At the threshold of human and AI healthcare.

This episode marks the end of Series 1 of the LiminalX podcast. The largest conference for medical physicists occurs next week in Vancouver. This episode is dedicated to attendees of the conference and provides an overview of ways to learn about the readiness and adoption of artificial intelligence in the radiation oncology field.

Healthcare has always been a team effort and the introduction to AI is no different. In this episode we explore what we can learn from engineers and physicists who maintain the performance of hardware and see what principles we can translate to challenges we currently face in AI.

History tends to repeat itself. In this episode, we'll investigate what lessons from other major technological adoptions like electricity or the Internet we can apply to the adoption of Artificial Intelligence, and establish three personal rules that can help each one of us navigate this technological change in a health care environment.

Most healthcare professionals think they're ready for AI. They're not — and the gap between what they know and how they actually behave when AI is in the room is costing patients. In this episode, Paul Naine draws on two decades of clinical technology adoption to make a case that nobody in healthcare is making clearly enough: AI isn't a product. It's a general-purpose technology. And we've been trying to introduce it like a toaster upgrade. We cover why the standard adoption playbook fails, what the research says about how humans actually respond to AI in clinical contexts, and why the behaviour pillar — not knowledge, not skill — is where healthcare AI lives or dies. One story at the end will make you question whether your trust in AI is as well-calibrated as you think.

After watching a autonomous taxi get stuck in a flood last week, we examine why training data is so important for AI and what three areas we have to be particularly vigilant about when it comes to AI in healthcare. Join us, ask better questions, and comment if you notice something interesting about this episode.

What does an FDA clearance label actually tell you? Efficacy vs effectiveness — the sticker problem in clinical AI, and why regulatory approval is necessary but not sufficient.

Healthcare is a resource-constrained system. AI that works but costs more than it saves doesn’t get adopted. Efficiency is the entry ticket — necessary, but getting in the door is not the endgame.

The disclaimer. “AI-generated content may be inaccurate.” Who put that there, and what does it actually mean? Accountability doesn’t disappear when AI enters the room — it just gets harder to locate.