Latest / CareTalk: Healthcare. Unfiltered. / Do We Really Need AI in Healthcare?

Transcript

Do We Really Need AI in Healthcare? · CareTalk: Healthcare. Unfiltered.

• 0:01
  David: Artificial intelligence is all the rage from open AI's Chat, G P T to Apples, Siri, and Tesla's self-driving cars.
• 0:08
  David: Consumers are gaining firsthand experience in what AI can do.
• 0:11
  David: But what about AI and healthcare?
• 0:13
  David: Do we really need ai?
• 0:15
  David: Will AI increase personalization, cut costs, and reduce staff burnout?
• 0:19
  David: The potential is there, but it's still early days.
• 0:22
  David: Welcome to Care Talk America's Home for incisive debate about healthcare, business and policy, not to mention a Top five Apple podcast.
• 0:29
  David: I'm David Williams, president of Health Business.
• 0:32
  John: And I'm John Driscoll, the president of Walgreens, south.
• 0:34
  John: David, you've been relying on fake intelligence for years.
• 0:38
  John: Artificial or not, like obviously you, you must be an expert in artificial intelligence, but maybe we start with why are we talking about artificial intelligence and healthcare?
• 0:50
  David: Well, John, I think it's pretty important, uh, to discuss, uh, artificial intelligence.
• 0:53
  David: Everybody's talking about it.
• 0:54
  David: As you said, there's not like a lot of real intelligence, uh, that's out there, but a big reason, John, is that the healthcare, uh, industry is at a bad spot.
• 1:02
  David: You can't get people, we talk about burnout from the tri demic, triple demic, et cetera.
• 1:07
  David: There's all sorts of potential of technology, but costs keep rising.
• 1:11
  David: And satisfaction quality aren't, aren't there.
• 1:13
  David: We gotta do something different.
• 1:14
  David: John and AI seems to have the potential to help us out, but we gotta get into it and understand what it really is.
• 1:22
  John: Well, what is it?
• 1:22
  John: I mean, you know, people have been throwing around these terms, gosh, since I went to college in the eighties.
• 1:28
  John: The notion of, of accelerating machine, I mean, I mean this is, this goes back to the Thomas, the the, the Alan touring days.
• 1:36
  John: Yeah.
• 1:36
  John: In the thirties when they were starting to put together machines that could, machines mirror.
• 1:42
  John: or actually improve Amer uh, intelligence, the promise of ai kind of having an impact has sort of been sort of, it, it, this, it's been sort of a false, it's a false flag for 30 years.
• 1:55
  John: What is this AI thing and what's different at this time?
• 2:00
  David: Yeah, so I mean, I think when we talk about artificial intelligence, it really is about a machine that's imitating intelligent human behavior.
• 2:05
  David: So that's kind of the touring test is, you know, is it a machine or.
• 2:09
  David: Uh, you know, is is it actually a person?
• 2:12
  David: I think it, it is important to understand some of the technologies.
• 2:15
  David: Uh, maybe before we, we jump into some of the specific examples.
• 2:18
  David: So sometimes you hear like, machine learning slash ai, it's used at the same time.
• 2:22
  David: Well, machine learning is using algorithms to process big amounts of data and to learn from it without being explicitly programmed.
• 2:29
  David: You know, you can program a computer to do this, that, or the.
• 2:31
  David: The difference here with machine learning is that the machine isn't being explicitly told what to do.
• 2:36
  David: It's learning.
• 2:37
  David: There's different types of learning.
• 2:38
  David: John, though.
• 2:39
  David: Well, well, but
• 2:39
  John: I, but I, but I'd be careful there, David.
• 2:41
  John: I mean, I think people throw around these terms to confuse people and our job is to make it clear.
• 2:46
  John: So let's, let's back up the truck a little bit and, and stop kind of leaning into these big terms.
• 2:52
  John: Machine learning is simply, More computer power applied to the same problems that will d that will allow, uh, computers to access these, these massive troves of data that are being created.
• 3:06
  John: And then, uh, uh, automate tasks, simplify tasks, and hopefully, you know, particularly in the consumer side, surprised and delight people with retail choices they didn't know they had.
• 3:17
  John: But it effectively, machine learning is simply.
• 3:20
  John: the machine doing more work and then, uh, using what it, what it spits out in practical ways in workflow.
• 3:27
  John: It, I, I just let, let's, let's, let's demystify this.
• 3:30
  John: Let's not, yeah, you may, you may be happy with these big terms, but, but let, let's demystify so we can use them.
• 3:37
  David: So John, I think under machine learning, you know, the, the key thing is, is to understand that there's a lot of, this is sort of supervised, so, which could be quite tedious.
• 3:45
  David: You have to label everything that you want and you train the model which limits how much can be done.
• 3:50
  David: So a lot of the reason that things have been slow is it takes a long time to train these, these models.
• 3:55
  David: For example, you'll say, Here's a cat.
• 3:57
  David: These are examples of a cat.
• 3:58
  David: This is an example of not a cat.
• 3:59
  David: And then you go and have the machine, uh, try it in other areas.
• 4:02
  David: Where we're going to start to get more excitement is in these unsupervised areas where the machine, the computer looks at tons of different data, like heartbeat data or financial trades, and then people can review it to see if it's useful.
• 4:14
  David: And then you've got deep learning.
• 4:16
  David: It's all under machine learning, which is neural networks, which inspired by how the human.
• 4:21
  David: works.
• 4:21
  David: So that's all under machine learning.
• 4:23
  David: So that's one part of artificial intelligence.
• 4:25
  David: There's also natural language processing,
• 4:27
  John: but, but slow, slow, slow down, slow down, slow down, slow.
• 4:30
  John: Let's start with the notion that, you know, supervised learning is simply, again, it's a form of programming, so that in the same way that language.
• 4:40
  John: Is coded for specific description.
• 4:43
  John: A machine can take a lot of data and understand that as the inputs come in, to your point, the image of a cat, the image of a dog, that it's a dog, not a donut.
• 4:51
  John: Uh, but the unsupervised is of similar form of programming where you don't necessarily know what it's going to, how, how it is going to solve a problem.
• 5:02
  John: But you lay out the problem, you leverage a lot of.
• 5:06
  John: , you program it through a program that's called an algorithm.
• 5:09
  John: An algorithm is, is simply a logical application of a, a series of if then statements to drive a, a particular piece of analysis.
• 5:18
  John: But let's, let's make sure we're using language that people understand.
• 5:20
  John: David, I mean, not everybody's a consultant here and uses big language to confuse people.
• 5:25
  John: Yeah.
• 5:25
  John: And then when you go down the road of, of of, of whether it's a neural network, People tend to talk about the methods of the models, like a a, a machine learning, um, which would be more of a regression analysis, and others are trying to accelerate the use of the application of.
• 5:48
  John: The computers effectively by coming up with different approaches to solving problems and analyzing them, whether it's more like or unlike the way the brain, the, the theory early on with ai, the Marvin Minsky, m i t thing that everybody thought was gonna solve.
• 6:02
  John: For, to kind of, kind of, kind of transform com computation and make it a lot faster and smarter, was to try to program a computer with a model of the mind.
• 6:12
  John: And that's where this neural part of the network and a network is simply the way the architecture, the way it's built.
• 6:19
  John: It turns out that that was really, really hard.
• 6:20
  John: Now there have been people there, people argue about what the right algorithm and architecture is, but I think if we explain it, these are simply different ways to approach how we take a lot of.
• 6:31
  John: And drive better insights.
• 6:32
  John: So based on the, the, the, the cough you have, yeah.
• 6:36
  John: A computer can take a lot of data and determine whether it's simply another example of you talking too much, David, or whether you've really gotten the cold.
• 6:44
  John: John, I
• 6:45
  David: think my problem is with listening too much and listening to you, although you may be the only person, uh, around who could get that 218 votes, so be careful what you say and how smart you want to be.
• 6:54
  David: But whether it's, you know, machine learning, Natural language process and computer vision, or robotic process automation.
• 7:00
  David: You know, these are all things that I think are, are relevant.
• 7:03
  David: Yeah.
• 7:03
  David: But
• 7:03
  John: let's, let's explain robotic process.
• 7:06
  John: A what, what the heck is that?
• 7:07
  John: Is that, is that you telling the people who work for you that they gotta work more like robots?
• 7:11
  John: Like what do you, what does that, what
• 7:12
  David: does that mean?
• 7:13
  David: So, robotic processor automation is something that.
• 7:16
  David: It is not always considered part of, uh, ai, but it's really about the automation of these very labor intensive tasks, high volume task, a lot of drudgery that you see in healthcare.
• 7:25
  David: It's more rules, rules based than typical ai, but it, it does incorporate machine learning
• 7:31
  John: sometimes.
• 7:31
  John: And let's, let's, let's, let's be a little bit specific.
• 7:33
  John: So when you're trying to pay a health, Care bill.
• 7:35
  John: Yeah, it's really complicated.
• 7:36
  John: Is someone covered?
• 7:37
  John: Is, is there, are they not covered, are they, where are they in their deductible or accumulator in their benefit?
• 7:42
  John: All these benefits are way too complicated.
• 7:44
  John: Mm-hmm.
• 7:44
  John: , that would be an example where automating the logic based on data and patterns we could do with robotic process.
• 7:52
  John: Automation, often called rpa.
• 7:54
  John: Again, one of the challenges with this AI RPA ml is people leverage the acronyms in the language to sort of push, push every, the rest of us away.
• 8:03
  John: If we're gonna really make use of this stuff in healthcare, we've gotta, we've gotta make it make it clearer so that people can then simplify it, apply to their own, like, what's this computer visualization thing?
• 8:14
  John: Yeah, yeah.
• 8:14
  John: Can it help?
• 8:15
  John: Can it help you and me look better?
• 8:16
  John: I mean, what is
• 8:16
  David: that?
• 8:17
  David: That, that's the way that we can look better, John, is I, I go like this and cover my eyes and that that tends to, that tends to do the job pretty well.
• 8:24
  David: Let's talk about actually how ai, how is AI used in healthcare?
• 8:29
  David: Cuz there's a few different ways that it's, that it's used that, that maybe make it more specific if we haven't already lost all the healthcare people.
• 8:35
  David: John, let's talk about what we can do with it.
• 8:38
  John: Great.
• 8:38
  John: Great.
• 8:38
  John: Well, well, well, perhaps you could come up with something that'll be simple and clear because you, I
• 8:43
  David: don't know, John, struggling throughout, whether, whether it's simple and clear or not.
• 8:46
  David: You give me the artificial, uh, you know, answer that it isn't.
• 8:49
  David: So the first thing that's maybe the most obvious is like an image analysis.
• 8:52
  David: So we're looking at like a mammogram or pathology slide.
• 8:56
  David: And you can get to faster, more accurate diagnoses or even think about at the dentist, you know, you get dental x-ray and you can help the patient and the dentist understand what they're seeing and what to do about it.
• 9:06
  David: So a computer can do a great job of looking at all these patents saying, is that a tumor?
• 9:10
  David: Does it look like other tumors?
• 9:11
  David: Uh, for example.
• 9:12
  David: So image analysis, I think is a real clear example.
• 9:15
  David: Um, Next one, John, is a little bit is, I think, really profound potentially, but it's, it's not so easy to get to, and that is about making healthcare more personalized and responsive.
• 9:26
  David: So part of it is something like building more like immersive 3D experiences like virtual reality.
• 9:32
  David: So for example, some companies are working on this.
• 9:34
  David: If you've got, uh, if you're working in physical therapy and you're doing need to get some specific exercises to work on your neck, Or your shoulder, uh, it gets to be very specific.
• 9:44
  David: And then precision medicine, you know, going beyond, uh, like saying a dia diabetic, you didn't get your eye exam to things that are much more personal and relate to a lot of the different data they have and can help with how someone's actually going to en engage.
• 9:56
  David: So that's sort of personalized, responsive healthcare, John.
• 9:59
  David: But, but that's a,
• 9:59
  John: but that's a really good example.
• 10:01
  John: Finally, you're starting to get into specifics here, David, but for example, In just the county of the, of the borough of Brooklyn.
• 10:09
  John: Yeah.
• 10:09
  John: There's over 200 different languages spoken.
• 10:11
  David: Mm-hmm.
• 10:12
  David: , if we could, they're all, all the, all the various hipster dialects or, uh, well,
• 10:15
  John: those are, yeah.
• 10:16
  John: And, and, and a, and a machine could help, uh, uh, you know, nurses and.
• 10:21
  John: P uh, PAs and PTs and doctors actually translate and contextualize what people are experiencing.
• 10:29
  John: You know, more than 30% of the, of, of a physician's evaluation of someone with a problem is really through back and forth contact.
• 10:36
  John: I, I think the computer assisted, uh, analysis particularly that solves for language and context could be incredibly powerful.
• 10:45
  John: Another.
• 10:46
  John: Exam.
• 10:46
  John: A very specific example to your point is in radiology where you can automate and simplify the scan.
• 10:54
  John: by taking, by by, by taking what a radiologist has developed 20 or 30 years, programming that into a computer, testing it against data, cuz that data is very standardized.
• 11:03
  John: One of the way, only ways that, and one of the hardest challenges now is not the power of the computer or the ability to come up with an algorithm.
• 11:11
  John: Again, a program that answers a question, but really cleaning up the data, making sure you've got a reasonable test data set to, to, to see whether what you're getting is.
• 11:23
  John: Or whether it's misleading.
• 11:24
  John: Uh, and then clean turns out cleaning up that data to, to, so that, so that you really have a, a great data set to work with is kind of hard unless everything is standardized and simple.
• 11:35
  John: Like, like, like, like, like for example, building codes or imagery that a, that a, that a dermatologist or a radiology.
• 11:42
  John: Can take a look at and say, yes, that's a tumor, or that's, that's a mold that doesn't look right versus not.
• 11:47
  John: That turns out to be one of the big tasks right now to really leverage the, the, the acceleration of computational power to make all of this machine learning, whether it's artificial or not.
• 11:58
  John: Power work.
• 11:59
  David: John, I don't love New York, but I do love your Brooklyn example.
• 12:03
  David: Going beyond language.
• 12:05
  David: Talk about culture and you talk about, you know, all these different languages, but sometimes what'll, what'll happen in terms of personalization, you might go into a diabetes clinic and they might have something in English and something in Spanish, right?
• 12:16
  David: But it doesn't account for, uh, how someone personally engages.
• 12:20
  David: So if you talk about.
• 12:21
  David: You know, food, talking about exercise, we talk about just housing.
• 12:26
  David: It's gonna be different for different people.
• 12:27
  David: And if you can, and in no way that any given clinician is gonna be able to understand all of that.
• 12:32
  David: But if you've got the right sort of data sets and that not trained in a biased way, you may actually be able to get the personalization beyond what any individual could do, and that could make a real difference to somebody's life.
• 12:44
  John: So David, I'm a techno.
• 12:46
  John: Unlike you, I am tech forward.
• 12:48
  John: I unlike you, I actually B buy into a lot of these new technologies.
• 12:53
  John: But I think starting with 2001 as Space Odyssey and a lot of the things that show up on movies, what can go wrong with the machines and leaning on machines and programming,
• 13:03
  David: well, nothing gone.
• 13:04
  David: Why don't what, what could go wrong?
• 13:06
  David: But I think there's just a few things.
• 13:07
  David: So, I mean, I, a lot of people do ask, you know, is AI in healthcare dangerous?
• 13:10
  David: That is a good question.
• 13:11
  David: So one of the issues is just.
• 13:14
  David: is bias.
• 13:15
  David: So we talked about how sure it could be culturally sensitive and it, and it could potentially personalize something for an individual.
• 13:22
  David: But what if it, what if it has the wrong information and it's very personalized and very specific, but you know, instead of, uh, Afghanistan that somebody put in Algeria, um, You have something that's just, you could end up reinforce.
• 13:35
  David: Let's,
• 13:35
  John: let's stick on the bias point because there's have been some great examples of unsupervised models where they program the inputs, throw a lot of data in, and then see what results they found.
• 13:45
  John: Some real racist bias.
• 13:46
  John: Yeah.
• 13:47
  John: That was implicit in the data.
• 13:49
  John: That then became one of the dominant features of the way the analysis is.
• 13:52
  John: And, and as, as you and I have pointed out a number of times, we've already got too much of a lack of, of a, of a deficient.
• 13:58
  John: And barriers to access to people of color, uh, to women in some cases.
• 14:03
  John: And the last thing we need is, is, is more bias in healthcare.
• 14:06
  John: So I think that's a significant, that's a big point issue that hasn't, hasn't really been solved for.
• 14:10
  David: No, but at least it's being recognized before it's baked into all the models.
• 14:13
  David: I'll give you another example, John.
• 14:15
  David: Um, which is that you could also have, we're talking about automation and doing all these tasks, but you can have the loss of human knowledge and the meaningfulness of work.
• 14:23
  David: Think about it, even in a, uh, like a self-driving car or autopilot, uh, even if you don't have like the full, full autopilot, you stop or just think about using a G P S.
• 14:34
  David: You lose
• 14:35
  John: some skills.
• 14:36
  John: You're str, you're struggling.
• 14:37
  John: Here.
• 14:37
  David: I am struggling, John, I'm always struggling.
• 14:39
  David: Jonathan, like this, a struggle.
• 14:40
  David: Struggling.
• 14:40
  David: You're struggling.
• 14:41
  David: But what happens is if you wait, wait, wait, wait.
• 14:42
  David: If you
• 14:43
  John: use the, you're telling me this, this is a worry.
• 14:45
  John: Yeah.
• 14:45
  John: That somehow we, we actually have doctors practice at the top of their license because they've got machines that are not biased, that are automating simply that somehow that's a problem.
• 14:56
  John: Like, no, I don't wanna, I don't wanna drive my car if my, I, if I'm, if my car, I don.
• 15:00
  John: Drive into a wall like some of the Teslas do, but . But I, I, I'd love to be driven around.
• 15:05
  John: In fact, I think you wanna be driven around.
• 15:07
  John: Like what do you, what do you what?
• 15:08
  David: Well, I don't, I dunno where we're going here.
• 15:10
  David: I did read about the Tesla that went off the cliff in California and surprisingly everybody was fine.
• 15:14
  David: So I'm not gonna, don't try to edit home when, but John, you missed those
• 15:17
  John: horse, those horse riding skills that used to be essential to moving transport, uh, in the early 20th century.
• 15:24
  John: . Um, do you, do you miss having to crack off big chunks of ice and throw them in an ice box?
• 15:29
  John: I mean, like, like, move, move, move,
• 15:31
  David: move on, dude.
• 15:32
  David: So, John, you're, I, I don't think you're right about practicing at the top of the license.
• 15:35
  David: So this, what, what happens is, let, let's say if you fly the plane on autopilot all the time, when you need to react to real emergency, you can't do it cuz you don't remember how in London where the drivers would have the knowledge, right?
• 15:45
  David: From knowing all the streets and all that, you know, their brain would actually grow.
• 15:48
  David: And if you have the gps, you just go around, you don't, you don't learn.
• 15:51
  David: It is a real issue, John, and they have to think about how to use AI in order to augment an individual and help them work at the top of their license as opposed to taking away, uh, you know, some of the things that the computer can do
• 16:02
  John: particularly well.
• 16:03
  John: I was, I was, I was giving you a hard time cuz I enjoyed doing that.
• 16:05
  John: But if you recall, one of the big challenges with Boeing and the crashes that they experienced is they were so relying on automat.
• 16:14
  John: That in many cases, challenged or overrode human interaction, that there were two big crashes because they took this, they took the controls, uh, until the absolute end out of the pilot's.
• 16:27
  John: Control and, and, and fought the pilot in, in, in two of the cases that, that, that I, that we read about the New York Times, and it's an example.
• 16:34
  John: I'm not sure they ever took complete responsibility or it's clear, but that you could easily see where you, you, where you, you, you, you, you, you actually overwhelm.
• 16:45
  John: The individual with a machine driven result, cuz you haven't thought about the corner cases and the challenges, which is I think what Tesla has run into.
• 16:53
  John: Um, so I, I, I think there are def and obviously the last place you want to be messing around with, um, over overriding, um, um, a doctor's judgment is in a, is in the case of a clinical crisis or a critical clinical clinical analysis.
• 17:07
  John: But I think that caution, does that mean.
• 17:09
  John: We should stop pushing for this, David.
• 17:11
  John: I mean, where do you land now on that?
• 17:12
  John: The, the,
• 17:13
  David: the artificial intelligence needs to be a slave, uh, to the real intelligence.
• 17:18
  David: And we need to make sure that we're not just doing what the computer can do or what the programmer came up with, but what we really need to help to augment, uh, both the productivity, but also the quality of life and the quality of work.
• 17:31
  David: We should spend a little more time, John.
• 17:33
  David: Whole question about reducing, reducing the burden on staff.
• 17:36
  David: I mean, I think about why would AI be a big deal in healthcare in 2023?
• 17:40
  David: It's because of these continued, uh, staff shortages and just the stress that's on the remaining workforce that's there and the ability to be able to take off the, you know, these very repetitive, uh, kind of things that, that people have to do to spend, uh, more time to be able to, uh, work at the top of the license.
• 17:57
  John: That's what we need to see.
• 17:58
  John: And, and David, we're seeing it at Walgreens.
• 18:00
  John: You know, we are reducing a lot of.
• 18:02
  John: Busy tasks and looking at metrics and automating through some basic robotic pro programming and sort of, uh, automating some of the, just the, the, the, the really dumb tasks so that pharmacists can practice at the top of their license.
• 18:15
  John: I think it's gonna be a much better.
• 18:16
  John: Experience for people like my 89 year old mom who's got some chronic conditions.
• 18:21
  John: So I, I thought she was
• 18:22
  David: 88.
• 18:24
  David: Happy birthday.
• 18:24
  David: She.
• 18:26
  David: John: Um, she, uh, yeah.
• 18:28
  David: No, but you're, you're distracting me from my, my point here.
• 18:33
  David: Um, which is that she, uh, she, she does continue to get older.
• 18:36
  David: Um, and that, that, She's just the kind of person who needs more care and attention from a pharmacist as opposed to having a pharmacist or a phar tech literally.
• 18:46
  David: Count pills.
• 18:47
  David: Right.
• 18:48
  David: And, and put them into little, from big boxes.
• 18:50
  David: Big, big, big, big, big bowls to, into, into, into small pill boxes.
• 18:55
  David: But I, I think what's what's most important, David, and I think we made modest progress even with your skepticism.
• 19:01
  David: Yeah.
• 19:01
  David: That people in healthcare really need to understand and simplify what's going on here because it's only when the, the brains and skills and care and heart of what people have, what, what brings people to healthcare is applied to.
• 19:13
  David: All of this programming and technology that we're gonna have the best outcomes.
• 19:17
  David: So I, you know, again, even though you're a skeptic, I, I think there's, the technology will continue to progress and it's super important that people really get their head around what's going on.
• 19:28
  David: I think the pharmacist is actually the best example, right?
• 19:30
  David: Where they've got people that are very highly trained, PhD, PhD level of training, a need to use that skill to interpret complex, you know, drug interactions, explain things, uh, to the patients.
• 19:41
  David: And at the same time, they end up doing things that are very rudimentary, but very important.
• 19:45
  David: Right?
• 19:45
  David: So you're scoffing about Yeah.
• 19:46
  David: Uh, counting the pills.
• 19:47
  David: They've gotta be counted, right?
• 19:48
  David: It's gotta be the right pill.
• 19:49
  David: So we need the automat.
• 19:51
  David: in order to help, but it's gotta be super reliable.
• 19:53
  David: Mm-hmm.
• 19:53
  David: . So that is a, that is a great, uh, example, John.
• 19:56
  David: And I do think that there, the danger is one of the other dangers of AI is that when we have to do some basic things on, like process automation and answering so-called basic questions, maybe to the patient, it's not always a basic question.
• 20:07
  David: And maybe there's some more nuance there that we can't just allow the computer say, that's a simple question.
• 20:11
  David: I'm gonna, I'm gonna answer it.
• 20:13
  David: We have to watch out for that as.
• 20:15
  John: Awesome.
• 20:15
  John: Well, David, I think that probably gets us close to a wrap because, because you're, you're running out of things to be better.
• 20:21
  John: I'm running
• 20:21
  David: out, running outta good ideas here, John.
• 20:23
  David: I gotta get back to the house floor and see if they're gonna give me the two 18 or whatever the number gets to be by the time this comes out.
• 20:28
  David: That's it for yet another edition of Care Talk.
• 20:30
  David: We've been talking about artificial intelligence.
• 20:32
  David: I'm David Williams, president of Health Business Group,
• 20:35
  John: and I'm John Driscoll, the president of Walgreen South.
• 20:37
  John: If you liked what you heard or you didn't, please listen to us and subscribe on your favorite service.