Since August, much of what the major AI systems produce carries an invisible watermark. In this episode I explain what that mark is, how it gets built into text and images, and why new European rules pushed companies to adopt it this summer. I separate watermarks from the AI detectors many educators already use, look at how both can be removed or faked, and consider what a mark can tell us when it turns up on a student essay, a manuscript figure, or a medical image - and perhaps more importantly, what it cannot tell us.

Links related to this episode

If a learner rehearses a difficult conversation with an AI, what exactly are they practising against? A listener asked, and the answer runs through two layers of embedded value: the persona an educator or student writes, and the model generating everything underneath it. This episode covers what the survey evidence shows about that second layer, why the stakes rise sharply as an encounter becomes more values-laden, and the case for teaching learners to correct for the bias rather than avoid the tools.

Links related to this episode

What problem did we say we were solving, and what would we have done about it if AI had never existed? That is the question we'll tackle this episode. Not whether AI tools work — they do, but that is what makes the cases interesting. Solutionism is the error of letting the available solution decide which problem we have, and once you know that it is easy to see AI solutionism all around us. So, why does it keep happening? Well, that has less to do with credulity and more with the fact that procurement is easier than reform.

Links related to this episode

Note: This episode was recorded on July 28, 2026 and discusses ongoing litigation which may have changed by release. On July 22, 2026, a Florida pastor filed suit against OpenAI, alleging ChatGPT talked him out of seeking care for what turned out to be a pulmonary embolism. On July 23, OpenAI made its health product generally available to every adult in the United States. Those two dates are a fair picture of where we are.

46% of Canadian adults have asked an AI chatbot for medical advice in the past year. 42% of the people who used one for a physical health question never went on to speak to a clinician at all. And nearly half of patients say they have hidden that use from their doctor, or would.

This episode looks at what the evidence actually shows about the public using generative AI for health information: where these tools are right, where they fail, why the failures are hard for patients to see, and what none of our professional bodies have told us yet.

Links related to this episode

Many of you have written in asking about OpenEvidence. In this episode I walk through what's actually inside the AI clinical search tool that didn't exist before 2023 and is now used by an estimated two-thirds of US physicians. I cover how it differs from ChatGPT, UpToDate, and Google Scholar, how its retrieval-augmented architecture works, where its content comes from, and how the two search modes, standard Consult and Deep Consult, perform in the emerging evaluation literature. I take on the trust question directly and land on a five-question framework for evaluating any AI evidence tool. And I close with what we as a profession aren't talking about enough: the enterprise Epic integrations at Mount Sinai, Sutter Health, and Cedars-Sinai, patient awareness, structural dependence on a single tool, and the business model.

The examples are clinical because OpenEvidence is a clinical tool. But the framework and the harder questions apply as much to the AI tools you use in research and teaching, and I've tried to make that portability explicit.

Most conversations about AI in medicine assume a patient in the room. This episode doesn't. Jessalyn talks with Joshua Samsoondar, a pathology resident at the University of Calgary, about what AI looks like from inside a non-patient-facing specialty — screening lymph nodes for cancer, using AI to catch up on unfamiliar terminology, and the tradeoffs built into decisions most patients never see happen. Josh makes the case that AI can sharpen efficiency and training without replacing the judgment pathologists bring to the work. For clinicians, researchers, and educators alike, it's a look at a part of medicine that AI is already reshaping, quietly.

Reach out to Dr. Samsoondar at [email protected]

Most conversations about AI in medical education are led by faculty. This one isn't. Hailee Rochon, a registered nurse and a student in the University of Calgary's Master of Physician Assistant Studies program, got in touch because she wanted to share how AI has changed the way she learns.

For Hailee, the value isn't speed or shortcuts. It's self-direction. She describes using AI to follow her own curiosity further than a syllabus allows, and to work on the specific areas where she knows she's struggling, at her own pace. It's a picture of a motivated learner setting the agenda, with AI as the tool that makes that practical.It's also a considered picture. Hailee is clear about where she doesn't rely on these tools, and how she checks what they tell her. This episode offers a chance for faculty to hear how at least one student is really learning with AI.

When a colleague was using an AI tool and it unexpectedly swore at him, his reaction caught him off guard. It wasn’t amusement or confusion — it was genuine discomfort. And that raised a question worth exploring: why do we have such strong emotional responses to AI behaviour?

In this episode, Dr. Kannin Osei-Tutu and I dig into the research on how humans relate to AI systems. We cover the CASA paradigm — the finding that we automatically apply social rules to computers the same way we do to people — and what happens when the “character” we’ve built for an AI tool suddenly breaks. We discuss the uncanny valley effect in text-based AI, the paradox that making AI feel more human-like can backfire, and the flip side of the coin: automation bias, where we trust AI too much.

Kannin reflects on what his experience revealed about his own assumptions, and we close with a challenge: pay attention to your emotional reactions when using AI tools this week. What patterns emerge? What character have you built?

For decades, we've understood human reasoning through two systems: the fast, intuitive one and the slow, deliberate one. But that framework was built before AI became a thinking partner. In this episode, I sit down with Dr. Steven Shaw — a Canadian scholar and postdoctoral fellow at the Wharton School — to talk about his new framework, Tri-System Theory, and what it means that AI now functions as a third cognitive system operating outside the brain.

Shaw coined the term "cognitive surrender" to describe what happens when we adopt AI outputs without critical evaluation — not as a deliberate choice, but as a quiet default. We get into how it differs from simply using AI as a tool and what it looks like across clinical documentation and graduate training. Plus a practical challenge to close.

Dr. Shaw's preprint

Dr. Shaw's website

Knowledge at Wharton podcast

Delegating to an AI agent is not the same as prompting a chatbot. In this episode, I walk through what good delegation actually looks like — scope, constraints, and checkpoints — and apply it to three concrete tasks: updating your CV, building a simple course assistant, and screening abstracts for a scoping review. The framework is practical, the failure modes are real, and the limits matter.