An AI habit more people should use: force disagreement. When a decision matters, I don’t ask one AI. I set up a few. The yes voice. The no voice. The numbers brain. The context brain. One AI is good at echoing your thinking. A small group that’s built to push back is how you get better answers.

The hardest part of AI transformation is not choosing the right tools. It is redesigning governance systems built for a slower, more static healthcare era.

I don’t use AI to get answers. I use it to argue with my ideas. I’ll toss out something half-baked and say, poke holes in this. What’s weak here? I’m not trying to win the argument. I’m trying to see what survives.

When AI enters healthcare, nothing stays where it was. How work flows, how data matters, and how responsibility is carried all shift. Delays are not inefficiency. They are renegotiation.

At this stage, healthcare AI involves unavoidable tradeoffs. So here’s the uncomfortable question: Which is worse, an alert that shouldn’t have fired, or a case the system never flagged?

AI won’t decide our future. Our habits around it will. Who experiments. Who stays curious. Who waits. Most outcomes won’t come from a single breakthrough. They’ll come from daily use, or daily avoidance.

In medicine, keeping a human in the loop is vital, but where does that loop begin and end? As AI becomes a 2nd opinion, mastering it may be part of competent care. The future standard: empathy from humans, data from machines. Ignoring AI? That might become negligence.

You can’t master a new paradigm while staring at a deadline. If you want your team to actually leverage AI, give them the breathing room to fail with it first. What would happen if you gave your team one hour this week to explore AI with no deliverables attached?

AI systems can now flag people as “high risk” for diseases long before they feel any symptoms. How does that shift who we treat as “sick” and what does that do to someone’s identity, stigma, and life choices before they’re actually ill?

This study exposes the 'autopilot trap,' but also the way out. While passive reliance degrades skill, active collaboration sharpens it. Doctors remain in the driver’s seat by treating AI as a coach rather than a crutch. The future isn't about being tethered by tech it’s about being amplified by it.

We must design AI that artificially re-introduces friction for learners, or we will inherit a generation of doctors who are merely liability sponges for algorithms.

If AI cheapens specialized medical knowledge (doctor's value), but elevates non-automatable human skills (nurse's value)... Will nurses soon earn more than doctors?

This older paper shows that matching human accuracy is not enough for AI in healthcare. AI systems face higher trust expectations, and their mistakes often trigger more concern than human errors. This is due to AI's lack of transparency, adaptability, and social presence.

AI-assisted tasks completed using Claude show an estimated 80% reduction in task time, which could potentially double US labor productivity growth over the next decade if widely adopted.

My new article provides a field report on AI's transition, confirming 2025 as the year progress accelerated from the lab towards the ward, now caught "Between Beta and Breakthrough." janbeger.substack.com/p/healthcare...

If you're using Al to save time, that's average. Use it to expand your ideas? Now you're dangerous.

Critical gaps in AI radiology research … #RSNA25

Healthcare AI hit a quiet tipping point in '25. Multimodal & early agentic systems moved from theory into pilots, shifting expectations across the field. This piece is a grounded snapshot shaped by my LinkedIn community. substack.com/@janbeger/no...

The future of care will belong to people who keep their minds active and let technology amplify their judgment rather than replace it. That’s where the real advantage will be, both for clinicians and for the patients who trust them.

In 2025, healthcare AI stopped being an experiment and became an environment.

Which one was your Halloween costume this year?

Fanboy moment today. Got to meet Ethan Mollick in person. I’ve been following his work for a while, so actually getting to say hi and thank him felt pretty great.

Scaling access to healthcare doesn’t erase the fact that people still need to be seen, heard, and held by humans. Capability isn’t care.

Just saying …

AI will increasingly handle cognitive labor: diagnosis, prediction, triage, documentation, pattern recognition. As machines absorb complexity, the value of the clinician doesn’t disappear. It shifts.

🚀 Working on a pro bono side project to teach kids about AI through an interactive book + website. Looking for someone to help build & lightly maintain the interactive site. Know any devs who might be interested? Great opportunity to do good + gain experience. DM me! 🙌

For years, everyone argued about whether AI would replace radiologists. They missed the point. The real shift isn’t about replacing the reader - it’s about rebuilding the reading room. Every step, from image acquisition to report, is being re-engineered around intelligence.

Efficiency gains don’t always lead to labor reduction; lower cost or faster throughput may expand use. This feedback loop is essential to model when estimating economic and workforce impacts of AI in healthcare.

Automation is a tool, not a goal. We should not automate just for the sake of it. Used well, automation can document processes and reduce human error, but it must serve a clear purpose. It should never be the default, no matter how clever or efficient it seems.

AI hasn’t crossed the line into independence on any radiologist task. Every core duty still requires human oversight.

With GPT-5, this is the official end of the “blank page” problem. Now the problem is having too many fully formed ideas to choose from!

AI isn't one magic box—it's layers stacked up: - Infrastructure (data, GPUs, PyTorch) - Perception (vision, hearing) - Cognition (understanding, language) - Decision-making (planning, acting) GenAI simplifies the stack—but underneath, infrastructure still matters more than ever. (Xu et al. 2021)

If someone tells you, “AI is better than doctors,” pause and ask: in what way? Then ask, does that mean it should replace them—or help them? Because better on paper doesn’t always mean better for people.

Reading built the modern world. Without AI literacy, we risk losing control of the next.

#HelloAI Basics is now completely free. And we’re offering the Advanced and Professional modules at 50% off. No catch. Just our way of helping more people get started — or go deeper — with practical, real-world AI education built for healthcare. Explore it here: www.helloaiprofessional.com

Scaling AI in healthcare feels a bit like navigating a global airport — lots of moving parts, different systems trying to talk to each other, and the need for clarity at every checkpoint. But we’re getting there. And the more we build for interoperability and trust, the smoother the journey.

AI doesn’t care what kind of doctor you are. It sees patterns, not departments. The old walls in medicine? Blurring. If we’re thoughtful, AI can bridge gaps and give time back to those who care. Let’s build tech that adapts to people — not the other way around.

Design like your user is tired, distracted, and scared. Because they are.

Al's biggest challenge in healthcare isn't accuracy. It's acceptance.

arXiv, the free research archive, has transformed science for 35+ years—hosting 2.6M+ papers & 20K new uploads/month. Despite its outdated UI, it fuels breakthroughs & open access. Big 2022 funding is modernizing it for the future. #OpenScience #arXiv

Rather than imitating human warmth, AI should earn trust through competence, reliability, and alignment with medical values. Artificial empathy is not the answer—responsibility is.

AI moves fast. Medicine moves slow. That friction is necessary. Deliberation, oversight, and caution aren’t barriers to innovation—they are what make AI in healthcare worth trusting.

The Imaging Wire just dropped its top 40 radiology resources for 2025—must-know websites, voices, and platforms shaping the field. A go-to guide for AI, trends, and expert insights. I learn from them every day. 🌐⇢ lnkd.in/dF5dkkxS

AI in healthcare cannot just be accurate—it must be accountable: Performance metrics alone don’t build trust. Transparent design, clear responsibility, and continuous monitoring do. AI must earn trust, not demand it.

Trust in traditional healthcare is built on the human soul; trust in AI is built on the human design of a non-human actor.