What "sample size is needed" for 10 different types of qualitative research -- new article by Wutich et al. Useful overview for those who value guidelines for saturation.
Wutich, A., Beresford, M., & Bernard, H. R. (2024). Sample sizes for 10 types of qualitative data analysis: an integrative review, empirical guidance, and next steps. International Journal of Qualitative Methods, 23, 16094069241296206.
Open-access article available here. https://lnkd.in/gmQd3RTJ | 70 comments on LinkedIn

An interesting opinion piece in the NYTimes.

Helen Ouyang is an ER doctor and Columbia professor. She typed her lab results into ChatGPT because her physician told her a phone call would require another appointment. That's what she wrote!

She is not a credulous patient who doesn't know better. She knows exactly what a chatbot can and cannot do. She used it anyway, and it worked. Her numbers improved. She credits the sustained back-and-forth, the patience, the absence of judgment.

She writes: "With low expectations, I typed my lab results into ChatGPT.
As both a physician and a patient, I found the experience startling. Not because ChatGPT dazzled me with its scientific knowledge, but because it behaved the way I wish modern medicine, and its practitioners, still would."

These last few words should stop every physician who reads them.

The usual framing treats patients using AI as a problem to manage, a liability to disclaim, a behavior to correct with the right guardrails. Ouyang's piece quietly dismantles that. The chatbot did not win because it knew more. It won because it had time. It asked follow-up questions. It remembered what she said five exchanges ago. It never seemed annoyed.

Those are not AI capabilities. Those are human capabilities that the current system has systematically squeezed out of clinical encounters.

There is a structural argument buried in her essay that she does not quite surface: patients are not going to AI because AI is good. They are going to AI because the alternative has been stripped of the things that make medicine work. Availability. Continuity. The permission to ask the same question twice.

Her proposed response is cautious and right as far as it goes: figure out how to support patients using AI tools, with clear guardrails.

For me, this is the judgment layer argument made visible from the physician side. Ouyang knows when to override the chatbot. Most patients do not. That asymmetry is not a reason to condemn AI use, it is the exact design problem we need to solve. The question is not whether patients will use AI. They will. The question is whether they will use it with enough contextual grounding to know what to do when it gets something wrong.

The Ouyang piece will be cited as evidence that AI is replacing doctors. That is the wrong read. It is evidence that something in medicine stopped working long before AI arrived, and AI is filling the vacuum.

Her last paragraph tells the whole story: "My experience with the chatbot has already shifted how I interact with patients in the E.R., with only minutes to piece together fragments of their circumstances. When a patient asks the same question repeatedly, I try to listen for what’s behind it. Maybe she’s not after more medical facts."

Dave deBronkart Hugo Campos Sara Riggare Jane Sarasohn-Kahn

#PatientsUseAI #CriticalAIHealthLiteracy
| 10 comments on LinkedIn

Across 31 studies, and 2,600+ medical students, AI tools made no statistically significant difference to clinical knowledge or skills for medicine students.

This is a systematic review and meta-analysis published this week in BMC Medical Education by colleagues across three Chinese medical institutions. They searched eleven databases, pooled every controlled study they could find comparing AI tools against traditional teaching for medical undergraduates, and came back with a standardised mean difference of 0.22 and a p-value of 0.22.

In plain terms, the effect is small and it is statistically indistinguishable from chance.

The overall evidence quality, the authors say, is “low”. That phrase appears in a peer-reviewed meta-analysis of thirty-one studies, and the quality is still low.
The universities buying AI tools for clinical education are not going to circulate this paper. The vendor is not going to send it round.

This paper is really helpful evidence for why we need to stop treating AI as a panacea for pedagogy, and perhaps think about investing that money elsewhere...

Link to article: https://lnkd.in/dnCb2fAy

#AI #Education #GenAI #HigherEd #SlowAI #Teaching| 15 commentaires sur LinkedIn

Ne manquez pas ce matin mon nouveau billet sur le blogue Convergence de Association des archivistes du Québec. La mort du texte? Certainement pas des miens !! Bonne semaine à tous!
https://lnkd.in/etbvskfk

How do you actually use your professional network to learn?

Most of us give some version of the same answer: we follow interesting people, scroll LinkedIn, catch a conference or two. If pressed, we say something like "I just… stay connected." That answer is not wrong. It is just not very useful.

The third post in my Linking Learning Advisory launch series introduces the LSA framework, which came directly out of my PhD research with practising teachers. It describes three distinct sets of connected learning practices: Linking (purposeful and goal-directed), Stretching (curious and expansive), and Amplifying (contributory and generative). These are not stages on a ladder. Most of us move between all three, depending on what our context and capacity allow at any given time.

What the framework offers is vocabulary. And vocabulary matters, because naming a practice is the first step towards being intentional about it.

The post also draws the connection between these learning practices and information literacy, and explains why the framework travels well beyond education into corporate and government contexts.

Read it at https://lnkd.in/gnY_yjJE

#linkinglearning #linkedlearningadvisory #professionallearning #PLN #connectedlearning #informationliteracy #LSAframework #knowledgemanagement

"𝗪𝗲 𝘀𝗽𝗲𝗻𝘁 𝗱𝗲𝗰𝗮𝗱𝗲𝘀 𝗽𝗲𝗿𝗳𝗲𝗰𝘁𝗶𝗻𝗴 𝗮𝘀𝘀𝗲𝘀𝘀𝗺𝗲𝗻𝘁𝘀 𝘁𝗵𝗮𝘁 𝗺𝗲𝗮𝘀𝘂𝗿𝗲 𝗺𝗲𝗺𝗼𝗿𝘆. 𝗧𝗵𝗲𝗻 𝗔𝗜 𝗮𝗿𝗿𝗶𝘃𝗲𝗱 𝗮𝗻𝗱 𝗺𝗮𝗱𝗲 𝗺𝗲𝗺𝗼𝗿𝘆 𝗶𝗿𝗿𝗲𝗹𝗲𝘃𝗮𝗻𝘁 𝗼𝘃𝗲𝗿𝗻𝗶𝗴𝗵𝘁."

So I built this poster to make that uncomfortable truth visible. It started with a question I could not stop asking: why do educators keep redesigning assessments around AI detection, when the real problem is that the assessments themselves were never designed to survive AI in the first place? Borrowing the logic of the Lippitt-Knoster change model, I mapped what actually happens when one critical element of AI-resilient assessment design is missing. The results were not surprising. They were sobering.

  1. 𝗡𝗼 𝗔𝗜 𝗣𝗼𝗹𝗶𝗰𝘆 𝗮𝗻𝗱 𝘀𝘁𝘂𝗱𝗲𝗻𝘁𝘀 𝗱𝗼 𝗻𝗼𝘁 𝗰𝗵𝗲𝗮𝘁. They fill the vacuum you left them.
  2. 𝗡𝗼 𝗙𝗮𝗰𝘂𝗹𝘁𝘆 𝗖𝗮𝗽𝗮𝗰𝗶𝘁𝘆 𝗮𝗻𝗱 𝘆𝗼𝘂𝗿 𝗲𝗱𝘂𝗰𝗮𝘁𝗼𝗿𝘀 𝗮𝗿𝗲 𝗻𝗼𝘁 𝗿𝗲𝘀𝗶𝘀𝘁𝗮𝗻𝘁. They are anxious, unsupported, and quietly overwhelmed.
  3. 𝗡𝗼 𝗔𝘂𝘁𝗵𝗲𝗻𝘁𝗶𝗰 𝗗𝗲𝘀𝗶𝗴𝗻 𝗮𝗻𝗱 𝗔𝗜 𝗱𝗼𝗲𝘀 𝗻𝗼𝘁 𝘂𝗻𝗱𝗲𝗿𝗺𝗶𝗻𝗲 𝘆𝗼𝘂𝗿 𝗮𝘀𝘀𝗲𝘀𝘀𝗺𝗲𝗻𝘁. Your assessment design invites it in.
  4. 𝗡𝗼 𝗔𝗜 𝗟𝗶𝘁𝗲𝗿𝗮𝗰𝘆 𝗮𝗻𝗱 𝗹𝗲𝗮𝗿𝗻𝗲𝗿𝘀 𝗱𝗼 𝗻𝗼𝘁 𝗯𝗲𝗰𝗼𝗺𝗲 𝗱𝗶𝘀𝗵𝗼𝗻𝗲𝘀𝘁. 𝘛𝘩𝘦𝘺 𝘴𝘪𝘮𝘱𝘭𝘺 𝘸𝘦𝘳𝘦 𝘯𝘦𝘷𝘦𝘳 𝘵𝘢𝘶𝘨𝘩𝘵 𝘩𝘰𝘸 𝘵𝘰 𝘦𝘯𝘨𝘢𝘨𝘦 𝘸𝘪𝘵𝘩 𝘈𝘐 𝘦𝘵𝘩𝘪𝘤𝘢𝘭𝘭𝘺 𝘢𝘯𝘥 𝘤𝘳𝘪𝘵𝘪𝘤𝘢𝘭𝘭𝘺.
  5. 𝗡𝗼 𝗙𝗲𝗲𝗱𝗯𝗮𝗰𝗸 𝗖𝘂𝗹𝘁𝘂𝗿𝗲 𝗮𝗻𝗱 𝗮𝗹𝗹 𝘆𝗼𝘂 𝗲𝘃𝗲𝗿 𝗮𝘀𝘀𝗲𝘀𝘀𝗲𝗱 𝘄𝗮𝘀 𝘁𝗵𝗲 𝗔𝗜-𝗽𝗼𝗹𝗶𝘀𝗵𝗲𝗱 𝗽𝗿𝗼𝗱𝘂𝗰𝘁. The reasoning, the struggle, the learning itself, all invisible.

This matters most in healthcare education. Dentistry. Medicine. Pharmacy. Nursing. We are not preparing students to produce outputs. We are preparing them to exercise clinical judgement in situations no AI model will ever fully own. AI-resilient assessment is not about locking AI out. It is about designing learning so rich in human reasoning that AI simply cannot do the work for you.

The framework draws on the AIAS (Perkins et al., 2024), the DRIVE Framework (Oliveira et al., 2025), and the ARAF (Harvinder Kaur and Siti Khadijah, 2025). It is evidence-grounded and built for practice, not just policy documents that gather dust.
Look at your last three written assignments. Ask honestly: could a capable student submit AI-generated work and pass? If the answer is yes, 𝘁𝗵𝗲 𝗾𝘂𝗲𝘀𝘁𝗶𝗼𝗻 𝗶𝘀 𝗻𝗼𝘁 𝗵𝗼𝘄 𝗱𝗼 𝘄𝗲 𝗰𝗮𝘁𝗰𝗵 𝘁𝗵𝗲𝗺. 𝗧𝗵𝗲 𝗾𝘂𝗲𝘀𝘁𝗶𝗼𝗻 𝗶𝘀 𝘄𝗵𝗮𝘁 𝗱𝗼 𝘄𝗲 𝗻𝗲𝗲𝗱 𝘁𝗼 𝗿𝗲𝗱𝗲𝘀𝗶𝗴𝗻.

𝘞𝘩𝘢𝘵 𝘪𝘴 𝘵𝘩𝘦 𝘰𝘯𝘦 𝘢𝘴𝘴𝘦𝘴𝘴𝘮𝘦𝘯𝘵 𝘤𝘩𝘢𝘯𝘨𝘦 𝘺𝘰𝘶 𝘢𝘳𝘦 𝘤𝘰𝘮𝘮𝘪𝘵𝘵𝘪𝘯𝘨 𝘵𝘰 𝘵𝘩𝘪𝘴 𝘺𝘦𝘢𝘳? 𝘐 𝘸𝘰𝘶𝘭𝘥 𝘨𝘦𝘯𝘶𝘪𝘯𝘦𝘭𝘺 𝘭𝘪𝘬𝘦 𝘵𝘰 𝘩𝘦𝘢𝘳 𝘪𝘵.

#AIinEducation #AssessmentDesign #HigherEducation #HealthcareEducation #AcademicIntegrity #FacultyDevelopment #GenerativeAI #DentalEducation #TeachingAndLearning #AIResilience #AIresilientAssessment

☕ Ce midi, nous recevions Lauranne Chaignon pour un nouveau "café doc" en ligne : Highly Cited Researchers : Anatomie d'une liste 📜

💡 Une présentation très éclairante basée sur l'article que Lauranne a publié dans Quantitative Science studies (QSS) pour retracer la trajectoire de la liste des HCR, ou comment d'une base de données destinée, entre autres, aux sociologues des sciences, cette liste est passée à un indicateur fragilisé par les méconduites scientifiques.

🎓 Lauranne Chaignon est doctorante au CSI Centre de sociologie de l'Innovation à Mines Paris - PSL. Elle rédige actuellement une thèse sous la direction de David Pontille : Contribution scientifique et manufacture de l’évaluation : le cas des chercheurs très cités.

ℹ️ A propos de son article :
La liste des chercheurs les plus cités (Highly Cited Researchers, HCR), publiée chaque année par Clarivate, occupe une place particulière dans le paysage universitaire, en raison de son utilisation dans le classement de Shanghai. L'article publié dans QSS examine l’évolution de cette liste, en s’appuyant sur les informations communiquées entre 2001 et 2023 par ses différents éditeurs (l’Institute for Scientific Information, Thomson Reuters et Clarivate) sur leurs sites web respectifs. Trois phases principales ont ainsi été identifiées dans son parcours. La première se caractérise par la création d’une base de données (2001-2011), la deuxième par l’affirmation d’un indicateur (2012-2018) et la troisième par l’affaiblissement d’une stratégie (2019-2023). L'analyse de cette trajectoire permet de mieux comprendre l'importance de cette liste et les défis auxquels elle est confrontée aujourd'hui, dans un contexte où certains des enjeux clés de l'évaluation de la recherche et de l'intégrité scientifique sont remis en question.

Vous voulez en savoir plus ?
Consultez l'article en ligne ici :
👉 Lauranne Chaignon; Highly Cited Researchers: Anatomy of a list. Quantitative Science Studies 2025; 6 305–327. doi: https://lnkd.in/eV9DXmgc

🫧 Pour suivre nos actualités, abonnez-vous à notre page ! 🫧

Les métriques Zenodo viennent de tomber : 260 vues, 58 téléchargements. Dans une démarche de recherche citoyenne et moins d'une semaine après la publication de mon preprint « Open Source 2.0 : From Open Source Software to Open Source Resources? », probablement un joli début ❤️‍🔥

Pour continuer la valorisation/promotion, lien article: https://lnkd.in/dVHusvF3

Avec déjà des échanges au sein de communautés qui ont plutôt tendance à confirmer ma thèse autour des « ressources open source », ces ressources numériques qui fournissent leurs fichiers source.

J'ai diffusé ces travaux dans une variété de communautés au sein de l'éducation ouverte, de la science ouverte et du logiciel ouvert en plus d'une publication sur LinkedIn et Mastodon.

Un bon intérêt sur ce concept de ressources open source du côté de l'open education où Alex Klinger, Alan Levine, Paola Corti et Italo Vignoli (cc Gilbert C. FAURE) sont alors venus me partager leurs problèmes/enjeux autour de la gestion des fichiers source et de leurs formats: https://lnkd.in/dKBQg5UR

Du côté science ouverte, 4-5 réactions du genre 👍 ou 👀. Des partages notamment auprès de la communauté Framework for Open and Reproducible Research Training (FORRT), des groupes linkedin comme l'Open Science Community Madrid ou Transform to Open Science de la NASA.

Espérons qu'ils soient en train d'analyser ça! Et comme c'est aujourd'hui le fonctionnement de la recherche toujours à la recherche d'évaluation par les pairs: https://lnkd.in/dXW8GH7B

Du côté des informaticiens, c'est intéressant. J'ai eu le droit à mon débat autour de l'Open Source Initiative et de leur Open Source Definition (OSD) dans la mailing list Teaching Open Source ! Lien échange: https://lnkd.in/d5pWxrZW

Avec un enseignant-chercheur, Dr. Bryan G. Behrenshausen, dont je viens de découvrir à la rédaction de cette publication qu'il est/était Senior Open Source Program Manager à GitLab. Je crois que j'ai bien défendu mon argumentaire dans cet échange intellectuel. Pas d'accords sur les terminologies, mais pas non plus de contradiction sur ma théorie autour des fichiers sources pour les autres typologies de ressources numériques.

Dans cette tradition, on m'a invité à appeler ça autrement ce à quoi je réplique que cela semble une évolution naturelle, déjà existante, avec une fin cordiale en concluant avec Descartes. Ils restent enfermés dans leurs convictions, je l'ai probablement déstabilisé, mais ces soutiens d'acteurs du logiciel open source à l'OSD me paraissent plutôt simples à contredire. Des arguments d'autorité avec un raisonnement un peu fragile...

Défi de ce vendredi : présenter ces réflexions au directeur de l'OSI qui a eu la mauvaise idée de proposer des créneaux de visio! Souhaitez-moi bon courage, j'ai quand même un peu peur du big boss 😅

Le meilleur prédicteur du succès d’un article scientifique n’est pas le prestige de l'université, l'expérience du chercheur, la méthodologie utilisée, ni même le talent de vos collaborateurs.

Le meilleur prédicteur... c'est les personnes que vous remerciez à la fin.

Cette découverte étonnante vient de paraître dans PNAS. Les chercheurs ont mesuré l'impact de 130 000 articles publiés par 86 000 chercheurs (à l'aide de différentes mesures comme le H-index ou le score Euclide).
Résultat : le meilleurs prédicteur de succès, c'est le réseau des gens qui apparaissent dans les remerciements – ce petit “merci à Jean-Michel pour ses précieux commentaires” que personne ne lit à la fin.

C'est interessant car ça montre à quel point les discussions de couloir, les retours rapides sur une intuition, les conseils donnés après un séminaire ou les échanges à la machine à café sont importants.

Je m'étais déjà fait la remarque à titre personnel. Ces petits bavardages informels, c'est comme une couche invisible de soutien intellectuel qui circule en dehors des structures officielles et qui fait émerger les meilleurs idées.

Bref, veillez sur votre réseau informel, c’est une véritable infrastructure cognitive. | 58 comments on LinkedIn

Congressman Greg Murphy, MD wrote: “ Unless our Medical Schools do a better job screening admissions candidates, we won’t have any doctors. If you don’t want to practice FULL time for at least 20-25 years, pick another profession.”

His comment was in response to a new study which shows 20% of physicians quit after 5 years.

My dad practiced gastroenterology for more than 40 years. His physician friends mostly did the same. Some talked about retiring for a decade. Some retired and came back multiple times. Some “cut back” to a schedule that still looked suspiciously like full time. You could not get them to stop.

That older model of medicine clearly had pride, purpose, identity, and endurance built into it.

But something has changed.

Maybe younger physicians are less willing to sacrifice their whole lives for the job. Maybe the job itself has become harder to recognize. Maybe full-time clinical medicine now includes so much inbox work, administrative friction, moral injury, and corporate pressure that comparing generations misses the point.

I don’t think the answer is simply screening medical students for who will tolerate the most pain.

The better question is why so many talented people are leaving so early.

What changed: the doctors, the job, or the deal medicine made with them?

Nisha Mehta, MD | 293 comments on LinkedIn

📊 Introducing the h* Index · Research Efficiency & Scientific Momentum

➡️ The classical h-index is widely used internationally as an indicator for assessing scientific impact by combining productivity and citation performance into a single metric.

➡️ However, research careers are dynamic by nature. Scientific activity evolves over time through new collaborations, emerging research directions, teaching commitments, leadership responsibilities, interdisciplinary projects, and many other forms of academic contribution. To better capture this evolving dimension of research activity, we developed the h* index, an extension of the traditional h-index that integrates temporal aspects of a researcher career. The h* index takes into account ⤵️

🔹 active publication periods
🔹 citation dynamics relative to career length
🔹 the continuity and momentum of scientific contributions over time

By incorporating these elements, the h* index provides a more dynamic and continuously updateable perspective on scientific impact and research activity.

➡️ Our objective is not to replace existing bibliometric indicators, but to complement them with an additional tool capable of offering a broader and more nuanced understanding of academic trajectories and research evolution. We believe that combining robustness, adaptability, and temporal analysis can contribute to more informed research evaluation and support decision-making processes within academic and scientific institutions.

📢 “An indicator does not say who you are, it says what others have remembered about you. The h or h* index does not measure the truth, it measures how many times an idea has managed to survive through time without fading into oblivion.”
Citation : Prof. Alexis Rusinek

📢 "Wskaźnik nie mówi, kim jesteś mówi, co inni o tobie zapamiętali. Czynnik h czy h* nie sprawdza prawdziwości, lecz pokazuje, ile razy dana myśl zdołała utrzymać się w obiegu, zamiast zniknąć w zapomnieniu."
Polish translation by Prof. Tomasz Jankowiak

🌐 If you would like to explore and visualize the evolution of your own scientific trajectory over time, you can now test it through our platform.

🌐 https://lnkd.in/euRk2xMK

Aurélien Besnard #Tomasz_Jankowiak

#Research #Scientometrics #Bibliometrics #Innovation #Academic_Research #Data_Science #Higher_Education #Science #Research_Evaluation

AI in medical education may be dangerous not only when it is wrong: it may be dangerous when it is "plausibly" wrong.

A very interesting study published in npj Digital Medicine examined the impact of AI-generated explanations on novice medical students, comparing correct explanations, misleading explanations, and no explanations.
The result is important because it touches one of the most delicate aspects of AI in medicine: the persuasive power of a well-structured explanation.
In clinical reasoning, an error that appears obviously wrong can often be recognized, questioned, and rejected. But an error that is coherent, fluent, and medically sophisticated is much more difficult to resist, especially for learners who have not yet developed strong internal models of disease.

Medical education is not only about reaching the correct answer: it is about learning how to evaluate the path that leads to that answer. If AI provides a wrong conclusion with a convincing explanation, the risk is not only diagnostic error. The deeper risk is that the learner may begin to trust the structure of reasoning without having the expertise to detect where that reasoning fails.

This is why AI literacy cannot simply mean knowing how to use AI; it must mean knowing how to interrogate AI.

Students and trainees will need to learn not only how to obtain explanations from artificial intelligence, but how to deconstruct them, compare them with biological plausibility, identify hidden assumptions, and recognize when confidence is not justified.
The future of medical education will not depend only on giving students access to better tools: it will depend on teaching them how to remain intellectually independent in front of tools that sound increasingly authoritative.
AI can support learning but it must not become a machine that teaches confidence before competence.
#ArtificialIntelligence #Medicine #MedicalEducation #ClinicalReasoning #DigitalHealth #MedicalAI #PatientSafety #FutureOfMedicine | 11 comments on LinkedIn

Today on Indicator: A handful of anonymous bots have taken over fact-checking on X.

This isn’t hyperbole. In the first three weeks of May, just eight AI contributors wrote 50.3% of all visible Community Notes on the platform.

Community Notes was supposed to bring scale and legitimacy to the fight against misinformation by empowering users to participate in fact-checking. Now that power is increasingly in the hands of a tiny group of hobbyists and researchers.

AI contributors aren’t just replacing the humans who are supposed to make up the community in Community Notes. They are publishing in languages and on topics that reflect the bias of their algorithmic setup.

My analysis suggests, for example, that AI supercontributors focus less on politicians and hyperpartisan accounts. This may be because they are optimizing for a bridging algorithm that prioritizes broad consensus over factuality.

For good and for bad, X is currently running the world's largest live experiment in automated fact-checking.

What happens next will matter not just for X, or for the copycat Community Notes features on Meta and TikTok. How it fares may define platform interventions against misleading content for years to come.

This 3,634-word article took a lot of time to report. To paraphrase Jeb Bush, please read: https://lnkd.in/ejBU9uAE

🧠 AI in Medical Training: Learning Tool or Shortcut?

A new Nature Medicine Perspective raises a cautious but important concern: “never-skilling” in medical education. The idea is distinct from deskilling, that may happens when experienced clinicians lose skills they already developed. Never-skilling would happen earlier: during formative training, when learners may fail to build foundational clinical reasoning.

→ AI is not i harmful to learning, however its effect depends on how and when it is introduced.

→ Answer-delivery AI may carry different risks from learning-mode AI.
A system that gives diagnoses directly may bypass cognitive effort, while a system that asks Socratic questions and gives feedback may support reasoning.

→ Assessment needs to change, as trainees are learning in AI-rich environments.

Caveats: direct causal evidence in clinical trainees is still lacking. This is a risk model and research agenda, not a proven phenomenon.

👉 How do we leverage AI as a learning accelerator without turning it into a shortcut around the skills physicians still need to master?

Ref: Ke et al Nan Liu, PhD, FAMIA . AI-induced never-skilling in medical education. Nature Medicine, 2026.