurance covers, the years of experience using insurance products, and whether they have switched from one insurance provider to the other. In the financial industry, SMB owners can be divided by the size and complexity of businesses they own, so I recruit research participants based on the types of SMB in which our organization is most interested.</p><p>From the organizational perspective, business grows with more customers in different market segments purchasing or subscribing to their products. In the market-led capitalistic system, <strong>a person is either already a customer or expected to become a customer for profits</strong>.</p><p>Eventually a human being is oversimplified into a figure of market value.</p><figure><img alt="Eventually a human being is oversimplified into a figure of market value —diagrams and tables with international data" src="https://cdn-images-1.medium.com/max/1024/0*4uIQpjuWsdTM-GIw" /></figure><p>The problem is that customers are still people with lived experience and dynamic individual context. <strong>They have social networks, they have different roles in the society, they come from different cultural groups, and they seek and fulfil the meaning of life under different worldviews</strong>.</p><p>Customers constitute the market, however, they also belong to different communities with different cultural, social, ethnic and political needs. In Merriam-Webster dictionary community is defined as:</p><figure><img alt="screenshot of the definition of “community” in the dictionary" src="https://cdn-images-1.medium.com/max/1024/0*6jPdiGQ9Vl-N_-5z" /></figure><p>The definition emphasized that a community consists of people who have common history or social, economic and political interests. They are not only a group that purchases and uses products, but more essentially they are individuals that belong to and behave within their shared cultures.</p><p>More importantly, a community is a range of individuals that <strong>are alike and gathered autonomously rather than being grouped by any organizations</strong>. “Single parent” or “new immigrants” is a cultural community, but “highly satisfied subscribers aged between 20–40 years old” is a market segment.</p><p><strong>A market segment is defined by what organizations think people are, and a cultural community depicts who people really are.</strong></p><p>Power comes from an organizational and market-led lens, through which humans are categorized into customer segments. The market-led views on humans will limit UX research to a tunnel vision instead of a holistic view on human experience.</p><p>The tunnel vision is a consequence of the power relationship between organizations and customers. It may ignore the diversity of communities and generate superficial product-centric insights rather than knowledge of real human beings’ needs, goals and experiences.</p><figure><img alt="Power in organization — customer relationship" src="https://cdn-images-1.medium.com/max/1024/0*hGYOEIVZ5NdwT_72" /></figure><p>I’ll elaborate why it is important to explore the community-based insights is no less important than the research findings of different market segments.</p><p><strong>3.4 Generate community-based knowledge</strong></p><p>Imagine there is a bike-sharing company in Australia. The company has seen the different needs between user A and user B. User A works remotely from home, while user B rides a bike to their office every day. They have different levels of dependency on bike riding and different frequencies of riding. This is a market segment disparity between user A and B.</p><figure><img alt="two different users defined by product and market segments" src="https://cdn-images-1.medium.com/max/984/0*UPotXC1tjI2vtg98" /></figure><p>However, a more basic and natural community-based difference is that user A and B are both young immigrants from the same country in South America, whereas user C is a local resident born in the 1950s. Maybe user C has similar frequencies of using bike sharing services with user B, which is very similar to user A; but user A and B belong to the same cultural community who hold<strong> shared values and interpretations on commute and transportations</strong>.</p><figure><img alt="The two different users defined by product and market segments can be similar based on cultural behaviours" src="https://cdn-images-1.medium.com/max/764/0*5jaIkbO-95wiu4nv" /></figure><p>From a market-led research perspective, what organizations can see are <strong>only customers within a product/service context, i.e. one user has a higher frequency of use than the other. </strong>This tunnel vision can easily miss the community factors that play a foundational role that composed the human experience. Community factors are foundational because they shaped how people believe, behave and think in most of their life based on interests and values. This is not what an interaction with a product can compare to.</p><p>A positive signal is that designers have realized the value of community-based design during co-creation activities. <a href="https://jnd.org/community-based-human-centered-design/">Community-based design</a> processes involve people of certain communities so that the design solutions are based on the local knowledge and serving the local people instead of from non-local experts. There are certainly limitations and risks in community-based design methods, but the awareness of communities is worth bringing into UX research practice.</p><p>Community is an important concept in traditional qualitative research because <strong>great storytelling and insights often reside in the disparity and commonality among people living in the community.</strong> In ethnography, researchers have demonstrated the importance of understanding a community by how they work. They are not an outsider to study the research participants defined by any organizations, they need to actively participate and engage in the community to make sense of the meanings behind behaviors to understand how people are defined by their own cultures.</p><p>As some ethnographers described their role in community-based research,</p><blockquote>“I evolved from being a cautious observer, learning the rules of the game to avoid exclusion, to being an adventurous participant who got involved in the everydayness of the community to uncover the tale of the village” (Elidrissi, Bouguerra &amp; D’Souza, 2020).</blockquote><p>During the engagement in communities, ethnographers usually do not have much consideration of a product, service or pre-exist solution to offer help but focus on the shared culture in it. The community-based mindset adds extra value and helps organizations see patterns of latent and foundational needs within and across communities beyond customer segments coined by the market.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/0*4Wpx26wKHwaUS3QC" /><figcaption>Community of the low-income young Chinese Migrant Families — A study on <a href="https://www.triciawang.com/projects/2011/8/31/digital-urbanisms-on-the-margins-chinese-migrants-and-intens.html">technology use</a> by Tricia Wang</figcaption></figure><p>There are certainly many ways to approach community-based research and insights because a community could refer to many aspects including languages, races, personal interests, socioeconomic status, physical conditions, religions and so on. It is worth a whole book to elaborate a community-based design methodology.</p><p><strong>What I suggest here is simply enhancing the awareness of community differences in UX research.</strong></p><p>It could be realized by long-term focus on studying one particular cultural community. In parallel with quick usability testing for a certain customer segment, the UX research team may also delve into the lived experience of a particular cultural community depending on the team’s interest, such as elderly with no children or people addicted to video games.</p><p>This can be one longitudinal research over a few months, or a few research that maintains the exploration into a community over a year. After a few community-based studies, the insights will tell the team how different people can be when they are placed in the cohorts that they naturally belong to.</p><p>In addition, I have seen companies categorize research insights with research dates, initiatives and teams. No doubt that this repository structure is easy for the team and stakeholders to find the insights they need. However, to help organizations understand the needs and beliefs of different communities, UX researchers could add extra tags to highlight the community-based knowledge in the research repository.</p><p>Ideally the team can filter through the insights not only by products, features or user demographics (ages, genders, income, education etc.), but also more diversely by cultural community tags like “skateboarders with children to take care of”, “anti-tech Generation Z” or “second generation of Malaysian immigrants working in tech industry” — just an example, I totally made them up here. The more that a customer was studied in their communities, the more fruitful findings the team can obtain after a period of research.</p><h4><strong>The end</strong></h4><p>Power is an enormous topic for a single article. Philosophers have dedicated their whole lives to understanding and explaining power. Philosophers like Foucault were challenging the power in political systems to strive for social equality. Even though political affairs are far away from UX research, the effect of power relationships exists in both areas.</p><p>The power in UX research <strong>stems from researchers and organizations interpreting reality through an inevitable subjective and intentional lens</strong>. The power is not created by anyone but by the way of thinking that is taken for granted.</p><p>In my earlier example, the foreign research participant who had a difficult time during the user interview. She was the vulnerable group in the power relationship between the researcher and the researched. It was not the researcher who exercised the power on purpose to suppress.</p><p>The power exists in the normalized way of portraying the qualified research participants in a mainstream culture, just like how I was portrayed as a coronavirus in the mainstream media context.</p><p>To mitigate this power caused by subjective intentions, we have to make an effort in UX research to be aware of people’s languages, communities and cultures that make them who they are, not simply labeling them as “customers”.</p><h4><strong>References</strong></h4><p>Elidrissi, Y. R., Bouguerra, N., &amp; D’Souza, R. C. (2020). Enacting Ethnography: Three Perspectives on Engagement with Political Communities. <em>Management</em>, <em>23</em>(3), 109–113.</p><p>Foucault, M. (2019). <em>The history of sexuality: the will to knowledge</em>. Penguin UK.</p><p>Foucault, M. (1990). The history of sexuality: An introduction, volume I. <em>Trans. Robert Hurley. New York: Vintage</em>, <em>95</em>.</p><p>Foucault, Michel. 1977. Discipline and Punish: The Birth of the Prison. London: Allen Lane.</p><p>Gaventa, J. (2003). Power after Lukes: an overview of theories of power since Lukes and their application to development. <em>Brighton: Participation Group, Institute of Development Studies</em>, 3–18.</p><p>Giddens, A. (1986). <em>Sociology: A Brief but Critical Introduction: A brief but critical introduction</em>. Macmillan International Higher Education.</p><p>Hasbrouck, J. (2017). <em>Ethnographic thinking: from method to mindset</em>. Routledge.</p><p>Miller, D. (2008). <em>The comfort of things</em>. Polity.</p><p>Russell, B. (2004). <em>Power: A new social analysis</em>. Routledge.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d77cba21d20c" width="1" height="1" alt=""><hr><p><a href="https://uxdesign.cc/the-invisible-relationship-how-power-misrepresents-knowledge-in-ux-research-d77cba21d20c">The invisible relationship: How power misrepresents knowledge in UX research</a> was originally published in <a href="https://uxdesign.cc/">UX Collective</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Traditionalization: Theoretical practice for industrial research]]></title>
            <link>https://medium.com/@weidanlee/traditionalization-theoretical-practice-for-industrial-research-152b06ae2b47?source=rss-870e9ba832c8------2</link>
            <guid isPermaLink="false">https://medium.com/p/152b06ae2b47</guid>
            <category><![CDATA[theory]]></category>
            <category><![CDATA[user-experience]]></category>
            <category><![CDATA[ux-research]]></category>
            <category><![CDATA[research]]></category>
            <category><![CDATA[design-research]]></category>
            <dc:creator><![CDATA[Weidan Li]]></dc:creator>
            <pubDate>Thu, 30 Dec 2021 06:21:00 GMT</pubDate>
            <atom:updated>2022-01-08T03:16:13.967Z</atom:updated>
            <content:encoded><![CDATA[<p>In the 1910s, Malinowski spent years on Trobriand Islands studying the Kula culture. He put himself in the local communities. He learnt the languages, participated in social encounters, drew the map of the local areas, took notes, photos, maybe also got sick for a while from the exotic food and natural environment. The whole process of being in the field and reflecting on the observation was a part of the research.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/464/0*jYGxw6gO3KV7Vw85" /><figcaption>Malinowski in the field</figcaption></figure><p>Malinowski might never have expected that someday across the globe people would use similar methods to empower products and businesses. In the 1990s, the term “user experience” was created. The psychologist Donald Norman at Apple who coined the term said “I invented the term because I thought Human Interface and usability were too narrow: I wanted to cover all aspects of the person’s experience with a system, including industrial design, graphics, the interface, the physical interaction, and the manual.”</p><p>User experience, including user experience research, were widely applied in the technology areas. Companies started to hire researchers. A great amount of researchers. From a few to hundreds. From junior researchers to research directors.</p><p>The traditional and academic way of doing research has been modernized into the more efficient UX research, design research or user research nowadays. I’ll call these seemingly different names — industrial research. Industrial research borrowed methods from the traditional research such as interviews, contextual enquiry, observations and surveys, which are common ones for sociologists and anthropologists. These methods stem from academia a century ago for the purpose of cultural and societal investigation, and now started to serve the capitalist mode of production process.</p><p>The reason for migrating the traditional research methods to the industry is obvious. We have a free and global market. Where a free market exists, there are competitions among companies who produce similar commodities. In order to surpass the other competitors, the companies need to know what the buyers (customers) want. The better they know it, the more likely they win the market. The superiority of “knowing what customers want” is important because it enhances the customers’ eagerness of possessing the product, and it converts the eagerness to attractive features, and further to cash.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/0*smusXj2QYA6EpHl7" /><figcaption>A focus group session in the TV series<em> Mad Men. Stories described the advertising industry in the 1960s US.</em></figcaption></figure><p>In this competition process, insights need to be produced rapidly and collectively to provide proper suggestions to product design. Research becomes quicker and digestible, but also cheaper and weaker than what it originally was. Many industrial research projects nowadays are echoing customers’ voices to be straightforward and “actionable” enough.</p><p>Many of them did not uncover why human behaviours and perspectives exist in a larger social and environmental system with what theories and conceptual frameworks in mind to interpret a phenomenon, which will very likely fail if it were in an academic world aiming for knowledge contribution. By “Phenomenon” I mean those observable events that drew researchers’ attention and intrigued them to explore the causations behind. Therefore, researchers find topics that they think worth studying, rather than those required by the business.</p><p>The trend of <a href="https://www.stripepartners.com/viewpoint/the-ux-ification-of-research/">UX-ification</a> of research is risky. When everyone is pursuing efficiency in research activities, seeking for less time cost and greater amount of findings, the thoughtfulness is gone.</p><p>Analogously the problem relates to what Martin Heidegger was concerned about the modern technology against nature. I put it in a very simple way but maybe an inaccurate way: the danger is humans are using technology in an extremely purposive manner to urge technology to exploit nature (i.e. how companies leverage research to succeed themselves in the market); nature (i.e. research) is no more something for humans to be in, to belong to, and to seek for meaning, but to utilize to pursue purposes (i.e. business success). Based on this large technology infrastructure that humans have established, humans (i.e. researchers) were turned into materials and resources to be utilized for purposes by themselves.</p><p>Fast industrial research is absolutely important. They provide a large amount of customer data to help companies survive and even live better. But when every company is doing research the same way trying to be quick and somewhat understand customers (let alone the limited concept “customer” or “user” as a research target itself has biased research), where are the areas for innovation to create real value to do good to human societies beyond just business figures like higher user growth rate or lower user churn rate?</p><h4><strong>Traditionalizing research with theory</strong></h4><p>It may be worth spending some time learning from what academic researchers do and think to make industrial research more meaningful. A good way to present us to the real value of research, in my opinion, is thinking about theory and using theoretical framework in research.</p><p>I quote my favorite <a href="https://dl.designresearchsociety.org/cgi/viewcontent.cgi?article=2702&amp;context=drs-conference-papers">definition</a> on theory, coined by a scholar in design and research field:</p><blockquote>In its most basic form, a theory is a model. It is an illustration describing how something works by showing its elements in their dynamic relationship to one another. It is the dynamic demonstration of working elements in action as part of a structure that distinguishes a model from a simple taxonomy or catalogue.</blockquote><p>There are a few ways to use theory in research. Firstly, theory can be an overarching framework to construct the whole research. With this method, theory is considered in the very early stage of research. It allows researchers to initiate topics and ideas, build research questions, frame methodologies, select methods, and provide solid references to insights.</p><p>I take feminism as an example. <a href="https://courses.lumenlearning.com/alamo-sociology/chapter/reading-feminist-theory/">Feminist theories</a> focus on the potential restrictions and inequality that women face in a dominant gender ideology toward sexual differences. By reviewing the literature on feminism, researchers may generate a series of study topics around the behavioural differences between female and male customers. Then convert the theory-informed topics to research questions that aim to study sexual differences or inequality.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*dzWU_i46xOZcuipE" /><figcaption>Feminism philosopher — Simone de Beauvoir and her husband Jean-Paul Sartre</figcaption></figure><p>Assuming a company has a music app, the research questions can be around how housewives’ preferences on music genres differ from their male partners. With a feminisit perspective, the researchers may find insights beyond the usage of the app to provide immediate suggestions to redesign the music genres categories.</p><p>More fundamentally, they may find what family roles and societal responsibilities are defined for housewives, whether they are reluctant to accept or not, and how inequality is created in the relationship with their male partners, and so on. The insights outside the quick design solutions are the space where innovative ideas present themselves to us.</p><p>The other way is to use theory as a lens to guide data analysis. In this manner, theory is a perspective through which researchers make sense of the data — usually qualitative data. There are numerous paths to the interpretation of qualitative data. Therefore, a theory adds a perspective, or angle, to approach data that helps researchers focus on important and interesting issues they think worth exploring deeply.</p><p>I was working on an academic research project in 2015. It was also a consulting project for a financial institute. We tried to study the Chinese Millennials’ perspective on money management. In the very beginning I didn’t know what theory to use for data synthesis, but I knew I had to use one or a few to meet the academic knowledge standard.</p><p>After a few months of study in the Shanghai urban area, I collected many hours of interview data and I read through the transcripts. I found it interesting that the young cohort in China considered spending and saving money in a <a href="https://medium.com/@weidanlee/as-researchers-we-use-multiple-ways-to-reduce-the-risks-in-product-and-service-design-9181fae7b359">large social context</a>. I stuck to this point and sought for social theories to help me further explore this phenomenon — the phenomenon that reveals the relation between social context and people’s interpretation of things. So I did more literature reviews and found <a href="https://courses.lumenlearning.com/alamo-sociology/chapter/reading-symbolic-interactionist-theory/">symbolic interactionism</a>. The theory defines that:</p><blockquote><em>Humans act toward things on the basis of the meanings they ascribe to those things.</em></blockquote><blockquote><em>The meaning of such things is derived from, or arises out of, the social interaction that one has with others and the society.</em></blockquote><blockquote><em>The Meanings are handled in, and modified through, an interpretative process used by the person in dealing with the things he/she encounters.</em></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/432/0*wkXXe64QEBvu3hWr" /><figcaption>Herbert Blumer, sociologist and author of symbolic interactionism</figcaption></figure><p>I drew a model of the theory and applied it to data analysis. The analysis was based on the perspective that meaning of objects is influenced by or created from collective and social activities. I knew the original theory allows people to understand it from many different angles, but I defined this one way to approach and interpret my data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*6Is84JDiNd1sGgay" /></figure><p>The theoretical lens prevented me from just coding the qualitative data with no focus and superficially echoing people’s voices. It helped me interpret data from a specific theoretical point of view to find the inner connections of human behaviors.</p><p>I found that the meanings of money are ever changing and meanings are made in the various social encounters such as gift giving over festivals, or lucky money games in WeChat. In these circumstances the value of money was enhanced by its social values beyond its original economic worth.</p><p>Every theory, especially social theory, provides a unique perspective to frame the research and approach data. For example, how does Saussure’s structural linguistics teach us about people communicating meanings through signs? Drawing on that, how does Claude Levi-Strauss tell us about the structuralists’ mode of analysing cultural artefacts? What’s the relation between individuals and the structure they live in? In George Simmel’s philosophy of money, he regarded the value of goods attached to subjective processes of exchange instead of the invested labour power. How does that inspire us on products’ prices and values? How does social functionalism tell us about the individual actions and how they are embedded in the whole functioning system for what purposes? And so on.</p><p>There are another two ways of applying theory in research. One of them is to use theory to build hypotheses. The whole purpose of the research is to validate the hypotheses and further to test the universality of the theory. This method is more often used in quantitative research than qualitative.</p><p>The other one is grounded theory. Although it’s often used for qualitative data collected from fieldwork or interviews, the goal of it is to generate new theories from in-depth data analysis.</p><p>I will not explain in detail because I think these two are more difficult to apply to industrial research than the two introduced earlier. They are more often used by academic scholars to contribute to a specific realm of knowledge, and uncover even a tidy bit of unknown knowledge space to create intellectual values.</p><h4><strong>Traditionalizing research in seeking talents</strong></h4><p>Using theory might be one of hundreds of different paths to traditionalizing research in the industry to preserve the value of research for obtaining new knowledge.</p><p>Companies that hire UX researchers can make changes too. The biggest problem happens at the organisational level. The way companies define research and the way they adopt research methods for business interest has made research activities less thoughtful.</p><p>When hiring researchers, the usual way is that the company tells candidates what products the team is working on, what the product visions are, and how they are expected to use research to solve problems that relate to products. Then the team will think about the candidates’ personalities and make a decision whether they fit in the team culture.</p><p>It is tricky, however, to reveal the candidate’s real research thinking and stances in this organization-centric hiring process, because it often ignores the researchers’ own personalities, writing styles, world views, values, cultural background and research interests. These factors matter a lot because they shape the way an individual thinks and creates, without which no one researcher differs from the other.</p><p>In order to let researchers show who they really are, companies can refer to the academic ways of hiring, such as applying for a role with a research proposal. A proposal is required for the application of all academic roles, from doctorate candidates to professor positions. It elaborates the value of the research, the methods, the theoretical framework, and the plan for data collection and analysis.</p><p>Most importantly, the research candidates decide by themselves what to propose as a future research topic. It gives free space to candidates and lets them show what phenomenon has intrigued them the most with which methods to approach it within which theoretical perspectives, and why it is valuable to study at all. It doesn’t involve an imaginary product or feature to test, so it won’t make candidate compromise to what the company wants to achieve, but only aims to elicit who the researcher is as a researcher.</p><h4><strong>Complicated research</strong></h4><p>Industrial research is pragmatic and efficient. It agile-lized the academic methodology, but oversimplified it at the same time.</p><p>Research is research because it takes more time for thinking and reflecting than facilitating and communicating. It’s not about keeping the team happy with research plans, or directing to insights the team would like to see. It’s about, being inspired with theories, researchers immersing themselves in the phenomenon they want to study and obtaining insights that interpret the phenomenon.</p><p>In a future where AI automates and even replaces a large amount of definite and repetitive work we are doing today, the enthusiasm for generating new knowledge through making sense of complexities might be what makes us more human than computers.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=152b06ae2b47" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>