9 to 5 Wellness

Dr. Michael Turken discusses AI limitations in longevity, wearable accuracy biases, and clinical safety risks.

1h 6m Listen to the original
LongevityAI in HealthcareWearable TechnologyClinical SafetyPreventive MedicineDigital HealthMedical BiasTelemedicine

Summary

Dr. Michael Turken, an internal medicine physician and founder of My Doctor Friend, joins the 9 to 5 Wellness Podcast to discuss the current state of AI in healthcare and longevity. He argues that while AI can assist with administrative tasks and early detection, it currently lacks evidence for improving patient longevity and poses risks such as hallucination and sycophancy. Turken warns against over-reliance on wearable metrics like VO2 max, citing skin tone bias and the superior importance of clinical biomarkers like blood pressure. He also highlights the dangers of using general-purpose AI for medication management, citing a fatal case of self-managed seizure medication withdrawal.

Key takeaways

  1. 1
    Sycophancy is this tendency of AI models to tell you what you want to hear.

    Unidentified speaker (speaker_0) · 0:23

  2. 2
    Apple Watches generally offer higher sensor accuracy than competitors because they use a stronger signal, though this results in significantly shorter battery life (1-1.5 days).

    Unidentified speaker (speaker_0) · 44:44

Episode sections

  1. 0:00 – 0:35

    Intro and AI Sycophancy Risk

    Speaker_0 warns that relying on AI for health metrics can be problematic if it replaces knowing critical vitals like blood pressure. He defines 'sycophancy' as the tendency of AI models to tell users what they want to hear, which poses a health risk.

  2. 0:35 – 5:11

    Podcast Intro and Guest Background

    Speaker_1 introduces the 9 to 5 Wellness Podcast and guest Dr. Michael Turken, an internal medicine physician trained at Stanford, former Google Health product lead, and founder of My Doctor Friend. Turken previously worked at Omada Health and Forward.

  3. 5:11 – 9:34

    AI and Longevity: Current State vs. Hype

    Speaker_1 asks if AI is currently capable of extending lifespan to 150-250 years, referencing Dr. David Sinclair's research. Speaker_0 states that while enthusiasm is high, the technology is not yet there to provide a guaranteed path to living to 150, but hopes it will accelerate research into aging biology.

  4. 9:34 – 15:31

    Lifespan vs. Health Span

    Speaker_1 asks for the difference between lifespan and health span. Speaker_0 defines lifespan as duration of life and health span as the period of healthy life. He emphasizes that most people want their health span to align with their lifespan, avoiding chronic illness and dependence.

  5. 15:31 – 18:15

    AI in Clinical Practice and Evidence Gaps

    Speaker_0 notes that while AI has existed for decades (e.g., fraud detection), new LLMs are recent. He claims most US doctors use AI tools like scribes or clinical decision support (OpenEvidence, UpToDate, Doximity). He states that while models ace medical licensing exams, there is no published evidence yet that they improve patient longevity or health outcomes.

  6. 18:15 – 21:20

    AI Screening and the PSA Test Analogy

    Speaker_0 discusses AI's potential in early detection via retinal photos, mammograms, and CT scans. He warns that earlier detection does not always improve outcomes, citing the PSA test for prostate cancer, where early diagnosis did not significantly change mortality because many detected cancers were indolent.

  7. 21:20 – 29:06

    Full-Body MRI Limitations and False Positives

    Speaker_0 critiques full-body MRIs as non-standard screening tests that can miss cancers (e.g., colon, cervical) and produce false positives. He shares that he signed up for Function Health to test these platforms and notes that finding incidental irregularities can lead to unnecessary anxiety and biopsies without improving health outcomes.

  8. 29:06 – 31:23

    AI for Caregiving and Complex Conditions

    Speaker_1 asks how AI can help caregivers of aging parents. Speaker_0 argues AI will transform caregiving by managing logistical, emotional, and cognitive burdens, such as tracking appointments and medication interactions, which currently lack sophisticated tools for laypeople compared to clinical EMRs.

  9. 31:23 – 43:12

    AI as a Clinical Collaborator and Visit Companion

    Speaker_0 argues that AI can help patients and caregivers prepare for appointments by organizing health issues and summarizing discussions, effectively acting as a 'clinician friend.' This mirrors the benefits of having a healthcare-professional family member present, which studies show leads to better care and longer life.

  10. 43:12 – 46:03

    Wearable Data Accuracy and Skin Tone Bias

    Speaker_0 warns that wearable devices using laser light for heart rate and oxygen measurements are less accurate for people with darker skin tones because darker skin absorbs more light. He notes that Apple Watches generally offer higher accuracy due to stronger signals, despite having shorter battery life.

  11. 46:03 – 49:25

    Prioritizing Clinical Biomarkers Over Wearable Metrics

    Speaker_0 cautions against focusing solely on wearable metrics like VO2 max or HRV while ignoring critical clinical biomarkers such as blood pressure, cholesterol, and A1C. He argues that marketing has shifted focus to measurable wearable data at the expense of more important, harder-to-measure health indicators.

  12. 49:25 – 55:36

    AI in EKG and Retinal Screening

    Speaker_0 highlights that AI algorithms can now detect major structural heart issues from EKGs and cardiovascular risks from retinal photos, patterns invisible to the human eye. He notes that while these models are currently in the research domain, they represent a significant shift in preventive risk calculation.

  13. 55:36 – 58:00

    Risks of General Purpose AI Chatbots

    Speaker_0 outlines four primary risks of using general purpose AI chatbots for health: privacy (lack of HIPAA coverage), hallucination (confabulation or 'bullshitting'), forgetting context, and sycophancy (telling users what they want to hear). He cites a New York Times story about a patient who reduced seizure medication based on AI advice without consulting her doctor.

  14. 58:00 – 1:06:04

    Safety Risks of AI in Healthcare

    Speaker_0 recounts a tragic case where a patient died after using AI to wean herself off seizure medication without medical supervision, highlighting the danger of relying on AI for critical health decisions. The speakers emphasize that AI is not suitable for determining medication dosing changes due to lack of context and potential for mathematical errors.

Claims worth verifying

Blood pressure is far more important to know than metrics from wearables like VO2 max.

Unidentified speaker (UNKNOWN) ·0:08

How to verify: Medical literature on cardiovascular risk factors; compare mortality/morbidity data for hypertension vs. VO2 max trends. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

AI models exhibit 'sycophancy,' telling users what they want to hear.

Unidentified speaker (speaker_0) ·0:23

How to verify: Academic papers on LLM alignment, specifically those discussing 'sycophancy' or 'yes-man' behavior in large language models. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

Most US doctors already use AI in healthcare via scribes and clinical decision support tools like OpenEvidence, UpToDate, and Doximity.

Unidentified speaker (UNKNOWN) ·17:12

How to verify: Surveys of physician AI adoption (e.g., AMA surveys, Deloitte reports) or company press releases from OpenEvidence, UpToDate, Doximity regarding market penetration. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

AI models can ace medical licensing exams, performing better than most clinicians.

Unidentified speaker (UNKNOWN) ·17:32

How to verify: Published studies comparing LLM scores on USMLE Step 1/2 or similar licensing exams to human physician averages. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

There is no published proof that AI clinical decision support tools help patients live longer or healthier.

Unidentified speaker (UNKNOWN) ·17:52

How to verify: Systematic reviews in medical journals (e.g., JAMA, NEJM) for randomized controlled trials measuring patient outcomes from AI clinical decision support. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

AI screening can find heart disease in an EKG or a retinal photo.

Unidentified speaker (speaker_0) ·18:15

How to verify: Peer-reviewed studies on AI analysis of EKGs for cardiac conditions and retinal scans for cardiovascular risk. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

Primary source

The PSA test for prostate cancer diagnosed more cancers but did not significantly change mortality outcomes because many detected cancers were indolent.

Unidentified speaker (speaker_0) ·19:19

How to verify: Clinical trials on PSA screening efficacy, such as the European Randomized Study of Screening for Prostate Cancer (ERSPC) or PLCO trial results. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

Full-body MRIs are not standard diagnostic tests and are limited screening tools that may miss colon, cervical, or breast cancers.

Unidentified speaker (UNKNOWN) ·22:01

How to verify: Radiology guidelines (e.g., ACR Appropriateness Criteria) comparing full-body MRI sensitivity/specificity against standard screenings (colonoscopy, mammogram, Pap smear). For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

Dr. Michael Turken is a Stanford-trained internist and former Google Health product lead.

Unidentified speaker (speaker_1) ·2:43

How to verify: Dr. Turken's professional biography on My Doctor Friend website, LinkedIn, or UCSF faculty profile. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

Dr. Michael Turken is the founder and CEO of My Doctor Friend.

Unidentified speaker (speaker_1) ·3:03

How to verify: Company press releases, SEC filings (if public, otherwise news articles), or My Doctor Friend website leadership page. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

Primary source

Patients and families are responsible for their health most of the time, spending only 0.5% to 1% of their life in a clinical setting.

Unidentified speaker (speaker_0) ·31:43

How to verify: Primary source: Speaker's own statement; verifiable via general healthcare utilization statistics from CDC or similar bodies. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

Having a clinician or healthcare professional family member accompany a patient to appointments leads to better care and longer life.

Unidentified speaker (UNKNOWN) ·34:26

How to verify: Primary source: Speaker claims 'this has been studied.' Verify via PubMed or medical journals for studies on 'family advocacy' or 'clinician companion' impact on mortality and care quality. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

Wearable devices using laser light are less accurate for people with darker skin tones because darker skin absorbs more light.

Unidentified speaker (UNKNOWN) ·43:50

How to verify: Primary source: Speaker claims he has written about this. Verify via peer-reviewed studies on 'photoplethysmography accuracy skin tone' or FDA reports on wearable device bias. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

Apple Watches use a stronger signal for more accurate readings, resulting in shorter battery life (1-1.5 days).

Unidentified speaker (UNKNOWN) ·44:38

How to verify: Primary source: Speaker's technical claim. Verify via Apple technical specifications or independent reviews comparing battery life and sensor accuracy of Apple Watch vs. competitors. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

Primary source

Most wearables do not accurately measure blood pressure.

Unidentified speaker (UNKNOWN) ·46:34

How to verify: Primary source: Speaker's claim. Verify via FDA clearance databases or clinical studies comparing wearable cuffless BP monitors against standard sphygmomanometers. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

AI algorithms can detect major structural heart issues from EKGs that previously required an ultrasound.

Unidentified speaker (speaker_0) ·49:56

How to verify: Primary source: Speaker's claim. Verify via published research papers on 'AI EKG structural heart disease detection' or FDA approvals for such AI tools. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

Retinal photos can be used to detect cardiovascular risk and past blood pressure damage.

Unidentified speaker (speaker_0) ·50:27

How to verify: Primary source: Speaker's claim. Verify via ophthalmology journals or studies on 'retinal imaging cardiovascular risk prediction'. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

Retinal scans are available at Warby Parker for approximately $70-80.

Unidentified speaker (speaker_0) ·51:18

How to verify: Primary source: Speaker's statement. Verify via Warby Parker website or appointment booking pages for 'retinal scan' pricing. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

Consumer health AI chatbots are generally not bound by HIPAA.

Unidentified speaker (UNKNOWN) ·55:42

How to verify: Primary source: Speaker's legal/regulatory claim. Verify via Federal Trade Commission (FTC) guidelines on health apps or HIPAA exemption lists for consumer wellness apps. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

A New York Times story reported on a young engineer who reduced her seizure medication dosage after consulting an AI chatbot without telling her doctor.

Unidentified speaker (UNKNOWN) ·57:14

How to verify: Primary source: Speaker's citation. Verify via New York Times archives for articles matching this description, likely published around 2026-2026. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

A patient died after using AI to wean herself off seizure medication without talking to her doctor.

Unidentified speaker (speaker_0) ·58:00

How to verify: Search news archives for reports of patient deaths related to AI-guided medication weaning or seizure disorder management errors. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

AI models make dosing calculation errors because they lack necessary clinical context.

Unidentified speaker (speaker_0) ·59:33

How to verify: Review medical literature or case studies on AI errors in clinical decision support, specifically regarding pharmacology and dosing. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

My Doctor Friend is available on the Apple App Store and at mydoctorfriend.one.

Unidentified speaker (speaker_0) ·1:03:59

How to verify: Visit the Apple App Store and the URL mydoctorfriend.one to confirm the app's existence and availability. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

Primary source

My Doctor Friend does not have a working Android app yet.

Unidentified speaker (UNKNOWN) ·1:04:07

How to verify: Search the Google Play Store for 'My Doctor Friend' to confirm its absence or beta status. For recent events, verify 2026-08-15 through 2026-11-15. Use another period only when the speaker explicitly names it.

Do your own research

  • Verify claims about wearable accuracy disparities by reviewing peer-reviewed studies on photoplethysmography and skin tone bias.
  • Check the ERSPC trial and other meta-analyses on PSA screening to assess whether early detection significantly changes mortality outcomes.
  • Review FDA clearance status for AI algorithms used in EKG and retinal screening to ensure they are approved for clinical use.
  • Investigate the specific case of the patient who died after AI-guided medication withdrawal to understand the legal and medical context.
  • Search for published studies comparing LLM performance on USMLE vs. actual patient outcomes to assess the 'evidence gap' in clinical AI.
  • Check for HIPAA compliance and data privacy policies of any AI health apps, noting that general-purpose chatbots often lack these protections.
  • Review clinical guidelines on full-body MRI screening to understand standard of care versus experimental screening.
  • Verify the developer's claims about 'My Doctor Friend' features and trial terms via the Apple App Store and official website.
Full transcriptexpand
0:00
Unidentified speaker (speaker_0): People are already starting to see
0:03
Unidentified speaker (UNKNOWN): their
0:03
Unidentified speaker (speaker_0): doctor less because they're using AI as a substitute, which
0:08
Unidentified speaker (UNKNOWN): can
0:08
Unidentified speaker (speaker_0): be problematic. Blood pressure is far more important to know than pretty much any metric you're gonna get from a wearable.
0:15
Unidentified speaker (UNKNOWN): So
0:15
Unidentified speaker (speaker_0): if you don't know your VO you know, if you know your VO2 max and what your VO2 max trends are, but you don't know what your blood pressure is, that's a problem.
0:23
Unidentified speaker (speaker_0): Sycophancy is this tendency of AI models to tell you what you want to hear.
0:29
Unidentified speaker (UNKNOWN): So
0:29
Unidentified speaker (speaker_0): they want to please you.
0:31
Unidentified speaker (UNKNOWN): Oh,
0:31
Unidentified speaker (speaker_1): yeah. And
0:32
Unidentified speaker (UNKNOWN): this can be
0:32
Unidentified speaker (speaker_0): a risk. This can be a risk for your health.
0:35
Unidentified speaker (speaker_1): Hello and
0:38
Unidentified speaker (speaker_1): welcome to the 925 Wellness Podcast. Today we are talking about something super
0:46
Unidentified speaker (speaker_1): relevant at this time, which is AI and how. AI
0:52
Unidentified speaker (speaker_1): will or is impacting our health. So we are probably all familiar by now that AI can now read your lab,
1:03
Unidentified speaker (speaker_1): it can uh track your sleep, uh, it can draft a workout plan for you guys in you know, literally seconds. You just have to give it a prompt, and you know, it will just throw out a great workout from. program. It can even progress you for overload if you wanted that. But the real question is, is it really
1:23
Unidentified speaker (speaker_1): making us healthier? So in this episode of Nine to Five Wellness Podcast, I am going to uncover some of the smart ways we can use AI for our health right now. We will also get into when to put it down. Because as much as I
1:43
Unidentified speaker (speaker_1): know we have started relying on AI and it's a great tool to have, but you know, in certain conditions and in certain uh situations, you probably do want the expert advice of a clinician. Cognition. So we are going to explore that. Before we get into today's conversation, and I introduce my
2:03
Unidentified speaker (speaker_1): guests to you guys, I do have a medical disclaimer that I want to share. This episode is meant to educate and inspire you, not to replace medical advice. Listening doesn't create a patient provider relationship. And everyone's health journey is different. We are all unique, right? You will all agree with that. So please check in
2:23
Unidentified speaker (speaker_1): with your physician or a trusted healthcare provider before making any changes to your diet, exercise, medications, or supplements, especially if you're managing a chronic health condition or a complex health condition. So let's get back into our conversation. And today, joining us
2:43
Unidentified speaker (speaker_1): is someone who has been exploring AI and it's, you know, who works at the intersection of AI and health and non-consciousness. longevity. And our guest today is Michael Turkin. He is an internal medicine physician who trained at Stanford and continues to see patients in urgent care at UCSF. He's the
3:03
Unidentified speaker (speaker_1): founder and CEO of My Doctor Friend, an AI health companion that helps people understand their medical information, prepare for appointments, and navigate care for themselves. and their families. He previously worked on clinical products at Google Health, Omada Health, and Forward. He also
3:22
Unidentified speaker (speaker_1): writes Sovereign Care, a newsletter about how patients can use AI to take more active role in their health care. Welcome to the podcast, Michael.
3:31
Unidentified speaker (speaker_0): Thank you so much. So thrilled to be here.
3:34
Unidentified speaker (speaker_1): Oh, I'm so excited for this conversation because I think healthcare is one of the areas where we're not sure, like, you know, th what role AI is gonna play as far as patient physician interaction goes and how much to trust it. So that's why I'm super excited about
3:54
Unidentified speaker (speaker_1): our conversation. Before we get into it, which you know we are going to, for anyone hitting play on this episode right now, whether they're listening on their commute or watching between meetings, what's the one big thing you hope they take away from our conversation?
4:13
Unidentified speaker (speaker_0): I think it would be that
4:16
Unidentified speaker (speaker_0): AI can be an incredibly powerful tool for understanding your health, for keeping healthy.
4:25
Unidentified speaker (speaker_1): Yes.
4:25
Unidentified speaker (speaker_0): But just as we with a lot of other powerful tools that we use, it has to be used carefully. And if you haven't uh really used it that much, you have to learn how to use it safely. But all of that said, it's really one of the most incredible opportunities we've all had to uh
4:45
Unidentified speaker (speaker_0): take a bigger role in understanding and managing our own health.
4:51
Unidentified speaker (speaker_1): Yeah, I couldn't agree more with you. And I would love to understand better like where it can help us and where we want to be a little cautious, right? And maybe along the road as things improve and technology improves, you know, there will be other areas that would emerge where it can
5:11
Unidentified speaker (speaker_1): start helping us. Here's the thing though, I was just having this conversation with one of my friends yesterday about, because I was talking to her, that hey,
5:20
Unidentified speaker (UNKNOWN): but
5:20
Unidentified speaker (speaker_1): Michael is coming on the podcast and we're going to have a conversation. Um, and she's like, you know, when I talk to people, one of the things I always hear from people when AI and longevity or AI and health is This being discussed, either there are people who are just talking about like chatbots, like the LLMs that we are very used to at this
5:40
Unidentified speaker (speaker_1): point. Or there are people who are talking like, oh, living up to 150 years or maybe 250 years. I was listening to David, Dr. David Sinclair's uh podcast the other day. And I mean, he has been working on longevity for such a long time, and his whole area of research is age
6:00
Unidentified speaker (speaker_1): reversal. So
6:03
Unidentified speaker (speaker_1): I want your take on this. Like, are we still with the chatbots? Are we literally at the point where we're gonna be living two 250 years? Where are we actually right now?
6:16
Unidentified speaker (speaker_0): Yeah.
6:18
Unidentified speaker (speaker_0): It's a great question. I'll do my best to answer it. I think there's uh probably some disconnect between where we are, where the technology is now, and where we would like it to be in the next 10, 15, 20 years and where we would dream it would take us. I think there's tremendous enthusiasm and excitement for
6:38
Unidentified speaker (speaker_0): AI to help us with drug discovery to help us understand parts of
6:45
Unidentified speaker (speaker_0): our biology that have been just mysterious to us. I I think
6:50
Unidentified speaker (UNKNOWN): maybe
6:50
Unidentified speaker (speaker_0): one of the places where I get most excited is when I think that just in the last few months, these chatbots, we can call them chatbots, but really they're they're more than just chatbots, and they can do more than just chat,
7:04
Unidentified speaker (UNKNOWN): have
7:04
Unidentified speaker (speaker_0): solved some of the hardest math problems that humanity knows about. And so I I think part of what we need to do is reframe. our experience of what this technology was when it first got released, which was, you're right, the chat bot, this thing that would be mildly amusing, and maybe you can get some good movie reviews and
7:24
Unidentified speaker (speaker_0): recipes and workouts out of it, and really
7:27
Unidentified speaker (UNKNOWN): start
7:27
Unidentified speaker (speaker_0): thinking about it as wow, this is
7:31
Unidentified speaker (UNKNOWN): you know,
7:32
Unidentified speaker (speaker_0): this represents something like a calculator for the mind. And it's across
7:37
Unidentified speaker (UNKNOWN): all
7:37
Unidentified speaker (speaker_0): the areas where we are. trying to learn and understand the science better. And so
7:44
Unidentified speaker (speaker_0): practically, for those of folks who are really excited about living to, I don't know, 150, 200,
7:51
Unidentified speaker (UNKNOWN): I
7:51
Unidentified speaker (speaker_0): don't think the technology is there yet. And I don't think that the AI models, I don't think there's a special prompt that you can give to the model and it's going to wink wink tell you wait a second, you know, here's this combination of lifestyle activities, supplements, medications you can take to be guaranteed to live to 150. We're not there yet, but
8:11
Unidentified speaker (speaker_0): my hope is that if this technology, and I don't think I'm alone here, by the way, if we can use this technology to accelerate our research and our science, we may start to understand
8:24
Unidentified speaker (UNKNOWN): the
8:24
Unidentified speaker (speaker_0): biology of aging and uh the biology of a lot of health conditions that have just eluded us. I mean, think about Alzheimer's, which we've been studying for decades and we haven't made all that much progress.
8:36
Unidentified speaker (UNKNOWN): If
8:36
Unidentified speaker (speaker_0): AI can help us, then we may have a credible path, a real path to extending not just our lifespan, but our health span.
8:46
Unidentified speaker (speaker_1): Yeah. So let's dive into that. Actually, as you're talking, I'm like thinking about it that it's true. Like we thought it was just like a cute little thing, you know, that's
9:00
Unidentified speaker (speaker_1): all with with maybe just uh recognizing patterns, doing our Excel sheets, right? You know, that's what it's gonna be for. And now today, like just within I wanna say two to three years of this rolling out, like it
9:14
Unidentified speaker (UNKNOWN): can
9:15
Unidentified speaker (speaker_1): do a lot more than what I we had thought. And especially with the integration of Deep Mind now within these LLMs, it's becoming more self-intelligent or automated intelligent model, which I'm excited to explore. I wanna understand
9:34
Unidentified speaker (speaker_1): what you were just touching upon, the lifespan and health span. Because I when I see patients and I talk to them, a lot of them are still kind of confused. Like, okay, is there a difference between the two? Uh what should we be striving for? Right? And and I normally throw that question just back to them and I say,
9:54
Unidentified speaker (speaker_1): hey, you tell me, like, what's your goal? Like if you were living in an ideal world, what goal would you have? And a lot of them come back to me with the fact, oh, I want to be able to enjoy the things that I do today, you know, for as long as I'm
10:14
Unidentified speaker (speaker_1): living. So help us understand the difference between a lifespan and health span.
10:21
Unidentified speaker (UNKNOWN): Sure,
10:21
Unidentified speaker (speaker_0): happy to. I can give you my perspective on it. Although I I like that you um you turned turn this question. around and ask your patients when the topic comes up. You mentioned something that I do want to acknowledge, which is that these models, um, whether they're Google's you know deep mind models or anthropic or open AIs, they are getting
10:41
Unidentified speaker (speaker_0): close to this point, which has been called recursive self-improvement, where these models are so smart that they can
10:47
Unidentified speaker (UNKNOWN): be
10:47
Unidentified speaker (speaker_0): instrumental in making themselves smarter. In other words, the loop of improvement. improvement accelerates and accelerates quickly.
10:54
Unidentified speaker (speaker_1): Absolutely.
10:55
Unidentified speaker (speaker_0): Yeah, to your question about lifespan and health span, I think that's a pretty accurate
11:01
Unidentified speaker (UNKNOWN): uh
11:02
Unidentified speaker (speaker_0): accurate way of thinking about it. Uh just stated another way.
11:07
Unidentified speaker (speaker_0): Lifespan is how long you live, health span is how many years of that lifespan are spent. in a state where you are healthy, however you define healthy. And that may be different for different people. And it may be different based on your degree of of activity. But to
11:27
Unidentified speaker (speaker_0): make it concrete,
11:29
Unidentified speaker (speaker_0): I think a lot of people
11:33
Unidentified speaker (speaker_0): uh would not be happy, you know, living into their 90s or hundreds if it meant that they were experiencing tremendous,
11:42
Unidentified speaker (UNKNOWN): you know,
11:42
Unidentified speaker (speaker_0): chronic pain and complexities of and uh you know effects of of multiple complex chronic illnesses. And if they were spending their time in and out of the hospital and they weren't able to do the things they wanted, whether that's gardening or spending time with their grandkids or walking outside,
11:59
Unidentified speaker (UNKNOWN): uh
12:00
Unidentified speaker (speaker_0): what everyone needs to be able to to do to to be
12:05
Unidentified speaker (UNKNOWN): to
12:05
Unidentified speaker (speaker_0): be happy is very different. Everyone's range of activity is very different. Some people, depending upon their abilities, may be totally content to not garden or not go jogging into their 90s and still would consider themselves quite healthy. So I think you're you're absolutely right. This is something that everyone has to define for themselves.
12:25
Unidentified speaker (speaker_0): What do they consider healthy and what is healthy enough? But I think most people want their health span and their lifespan to align as much as possible. And when I've talked to cardiologists and other clinicians, when they think about their health span, most of them
12:42
Unidentified speaker (UNKNOWN): I
12:42
Unidentified speaker (speaker_0): ask them, well, how do you want to live and how how would you like to like if you could, you know, if you could plan when you pass away. what would that look like? And I I can't tell you how many of them have said, well, I would love it if
12:57
Unidentified speaker (UNKNOWN): you know
12:58
Unidentified speaker (speaker_0): I live into my 90s and I'm in my late 90s on a bike ride through the woods, and then you know, something sudden happens. And then, you know, that that is that that might not be a bad way, a bad way for me to have lived my life, meaning that they're able to do all these things that they really
13:14
Unidentified speaker (UNKNOWN): love
13:14
Unidentified speaker (speaker_0): to do. right into the moment where their lifespan ends.
13:18
Unidentified speaker (speaker_1): Yeah.
13:19
Unidentified speaker (speaker_0): Um
13:20
Unidentified speaker (speaker_1): yeah. Absolutely. I agree. And I want to highlight something you just said that what we really are we all are um unconsciously or consciously trying to do is have both of them like come together, or there should be like a merger of the lifespan and health span. at you know that
13:40
Unidentified speaker (speaker_1): intersection, we want to find that intersection for ourselves. And as you said, nobody wants to probably, at least I don't, and I think I'm speaking for many people here who I've talked to, that they don't want to necessarily uh want to live long a longer life if that
13:59
Unidentified speaker (speaker_1): means, you know, chronic illness or complex illnesses.
14:03
Unidentified speaker (speaker_1): and as you said being going in and out of the hospital or being dependent on others. This is one thing I hear a lot from my patients. Like I want I don't want to be a burden on someone, you know? Yeah. So it's like how can we move the needle from
14:23
Unidentified speaker (speaker_1): dependence to independence for uh our patients. I think that it it comes down to that goal. And yes, and if at this point I believe it's more of a health span thing. As we a as all the advances in medicine and science are happening, we know that lifespan is getting
14:43
Unidentified speaker (speaker_1): longer already. But we want that health span to match to that uh life lifespan. span. Ideally,
14:49
Unidentified speaker (speaker_1): yeah, yeah. So so yeah no that was a great insight. I wanna um
14:56
Unidentified speaker (speaker_1): switch gears a little bit and understand
15:01
Unidentified speaker (speaker_1): where do we have good evidence that AI is going to help us with our health. Maybe that looks like early detection. Maybe that looks like uh reduction of risk or um, you know, uh prevention and management of
15:21
Unidentified speaker (speaker_1): diseases. Where is it already kind of like functional and what is still like in the experimental stages?
15:31
Unidentified speaker (speaker_0): It's a it's an excellent question. I think
15:36
Unidentified speaker (speaker_0): AI in in the form that we've been discussing it, so large language model specifically, because AI, it's good to remind people AI has been around for a long time. You know, everything from face ID to you know bank fraud detection, these are all forms of AI. Uh chess games that d defeat our grandmasters. These are all AI that that has been around for a while.
15:56
Unidentified speaker (UNKNOWN): In
15:56
Unidentified speaker (speaker_0): the healthcare space, this new kind of AI is still so new that we don't really have that much evidence yet to know what it's going to do. Meaning,
16:09
Unidentified speaker (UNKNOWN): if
16:09
Unidentified speaker (speaker_0): you went back four years ago, there
16:12
Unidentified speaker (speaker_0): was no chat GPT in healthcare. It's it's so new. And the
16:19
Unidentified speaker (UNKNOWN): uh
16:20
Unidentified speaker (speaker_0): the thing we have to remind ourselves is that
16:23
Unidentified speaker (speaker_0): especially when it comes to looking at outcomes like longevity. or prevention, the timeline for understanding this is measured in years, not months. And so
16:36
Unidentified speaker (speaker_0): I will say that in general, the body of evidence around what AI does for health outcomes is really in its infancy. Now, it's still worth separating where is the evidence from functionally, where is this technology being used, or
16:53
Unidentified speaker (UNKNOWN): how
16:53
Unidentified speaker (speaker_0): are how are clinicians and non-clinicians using it? Because what's fascinating is that there's so much excitement and so much enthusiasm for what AI can do that I believe, at least in the US, the majority of doctors are using AI in their clinical practice, whether it's
17:12
Unidentified speaker (UNKNOWN): an
17:12
Unidentified speaker (speaker_0): AI scribe that helps them with documentation and notes. or it's an AI reference like open evidence or up to date or Doximity or ChatGPT for clinicians, there are now quite a few of these AI references, and AI, they're called clinical decision support. So they help clinicians make decisions.
17:29
Unidentified speaker (speaker_0): Now there is evidence
17:32
Unidentified speaker (UNKNOWN): that
17:32
Unidentified speaker (speaker_0): these models perform really well on medical licensing exams. They can ace them, they do better than most clinicians do.
17:39
Unidentified speaker (UNKNOWN): But yeah, I read
17:40
Unidentified speaker (speaker_1): that actually in an article. Yeah.
17:43
Unidentified speaker (speaker_0): But in terms of measuring, well, what does this mean for patients? Are patients living longer? Are they healthier because of this technology? At least for the clinical decision support tools,
17:52
Unidentified speaker (UNKNOWN): no
17:53
Unidentified speaker (speaker_0): one has published that evidence yet. We still don't know.
17:55
Unidentified speaker (UNKNOWN): And
17:55
Unidentified speaker (speaker_0): so there's been a lot of enthusiasm, maybe before we have conclusive data. Now you mentioned something, I think. maybe this will reach folks sooner, which is the early detection piece. Um, there are AI models being tested across many areas of uh diagnostic imaging, everything
18:15
Unidentified speaker (speaker_0): from being able to diagnose early heart disease from pictures just using pictures of your eye and your retina to
18:23
Unidentified speaker (UNKNOWN): uh
18:24
Unidentified speaker (speaker_0): better screening and uh interpretation of mammograms, CT scans. And so it could very well be that within a few years we're gonna be at a place where not only are screening diagnostic tests cheaper because we have AI
18:44
Unidentified speaker (speaker_0): helping us with some of the interpretation, but they're also a lot more accurate. Again, still remains to be seen. what this does to
18:54
Unidentified speaker (speaker_0): longevity and outcomes. Because one of the things that is I think folks don't always appreciate is that just because we're able to detect things earlier doesn't necessarily mean that it makes a difference in terms of our outcome. outcomes. So this is this is a place where I think screening and testing often
19:14
Unidentified speaker (speaker_0): get confused. And and a good example I I tell people about this is
19:18
Unidentified speaker (UNKNOWN): when
19:19
Unidentified speaker (speaker_0): we first started developing and using PSA tests. So this is a test for prostate cancer
19:26
Unidentified speaker (speaker_0): for men. And when this test first came out, we were diagnosing so many more prostate cancers than before. We were catching earlier prostate cancers. Surprisingly,
19:36
Unidentified speaker (UNKNOWN): the
19:36
Unidentified speaker (speaker_0): outcomes didn't change all that much because one of the things it took us, it ended up taking us 10, 15, maybe even 20 years to realize that more people die with prostate cancer than of prostate cancer. And so we diagnosed a lot of these cancers that never really were going to cause people that much trouble. And so that's the other piece that
19:56
Unidentified speaker (speaker_0): I think folks should start to think about, which is that the road to better outcomes may be bumped. bumpy and there may be some unexpected turns, meaning we may start being able to detect things really early on before we have
20:09
Unidentified speaker (UNKNOWN): the
20:09
Unidentified speaker (speaker_0): treatments to actually change the trajectory of the outcomes.
20:14
Unidentified speaker (speaker_1): Absolutely. And I just want to add to a couple of things that you just mentioned. First, you said, oh, the accessibility might improve to a lot of early detection skills. scans or technology. Like today, you can go and get an MRI if you were paying out of pocket for $900. I'm
20:34
Unidentified speaker (speaker_1): not saying that $900 is not a big deal, but it is way less than what it was three or four years ago. Like it was in thousands of dollars, right? Today you can do get a whole body MRI if you wanted want it to. But then the second aspect is okay. if it does detect, let's say uh
20:53
Unidentified speaker (speaker_1): tumors or whatever, are you gonna now start going and like literally asking your doctor to do biopsies of every single one of them? Like, you know, that's where it becomes really hard to understand. Okay, now that you have this ability to have early detection,
21:12
Unidentified speaker (speaker_1): does that actually lead to also
21:15
Unidentified speaker (speaker_1): a better life or extended lifespan?
21:19
Unidentified speaker (speaker_0): Yeah.
21:19
Unidentified speaker (UNKNOWN): Yeah.
21:20
Unidentified speaker (speaker_0): I think that's I'm glad you brought that up. And the full body MRI is maybe one of one of I think the least understood diagnostic tests. Because you're right, you can get,
21:30
Unidentified speaker (UNKNOWN): I
21:31
Unidentified speaker (speaker_0): think you can get it cheaper than $900 now.
21:33
Unidentified speaker (UNKNOWN): S
21:33
Unidentified speaker (speaker_1): Some places, yes. Yes.
21:35
Unidentified speaker (speaker_0): I I um I recently signed up for function health mostly out of curiosity. And I'm I'm starting to test a lot of these direct consumer health platforms. And one of the things they've been trying to sell me is this full body MRI. And uh I think they were trying to incentivize me by by you know lopping off a few hundred dollars to make it cheaper. Your point is exactly
21:55
Unidentified speaker (speaker_0): right. Well, there and there are some additional points. First, the full body, first of all, the
22:01
Unidentified speaker (UNKNOWN): a full
22:02
Unidentified speaker (speaker_0): body MRI. is not a standard diagnostic. There's not one full body MRI, meaning there are many ways that you can go into an MRI scanner and get an image out at the other end. And my understanding is that their MRI is somewhat limited. It's a it's a screening MRI.
22:20
Unidentified speaker (speaker_1): Yes.
22:21
Unidentified speaker (speaker_0): So it is not going to capture everything. It is not the best screening test. for most kinds of cancer.
22:29
Unidentified speaker (UNKNOWN): Right.
22:29
Unidentified speaker (speaker_1): Because
22:29
Unidentified speaker (speaker_0): I think that's the other thing is I've heard some people they get an MRI, they get one of these full body MRIs, it's clear and they're like, Thank goodness I don't have, you know, I don't have any cancer. And then part of me is as a clinician thinks, Oh my gosh, I hope you are still doing your colon cancer screening.
22:43
Unidentified speaker (UNKNOWN): Right.
22:44
Unidentified speaker (speaker_0): I hope you didn't uh skip your mammogram or your cervical cancer screening because that MRI is not going to catch it. Like it's not designed. to catch cancers at those early stages. So that's the other piece, is that
22:56
Unidentified speaker (speaker_0): they're imperfect. They're imperfect. They will miss things. But then when they do catch things, and you brought this point up, well, what do you do with that information? What people don't realize is that just like on the outside of our bodies, we have, you know, we have we have moles, we have spots, we have all these irregularities. on our outside. Most of
23:16
Unidentified speaker (speaker_0): us have those irregularities on the inside, and you don't find out about them until you do something like an MRI or something.
23:24
Unidentified speaker (UNKNOWN): So just
23:25
Unidentified speaker (speaker_0): like you know, just like you wouldn't want to get every single mole biopsied. Like, you know, uh it's much harder to do that kind of
23:34
Unidentified speaker (UNKNOWN): uh
23:35
Unidentified speaker (speaker_0): check when you when you get a full-body full body MRI, meaning if you have an irregular spot. somewhere inside on your kidneys or in your pancreas, you may have signed up for years of anxiety watching that thing, you know, watching these spots that otherwise are are gonna be pretty, pretty benign. Oh, maybe maybe in the future AI will help us
23:53
Unidentified speaker (UNKNOWN): understand
23:54
Unidentified speaker (speaker_0): understand this and help us make more use
23:56
Unidentified speaker (UNKNOWN): of
23:57
Unidentified speaker (speaker_0): of this kind of imaging. But I I think the science still has a lot of catching up to
24:01
Unidentified speaker (UNKNOWN): do.
24:02
Unidentified speaker (speaker_1): So true. I actually want to give an example from my real life here. It's not an MRI example, just a diagnostic example. Like my youngest son, he's in middle school and you know, they do like scoliosis screening at middle school
24:18
Unidentified speaker (speaker_1): level. And they screen him and they sent a note out to me at home that hey, he needs to be seen by a doctor. We flagged him in sixth grade, too, and you didn't take him. I'm like, okay. And and there's nothing like it's not visibly apparent, and I've seen his fine
24:39
Unidentified speaker (speaker_1): and everything. And it he looks good. And he's a runner. Like he's a track athlete. He has never complained of anything, um, other than some shin spleen. spleents. And so I'm like, okay, you know, since they are saying maybe. it's a good good idea. So I took him like to an orthopedist and they check him and you know, she just wrote me a note. She's like, no,
24:58
Unidentified speaker (speaker_1): he's all good. You know, if if it's not impeding his functionality, like there's nothing else he has ever complained about, like it's almost like creating an issue out of something that you just saw, right?
25:12
Unidentified speaker (speaker_0): Yeah.
25:12
Unidentified speaker (speaker_1): On the a scan. And um so yeah, I mean, it was all good, but it's exactly similar to that. Like, okay, so you do a screening MRI scan, and all of a sudden, just like you said, there are some irregularities in our bodies, which probably don't impact us ever, right?
25:31
Unidentified speaker (UNKNOWN): And
25:31
Unidentified speaker (speaker_1): then now all of a sudden That and you start just um uh gig g digging into it and creating it a problem when it's not supposed to be.
25:42
Unidentified speaker (speaker_0): Right, right. And yeah, I I think it's a it's a fantastic it's a fantastic example. And
25:47
Unidentified speaker (UNKNOWN): to
25:47
Unidentified speaker (speaker_0): be clear, there there are there
25:50
Unidentified speaker (speaker_0): not all you know, screening by its nature,
25:54
Unidentified speaker (UNKNOWN): uh
25:55
Unidentified speaker (speaker_0): if it's designed the right way, you you are gonna have some false positives. Meaning every screening test has some amount of false positives, right? Because generally we design screening tests so that we don't miss the real things. We'd rather most of the time, we'd rather have a few false alarms so that we don't miss the real alarm. And getting
26:14
Unidentified speaker (speaker_0): that to be calibrated properly,
26:17
Unidentified speaker (UNKNOWN): it
26:17
Unidentified speaker (speaker_0): takes a lot of time and a lot of years. I mean, there's reason that uh there's been so much argument and controversy about how when do we start doing mammograms and when how often do we do them? Do we do them every year? Do we do them every two years? How late do we do them? Because there's a lot of nuance to
26:36
Unidentified speaker (UNKNOWN): well,
26:36
Unidentified speaker (speaker_0): we don't want to biopsy too many people, we want to save lives, but do we want to subject 2,000 women to biopsies they don't need and all the pain and fear and cost that's associated with it? that it's not easy to make those trade-offs. Um, and really smart
26:57
Unidentified speaker (UNKNOWN): uh you know
26:58
Unidentified speaker (speaker_0): groups of clinicians get together and they argue over it, you know, for a long time to try to give us guidelines for how to approach these kinds of of tests. I I think one of the things that that's interesting about where I think the longevity health space is moving and a lot of this direct to consumer
27:18
Unidentified speaker (speaker_0): testing, like the full body MRIs, is moving, is that, you know, traditionally there really weren't easy ways you could get a full body MRI even if you wanted it. And now we're at a place where people can sell you things
27:32
Unidentified speaker (speaker_0): that are interesting to you and they can market them to you in a way that's very appealing before anyone has had a chance to study it and say, well,
27:40
Unidentified speaker (speaker_0): does this help more people than it hurts? And so
27:44
Unidentified speaker (speaker_0): I I don't think it's all I don't to be clear, I don't want to I don't want folks to think that I'm totally against uh this because I'm not. I actually think one of the areas where AI is going to do the most good is by empowering people to take control of their health and to think more
28:02
Unidentified speaker (UNKNOWN): critically.
28:03
Unidentified speaker (speaker_0): about which tests may be right for them, even if they're not right for other people.
28:09
Unidentified speaker (speaker_1): Right.
28:10
Unidentified speaker (speaker_0): And
28:11
Unidentified speaker (speaker_0): again, traditionally we have to think that the guidelines were written for population health management, not for personal health management. And so I do think if we get this right, there is a world where AI helps the right person who definitely would benefit from. a full body MRI, but whose insurance may not cover it because
28:31
Unidentified speaker (speaker_0): insurance and public health may take years to catch up. But that person will know hey, this is actually a good test for me.
28:38
Unidentified speaker (UNKNOWN): And
28:38
Unidentified speaker (speaker_0): it's it's worth the money that I'm going to spend out of pocket for it. Even if my insurance wouldn't pay for it, this is worth doing for me.
28:45
Unidentified speaker (speaker_1): Yeah.
28:46
Unidentified speaker (UNKNOWN): No,
28:47
Unidentified speaker (speaker_1): that's great. So we'll mark that as like a caution and instead. in certain cases and conditions, you know, assessment tests that are coming out, which are AI enabled, would be a good idea. Okay. So let's talk about then how you believe can AI help someone living with a complex
29:06
Unidentified speaker (speaker_1): condition or a family member, especially like caregiving is a big thing. If you have aging parents,
29:14
Unidentified speaker (speaker_1): and and a lot of people I talk to, they are in that stage of life. And you know, it's it's hard. So how can caretakers or family members who are helping coordinate their care, how will AI impact that part of the healthcare equation?
29:31
Unidentified speaker (speaker_0): Yeah, I think it's going to transform it in And mostly almost entirely for the good. Because you're right, if you are managing
29:41
Unidentified speaker (UNKNOWN): the
29:42
Unidentified speaker (speaker_0): health of aging parents, it can be, it can be a full-time job. And it's not an easy job. It's a job that requires you to
29:52
Unidentified speaker (speaker_0): keep track of appointments, keep track of health information, follow up on details. It's both logistically challenging, emotionally challenging,
30:02
Unidentified speaker (speaker_0): cognitive. cognitively challenging, especially if you're trying to figure out,
30:07
Unidentified speaker (UNKNOWN): you
30:07
Unidentified speaker (speaker_0): know, a parent has has you know a dozen and a half different medications, and you have to remember
30:14
Unidentified speaker (speaker_0): how to help them take those medications. And so this was one of the reasons why, probably the biggest reason why I created my doctor friend was recognizing that a lot of people are doing this. this job and it really is. It's a a like for some folks, a full time job to manage all of this complexity.
30:34
Unidentified speaker (speaker_0): And we just until recently did not have the kinds of tools to help people with this. So I think about the tools that I have as a clinician in my in my clinic. So when I if I see someone who's medically complex in my clinic, I'm not doing it by my own, you know, by myself with just, you know, with with uh rudimentary tools. I have a whole staff
30:54
Unidentified speaker (speaker_0): of people helping me with logistics, scheduling, taking vital signs, drawing blood. I have a lab technician. I have an imaging center if I need it.
31:04
Unidentified speaker (UNKNOWN): And
31:04
Unidentified speaker (speaker_0): then I have really sophisticated software. I have an electronic medical record that has billions of dollars of investment that helps me keep track of medical details, allergies, medication interactions.
31:15
Unidentified speaker (UNKNOWN): And
31:15
Unidentified speaker (speaker_0): yet most patients and their families, up until fairly recently, all they had was a patient portal, maybe a notebook,
31:23
Unidentified speaker (UNKNOWN): you know,
31:23
Unidentified speaker (speaker_0): maybe a spreadsheet on their computer. And this is where I think AI is going to allow us to design tools for patients and their caregivers that really match the complexity of the work that they need to do. Because the work they have to do. you know, ultimately, patients and their families are responsible for their
31:43
Unidentified speaker (speaker_0): health most of the time. Only a tiny, you know, half a percent, 1% of their life is going to be spent in a,
31:50
Unidentified speaker (UNKNOWN): you know,
31:51
Unidentified speaker (speaker_0): in a clinic,
31:53
Unidentified speaker (UNKNOWN): uh,
31:53
Unidentified speaker (speaker_0): in some sort of uh sort of clinical setting. Most of the time, they're the ones who are responsible. And so where I think AI can help tremendously, they're in a few, areas. One is understanding, and you said this at the start, which is helping people understand the jargon, the labs, the diagnostic test results. It can help them understand what was
32:13
Unidentified speaker (speaker_0): discussed at an appointment. But maybe even more helpful than that, it can help them prepare and get the most out of those appointments. Meaning
32:24
Unidentified speaker (speaker_0): as people have questions between visits, being able to summarize
32:29
Unidentified speaker (UNKNOWN): all
32:30
Unidentified speaker (speaker_0): of the issues that they may need to discuss with a clinician, present that in an organized way with a clinician means that
32:37
Unidentified speaker (speaker_0): clinicians, when they when they see you or they see you and and your you know and your parent, instead of spending most of that appointment time trying to get the information from you, if you are using the right tool that helps you come up with an agenda, helps you throw all the information in an organized way, you're going to get a lot more out of that that
32:57
Unidentified speaker (speaker_0): appointment time with a clinician than you otherwise would. And it's it's not just abstract. It's not like, oh, you're going to get more out of this. You will be able to discuss more of the health issues that are concerning, right? So you're not just going to cover the top two concerns. You may be able to cover the top five.
33:15
Unidentified speaker (speaker_0): Or
33:16
Unidentified speaker (UNKNOWN): you
33:17
Unidentified speaker (speaker_0): may have time not just to discuss
33:19
Unidentified speaker (UNKNOWN): the
33:20
Unidentified speaker (speaker_0): treatment plan for one, two, or three of those conditions, maybe five or six conditions. So your healthcare can move more efficiently, more quickly if you, and/or the folks you're providing care for, if you become more active collaborators in your healthcare. And this is something that
33:39
Unidentified speaker (speaker_0): that the reason we called our our
33:43
Unidentified speaker (UNKNOWN): uh
33:43
Unidentified speaker (speaker_0): we called it my doctor friend is that this in many ways is is the same thing as having
33:49
Unidentified speaker (UNKNOWN): a
33:49
Unidentified speaker (speaker_0): doctor come with you to a doctor's appointment. And I see this all the time in clinic, which is that patients who come with either a child who's a clinician, doesn't have to be a doctor, a nurse, uh someone who has that
34:01
Unidentified speaker (UNKNOWN): healthcare professional,
34:02
Unidentified speaker (speaker_1): yeah.
34:03
Unidentified speaker (speaker_0): Professional knowledge,
34:05
Unidentified speaker (UNKNOWN): they
34:05
Unidentified speaker (speaker_0): know the right questions to ask. They come with the agenda, they know when to push on the system and they get better care. And this has been studied. You asked at the start, like, has AI been studied? Not as well as uh doctor friends and clinician friends and families have been studied Dude, because this has been shown that people who have a clinician who's a close family
34:25
Unidentified speaker (speaker_0): member.
34:26
Unidentified speaker (UNKNOWN): They
34:26
Unidentified speaker (speaker_1): can add
34:26
Unidentified speaker (UNKNOWN): advocate,
34:27
Unidentified speaker (speaker_1): you know, they can advocate
34:28
Unidentified speaker (UNKNOWN): for the right things.
34:30
Unidentified speaker (speaker_0): They get better care and they live longer.
34:32
Unidentified speaker (UNKNOWN): Oh
34:33
Unidentified speaker (speaker_1): wow.
34:34
Unidentified speaker (speaker_0): Yeah, so so the evidence is clear. And so this is where, you know, it's not
34:39
Unidentified speaker (UNKNOWN): it's
34:39
Unidentified speaker (speaker_0): not a d you can't make a direct uh logical connection. Well, AI is going to do the same thing as a as a clinician in the family, but that was our aspiration at least. And that's where I think um that's one of the best places where I think AI can help right now.
34:54
Unidentified speaker (speaker_1): Yeah. Yeah. What I'm hearing from you is it makes the whole process of that appointment that you have with the physician, uh, where you're taking your um parent or and or family member, or just going yourself more efficient. More efficient. Uh, we all know doctors
35:14
Unidentified speaker (speaker_1): only have like 15 minutes per patient, right? So if you can get the most important thing, if you can talk to your physician in those 15 minutes about the most important thing. And or maybe rather than just the the thing that's on top of your mind. And so you, you know, you're not going back in another four
35:34
Unidentified speaker (speaker_1): weeks unless it's necessary, asking, okay, so how what can I do to, you know make sure that I'm staying well for the next two months, something like that. You know, all of that, like follow up questions that you have. A lot of times like patients would go home and they would think of thou those. If you already have those organized in uh my
35:54
Unidentified speaker (speaker_1): doctor friend um app, then you know you're not scrambling for those and you're not like, oh, I totally forgot to ask her this. Like I should have asked him. You know, stuff like that. And um like my sister is in the same boat because my parents are in SF. And um so she takes them to their doctor's appointments. And when
36:14
Unidentified speaker (speaker_1): I visit, I do the same thing there at um the their healthcare is at Kaiser.
36:19
Unidentified speaker (speaker_1): But that's the So thing my parents are very old school. So they have like this folder with everything still like all the EMR printouts in their oh this time when we go, make sure you take this form with you because we this is what this doctor needs. And I've been kind of like
36:39
Unidentified speaker (speaker_1): trying to convince them to like just move to everything like digital. Um, but I'm gonna share uh the portal with my sister. Like it would really help her.
36:49
Unidentified speaker (UNKNOWN): Yeah,
36:49
Unidentified speaker (speaker_0): please do. And that that was another, I mean,
36:52
Unidentified speaker (UNKNOWN): that
36:53
Unidentified speaker (speaker_0): I think is another place where this can be really helpful. Is
36:56
Unidentified speaker (UNKNOWN): um, you know,
36:57
Unidentified speaker (speaker_0): my doctor friend certainly can pull records from multiple places and it can allow you to share these records with family members. So we do have quite a few folks who are managing, you know, where you know, three siblings are managing two parents, you know, health issues. And everyone can be on the same page because they can share all these tests. test results, they can share the conversations.
37:17
Unidentified speaker (UNKNOWN): Um,
37:18
Unidentified speaker (speaker_0): and rather than having to shuffle that stack of papers from one person to the next, I will add one additional piece, and that is, you know, I think a lot of folks have become accustomed to AI note takers, whether in their meetings or in their professional lives, but increasingly these are now available to patients for visits with their clinicians. clinician. So, you know, we have one in in my doctor friend.
37:38
Unidentified speaker (speaker_0): we call it the visit companion. Not only does it take notes during the visit, but it can suggest questions in real time, meaning it will listen and say, hey, you should ask this question while you're there in the appointment, so you don't forget. And so you don't think about it when you get home. So
37:56
Unidentified speaker (UNKNOWN): I
37:56
Unidentified speaker (speaker_0): think the tools are only going to get better and and more sophisticated and
38:02
Unidentified speaker (UNKNOWN): uh
38:03
Unidentified speaker (speaker_0): the
38:05
Unidentified speaker (speaker_0): where I get so excited as a clinician is that it's it's so much better to be able to
38:15
Unidentified speaker (UNKNOWN): um
38:15
Unidentified speaker (speaker_0): to have a collaborative relationship with your patient and/or the the folks that are
38:20
Unidentified speaker (UNKNOWN): um
38:21
Unidentified speaker (speaker_0): taking care
38:21
Unidentified speaker (UNKNOWN): giving care to them
38:22
Unidentified speaker (speaker_0): because it
38:23
Unidentified speaker (UNKNOWN): I
38:23
Unidentified speaker (speaker_0): think everyone everyone does better. Like patients do better. I've I feel less anxious when I know that I'm working with family that is on top of the details, that are they're keeping track of things, that they're curious, they're trying to learn.
38:36
Unidentified speaker (UNKNOWN): Um
38:38
Unidentified speaker (speaker_0): learn about their condition. They're they're you know, they're keeping a close eye on things.
38:44
Unidentified speaker (speaker_2): Yeah, absolutely. Absolutely. I love the uh fact that your app has a note note-taker taker component. And yeah, you can just have it active. Um I I take it like there would be like a button, something to activate it while w while you're talking to your physician or where while you're at the appointment and it can suggest
39:03
Unidentified speaker (speaker_2): stuff, right? Um that's that's amazing. That's great.
39:07
Unidentified speaker (speaker_0): Oh, thank you. Yeah, and and uh to be clear, I think first of all, I recommend to folks anyone, whether you're using uh, you know, our tool, no tool, any tool, if you have a complex visit with a clinician, you can always I mean, some different states have different rules around recording. Many of them, including
39:27
Unidentified speaker (speaker_0): California, do require that you get consent. So you do need consent.
39:31
Unidentified speaker (speaker_2): Yeah, I remember last time I took my mom, position was asking that I'm gonna use scribe and I have a note taker. Is
39:37
Unidentified speaker (UNKNOWN): that okay?
39:37
Unidentified speaker (speaker_2): And I'm like,
39:38
Unidentified speaker (UNKNOWN): Yeah, that's
39:39
Unidentified speaker (speaker_2): fine.
39:39
Unidentified speaker (speaker_0): And that should give you permission to ask, well, okay, if I, you know, if I record and whether you use an AI tool or just an old fashioned, you know, uh voice recorder on your phone, you can still take those notes, share them with AI,
39:55
Unidentified speaker (UNKNOWN): uh, to
39:56
Unidentified speaker (speaker_0): help you understand what happened during the visit. Meaning, I think
39:59
Unidentified speaker (UNKNOWN): people
40:00
Unidentified speaker (speaker_0): should feel more empowered to do more in their health because of these tools.
40:05
Unidentified speaker (speaker_2): Yeah, no, absolutely. I I think it's a great tool to have
40:09
Unidentified speaker (UNKNOWN): uh
40:09
Unidentified speaker (speaker_2): for patients. And I want to say, like, once the physician probably s sees it and they're like, Oh, this is so good. Like, you know, this actually helped us with the conversation we were gonna have with this patient too, and with the visit. So yeah, no, that uh all of that sounds great. And you m actually made a great point that yeah, in
40:29
Unidentified speaker (speaker_2): California, especially in Kaiser system, they still ask. So uh that's a great uh place to actually ask the physician, hey, I have this app, I'm gonna just activate it. Are you fine with it? So yeah, no, that's great. Okay. I wanna talk to you a little bit about
40:48
Unidentified speaker (speaker_1): all the technology we have on our hands. which might not even be medical technology, but it is g giving us data like you know, I wear my garment watch and Whoop now has like it has really good data about it gives you a good data about your HRV or sleep
41:08
Unidentified speaker (speaker_1): patterns, sleep speech score, right? A lot of people are actually now CGMs, right? People wear CGMs all the time, which I
41:17
Unidentified speaker (UNKNOWN): think is very
41:17
Unidentified speaker (speaker_2): beneficial,
41:18
Unidentified speaker (speaker_1): honestly. But
41:20
Unidentified speaker (speaker_1): um so people are coming to appointment with all this data. They've done this research with their AI, uh, you know, LLMs and whatnot, and they're coming to the physician informed, much m better informed than they were before. And yes, it's a great thing to have. Um, what do you think of
41:40
Unidentified speaker (speaker_1): how this aspect
41:42
Unidentified speaker (speaker_2): of what how AI is changing uh the physician patient interaction and with all the health data that patients are bringing in with them?
41:52
Unidentified speaker (UNKNOWN): Yeah,
41:53
Unidentified speaker (speaker_0): I love that. I love that question. Um so one of the things I maybe didn't share is that I worked on wearable one of the things I did when I was at Google was work on wearable devices. So I'm very familiar with wearable data
42:04
Unidentified speaker (speaker_0): and all the things you can and do with it and all the things people would like to do with it. I think traditionally the data was overwhelming for clinicians, meaning, I mean, this is now we've had wearables for a long time. Yeah. And I can't tell you, I've had many patients show up to appointments with whether it was like
42:23
Unidentified speaker (speaker_0): printed pages of like heart rate data and HRV data and you know, you name it. And as a clinician, we just don't I think traditionally it's been hard to make much of this data, partly because there are a few reasons. One is you're we're not trained. Like wearable data is not part of medical training,
42:43
Unidentified speaker (speaker_0): generally speaking. Um, so we don't we don't really have even a place for it in our medical record system. There's no real easy place for us to most medical record systems don't even have a place for wearable data. Um
42:57
Unidentified speaker (UNKNOWN): I
42:57
Unidentified speaker (speaker_0): I do think
42:59
Unidentified speaker (UNKNOWN): uh
42:59
Unidentified speaker (speaker_0): I'll I'll put a like a few things out there. One is
43:04
Unidentified speaker (speaker_0): uh not all wearable data is is equal. Not all wearables are equal. That's
43:10
Unidentified speaker (UNKNOWN): a
43:10
Unidentified speaker (speaker_2): great
43:10
Unidentified speaker (speaker_1): point.
43:12
Unidentified speaker (UNKNOWN): I want to
43:12
Unidentified speaker (speaker_1): dig into it.
43:13
Unidentified speaker (UNKNOWN): Yeah,
43:13
Unidentified speaker (speaker_0): yeah, yeah. So so this is just more like a public service announcement because I don't think folks are aware of it. I've written about this, but um
43:22
Unidentified speaker (speaker_0): especially, you know, many of the wearable devices use laser light to measure things like heart rate and blood oxygenation, which can be used for measuring sleep, um, and all the all the things that sort of come from those measurements. One thing people don't know is that um many of where wearable devices are not as accurate
43:42
Unidentified speaker (speaker_0): for people with darker skin tone. So No,
43:45
Unidentified speaker (UNKNOWN): I
43:45
Unidentified speaker (speaker_2): didn't know that. That's
43:47
Unidentified speaker (speaker_0): yeah, okay,
43:47
Unidentified speaker (UNKNOWN): tell
43:48
Unidentified speaker (speaker_0): me
43:48
Unidentified speaker (speaker_2): more.
43:48
Unidentified speaker (UNKNOWN): Yeah, so this is
43:49
Unidentified speaker (speaker_0): this is this is
43:50
Unidentified speaker (UNKNOWN): I have like
43:51
Unidentified speaker (speaker_1): this tan
43:51
Unidentified speaker (UNKNOWN): skin, olive skin will work.
43:53
Unidentified speaker (speaker_0): So it's not it's not published enough, it's not published enough, but
43:57
Unidentified speaker (speaker_0): because uh darker skin absorbs more light,
44:01
Unidentified speaker (UNKNOWN): it
44:02
Unidentified speaker (speaker_0): means that. the the uh the sensors on these devices, I'll show you my Apple Watch, which are on the back.
44:08
Unidentified speaker (UNKNOWN): Sure.
44:09
Unidentified speaker (speaker_0): If you have darker skin, they don't get as much signal because more of the signal is absorbed by darker skin.
44:15
Unidentified speaker (UNKNOWN): And
44:15
Unidentified speaker (speaker_0): so the way device makers so true.
44:18
Unidentified speaker (speaker_2): Look how uh my mine is already blinking, like it's it keeps blinking, but that's how it's
44:21
Unidentified speaker (UNKNOWN): like. Well, yours. looks like it
44:22
Unidentified speaker (speaker_0): has. I don't know if it has two, two, Um
44:25
Unidentified speaker (UNKNOWN): uh two,
44:26
Unidentified speaker (speaker_0): I can't. see closely enough. And granted, I am not I am not a hardware.
44:29
Unidentified speaker (UNKNOWN): Yeah,
44:29
Unidentified speaker (speaker_2): mine is two blinkers.
44:31
Unidentified speaker (speaker_0): Yeah. You know, I think some devices have more, and and some of the devices, I think Apple Watch, I'm a
44:38
Unidentified speaker (UNKNOWN): uh
44:38
Unidentified speaker (speaker_0): I do not work for Apple. So no,
44:40
Unidentified speaker (UNKNOWN): no, I I get
44:41
Unidentified speaker (speaker_1): it.
44:42
Unidentified speaker (UNKNOWN): It's not
44:42
Unidentified speaker (speaker_3): a product.
44:43
Unidentified speaker (UNKNOWN): We shared the
44:43
Unidentified speaker (speaker_1): disclaimer, we are good.
44:44
Unidentified speaker (speaker_0): Yes, yes. But but um Apple Watches traditionally have had terrible battery life, they only last about a day, day and a half. But one of the reasons why that's the case is they use a much stronger signal to get much more accurate readings. The devices that have much longer battery lives don't have as strong signals and they tend to be not not be as accurate as a result.
45:04
Unidentified speaker (speaker_0): So that's just like a word of caution to your to your your audience. Like I I tell people if you have darker skin tone, like strongly consider
45:12
Unidentified speaker (UNKNOWN): uh
45:13
Unidentified speaker (speaker_0): getting an Apple device. The Apple devices are the most accurate, are I think Garnet has some that are quite accurate, and there are some exceptions, but generally speaking, Apple watches are the most accurate when it comes to sensors.
45:25
Unidentified speaker (UNKNOWN): Um the
45:26
Unidentified speaker (speaker_0): second thing I'll say, my second PSA is that I think
45:32
Unidentified speaker (speaker_0): while a lot of the data that we get from wearables is is great. And I think um, for example, the Apple Watch has AFib detection. The sleep monitoring is getting a lot more sophisticated and helping people to identify certain sleep issues. I think for um
45:48
Unidentified speaker (speaker_0): for people who are trying to
45:53
Unidentified speaker (UNKNOWN): uh
45:53
Unidentified speaker (speaker_0): you know trying to optimize their fitness, I think HRV can be a wonderful way for people. And and you know a lot of these these wearable devices have their own sort of scores that they develop
46:03
Unidentified speaker (speaker_0): But I try to convince folks to not forget about some of the things that wearables don't measure. So this is especially for folks who are interested in longevity, blood pressure is far more important to know than pretty much any metric you're going to get from a wearable.
46:18
Unidentified speaker (UNKNOWN): So
46:19
Unidentified speaker (speaker_0): if you don't know your VO2, you know, if you know your VO2 max and what your VO2 max trends are, but you don't know what your blood pressure is, that's a problem.
46:27
Unidentified speaker (speaker_0): That's a problem because
46:29
Unidentified speaker (UNKNOWN): uh
46:30
Unidentified speaker (speaker_0): you could have very good VO2 max, but your blood pressure can be out of control,
46:34
Unidentified speaker (UNKNOWN): you know,
46:34
Unidentified speaker (speaker_0): can be out of be not in a healthy place. Unfortunately, none of these wearables does a really good job of measuring blood pressure. And so that's one of the downsides. The marketing for wearable devices has made people care less about some of these things that are probably from a longevity standpoint. more important to know, but just harder and and more
46:54
Unidentified speaker (speaker_0): inconvenient to measure, whether that's blood pressure, cholesterol. So
46:59
Unidentified speaker (speaker_0): um
47:00
Unidentified speaker (UNKNOWN): that's that was a bit of a
47:01
Unidentified speaker (speaker_0): digression. It was a bit of a digression from your question.
47:04
Unidentified speaker (speaker_2): No, I like it. I like it you shared that. And I agree because a lot of people are now so obsessed with all the um physiologic data we can say. Um like you know uh just like you said sleep quality and score hrv uh vo two max all of those are important
47:24
Unidentified speaker (speaker_2): in the grand scheme of longevity but not as important the the first basis or foundation should be do you know your cholesterol total cholesterol LDL, l dl hdl um and then you can dive deeper into LP little A and um right. Um,
47:43
Unidentified speaker (speaker_1): and do you know your A one C? I know a lot. Yes,
47:46
Unidentified speaker (UNKNOWN): your
47:46
Unidentified speaker (speaker_0): A one C, yes, not to forget.
47:47
Unidentified speaker (UNKNOWN): Yeah.
47:48
Unidentified speaker (speaker_1): They they're coming to me with all of this, and then I'm like, Do you know your pre-diabetic? And they're like, What? What what? That's
47:55
Unidentified speaker (speaker_0): right. That's right.
47:56
Unidentified speaker (speaker_1): Right?
47:57
Unidentified speaker (speaker_0): And I think that's I think, you know, to your question, which is like, well, what do people how do people present this to clinicians? I do think. that's part of where there's sometimes a disconnect between what clinicians think and what patients think about wearable data, because some patients will bring in all this wearable data and they they feel very proud of all the data that they've been collecting and they've been following and they've been using it. And the doctors in their
48:17
Unidentified speaker (speaker_0): mind are thinking, Oh, I gotta know what this person's cholesterol, their A1C is, their blood pressure is. Like, like they're thinking the metrics and and measurements that they're thinking about are really different. Part of it is medical training, but part of it is also um I don't think again, I I think the marketing industry the uh wearable industry through its
48:37
Unidentified speaker (speaker_0): marketing has done kind of a disservice to a lot of areas of preventive care by focusing so much on these things that they can measure, um, to the exclusion of the things that they can't measure, but which are again often more important. Um, what I what I do what we're there is like
48:57
Unidentified speaker (speaker_0): there is a positive part of this, which is I do think there are patterns in this wearable data that
49:05
Unidentified speaker (speaker_0): over the next few years, AI may help us understand signals that we were never able to detect in this wearable data that may be really important. And I I think. the the hint of this uh comes from some of the recent AI models that were developed to interpret EKGs. I don't know if you know about this. Okay, but for your audience members who
49:25
Unidentified speaker (speaker_0): don't know what an EK, EKG
49:26
Unidentified speaker (UNKNOWN): is
49:26
Unidentified speaker (speaker_0): the thing where they put the sticky things on your chest and you
49:28
Unidentified speaker (UNKNOWN): stress tests.
49:29
Unidentified speaker (speaker_0): Yeah, well, the swe you get those sort of uh pink sheet with the squiggly lines.
49:34
Unidentified speaker (speaker_2): That's what I do.
49:35
Unidentified speaker (UNKNOWN): The
49:35
Unidentified speaker (speaker_2): squiggly spiking
49:36
Unidentified speaker (speaker_0): lines. Well, well, traditionally. traditionally we could use that. You know, we could use that to detect abnormal heart rhythms. We could use it to detect, you know, acute heart attacks and and certain kinds of heart disease. I mean, there were there were a fair number of things we could detect, but now using AI algorithms, they're able to just in these micro patterns that we
49:56
Unidentified speaker (speaker_0): never saw.
49:57
Unidentified speaker (UNKNOWN): Um,
49:58
Unidentified speaker (speaker_0): we've been reading AKGs for decades, but these algorithms can now find major structural heart issues that previously you needed an ultrasound of the heart to detect.
50:10
Unidentified speaker (speaker_0): So this is where there are, and it similarly, this goes for the photos of the eye that previously, you know, we've been looking at retinas in the eye to look for signs of diabetic retinopathy, but now we're discovering that wow, this can be used for cardiovascular risk detection because the AI is able to detect these.
50:27
Unidentified speaker (speaker_0): really just unbelievable patterns that were invisible to our eyes. And that's where I get excited about this, about sort of wearable and continuously gathered data is I think over the next five, 10 years, we're gonna learn things that will um, you know, may just transform our understanding of what this wearable technology can
50:46
Unidentified speaker (UNKNOWN): do.
50:47
Unidentified speaker (speaker_2): Absolutely. And I wanna want to underscore something you just said. I gave a presentation at uh Medical Fitness Foundation earlier this year, and retinal scans can tell you a lot about your health. Like if you are
51:04
Unidentified speaker (UNKNOWN): in
51:04
Unidentified speaker (speaker_2): the process of developing a chronic condition, especially like your blood pressure, you know, if it has been elevated, like your retina has uh small vessels,
51:13
Unidentified speaker (speaker_1): if they've been bursting. bursts in the past, like they can see that on the skin. Oh
51:18
Unidentified speaker (speaker_0): yeah.
51:18
Unidentified speaker (speaker_1): And if I'm not wrong, it the price might have changed, but when I was gonna give that talk, it was it runs around seventy to eighty dollars. You can go to Warby Parker and schedule it for yourself. And I'm like, see how accessible the technology now is.
51:36
Unidentified speaker (speaker_0): Yeah. I think that's yeah, so you can. I
51:39
Unidentified speaker (UNKNOWN): mean,
51:39
Unidentified speaker (speaker_0): a lot of optometrists' offices, they do this, you know, they didn't do it for for longevity screening. They did it to screen for, you know, for braces,
51:46
Unidentified speaker (UNKNOWN): but
51:47
Unidentified speaker (speaker_0): eye conditions, you
51:48
Unidentified speaker (UNKNOWN): know,
51:48
Unidentified speaker (speaker_0): for glaucoma or uh macular degeneration. But you're right. I mean, they're very accessible. I will say some of those AI models and algorithms that you need to to be able to are are still in the research domain, meaning you can't just go to you know, uh AI retinal scan.com and upload your photo. I mean, someone correct me.
52:08
Unidentified speaker (speaker_0): Your audience
52:10
Unidentified speaker (speaker_0): it was already out there. Because I do think people are going to commercialize these models, but I think that's that's one of the places where I'm I'm really excited to see, you know, um AI start to to transform both prevention, risk calculation
52:27
Unidentified speaker (UNKNOWN): in
52:27
Unidentified speaker (speaker_0): a way that's distinct from what ChatGPT and and these other large language models are doing, because these are different kinds of AI models. They're not, they're not models that you can have a lengthy conversation with about your health and then,
52:39
Unidentified speaker (UNKNOWN): you know,
52:39
Unidentified speaker (speaker_0): do your do your European travel planning with. Like they're really like hyper focused on, you know, just your eye, just just your health from from your eye data.
52:49
Unidentified speaker (speaker_2): Absolutely. Absolutely. No, that's what that was a great addition, actually. Thanks for highlighting that.
52:55
Unidentified speaker (UNKNOWN): Um
52:56
Unidentified speaker (speaker_1): okay, I want to talk about so wearable data. Thanks for adding all the information to my knowledge too. So proceed with caution to try to prioritize the the message here for the listener and audience is prioritize your biomarkers. Uh with your blood test. Like,
53:16
Unidentified speaker (speaker_1): you know, if you have a good handle of that, make sure you're getting those every year, like going to your physician, getting those, uh and um have your physician read those labs for you. And then on top of it, then you have good data from here, the the wearables. That's good. But don't ignore that data. So that
53:36
Unidentified speaker (speaker_1): that's great. Also, retinal scan is a great way to assess some of the health risks. So go get that.
53:44
Unidentified speaker (UNKNOWN): I
53:44
Unidentified speaker (speaker_1): want to talk about safety now. So that means like, okay, wearables, we we'll just say, okay, don't believe it too much. Like still proceed with caution. Prioritize your um annual blood test.
53:59
Unidentified speaker (speaker_1): What are the risks of asking the
54:03
Unidentified speaker (speaker_1): general purpose like AI chatbot or LLM health questions? Because a lot of patients are doing that now. It has replaced Google, right? We used to have Dr. Google. Now it's now it's Dr. Claude or Dr. Gemini, right? So how can someone safely do it? And I'm going gonna to just um tag the
54:23
Unidentified speaker (speaker_1): next question that I have because I think it goes in conjunction with this. question. Is when should someone just actually stop relying on that information that they're getting from that AI app or and or LLM and you know, actually see the physician do not delay the care. Like,
54:41
Unidentified speaker (UNKNOWN): you
54:41
Unidentified speaker (speaker_1): know,
54:42
Unidentified speaker (speaker_0): yeah.
54:42
Unidentified speaker (speaker_1): I'm actually more afraid that people are gonna Going start to relying on this information so much that now they're not getting help
54:49
Unidentified speaker (speaker_0): at
54:49
Unidentified speaker (speaker_1): all.
54:50
Unidentified speaker (speaker_0): Good. good questions. I'm glad we're getting to this because this is, you know, maybe maybe we're saving the mediest part for the end of our chat. Uh, because I think you're right. I mean, there were surveys done, I think this was within the last year, but a significant percentage of people are substituting care with AI, meaning people are already
55:10
Unidentified speaker (speaker_0): starting to see
55:11
Unidentified speaker (UNKNOWN): their
55:12
Unidentified speaker (speaker_0): doctor less because they're using AI as a substitute with. which
55:16
Unidentified speaker (speaker_0): can be problematic. So the the there are a few sort of categories of safety issues that I think people should be aware of. And I'll name them and then I'll go through well, I'll name them, explain, you know, uh define them, and then I can go in into a bit more detail. But the the first sort of concern about AI for some people, not for everyone, is privacy. So do
55:36
Unidentified speaker (speaker_0): you know
55:37
Unidentified speaker (UNKNOWN): what's
55:38
Unidentified speaker (speaker_0): happening with your data? when you share it with
55:42
Unidentified speaker (UNKNOWN): people or
55:42
Unidentified speaker (speaker_0): Demon?
55:42
Unidentified speaker (UNKNOWN): Yeah,
55:42
Unidentified speaker (speaker_1): I mean, there's no HIPAA.
55:44
Unidentified speaker (speaker_0): Yeah, they're not, they're not generally bound by HIPAA. And and different kinds of consumer health AI companies have different policies around privacy and your data. The second thing is um hallucination. So models, AI models making things up that are not true.
56:02
Unidentified speaker (UNKNOWN): Um
56:03
Unidentified speaker (speaker_0): all AI models have a risk of making things up.
56:07
Unidentified speaker (UNKNOWN): Um,
56:07
Unidentified speaker (speaker_0): people have called it hallucination. I think a better term is confabulation.
56:11
Unidentified speaker (UNKNOWN): Um if
56:12
Unidentified speaker (speaker_0): you'll pardon my language, but the you
56:13
Unidentified speaker (UNKNOWN): know,
56:13
Unidentified speaker (speaker_0): the the chief of medicine at UCSF has a really a much plainer, maybe better term, which is bullshitting. You know, he says all models have this tendency and it's just how they're built. That's Bob Wachter, who's a mentor of mine. And then the the third thing I would say is, and these are these are sort of two related things, which is models forget.
56:34
Unidentified speaker (speaker_0): They forget things that you may have told
56:36
Unidentified speaker (UNKNOWN): them,
56:37
Unidentified speaker (speaker_0): and they may not have context, meaning they don't necessarily know the things about you that maybe you think they should.
56:46
Unidentified speaker (UNKNOWN): Um and
56:47
Unidentified speaker (speaker_0): I will say the last the last one here is sycophant. sycophancy. For people who don't know what that term is, sycophancy is this tendency of AI models to tell you what you want to hear. So they want to please you. Oh
57:00
Unidentified speaker (speaker_2): yeah. And
57:01
Unidentified speaker (UNKNOWN): this can be
57:01
Unidentified speaker (speaker_0): a risk. This can be a risk for your health. And there there have been a few stories now in the national press. Unfortunately there was a story in the New York Times that was just just um just devastating
57:14
Unidentified speaker (UNKNOWN): of a
57:14
Unidentified speaker (speaker_0): young young woman, an engineer
57:17
Unidentified speaker (speaker_0): who was struggling with a seizure disorder. And she was, she had a doctor, she had a neurologist. I think it's this was the chief of neurology at our hospital that she was seeing for her seizures. Unfortunately, her seizure medication
57:34
Unidentified speaker (UNKNOWN): uh
57:34
Unidentified speaker (speaker_0): came with some side effects that were affecting her work performance. And so without telling her doctor, and in consultation with Dr. GPT,
57:43
Unidentified speaker (UNKNOWN): um, she
57:44
Unidentified speaker (speaker_0): started lowering the dose of her seizure medication.
57:47
Unidentified speaker (speaker_2): Oh, no.
57:48
Unidentified speaker (speaker_0): Without talking to her doctor. And they had these extensive conversations. And I think the model at some point said, you should talk to your doctor about it, but then would sort of go along with
57:59
Unidentified speaker (UNKNOWN): it.
58:00
Unidentified speaker (speaker_0): She kept reducing her. her dose, I think with the goal of trying to get off the medication, didn't talk to her doctor and then
58:07
Unidentified speaker (speaker_2): Yeah, to wean herself off. Yeah. To
58:09
Unidentified speaker (speaker_0): wean herself off. And tragically she had she had a major seizure event in her apartment by herself and she died.
58:17
Unidentified speaker (UNKNOWN): Oh
58:17
Unidentified speaker (speaker_2): no.
58:17
Unidentified speaker (speaker_0): She died. And and so it's it's just a horrifying example of how
58:22
Unidentified speaker (UNKNOWN): um you know
58:24
Unidentified speaker (speaker_0): you raise the question, you know,
58:26
Unidentified speaker (UNKNOWN): this
58:26
Unidentified speaker (speaker_0): is this is the real safety concern, which is that people use this technology in lieu of seeing
58:35
Unidentified speaker (UNKNOWN): a
58:35
Unidentified speaker (speaker_0): doctor. Now, I can understand the temptation and I think we have to put that out there. It's not easy for people to see a doctor. People don't always have the luxury of being able to see a doctor about a health issue that they have at 10 PM on a Friday night.
58:48
Unidentified speaker (UNKNOWN): Sure.
58:49
Unidentified speaker (speaker_0): And or they may be afraid of going to a an emergency room or an urgent care clinic because of the expense. So totally get it. Where I think
58:58
Unidentified speaker (UNKNOWN): you
58:58
Unidentified speaker (speaker_0): know it it's important to draw the line is
59:01
Unidentified speaker (UNKNOWN): if
59:02
Unidentified speaker (speaker_0): you're someone who has serious, a serious health issue, and I think seizure disorder is one example of it,
59:09
Unidentified speaker (speaker_0): you have to be really careful when it comes to questions
59:13
Unidentified speaker (UNKNOWN): um
59:13
Unidentified speaker (speaker_0): that are going to influence. a decision that you make. You said this at the start of our, you know, this disclaimer at the start of our our talk, but AI is not good at uh suggesting medication dosing changes. It usually does not have the context it needs. They make math problems, dosing prop dosing
59:33
Unidentified speaker (speaker_0): mistakes.
59:34
Unidentified speaker (UNKNOWN): Um
59:35
Unidentified speaker (speaker_0): the
59:35
Unidentified speaker (UNKNOWN): I'll
59:36
Unidentified speaker (speaker_0): I'll be really brief. I'll I'll go through just a few of a few. things that you can do to use this technology more safely. I think
59:43
Unidentified speaker (speaker_0): for hallucination and making things up, if it's an important piece of information, ask the model to look it up. Say, hey, can you confirm this? Or if you use if you use Dr. Claude,
59:55
Unidentified speaker (UNKNOWN): go
59:55
Unidentified speaker (speaker_0): to Dr. Gemini and
59:57
Unidentified speaker (UNKNOWN): say,
59:57
Unidentified speaker (speaker_0): Hey, I'm having this conversation, copy and paste what you said. Is this right? Or is this model is this model giving me a bunch of BS? Like you can get a second AI opinion. In my doctor friend, we make this easy. You just press a button. You can get a second opinion, third opinion, you can get opinions from a
1:00:13
Unidentified speaker (speaker_0): lot. I
1:00:13
Unidentified speaker (UNKNOWN): love that.
1:00:14
Unidentified speaker (speaker_0): It makes it a lot easier and and faster. Um, in terms of the context and forgetting, ask the AI model, what do you know about my health? Summarize everything that you know about me and periodically ask it every If you're having a very long conversation with AI, make sure you ask this at least once a week. What do you remember from what we've been talking about? Do you
1:00:34
Unidentified speaker (speaker_0): remember my medications? Do you remember my allergies?
1:00:37
Unidentified speaker (UNKNOWN): Do
1:00:37
Unidentified speaker (speaker_0): you remember my? And that will help you identify if the model has forgotten something about you that's really important or critical. Certain systems are designed to have better memory,
1:00:47
Unidentified speaker (UNKNOWN): uh,
1:00:47
Unidentified speaker (speaker_0): so that they don't they don't forget. Um, and then for the Sika sycophancy, family. you know, the telling you stuff you want to hear, this may be the most important piece. And that is
1:00:58
Unidentified speaker (UNKNOWN): uh
1:00:59
Unidentified speaker (speaker_0): second opinion. So get a second opinion from a different model.
1:01:03
Unidentified speaker (UNKNOWN): Um,
1:01:04
Unidentified speaker (speaker_0): ask the model why might you be wrong? Argue the opposite point of view. Why might I be wrong? Are you telling me what I want to hear?
1:01:13
Unidentified speaker (UNKNOWN): Like
1:01:13
Unidentified speaker (speaker_0): these are things you can ask. the AI model to reset itself. And oftentimes they'll be able to catch themselves and say, Yep, you know, I've been kind of telling you what you want to hear. They'll often admit to it, this bad
1:01:24
Unidentified speaker (UNKNOWN): behavior if
1:01:24
Unidentified speaker (speaker_0): you call them on it. But you have to be, again, this is this gets back to this is a powerful tool, but you have to learn how to use it safely. This is one of those things you have to be willing to ask the model, hey, is this
1:01:37
Unidentified speaker (UNKNOWN): are
1:01:37
Unidentified speaker (speaker_0): like, are you telling me just you know, are you trying to make me happy?
1:01:42
Unidentified speaker (speaker_0): Are you trying to convince me that I'm only full of good ideas? Which of these ideas that I've presented is maybe bad or unsafe? And and that may that can sometimes be just a really easy, quick way to protect yourself from some of these bad tendencies that all the AI models animals have.
1:01:58
Unidentified speaker (speaker_2): Yeah. Yeah. Oh, I wanna want to highlight everything you said because it's all so wonderful. Uh this is a great way to end our conversation, honestly,
1:02:06
Unidentified speaker (speaker_1): that, you know, um
1:02:09
Unidentified speaker (speaker_1): that the sycophancy, I didn't know the term. I knew that it actually tries to please you for whatever you're asking. So yeah, double check it, ask it again. Can you can you check it for me? Um ask it to argue in the opposite uh direction to and at the end of the day, the bottom line is like
1:02:29
Unidentified speaker (speaker_1): it's your health, it's your life. Um use some common sense. Yeah,
1:02:35
Unidentified speaker (UNKNOWN): go to
1:02:35
Unidentified speaker (speaker_0): yes, yeah, absolutely.
1:02:37
Unidentified speaker (speaker_0): Yeah, if you're having a serious, if you're feeling really sick or really bad, go
1:02:41
Unidentified speaker (speaker_1): to the doctor,
1:02:42
Unidentified speaker (UNKNOWN): please.
1:02:43
Unidentified speaker (UNKNOWN): Talk to the doctor,
1:02:44
Unidentified speaker (speaker_0): talk to a person. Yes, that's
1:02:45
Unidentified speaker (UNKNOWN): right.
1:02:45
Unidentified speaker (speaker_1): Exactly. And even like the example you just gave me,
1:02:50
Unidentified speaker (UNKNOWN): do
1:02:51
Unidentified speaker (speaker_1): not try to adjust your medication on your own. Like, come on.
1:02:57
Unidentified speaker (speaker_0): Yeah.
1:02:58
Unidentified speaker (speaker_1): Right? This is something you wanna run by a healthcare practitioner, preferably your physician who prescribed it. You don't want to take those kind of actions into your own hands. So, no, this is great. This is so great. And I hope, you know, now listeners of this podcast, they have
1:03:17
Unidentified speaker (speaker_1): good concrete like examples of how AI is being used at this time in healthcare, how it can help them, where to be a little cautious, and where to actually just go to your doctor. Just, you know, yeah, just uh have someone take you if you are not unable to uh drive
1:03:37
Unidentified speaker (speaker_1): yourself. or just go to the nearest ER and yeah get checked. So I want to ask you about my doctor friend. How can our listeners and viewers find out about it? Where is it available? Um, is it on the App Store? Please help us understand that. And where can our viewers and
1:03:57
Unidentified speaker (speaker_1): listeners connect with you?
1:03:59
Unidentified speaker (speaker_0): Ah, okay. Uh well, uh my doctor friend is available. It's in the App Store, the Apple App Store. We don't have an Android app yet working on it,
1:04:07
Unidentified speaker (UNKNOWN): but
1:04:07
Unidentified speaker (speaker_0): it's also available as a website and it and as a mobile website. So my Doctorfriend OneWord dot AI. You can learn more about uh about my Doctor Friend, how it works, how it can help you. And uh it's free to try. So you can try it for free. We also have like a free trial for a week, and you can use all the premium. features
1:04:27
Unidentified speaker (speaker_0): just to see if
1:04:28
Unidentified speaker (UNKNOWN): it's
1:04:28
Unidentified speaker (speaker_0): something that's that's going to help you. And
1:04:31
Unidentified speaker (UNKNOWN): uh
1:04:32
Unidentified speaker (speaker_0): folks can reach out to me. I think we can probably put my sort of contact information in the notes, but you know, we'll have links to both my doctor friend, Instagram, Substack. I'm on a lot of the different platforms and YouTube. I'd love to hear what people think, what their experiences have been with AI suggestions, thoughts. I know. people have a lot of
1:04:51
Unidentified speaker (speaker_0): opinions about AI. One of the things we didn't talk about, but but folks may have questions about maybe this is a follow up topic, but the social environmental costs of AI, which I know are a lot of people's minds.
1:05:04
Unidentified speaker (speaker_1): Yeah. I've
1:05:05
Unidentified speaker (UNKNOWN): been thinking
1:05:06
Unidentified speaker (speaker_1): about it. Yeah, with all the data centers. And also, you know, just the end product of it all, because it's gonna change our society. like in different ways. You know, we are not really thinking of those yet. Not yet. We'll we'll see though. Yeah, but definitely wanna have that follow up conversation with you at
1:05:26
Unidentified speaker (speaker_1): at some point. And I will include the links to my doctorfriend uh dot AI and your Substack, your LinkedIn and YouTube I have in front of me, and that will be in the show notes. So please reach out to uh Michael uh for more information on all the topics that we discussed and questions we discussed today.
1:05:46
Unidentified speaker (speaker_1): Thank you, Michael. Thanks a lot for your time today.
1:05:48
Unidentified speaker (speaker_0): My gosh, thanks for having me on. This was so much fun.
1:05:51
Unidentified speaker (speaker_1): Of course. Thank you so much. And thanks to everyone if you joined us live today and or if you're listening to us afterwards in the podcast format or playing a recording of this on LinkedIn. Bye.
This page was produced automatically: the audio was transcribed with nvidia/parakeet-tdt-0.6b-v3+sortformer-v2.1 and interpreted by Qwen3.6-35B-A3B-UD-IQ3_S.gguf. Transcription and interpretation can contain errors; always confirm against the original episode. Nothing here is investment advice.

More from 9 to 5 Wellness

All insights