Latest / Paths of Curiosity in an AI World / Paths of Curiosity in an AI World: AI in Healthcare Recap

Transcript

Paths of Curiosity in an AI World: AI in Healthcare Recap · Paths of Curiosity in an AI World

• 0:04
  SPEAKER_05: This is Paths of Curiosity in an AI World, where we explore the careers being built right now at the intersection of AI and the real world.
• 0:21
  SPEAKER_04: Five lives a month.
• 0:23
  SPEAKER_04: That's what one algorithm is saving right now in a single emergency department in San Diego.
• 0:31
  SPEAKER_04: Five lives per month from one hospital.
• 0:36
  SPEAKER_04: One hospital.
• 0:42
  SPEAKER_04: Those are five people who would not have walked out of the hospital before.
• 0:48
  SPEAKER_04: Now they make it home.
• 0:52
  SPEAKER_05: That's not a research finding.
• 0:54
  SPEAKER_05: That's someone's parent.
• 0:56
  SPEAKER_05: Or someone's kid.
• 1:00
  SPEAKER_04: Over the last several months, I've sat down with six people working at the intersection of AI and healthcare: a hospital CEO, an emergency physician, a Stanford pediatrician, a policy researcher who helped write federal privacy law, a UCSF professor who studies whether AI tools actually work once they're in the real world, and a Carnegie Mellon engineer who has never once treated a patient and his work might just end up in your AirPods.
• 1:35
  SPEAKER_04: I'm Malora and I'm Acadia, and this is the AI and Healthcare mini-series recap.
• 1:50
  SPEAKER_05: Before we get into the guests, give me the honest version because AI and healthcare could mean almost anything.
• 1:56
  SPEAKER_05: What is actually happening?
• 2:00
  SPEAKER_04: The short version, hospitals have been digitizing everything for about 25 years.
• 2:07
  SPEAKER_04: Every patient visit, every lab result, every prescription, every doctor's note.
• 2:13
  SPEAKER_04: All of it turned into data.
• 2:15
  SPEAKER_04: Infrastructure, regulation, sheer volume.
• 2:19
  SPEAKER_04: For most of that time, it was collected and not used to anywhere near its potential.
• 2:25
  SPEAKER_04: But now AI can process that data in real time at a scale that wasn't possible before.
• 2:32
  SPEAKER_04: And that changes what's actually achievable.
• 2:40
  SPEAKER_04: Dr.
• 2:40
  SPEAKER_04: Chris Longhurst, he's now the CEO of Seattle Children's Hospital, and before that, he led digital health at UC San Diego for years.
• 2:50
  SPEAKER_04: He made a point that reframes the whole conversation for me.
• 2:56
  SPEAKER_08: And so I'd say one of the most exciting things about AI is restoring humanism in medicine.
• 3:03
  SPEAKER_08: The ability to take an AI recording, have it smartly transcribe, not just word for word, but actually put together an intelligent clinician note that saves the doctor that time at the keyboard.
• 3:16
  SPEAKER_08: Matt even tee up orders, make them more efficient.
• 3:19
  SPEAKER_08: And ultimately improves the patient experience because you're restoring humanism that we've taken away as we've tried to digitize and make care safer.
• 3:31
  SPEAKER_05: So the argument is that digitizing medicine actually made it less human, and AI might fix a problem that earlier technology created.
• 3:42
  SPEAKER_04: Which is the opposite of how most people frame it.
• 3:46
  SPEAKER_04: Most people assume AI is the threat to human connection in medicine.
• 3:51
  SPEAKER_04: He's arguing it might be the restoration of it.
• 3:55
  SPEAKER_05: What does that look like day to day?
• 3:58
  SPEAKER_04: Doctors finishing their notes before they leave the building instead of staying up until midnight.
• 4:05
  SPEAKER_04: Being able to make eye contact during an appointment instead of staring at a keyboard.
• 4:11
  SPEAKER_04: He runs a hospital that operates almost like a NASA command center.
• 4:16
  SPEAKER_04: Walls of screens showing every patient, every bed, every staff movement in real time, with an AI layer on top to help route patients, flag risks, and catch things early before they become emergencies.
• 4:33
  SPEAKER_04: That sounds like a television show.
• 4:36
  SPEAKER_04: It kind of is, except when it catches something the doctors would have missed at 2 in the morning.
• 4:42
  SPEAKER_04: The stakes are very real.
• 4:48
  SPEAKER_05: Let's talk about the emergency department, because that's where the number from the cold open comes from.
• 4:56
  SPEAKER_04: Episode 3.
• 4:57
  SPEAKER_04: Dr.
• 4:58
  SPEAKER_04: Gabriel Warty.
• 4:59
  SPEAKER_04: Emergency and critical care physician at UC San Diego.
• 5:04
  SPEAKER_04: He works in both the ER and the ICU.
• 5:07
  SPEAKER_04: And his entry into medicine is worth noting for students because it was not planned.
• 5:13
  SPEAKER_04: He was studying biomedical engineering at Georgia Tech sophomore year, and someone said, You should look at medicine.
• 5:21
  SPEAKER_04: He volunteered at a hospital in Atlanta with no connections, no family in the field, just walked in and asked.
• 5:31
  SPEAKER_04: And that was it.
• 5:34
  SPEAKER_04: That was it.
• 5:35
  SPEAKER_04: Now he's one of the people leading AI implementation in emergency care.
• 5:40
  SPEAKER_04: The condition he focused on is sepsis, which I think most people underestimate.
• 5:48
  SPEAKER_05: Explain it for the people who haven't heard the episodes.
• 5:53
  SPEAKER_04: Sepsis is what happens when your body gets an infection and your own immune system starts attacking your organs.
• 6:00
  SPEAKER_04: Your heart, your lungs, your kidneys.
• 6:03
  SPEAKER_04: It kills about 350,000 people a year in the United States alone.
• 6:10
  SPEAKER_04: Globally, it's closer to 10 million.
• 6:13
  SPEAKER_04: And the reason it's so hard to catch is that it doesn't have one obvious symptom.
• 6:19
  SPEAKER_04: There's no movement like a heart attack where someone clutches their chest.
• 6:24
  SPEAKER_04: It creeps in, and by the time it's obvious, serious damage is already done.
• 6:31
  SPEAKER_05: So timing is everything.
• 6:36
  SPEAKER_04: His team built a machine learning model that feeds into patients' vitals, their labs, their demographics, their history, and flags someone as likely developing sepsis for the six hours before a doctor would otherwise catch it.
• 6:52
  SPEAKER_04: They deployed it and they measured the outcomes.
• 7:11
  SPEAKER_05: That's not a simulation.
• 7:13
  SPEAKER_05: That's a measured result from a real hospital.
• 7:17
  SPEAKER_04: And he was honest about what could go wrong because he's seen it.
• 7:21
  SPEAKER_04: The failure mode isn't usually the algorithm, it's the implementation.
• 7:26
  SPEAKER_04: He said it plainly.
• 7:58
  SPEAKER_04: The comparison he made was using an AI tool to do your homework.
• 8:04
  SPEAKER_04: You might get the right answer, but you haven't built the thinking.
• 8:08
  SPEAKER_04: In medicine, when the AI is wrong, you need something to fall back on.
• 8:16
  SPEAKER_05: Which is probably true in every field right now, not just in medicine.
• 8:32
  SPEAKER_04: Dr.
• 8:32
  SPEAKER_04: Julia Adler Milstein at UCSF.
• 8:35
  SPEAKER_04: She studies whether AI tools actually work once they're in the real health system with real patients and real clinicians under pressure, not in a controlled lab.
• 8:48
  SPEAKER_04: And her research is surfacing something uncomfortable about the assumption behind most AI deployment right now.
• 8:56
  SPEAKER_05: Which is that a human will catch it when AI is wrong.
• 9:01
  SPEAKER_04: Right.
• 9:01
  SPEAKER_04: The safety net in most systems is the doctor reviews the AI output and intervenes if something's off.
• 9:09
  SPEAKER_04: But she's been researching whether humans are actually capable of doing that reliably.
• 9:14
  SPEAKER_04: And the evidence is not reassuring.
• 9:35
  SPEAKER_05: The field is actively working on this, which feels important to say out loud.
• 9:40
  SPEAKER_04: She was very clear about that.
• 9:42
  SPEAKER_04: It's called automation bias, and Dr.
• 9:44
  SPEAKER_04: Pagelard named the mirror problem too: automation neglect, where experienced physicians dismiss what the AI flags because they trust their own instincts more.
• 9:55
  SPEAKER_04: Both directions create risk.
• 9:58
  SPEAKER_04: And somewhat comically, her leading solution is using AI to check the AI.
• 10:04
  SPEAKER_04: One model acting as a judge of another model's output.
• 10:08
  SPEAKER_04: Not perfect, but neither is a tired clinician at the end of a long shift.
• 10:26
  SPEAKER_05: Episode 4 brought in a dimension that I don't think gets nearly enough attention in these conversations.
• 10:35
  SPEAKER_04: Dr.
• 10:35
  SPEAKER_04: Natalie Pagler at Stanford, pediatric intensivist and clinical informaticist, meaning she spent years in pediatric ICUs and now runs the Clinical Informatics Division.
• 10:48
  SPEAKER_04: AI in Pediatrics is not just a scaled-down version of AI in adult medicine.
• 10:54
  SPEAKER_04: It's genuinely different.
• 10:57
  SPEAKER_05: Because kids are not just small adults.
• 11:01
  SPEAKER_04: Not even close.
• 11:03
  SPEAKER_04: The data problem alone is significant.
• 11:06
  SPEAKER_04: Kids are healthier than adults generally, so there is far less clinical data to train models on.
• 11:12
  SPEAKER_04: And within pediatrics, a six-month-old and a 15-year-old are physiologically so different that a model trained on one age group can fail completely on another.
• 11:24
  SPEAKER_04: She gave this example.
• 11:34
  SPEAKER_04: Now teenagers are walking into clinics with arrhythmia alerts on their watches, and the clinicians have to figure out whether that signal is meaningful for that age.
• 11:47
  SPEAKER_04: Which is happening constantly right now.
• 11:50
  SPEAKER_04: But the bigger point she kept returning to is the timeline.
• 11:54
  SPEAKER_04: For pediatrics, a decision made when someone is eight years old can affect their health at 40.
• 12:00
  SPEAKER_04: This raises the stakes on getting it wrong.
• 12:03
  SPEAKER_04: But that also means the upside is enormous.
• 12:08
  SPEAKER_02: Interventions that we make in childhood can last a lifetime.
• 12:12
  SPEAKER_02: And so that can be, you know, you have to think about it cautiously because it could potentially negatively affect a lifetime.
• 12:20
  SPEAKER_02: But also it's a huge opportunity.
• 12:22
  SPEAKER_02: So if we make interventions in childhood, that could lead to years of healthier and more productive lives.
• 12:31
  SPEAKER_05: She also had that example about the appointment no-show model that I keep coming back to.
• 12:38
  SPEAKER_04: It's the clearest illustration of how identical AI tools can produce completely opposite outcomes depending on how you choose to use them.
• 12:48
  SPEAKER_04: Some health systems built a model that predicts which patients are likely to miss appointments, and use it to overbook their clinical schedules, keeping things efficient.
• 13:02
  SPEAKER_05: Which sounds reasonable until you ask who is most likely to miss the appointment.
• 13:18
  SPEAKER_04: So you've taken an AI model and used it to make care harder to access for exactly the people who already have the least.
• 13:27
  SPEAKER_04: At Sanford Children's, they took the same model and flipped it.
• 13:31
  SPEAKER_04: They call those families and ask what's in the way.
• 13:35
  SPEAKER_04: Can we convert this to a telehealth visit?
• 13:38
  SPEAKER_04: Can we help with transportation?
• 13:41
  SPEAKER_05: Same data, completely different impact, depending on the values behind the design decision.
• 13:50
  SPEAKER_04: And someone has to make that decision.
• 13:53
  SPEAKER_04: It's not a technology question.
• 13:56
  SPEAKER_04: That's a people question.
• 13:58
  SPEAKER_04: And that person could be a student listening to this right now.
• 14:05
  SPEAKER_05: There is one more piece of this that I don't think we can skip because it's the most tangible thing in the whole series for a student right now.
• 14:15
  SPEAKER_04: Dr.
• 14:15
  SPEAKER_04: Mayan Coel at Carnegie Mellon.
• 14:18
  SPEAKER_04: He's never treated a patient.
• 14:20
  SPEAKER_04: He's a computer scientist whose entire career is about turning consumer devices into medical tools.
• 14:28
  SPEAKER_04: And his honest assessment of where wearables are right now is a little deflating.
• 14:33
  SPEAKER_04: He said most people stop looking at their step count within a few months.
• 14:38
  SPEAKER_04: The heart rate data sits unused.
• 14:41
  SPEAKER_04: There is enormous potential in the devices we already carry, and we are barely using it.
• 14:49
  SPEAKER_05: So what is actually possible with what already exists?
• 15:00
  SPEAKER_04: A smartwatch that detects medication levels through your skin, sensors that catch early signs of illness before you feel any symptoms.
• 15:09
  SPEAKER_04: And right now, he's working on using AirPods and smartwatches to monitor children with ADHD during school.
• 15:16
  SPEAKER_04: Not cameras, just movement and posture data, giving doctors a real window into what's actually happening in a classroom.
• 15:32
  SPEAKER_04: That constraint comes up everywhere in this field.
• 15:35
  SPEAKER_04: Health data is inherently private, which means sharing it across research labs to build better models is genuinely difficult.
• 15:44
  SPEAKER_04: There are labs all over the world working on nearly identical problems and very little data flows between them.
• 15:51
  SPEAKER_04: It's one of the biggest bottlenecks right now.
• 16:06
  SPEAKER_00: So because we knew that okay, there is a champion in the field, a practitioner, who believes that there is value in solving this problem.
• 16:14
  SPEAKER_00: We didn't have a solution at that point, but we had some interesting ideas, and then we just ran with that.
• 16:20
  SPEAKER_05: Fifteen years running towards a problem you don't have a solution to yet.
• 16:25
  SPEAKER_05: That's a different kind of motivation than anything we've heard from the clinical guests.
• 16:31
  SPEAKER_04: And there's one more guest, Dr.
• 16:33
  SPEAKER_04: Erin Holvey.
• 16:34
  SPEAKER_04: She spent decades building the data system and writing the policy frameworks that make everything we just described possible.
• 16:42
  SPEAKER_04: She helped write the laws protecting your medical records.
• 16:46
  SPEAKER_04: We'll get to her later.
• 16:47
  SPEAKER_04: Her path and what she says about skills is some of the most practical advice in the entire series.
• 16:56
  SPEAKER_05: So that's what's happening right now.
• 16:58
  SPEAKER_05: The question is, who's doing it?
• 17:00
  SPEAKER_05: And none of the answers look like what you'd expect.
• 17:10
  SPEAKER_05: Let's talk careers.
• 17:12
  SPEAKER_05: Because I think this is where most people have the biggest blind spots.
• 17:18
  SPEAKER_04: Some of you might end up doing a job that doesn't exist yet.
• 17:23
  SPEAKER_04: What do you mean?
• 17:25
  SPEAKER_04: Well, Dr.
• 17:26
  SPEAKER_04: Longhurst and Dr.
• 17:27
  SPEAKER_04: Pagler are both in a field called clinical informatics.
• 17:32
  SPEAKER_04: The first board certification exam was in 2013.
• 17:36
  SPEAKER_04: Before that, the field didn't officially exist.
• 17:40
  SPEAKER_04: They didn't pick it from a list of options.
• 17:42
  SPEAKER_04: It emerged and they were ready for it.
• 17:46
  SPEAKER_05: And that's still happening right now.
• 17:51
  SPEAKER_04: AI governance.
• 17:53
  SPEAKER_04: People hired specifically to oversee whether AI tools are working fairly across different patients.
• 18:00
  SPEAKER_04: That role barely existed five years ago.
• 18:03
  SPEAKER_04: Nobody planned for that career at 17.
• 18:08
  SPEAKER_05: Which means there are fields being defined right now that will be fully established by the time students in high school today reach mid-career.
• 18:18
  SPEAKER_04: That's the thing I most want students to sit with.
• 18:21
  SPEAKER_04: This specific job you might end up doing in 35 years may not have a clear title today.
• 18:28
  SPEAKER_04: That is not a reason to panic.
• 18:31
  SPEAKER_04: It's a reason to build a broad, adaptable foundation rather than optimizing for a single track.
• 18:39
  SPEAKER_05: And what does that foundation actually look like?
• 18:43
  SPEAKER_05: Let's go through the guests.
• 18:45
  SPEAKER_05: Because how they got there matters as much as where they ended up.
• 18:49
  SPEAKER_05: None of these paths were straight.
• 18:54
  SPEAKER_04: Dr.
• 18:54
  SPEAKER_04: Longhurst, a pediatrician who took a year off between his third and fourth year of medical school to get a master's in healthcare informatics.
• 19:03
  SPEAKER_04: His school had just started the program.
• 19:06
  SPEAKER_04: He found a door that had just been unlocked and walked through it.
• 19:10
  SPEAKER_04: He didn't know it would define his career.
• 19:17
  SPEAKER_05: Dr.
• 19:18
  SPEAKER_04: Adler Milstein started in management consulting.
• 19:23
  SPEAKER_04: Knew she cared about healthcare, but couldn't have told you what that meant as a job.
• 19:29
  SPEAKER_04: Ended up at a health system in a research group almost by accident, realized she loved the research, went back for a PhD in a field that didn't have a graduate program specifically for what she wanted to study.
• 19:42
  SPEAKER_04: She built the path as she walked it.
• 19:49
  SPEAKER_05: Dr.
• 19:50
  SPEAKER_04: Warty by a medical engineer who didn't feel engineering was his calling, volunteered at a hospital with no connections and fell in love with emergency medicine.
• 20:03
  SPEAKER_04: His engineering background turned out to be exactly what he needed to build the AI models he's now known for.
• 20:10
  SPEAKER_04: He just didn't know that when he made those earlier choices.
• 20:18
  SPEAKER_05: People in it know each other.
• 20:20
  SPEAKER_05: Longhurst and Pagelor trained in overlapping worlds.
• 20:23
  SPEAKER_05: Warty's engineering background put him in the same research circles as the people building the tools.
• 20:30
  SPEAKER_05: Networks form early, and getting into them matters.
• 20:37
  SPEAKER_04: More than most students realize, which is why they volunteer ask.
• 20:45
  SPEAKER_04: That's not desperation.
• 20:47
  SPEAKER_04: That's how this field actually works.
• 20:53
  SPEAKER_05: Dr.
• 20:54
  SPEAKER_05: Aaron Holby.
• 20:59
  SPEAKER_04: She helped write HIPAA, the federal law governing your medical privacy.
• 21:04
  SPEAKER_04: But she started in high school being drawn to what she calls big messy questions, things that don't resolve cleanly.
• 21:13
  SPEAKER_04: She worked in environmental policy, moved towards public health, and spent decades at the intersection of data, research, and legislation.
• 21:22
  SPEAKER_04: She described her core skill as translation.
• 21:26
  SPEAKER_04: She can talk to highly technical people and highly non-technical people and help them understand each other.
• 21:34
  SPEAKER_05: That is almost impossible to automate.
• 21:53
  SPEAKER_04: His entire career is about turning consumer devices into medical tools.
• 22:00
  SPEAKER_00: Far as I can recall, even in middle school, the projects I used to do were always about doing more from less.
• 22:08
  SPEAKER_00: Instead of learning something new, what I always wanted to do was whatever I know so far, what's the best way I can use this to solve an interesting problem?
• 22:16
  SPEAKER_05: That framing applies to the whole field.
• 22:19
  SPEAKER_05: These are people who looked at what already existed and asked better questions about it.
• 22:26
  SPEAKER_04: The earbuds already exist.
• 22:28
  SPEAKER_04: The smartwatches already exist.
• 22:31
  SPEAKER_04: The electronic health record already exists.
• 22:34
  SPEAKER_04: The question is, what are you doing with what's already there?
• 22:41
  SPEAKER_05: Which is actually a useful frame for students who don't know where to start.
• 22:46
  SPEAKER_05: You don't need to invent the device, you need to ask a better question about one you already have.
• 22:53
  SPEAKER_04: And his career advice for anyone at the beginning of this: if a problem excites you and you don't have the solution yet, keep it in your back pocket.
• 23:04
  SPEAKER_04: When the technical breakthrough comes, whether it comes to you or to someone you know, you're already thinking about it.
• 23:13
  SPEAKER_04: Which brings us to the main question for someone listening to this right now.
• 23:26
  SPEAKER_05: Concrete skills.
• 23:29
  SPEAKER_05: What do you actually need to build for this field?
• 23:33
  SPEAKER_04: And some of the answers surprised me.
• 23:36
  SPEAKER_04: Dr.
• 23:36
  SPEAKER_04: Goel said physics, not AI, not machine learning, not a specific coding language.
• 23:43
  SPEAKER_04: Physics.
• 23:44
  SPEAKER_04: And Dr.
• 23:45
  SPEAKER_04: Aaron Holvey said reading, not a subject, not a tool, just the practice of reading carefully and thinking critically about what you've read.
• 23:56
  SPEAKER_04: His reasoning is worth understanding.
• 23:59
  SPEAKER_04: Physics is the part that doesn't change.
• 24:02
  SPEAKER_04: The tools change, the frameworks change, the languages change, but the physical principles that allow you to measure blood pressure from sound waves or detect medication levels through the skin, those are grounded in physics, established a century ago.
• 24:19
  SPEAKER_04: If you understand why two things are related, you can find that relationship in data regardless of what tools exist.
• 24:29
  SPEAKER_04: If you only know how to run the current algorithm, you're depending on whoever designs the next one.
• 24:37
  SPEAKER_05: And Dr.
• 24:37
  SPEAKER_05: Holby's answer, reading critically, is about something similar.
• 24:42
  SPEAKER_05: If you can assess whether a source is credible, notice when a claim is oversimplified, evaluate whether the evidence actually supports the conclusion.
• 24:52
  SPEAKER_05: That skill applies everywhere, including AI outputs.
• 24:58
  SPEAKER_03: Read a lot.
• 24:59
  SPEAKER_03: Be a critical thinker.
• 25:01
  SPEAKER_03: There's just no substitute for reading.
• 25:03
  SPEAKER_03: I know you've joked about the fact that I'm a big reader, right?
• 25:05
  SPEAKER_03: You see those books that I bring with me everywhere, digital books, right?
• 25:09
  SPEAKER_03: And so if you read primary sources, if you are really engaged in what you're hearing and learning about, it's a lot easier to discern or understand some of the things that might come out of an AI summary.
• 25:24
  SPEAKER_05: Dr.
• 25:25
  SPEAKER_05: Adler Milstein named writing, which surprised me in a technical field.
• 25:30
  SPEAKER_04: She was emphatic about it.
• 25:33
  SPEAKER_04: The ability to get a research grant, publish in a significant journal, make your ideas matter beyond your own desk.
• 25:41
  SPEAKER_04: All of it depends on whether you can write clearly.
• 25:45
  SPEAKER_04: She said it's the most underappreciated skill in research careers and has spent a significant portion of her doctoral training on it.
• 25:56
  SPEAKER_05: Dr.
• 25:57
  SPEAKER_05: Longhurst talked about systems thinking.
• 26:01
  SPEAKER_04: He said medicine trains people to treat the individual in front of them, which remains essential.
• 26:07
  SPEAKER_04: But healthcare increasingly needs people who can design processes that produce good outcomes across entire populations.
• 26:16
  SPEAKER_04: Not just how do I treat this patient, but how do I build the system so every patient gets treated well?
• 26:24
  SPEAKER_04: He said new medical students are building this into their curriculum explicitly because they see the gap.
• 26:31
  SPEAKER_04: And it connects directly to what Dr.
• 26:33
  SPEAKER_04: Pagler does.
• 26:35
  SPEAKER_04: She described her shift from cheating one patient at a time to redesigning the system so every pediatrician can provide better care.
• 26:45
  SPEAKER_05: None of these are technical skills in the narrow sense.
• 26:50
  SPEAKER_04: Physics, writing, critical reading, and systems thinking.
• 26:56
  SPEAKER_05: Tools are changing too fast for any specific technical skill to be the whole answer.
• 27:02
  SPEAKER_05: What doesn't change is the ability to understand why something works, communicate it clearly, and see how it fits into a larger system.
• 27:11
  SPEAKER_05: I want to name something before we get to the action piece.
• 27:16
  SPEAKER_05: Because I don't want students, parents, or counselors walking away thinking this field is purely exciting.
• 27:23
  SPEAKER_05: There are real tensions here that show up across almost every conversation.
• 27:30
  SPEAKER_04: Go ahead.
• 27:33
  SPEAKER_05: Privacy.
• 27:34
  SPEAKER_05: Your health data is some of the most sensitive information that exists about you.
• 27:39
  SPEAKER_05: And the more powerful AI gets at using this data, the more valuable it becomes.
• 27:45
  SPEAKER_05: To researchers, to health systems, and to companies.
• 27:49
  SPEAKER_05: Dr.
• 27:49
  SPEAKER_05: Holvey helped write the regulations meant to protect it.
• 27:54
  SPEAKER_05: But she raised something that I think is genuinely underappreciated.
• 27:59
  SPEAKER_04: The chatbot problem is right.
• 28:04
  SPEAKER_05: If you enter your health information into a commercial AI product, a chatbot, an app, a large language model, that information may not be protected by HIPAA.
• 28:15
  SPEAKER_05: Because you chose to put it there.
• 28:17
  SPEAKER_05: The law protecting your medical records at a doctor's office does not automatically follow you into a consumer application.
• 28:26
  SPEAKER_04: This is happening right now.
• 28:28
  SPEAKER_04: People are using these tools for health questions constantly.
• 28:32
  SPEAKER_04: It's not just a hypothetical future risk.
• 28:36
  SPEAKER_05: And then there's the equity issue.
• 28:39
  SPEAKER_05: Dr.
• 28:39
  SPEAKER_05: Pagler's no-show model is the clearest example, but it came up everywhere.
• 28:45
  SPEAKER_05: The same AI tool can widen disparities and care or close them, depending entirely on what the people using it chose to do with it.
• 28:53
  SPEAKER_05: That is not a technology question.
• 28:56
  SPEAKER_05: It's a values question, and someone has to answer it.
• 29:00
  SPEAKER_04: Dr.
• 29:01
  SPEAKER_04: Worthy made a related point.
• 29:03
  SPEAKER_04: The hospitals most likely to successfully implement AI are also the most resourced to do it carefully.
• 29:11
  SPEAKER_04: To measure outcomes, to have physicians involved in design, to catch problems and correct them.
• 29:18
  SPEAKER_04: The hospitals serving communities with the greatest need are often the least equipped to do all of that.
• 29:25
  SPEAKER_04: So the technology exists, whether it reaches everyone equitably, is a separate problem.
• 29:32
  SPEAKER_04: And that problem needs people who understand both the technology and the policy.
• 29:42
  SPEAKER_05: What do you actually do if this has sparked something?
• 29:46
  SPEAKER_04: I asked every guest some version of this, and I want to frame it as a real progression, not a list of vague suggestions.
• 29:55
  SPEAKER_04: Dr.
• 29:56
  SPEAKER_04: Worty walked into a hospital with no connections and asked to volunteer.
• 30:01
  SPEAKER_04: That is still the most direct way to find out whether the clinical environment is somewhere you want to spend your time.
• 30:08
  SPEAKER_04: And it is the start of building a network that will matter later.
• 30:13
  SPEAKER_04: Don't wait for a formal program to exist.
• 30:16
  SPEAKER_04: Find a hospital, a clinic, a research lab, and ask.
• 30:20
  SPEAKER_04: He said explicitly, if someone doesn't respond, don't read that as a rejection.
• 30:26
  SPEAKER_04: Everyone in medicine is overwhelmed.
• 30:29
  SPEAKER_04: Try again.
• 30:30
  SPEAKER_04: Ask someone else.
• 30:31
  SPEAKER_04: Keep going.
• 30:34
  unknown: Dr.
• 30:34
  SPEAKER_05: Holby's version of that for research and policy is to start noticing problems in your own environment.
• 30:41
  SPEAKER_05: What is inefficient, inequitable, or broken at your school, your community, or your neighborhood?
• 30:47
  SPEAKER_05: What would you want to understand better?
• 30:50
  SPEAKER_05: Research careers begin with a habit of noticing things and wanting to know why.
• 30:58
  unknown: Dr.
• 30:58
  SPEAKER_04: Paichler's advice was about breadth.
• 31:01
  SPEAKER_04: She had an engineering background, went to medical school, and found intersections she never anticipated.
• 31:07
  SPEAKER_04: Don't prune your interests too early.
• 31:11
  SPEAKER_04: The combination that feels random right now may be exactly what makes you valuable later.
• 31:19
  SPEAKER_05: And for anyone who's already in a degree and worried they chose wrong, every single person we spoke to made at least one major pivot.
• 31:27
  SPEAKER_05: None of them started where they ended up.
• 31:30
  SPEAKER_04: That's not a consolation.
• 31:32
  SPEAKER_04: That's the pattern.
• 31:42
  SPEAKER_00: Build more bridges, learn more from others, and then see when you do that, you are able to amplify the kind of impact you can have, the kind of solutions you can build.
• 31:50
  SPEAKER_00: And that is something that I learned more in grad school.
• 31:53
  SPEAKER_00: And after coming to CMU, I would have ideally liked to learn that slightly sooner.
• 31:59
  SPEAKER_05: The work that matters in this field is almost never done alone.
• 32:03
  SPEAKER_05: It's done at the seams between disciplines.
• 32:08
  SPEAKER_04: Policy researchers working with engineers, data scientists working with ethicists, the people doing the most interesting work are the ones who know how to move between those worlds.
• 32:25
  SPEAKER_05: The entry points into this field are genuinely diverse.
• 32:29
  SPEAKER_05: Computer science, biology, public health, writing, law, ethics, and nursing.
• 32:36
  SPEAKER_05: There is meaningful work at every one of those intersections.
• 32:40
  SPEAKER_05: The students who will do well here are not necessarily the ones who know the most about AI right now.
• 32:47
  SPEAKER_05: They are the ones who are curious, who can communicate across disciplines, and who have built enough of a foundation in science, in writing, and systems thinking that they can keep learning as the tools change.
• 33:02
  SPEAKER_04: And if you're an educator building curriculum around any of this, or a counselor helping students navigate it, we have free resources at paladsofuriosity.com.
• 33:13
  SPEAKER_04: Guides built around every episode of this series, ready to be used in a classroom or advisory session.
• 33:54
  SPEAKER_08: I think what's much more important is being driven by that passion and curiosity.
• 33:59
  SPEAKER_08: And so, as your high school counselor would say, I think the most important thing is choosing something that you love.
• 34:05
  SPEAKER_08: Go into it for the right reasons, not because your parents tell you or because you think you might be interested, but explore it.
• 34:12
  SPEAKER_08: And whatever it is that you love so much that you would do it for free, if you can figure out how to get paid to do that, then you'll be a happy camper your whole career.
• 34:32
  SPEAKER_05: She was talking about where AI in healthcare is ultimately headed.
• 34:37
  SPEAKER_05: And she said the biggest question isn't what AI can do, it's about who the decision maker is.
• 34:44
  SPEAKER_05: Right now, humans are still the gatekeepers.
• 34:47
  SPEAKER_05: AI helps them do their jobs better.
• 34:50
  SPEAKER_05: But the real transformation, the version that changes healthcare fundamentally, is when AI operates at a scale no individual human can manage.
• 35:00
  SPEAKER_05: Running a health system, thinking across an entire population.
• 35:04
  SPEAKER_05: And our question was, are we ready for that?
• 35:07
  SPEAKER_05: Do we have the governance structures, the ethical frameworks, the policy tools to manage that responsibly?
• 35:16
  SPEAKER_04: And the honest answer right now is not entirely.
• 35:20
  SPEAKER_05: Which means the people who build those structures are either in school right now or just starting their careers.
• 35:27
  SPEAKER_05: That's not abstract, that's a real opening.
• 35:31
  SPEAKER_04: That's the thing that stays with me most across all six conversations.
• 35:37
  SPEAKER_04: It's not just that the field is exciting, it's that it's genuinely unfinished.
• 35:44
  SPEAKER_04: The most important work hasn't been done yet.
• 35:47
  SPEAKER_04: All six episodes of the AI and healthcare miniseries are linked in the show notes.
• 35:53
  SPEAKER_04: If you're not sure where to start, pick the guest whose work connects with something you already care about.
• 35:59
  SPEAKER_04: Clinical Medicine, start with Dr.
• 36:02
  SPEAKER_04: Worty or Dr.
• 36:03
  SPEAKER_04: Longhurst.
• 36:04
  SPEAKER_04: Policy and Data, Dr.
• 36:05
  SPEAKER_04: Adler Milstein or Dr.
• 36:07
  SPEAKER_04: Holvey.
• 36:08
  SPEAKER_04: Building the technology, Dr.
• 36:10
  SPEAKER_04: Goel, Pediatrics or System Design, Dr.
• 36:14
  SPEAKER_04: Pagler.
• 36:17
  SPEAKER_05: And everything we mentioned today the career landscape map, what's your spark, real path cards for all six guests, first steps, and teach this for our educators.
• 36:28
  SPEAKER_05: All of it is free at pathsofuriosity.com.
• 36:31
  SPEAKER_05: Every link is now in the show notes.
• 36:36
  SPEAKER_04: We also want to hear from you.
• 36:38
  SPEAKER_04: If something in this series sparked a question, if you reached out to someone, if you signed up to volunteer somewhere, let us know.
• 36:48
  SPEAKER_04: That's the whole point.
• 36:52
  SPEAKER_05: This has been Paths of Curiosity in an AI World.
• 36:56
  SPEAKER_05: I'm Alora.
• 36:58
  SPEAKER_04: And I'm Acadia.
• 37:00
  SPEAKER_04: Until then, stay curious.