WEBVTT 00:00:13.580 --> 00:00:16.460 Hello and welcome to Listen Up People, a podcast 00:00:16.460 --> 00:00:19.399 of the USC Suzanne Dworak-Peck School of Social 00:00:19.399 --> 00:00:22.399 Work. I'm Dr. Eric Rice, professor and associate 00:00:22.399 --> 00:00:25.719 dean for research. The impact of artificial intelligence 00:00:25.719 --> 00:00:29.320 has exploded in recent years. Intellectual leaders 00:00:29.320 --> 00:00:31.879 from the technology space like Bill Gates have 00:00:31.879 --> 00:00:35.420 said, I quote, AI will lead to breakthrough treatments 00:00:35.420 --> 00:00:38.420 for deadly diseases, innovative solutions for 00:00:38.420 --> 00:00:40.740 climate change, and high quality education for 00:00:40.740 --> 00:00:43.840 everyone. Not everyone agrees with him. Intellectuals 00:00:43.840 --> 00:00:47.100 like Uval Noah Harari has said, quote, for the 00:00:47.100 --> 00:00:49.799 first time, we have real competition on the planet. 00:00:49.920 --> 00:00:52.460 We have been the most intelligent species by 00:00:52.460 --> 00:00:55.679 far for tens of thousands of years. And now we 00:00:55.679 --> 00:00:57.700 are creating something that can compete with 00:00:57.700 --> 00:01:00.509 us in the very near future. So my guests today 00:01:00.509 --> 00:01:02.429 are at the center of this space and the center 00:01:02.429 --> 00:01:05.989 of this debate. They are working with AI to improve 00:01:05.989 --> 00:01:08.549 health and well -being, yet they are deeply concerned 00:01:08.549 --> 00:01:11.170 about the role that AI will have in our workplace 00:01:11.170 --> 00:01:13.430 and in our daily lives. So I'd like to welcome 00:01:13.430 --> 00:01:16.510 my guests today, Dr. John Oberg, a USC School 00:01:16.510 --> 00:01:19.629 of Social Work DSW alumni, who is founder and 00:01:19.629 --> 00:01:23.420 CEO of Precina Health, and Josiah Bryan. who 00:01:23.420 --> 00:01:26.620 is the Chief Technology Officer and Lead AI Researcher 00:01:26.620 --> 00:01:28.680 at Precina Health. It's great to have you both 00:01:28.680 --> 00:01:30.260 here. Yeah, thanks, Eric. Thanks for having us. 00:01:30.280 --> 00:01:32.739 We're really excited to be here. So John, I know 00:01:32.739 --> 00:01:35.579 you got a doctorate degree here. I was here to 00:01:35.579 --> 00:01:37.700 see it happen. So of all the places you could 00:01:37.700 --> 00:01:40.019 have gone, why did you decide to pursue a doctorate 00:01:40.019 --> 00:01:42.939 here at USC? My grandfather taught at USC. My 00:01:42.939 --> 00:01:45.959 mom went to USC as an undergrad. I was there 00:01:45.959 --> 00:01:49.670 as an undergrad. and then it was really the natural 00:01:49.670 --> 00:01:53.069 place to come home. My undergrad degree, I left 00:01:53.069 --> 00:01:54.670 after a couple of years and finished up at New 00:01:54.670 --> 00:01:57.069 Mexico, and I told people I'd be back to finish 00:01:57.069 --> 00:01:58.870 my degree. I didn't finish my undergrad degree 00:01:58.870 --> 00:02:00.489 at USC, but I did come back when it was time 00:02:00.489 --> 00:02:04.329 to get a doctorate. Nice. Why the DSW program? 00:02:04.519 --> 00:02:07.000 I think that social workers are really misunderstood 00:02:07.000 --> 00:02:09.000 in the world. I think people think about social 00:02:09.000 --> 00:02:10.460 work and whatever experience they've had with 00:02:10.460 --> 00:02:12.560 them, whether it's CPS or medical social work 00:02:12.560 --> 00:02:14.819 or whatever experience someone's had, they just 00:02:14.819 --> 00:02:16.699 kind of put social workers into that bucket. 00:02:16.719 --> 00:02:19.080 And that's not how I saw social work. I saw social 00:02:19.080 --> 00:02:21.699 work as the people that use systems thinking 00:02:21.699 --> 00:02:24.719 to draw the really big picture and connect it 00:02:24.719 --> 00:02:26.680 to the micro level. So in social work, we call 00:02:26.680 --> 00:02:28.860 that the macro level, the meso level or the micro 00:02:28.860 --> 00:02:33.599 level. But I think most disciplines don't so 00:02:33.599 --> 00:02:36.240 deliberate. connect the individual to the system 00:02:36.240 --> 00:02:38.180 with the community in the middle. And I think 00:02:38.180 --> 00:02:40.020 that's such an important framework for us as 00:02:40.020 --> 00:02:43.270 we deal with these really complicated problems 00:02:43.270 --> 00:02:46.310 that seem intractable, but they're not intractable 00:02:46.310 --> 00:02:48.770 problems when you can actually get your mental 00:02:48.770 --> 00:02:51.830 model around how the system plays with the community, 00:02:51.830 --> 00:02:53.789 plays with the individual. And that's a really 00:02:53.789 --> 00:02:56.370 important part of how we've actually impacted 00:02:56.370 --> 00:02:58.490 type two diabetes. Like our first research project 00:02:58.490 --> 00:03:01.030 during our doctorate, you know, 98 % success 00:03:01.030 --> 00:03:03.090 in reversing type two diabetes. I'll share with 00:03:03.090 --> 00:03:05.009 you later in the podcast, some of our more recent 00:03:05.009 --> 00:03:08.229 research, but it's because we're able to look 00:03:08.229 --> 00:03:10.590 at that systems thinking and that social workers 00:03:10.590 --> 00:03:14.430 have this amazing view of the broad world and 00:03:14.430 --> 00:03:16.150 how it plays with the community and how it plays 00:03:16.150 --> 00:03:18.069 with the individual. I really felt it was the 00:03:18.069 --> 00:03:20.409 right place to go and then take the research 00:03:20.409 --> 00:03:22.210 and apply it in a way that was really meaningful 00:03:22.210 --> 00:03:23.949 in the world. And, you know, I think all of our 00:03:23.949 --> 00:03:26.169 clinical results are speaking to that. Yeah. 00:03:26.270 --> 00:03:28.969 I mean, it goes without saying I'm a huge social 00:03:28.969 --> 00:03:32.169 work fan as well. I wouldn't be here doing this 00:03:32.169 --> 00:03:35.469 if I wasn't, but it's great to hear the value 00:03:35.469 --> 00:03:38.270 you place on this. Josiah, I want to ask you, 00:03:38.569 --> 00:03:41.520 I know you're a tech guy. How did you first get 00:03:41.520 --> 00:03:44.560 involved in AI research? I've always been fascinated 00:03:44.560 --> 00:03:47.560 with the way the computers have the ability to 00:03:47.560 --> 00:03:50.099 make our lives better and make our lives easier. 00:03:50.479 --> 00:03:52.919 I've inherently very lazy. I started working 00:03:52.919 --> 00:03:55.460 with my hands, taking apart radios, my dad's 00:03:55.460 --> 00:03:58.060 old reel to reel player. Then when I discovered 00:03:58.060 --> 00:04:01.639 that I could be creative with computers, I found 00:04:01.639 --> 00:04:03.860 that I didn't want to do the hardware thinking 00:04:03.860 --> 00:04:06.340 anymore. I wanted to work with software. And 00:04:06.340 --> 00:04:08.460 very early on when I was about 12 years old, 00:04:08.500 --> 00:04:10.340 my dad started taking me to the library and I 00:04:10.340 --> 00:04:12.240 started reading graciously all the books I could. 00:04:12.689 --> 00:04:16.209 One of them stood out to me was this book, I 00:04:16.209 --> 00:04:17.269 haven't been able to find anymore, it's called 00:04:17.269 --> 00:04:19.250 Sierra Circuit Cellar. In it, it had an article, 00:04:19.370 --> 00:04:21.490 teach computers to think for themselves. Really, 00:04:21.670 --> 00:04:24.290 it was just this, pick from a menu what you want 00:04:24.290 --> 00:04:26.110 the computer to do and we'll paste together, 00:04:26.250 --> 00:04:27.889 think it's a basic code and we'll do its thing. 00:04:28.430 --> 00:04:29.970 But ever since then, that sparked something in 00:04:29.970 --> 00:04:31.850 me at 12 years old thinking, wow, I can get my 00:04:31.850 --> 00:04:33.870 computer to help make my life better with a little 00:04:33.870 --> 00:04:35.990 bit of intelligence and think intelligently. 00:04:36.220 --> 00:04:38.600 And that's put me on this lifelong trajectory 00:04:38.600 --> 00:04:41.680 of learning about AI, experimenting with it, 00:04:41.779 --> 00:04:44.240 trying to find ways to integrate it into my life, 00:04:44.600 --> 00:04:48.019 and to make human lives better, connect us more 00:04:48.019 --> 00:04:50.779 deeply as humans while still enjoying learning 00:04:50.779 --> 00:04:53.879 about what makes us truly human so I can make 00:04:53.879 --> 00:04:57.500 AI better. I am very, very sympathetic to this 00:04:57.500 --> 00:05:00.560 perspective. I think... Some of the listeners 00:05:00.560 --> 00:05:02.779 that we have may know this, but some of them 00:05:02.779 --> 00:05:05.500 who are turning in to listen to the two of you 00:05:05.500 --> 00:05:07.379 may not know this. I started a center called 00:05:07.379 --> 00:05:10.319 the USC Center for AI and Society back in 2016. 00:05:10.800 --> 00:05:13.600 And like John, I'm really convinced that really 00:05:13.600 --> 00:05:17.259 good social work -driven systemic thinking about 00:05:17.259 --> 00:05:20.779 complex problems can help us to come up with 00:05:20.779 --> 00:05:24.310 real solutions. But like you, Josiah, I really 00:05:24.310 --> 00:05:28.449 think that AI is this incredible opportunity 00:05:28.449 --> 00:05:32.069 for building tools that can be as maybe complex 00:05:32.069 --> 00:05:35.250 as the problems that we're dealing with and have 00:05:35.250 --> 00:05:37.329 some real opportunities to move the needle on 00:05:37.329 --> 00:05:39.470 some stuff that we've had trouble moving the 00:05:39.470 --> 00:05:42.470 needle on. And so I guess with that in mind, 00:05:42.629 --> 00:05:46.149 I'll let the spotlight go back to the two of 00:05:46.149 --> 00:05:48.290 you, which is to say, John, you know, maybe you 00:05:48.290 --> 00:05:50.230 could just give us an overview of Christina Health 00:05:50.230 --> 00:05:52.769 and what are the goals? of the company and what 00:05:52.769 --> 00:05:54.850 kind of work are you doing there? Yeah, we're 00:05:54.850 --> 00:05:56.810 a medical practice. So it's, you know, you bring 00:05:56.810 --> 00:05:59.470 on two guys, you know, my master's degree is 00:05:59.470 --> 00:06:02.410 in policy. And then I have a second specialization 00:06:02.410 --> 00:06:04.389 in the management of technology where we researched, 00:06:04.389 --> 00:06:06.629 you know, the history of technology and what's 00:06:06.629 --> 00:06:08.529 happened in terms of pace and speed. That was 00:06:08.529 --> 00:06:10.949 one of my capstone projects there. But we opened 00:06:10.949 --> 00:06:12.790 up a medical practice because you've got this 00:06:12.790 --> 00:06:15.370 really nasty problem with type two diabetes that 00:06:15.370 --> 00:06:16.910 started with my mother -in -law getting sick, 00:06:17.350 --> 00:06:19.730 ended up in the hospital. And I got really frustrated 00:06:19.730 --> 00:06:22.990 that she was in the hospital in my mind unnecessarily 00:06:22.990 --> 00:06:25.129 if she had just taken the medication the right 00:06:25.129 --> 00:06:27.889 way. I've met her doctor. I have the highest 00:06:27.889 --> 00:06:30.290 respect for her doctor. She's an amazing primary 00:06:30.290 --> 00:06:33.350 care physician. And so I called my friends at 00:06:33.350 --> 00:06:34.889 USC basically and said, I'm coming back to get 00:06:34.889 --> 00:06:37.029 my doctorate. I want to solve type two diabetes. 00:06:37.490 --> 00:06:39.949 And I think the nicest way to say it is that 00:06:39.949 --> 00:06:41.589 the people at USC kind of patted me on the head 00:06:41.589 --> 00:06:43.740 and said, hey, We've been dealing with diabetes 00:06:43.740 --> 00:06:46.600 for the last 60 years as a world, and we haven't 00:06:46.600 --> 00:06:48.860 solved it. That's cute. Come on back. We'll take 00:06:48.860 --> 00:06:51.060 your tuition money. We'll educate you. But I 00:06:51.060 --> 00:06:52.139 don't know if you're going to solve the problem. 00:06:52.329 --> 00:06:54.790 And so in our first pilot study, we took 50 patients 00:06:54.790 --> 00:06:58.149 who were super sick, average A1C of 9 .6 if you're 00:06:58.149 --> 00:07:01.149 a clinician, and we reduced that average A1C 00:07:01.149 --> 00:07:04.589 to 6 .4 in 12 weeks. And so most doctors would 00:07:04.589 --> 00:07:06.689 tell you if you can reduce A1C by one percentage 00:07:06.689 --> 00:07:08.589 point in a year, you have a billion dollar drug. 00:07:08.970 --> 00:07:11.029 And using just a different set of behavioral 00:07:11.029 --> 00:07:13.889 interventions, medical interventions, using devices, 00:07:14.170 --> 00:07:16.449 pharmacology, but also most importantly, behavior 00:07:16.449 --> 00:07:19.889 change, we reduced people's A1C by 3 .2 % in 00:07:19.889 --> 00:07:22.120 12 weeks, which is super impressive. But then 00:07:22.120 --> 00:07:24.459 we kept people in clinical control for the next 00:07:24.459 --> 00:07:27.980 two years in 83 % of our cases. And I think people 00:07:27.980 --> 00:07:30.160 were just, their jaws hit the ground. How did 00:07:30.160 --> 00:07:32.680 you do that with this group that was low income? 00:07:33.060 --> 00:07:35.759 They had low access to care. And most providers 00:07:35.759 --> 00:07:37.500 who had worked with them said they've had uncontrolled 00:07:37.500 --> 00:07:39.899 diabetes for years. And some doctors were calling 00:07:39.899 --> 00:07:43.720 these patients lazy or uneducated. It was really 00:07:43.720 --> 00:07:46.120 judgmental and it was horrifying for us to see. 00:07:46.240 --> 00:07:48.879 And so we said no. it's not the right, it wasn't 00:07:48.879 --> 00:07:50.639 the right intervention. So we stripped down the 00:07:50.639 --> 00:07:53.360 medical practice to kind of its most basic units 00:07:53.360 --> 00:07:55.959 of clinical intervention, of administration. 00:07:56.480 --> 00:07:58.620 We really took apart the, like including the 00:07:58.620 --> 00:08:00.540 paperwork you sign when you walk in, like everything 00:08:00.540 --> 00:08:03.199 got ripped apart. And we created what we call 00:08:03.199 --> 00:08:05.889 the systemic. therapeutic alignment and rapid 00:08:05.889 --> 00:08:08.449 titration model that looks at all of the medicine, 00:08:08.790 --> 00:08:11.149 all of the mental health issues, all of the behavioral 00:08:11.149 --> 00:08:13.689 non -clinical issues, all the social determinants 00:08:13.689 --> 00:08:15.949 of health, all of the health beliefs. But then 00:08:15.949 --> 00:08:18.810 we layer across a patient's readiness for change. 00:08:19.290 --> 00:08:21.970 And then we make these really, really, really 00:08:21.970 --> 00:08:24.569 small steps to help patients get better a little 00:08:24.569 --> 00:08:27.250 bit at a time. And we use AI to do things like 00:08:27.250 --> 00:08:30.670 taking in thousands of patients, you know, pages 00:08:30.670 --> 00:08:33.269 of patient charting, and then highlighting for 00:08:33.269 --> 00:08:35.830 the physician, like which ones are most important 00:08:35.830 --> 00:08:38.429 so that our physicians can do crazy things like 00:08:38.429 --> 00:08:40.750 reading the chart before they meet with a patient. 00:08:41.169 --> 00:08:42.809 And I think a lot of, you know, providers want 00:08:42.809 --> 00:08:44.750 to do that. But when you have a thousand pages 00:08:44.750 --> 00:08:47.049 of history, and I'm being literal, I'm not figurative 00:08:47.049 --> 00:08:50.629 in some cases, right? So we can use AI to do 00:08:50.629 --> 00:08:52.690 that. We can use AI for things that I would call 00:08:52.690 --> 00:08:54.830 fairly vanilla, like dictation and all those 00:08:54.830 --> 00:08:56.690 types of things. Sure. More importantly, we can 00:08:56.690 --> 00:08:59.529 take 1 ,000 pages of history for 20 years, and 00:08:59.529 --> 00:09:01.769 we can find patterns where it says to the provider, 00:09:01.990 --> 00:09:04.389 hey, have you considered looking at this issue 00:09:04.389 --> 00:09:07.269 that shows up like arthroscorosis, but actually... 00:09:07.070 --> 00:09:09.830 over in this system hasn't been identified, does 00:09:09.830 --> 00:09:13.029 that require a lab draw or not? Or, hey, in this 00:09:13.029 --> 00:09:15.070 case, they've been pulled off of this particular 00:09:15.070 --> 00:09:18.070 drug, and SGLT2 is one that was recent, and it's 00:09:18.070 --> 00:09:19.750 actually no longer clinically indicated they 00:09:19.750 --> 00:09:21.590 get pulled off. They should actually be put on 00:09:21.590 --> 00:09:23.690 and actually have as much as they can of that 00:09:23.690 --> 00:09:26.320 drug, and that... guidance is only four weeks 00:09:26.320 --> 00:09:29.200 old. And then what we can do is say, hey, this 00:09:29.200 --> 00:09:31.940 doctor wants to see the data in this way externally. 00:09:32.379 --> 00:09:34.919 This other external wants to see the data a different 00:09:34.919 --> 00:09:37.460 way. Can we use AI to format exactly the way 00:09:37.460 --> 00:09:39.820 they want, review it and send it out in a way 00:09:39.820 --> 00:09:42.549 that's not administratively burdensome? And so 00:09:42.549 --> 00:09:44.990 we use AI and Josiah, correct me if I'm wrong, 00:09:45.169 --> 00:09:47.830 12 or 13 different discrete places in our medical 00:09:47.830 --> 00:09:50.049 practice without being a tech company. My next 00:09:50.049 --> 00:09:52.529 question to Josiah is like, John's described 00:09:52.529 --> 00:09:56.250 a bunch of applications of AI. What are you doing 00:09:56.250 --> 00:09:58.950 with AI? to make all this stuff, this amazing 00:09:58.950 --> 00:10:01.929 stuff happen. We've put our trust in a platform 00:10:01.929 --> 00:10:04.049 called Salesforce. It's a small little company. 00:10:04.190 --> 00:10:05.830 I'm sure you might've heard of it, but they've 00:10:05.830 --> 00:10:07.889 done a great job. They've done a great job making 00:10:07.889 --> 00:10:11.090 it so that I can move faster by putting my trust 00:10:11.090 --> 00:10:13.610 in the, keeping my data there and then building 00:10:13.610 --> 00:10:16.009 on top of their platform. to that end there, 00:10:16.230 --> 00:10:17.889 there's some things that they're incredibly good 00:10:17.889 --> 00:10:19.789 at and some things that they just aren't going 00:10:19.789 --> 00:10:21.610 to focus on because that doesn't make sense for 00:10:21.610 --> 00:10:24.470 their market. And so we build into their platform 00:10:24.470 --> 00:10:26.889 and provide our providers with things like a 00:10:26.889 --> 00:10:29.200 provider coach. to help them reset emotionally 00:10:29.200 --> 00:10:31.899 between calls and to coach them up in Prasena 00:10:31.899 --> 00:10:34.960 way of thinking. And it all is oriented around 00:10:34.960 --> 00:10:38.460 adding in emotional context, adding in wisdom 00:10:38.460 --> 00:10:42.059 in some form or fashion into the AI systems that 00:10:42.059 --> 00:10:44.600 we deploy into our practice. That's the provider 00:10:44.600 --> 00:10:47.779 facing one. We're currently in beta testing with 00:10:47.779 --> 00:10:51.139 a patient facing AI, which will enable them to 00:10:51.139 --> 00:10:53.659 advocate for themselves and to know what questions 00:10:53.659 --> 00:10:56.299 to ask their doctors for that Prasena, as well 00:10:56.299 --> 00:10:58.259 as their other health. care providers. So those 00:10:58.259 --> 00:11:01.059 are two of the AIs out of a large plethora of 00:11:01.059 --> 00:11:03.240 AIs that we have deployed. Some are more boring 00:11:03.240 --> 00:11:06.519 behind the scenes, such as the document filtration 00:11:06.519 --> 00:11:09.639 AI, which is a workflow that goes through, takes 00:11:09.639 --> 00:11:12.080 in and process the documents for our providers. 00:11:12.360 --> 00:11:14.559 It lifts the load off of the human providers. 00:11:14.740 --> 00:11:17.019 There's a number of agents like that behind the 00:11:17.019 --> 00:11:19.820 scenes that help us accelerate and deliver more 00:11:19.820 --> 00:11:23.379 human care at a lower cost and save more lives 00:11:23.379 --> 00:11:25.879 and save and help more people. I want to make 00:11:25.879 --> 00:11:28.279 it real for the listeners. So two really, really 00:11:28.279 --> 00:11:32.360 use cases. We've essentially taken doctors and 00:11:32.360 --> 00:11:34.120 nurse practitioners and physicians associates 00:11:34.120 --> 00:11:36.039 and said, hey, look, you've got to learn an entirely 00:11:36.039 --> 00:11:37.639 different way of doing medicine if you're going 00:11:37.639 --> 00:11:40.259 to work with us, similar to a fellowship in terms 00:11:40.259 --> 00:11:44.100 of the level of training that we give. And the 00:11:44.100 --> 00:11:46.080 medical doctor, my partner, Dr. Dustin Williams 00:11:46.080 --> 00:11:48.100 is a long -term medical educator, has written 00:11:48.100 --> 00:11:50.899 medical curriculum for 80 different medical schools 00:11:50.899 --> 00:11:53.120 across the world. So we know that we're asking 00:11:53.120 --> 00:11:54.940 a lot of a provider to come in and work with 00:11:54.940 --> 00:11:56.679 us, very similar to what he asked fellows to 00:11:56.679 --> 00:11:58.960 do when he was training fellows. And so the AI 00:11:58.960 --> 00:12:01.759 will actively help support us to deliver that 00:12:01.759 --> 00:12:04.879 training in a lower cost way by having our providers 00:12:04.879 --> 00:12:07.740 actually practice what we want them to say with 00:12:07.740 --> 00:12:10.820 the patient, because it's so nuanced and the 00:12:10.820 --> 00:12:13.039 AI never gets tired. So if somebody wants to 00:12:13.039 --> 00:12:15.820 practice 20 times before work, they can do that 00:12:15.820 --> 00:12:18.299 any hour of the day, the AI gives that feedback 00:12:18.299 --> 00:12:20.740 to the provider. And if the provider wants to 00:12:20.740 --> 00:12:22.440 share that with their manager, the manager can 00:12:22.440 --> 00:12:25.059 come in and be even more new. So that's a really 00:12:25.059 --> 00:12:27.600 helpful tool. On the flip side of that coin, 00:12:27.919 --> 00:12:30.259 if a patient says to their provider, I know I 00:12:30.259 --> 00:12:32.440 need to eat better, but I need a meal plan. It 00:12:32.440 --> 00:12:34.779 used to take hours to create a meal plan. In 00:12:34.779 --> 00:12:36.639 five or 10 minutes, our providers can talk to 00:12:36.639 --> 00:12:39.500 a patient. in the video chat, create a meal plan, 00:12:40.320 --> 00:12:42.840 and feed into the AI what food is in the patient's 00:12:42.840 --> 00:12:44.940 house today, ask what the shopping list should 00:12:44.940 --> 00:12:47.480 be based on the size of the family and how many 00:12:47.480 --> 00:12:49.559 times that person wants to eat the same leftovers, 00:12:50.019 --> 00:12:51.879 based on their own preferences for food, their 00:12:51.879 --> 00:12:54.759 own allergies, the actual store that's nearby, 00:12:55.299 --> 00:12:57.320 and Josiah has told me that very shortly we'll 00:12:57.320 --> 00:12:58.960 be able to wire it up so we can actually send 00:12:58.960 --> 00:13:01.639 that shopping list to that store for the patient 00:13:01.639 --> 00:13:03.340 and have them just pick it up at the curb if 00:13:03.340 --> 00:13:05.490 they want to. And that's all taking like five 00:13:05.490 --> 00:13:07.710 minutes now with a well -trained provider versus 00:13:07.710 --> 00:13:09.889 the hours of time it would take, or it just not 00:13:09.889 --> 00:13:12.049 getting done at all and the patient not following 00:13:12.049 --> 00:13:14.029 the plan because it's just too hard to follow 00:13:14.029 --> 00:13:17.309 the plan. Very, very small movements sometimes 00:13:17.309 --> 00:13:19.049 require a lot of planning. Well, yeah, I mean, 00:13:19.049 --> 00:13:22.809 making a plan seems like not a hard thing to 00:13:22.809 --> 00:13:26.169 do when you are the person who is instructing 00:13:26.169 --> 00:13:28.970 somebody else on the plan. Enacting a plan when 00:13:28.970 --> 00:13:30.750 it creates, you know, all of them, there's all 00:13:30.750 --> 00:13:32.929 these logistic barriers in real life is really 00:13:32.929 --> 00:13:35.049 hard. I mean, that makes perfect sense to me. 00:13:35.289 --> 00:13:36.889 And I gotta say, Josiah, you'd said something 00:13:36.889 --> 00:13:40.070 that I just wanna throw my two cents on, which 00:13:40.070 --> 00:13:41.950 is, you'd said, oh, well, there's some of this 00:13:41.950 --> 00:13:45.029 stuff that's not that exciting, that's the background 00:13:45.029 --> 00:13:48.370 thing that is digesting all of these thousands 00:13:48.370 --> 00:13:51.350 of pages of information. I actually think that 00:13:51.350 --> 00:13:54.690 that stuff is incredibly exciting. I mean, it 00:13:54.690 --> 00:13:57.269 doesn't sound sexy, right? I think sometimes, 00:13:57.269 --> 00:14:00.590 you know, AIs that talk back to you or that allow 00:14:00.590 --> 00:14:04.169 you to talk to them sound in some ways more like 00:14:04.169 --> 00:14:06.169 the science fiction robots that we envisioned 00:14:06.169 --> 00:14:09.370 when we were kids. But the fact that a real human 00:14:09.370 --> 00:14:13.669 being doesn't have the time. to read 1 ,000 pages 00:14:13.669 --> 00:14:17.009 of patient notes. But an AI might be able to 00:14:17.009 --> 00:14:20.009 digest some of those patient notes for that provider 00:14:20.009 --> 00:14:23.610 and put in front of that provider some possible 00:14:23.610 --> 00:14:27.070 things to think about that then doesn't replace 00:14:27.070 --> 00:14:29.230 the provider as the provider, but just makes 00:14:29.230 --> 00:14:31.649 the provider better. at what the provider does. 00:14:32.529 --> 00:14:35.549 That's so exciting. I mean, it's hard to, I mean, 00:14:35.549 --> 00:14:37.490 you know, it takes 30 seconds to explain it. 00:14:37.509 --> 00:14:39.450 So it's not, it doesn't, it's not pithy in the 00:14:39.450 --> 00:14:42.350 same way, but that's a game changer, right? It 00:14:42.350 --> 00:14:45.370 is. It helps us be connect better as humans. 00:14:45.730 --> 00:14:47.470 I think with some of the AI we've done with Dustin, 00:14:47.750 --> 00:14:50.470 he said it takes his work, his prep time down 00:14:50.470 --> 00:14:53.590 from 45 minutes to 15 minutes, but it enables 00:14:53.590 --> 00:14:55.909 him that 45 minutes he gets back, he can then 00:14:55.909 --> 00:14:58.799 invest that with the patient. Dustin's been doing 00:14:58.799 --> 00:15:02.360 this for years. So that's 45 to 15 for a world 00:15:02.360 --> 00:15:04.220 -class expert at the person that designed the 00:15:04.220 --> 00:15:07.139 protocols. For other people, it's 90 minutes 00:15:07.139 --> 00:15:12.720 to 15 minutes. It would be hard to overstate 00:15:12.720 --> 00:15:15.379 the impact, but it's also hard to overstate the 00:15:15.379 --> 00:15:18.340 importance of the human in the loop because, 00:15:19.139 --> 00:15:21.639 again, let me use a real example. When Josiah 00:15:21.639 --> 00:15:25.019 uses AI to code, he's a domain expert and knows 00:15:25.019 --> 00:15:28.120 exactly how to use AI to help us move faster 00:15:28.120 --> 00:15:31.720 in burning down our backlog. Amazing. If I were 00:15:31.720 --> 00:15:34.659 to try to use AI to code, not a domain expert, 00:15:34.919 --> 00:15:38.580 I would create additional work for quality assurance 00:15:38.580 --> 00:15:41.299 from Josiah that would likely be faster for him 00:15:41.299 --> 00:15:43.340 just to do the work himself than trying to oversee 00:15:43.340 --> 00:15:47.809 the garbage I would create. Right, so technology 00:15:47.809 --> 00:15:50.190 solves big problems or it can create big problems 00:15:50.190 --> 00:15:52.629 too, right? So a car can get you from A to B 00:15:52.629 --> 00:15:53.850 really quickly, but if you don't know how to 00:15:53.850 --> 00:15:56.730 drive, it can cause an 18 car pileup too. 100%. 00:15:56.730 --> 00:15:59.129 And I think that really kind of gets to the heart 00:15:59.129 --> 00:16:01.250 of some of the debate that I was mentioning at 00:16:01.250 --> 00:16:04.529 the beginning of our podcast, which is, it seems 00:16:04.529 --> 00:16:06.850 really exciting and wonderful, all of these things 00:16:06.850 --> 00:16:09.700 that AI can do, but people... you know, really 00:16:09.700 --> 00:16:12.279 thoughtful people like Harari. I mean, he's not 00:16:12.279 --> 00:16:15.059 a he's not a clown. Like, he's freaked out that, 00:16:15.059 --> 00:16:18.309 you know, we've kind of created the our own demise. 00:16:18.509 --> 00:16:20.789 I mean, and so when you're thinking about these 00:16:20.789 --> 00:16:23.250 things, I mean, how are you as a company thinking 00:16:23.250 --> 00:16:26.090 about making the decisions about what AI should 00:16:26.090 --> 00:16:28.950 do, maybe what AI shouldn't do? Yeah, I mean, 00:16:28.950 --> 00:16:31.490 I'll say where we start with everybody. Sometime 00:16:31.490 --> 00:16:33.570 in the late 1800s, the guy that ran the patent 00:16:33.570 --> 00:16:35.250 office for the US said it was time to shut the 00:16:35.250 --> 00:16:36.769 whole thing down because everything that ever 00:16:36.769 --> 00:16:38.730 could have been invented already had been invented. 00:16:39.549 --> 00:16:41.789 That was in the late 1800s, that was said. Now 00:16:41.789 --> 00:16:44.289 just think about the number of, we didn't have 00:16:44.289 --> 00:16:46.860 air flight. commercially. In fact, you know, 00:16:46.940 --> 00:16:49.340 Kitty Hawk was like 1914 or something like that. 00:16:49.340 --> 00:16:52.179 Yeah, give or take. So I mean, and then we've 00:16:52.179 --> 00:16:54.500 gone from like the Wright brothers to the moon 00:16:54.500 --> 00:16:57.100 to now talking about Mars in the last hundred 00:16:57.100 --> 00:17:00.860 and 25 years. And so the reality is that humans 00:17:00.860 --> 00:17:03.370 don't love change. If you were to kind of look 00:17:03.370 --> 00:17:05.769 at like a behavioral chart, like go look at some 00:17:05.769 --> 00:17:07.710 of the peer reviewed disk models and how people, 00:17:07.910 --> 00:17:09.990 you know, there's, there's a small percentage, 00:17:09.990 --> 00:17:12.549 let's call it give or take less than 20 % heuristically 00:17:12.549 --> 00:17:15.049 of the population in the U S that loves the idea 00:17:15.049 --> 00:17:17.630 of change as a starting place. The other 80 % 00:17:17.630 --> 00:17:20.569 not so excited, right? Right. And so, you know, 00:17:20.650 --> 00:17:23.589 we're calculators bad for people in the, in the 00:17:23.589 --> 00:17:26.170 large floors of rooms in corporate America where 00:17:26.170 --> 00:17:28.509 they had 10 foot pieces of paper doing spreadsheets 00:17:28.509 --> 00:17:30.869 by hand with slide rules. Yes, that was very 00:17:30.730 --> 00:17:32.950 bad for that part of the industry. And people 00:17:32.950 --> 00:17:36.009 had to be retrained to do other things. And there 00:17:36.009 --> 00:17:37.890 were some individuals that were really negatively 00:17:37.890 --> 00:17:40.230 impacted by that. And am I worried about how 00:17:40.230 --> 00:17:43.170 my son is going to have to tool up for the job 00:17:43.170 --> 00:17:44.910 world relative to how I did? It's going to have 00:17:44.910 --> 00:17:47.130 to be totally different. Our people have to be 00:17:47.130 --> 00:17:49.150 OK with the fact that things are changing. And 00:17:49.150 --> 00:17:51.430 for us, the way medicine is being practiced is 00:17:51.430 --> 00:17:54.089 changing. And that's OK. And we're going to lean 00:17:54.089 --> 00:17:56.769 into that. But we're really, really thoughtful 00:17:56.769 --> 00:17:59.769 about clinical outcomes first and foremost, measurable 00:17:59.769 --> 00:18:02.910 clinical outcomes. And then we also want sustainability, 00:18:03.309 --> 00:18:05.809 not just financial sustainability, but like behavioral 00:18:05.809 --> 00:18:08.289 change, sustainability, financial sustainability 00:18:08.289 --> 00:18:11.490 for sure. But, but how do you do all these things 00:18:11.490 --> 00:18:14.230 so that the problem that's spinning out of control 00:18:14.230 --> 00:18:16.690 with healthcare in America in particular, like 00:18:16.690 --> 00:18:18.529 we can actually tell individuals and we say this 00:18:18.529 --> 00:18:20.829 to our patients, the system may be working against 00:18:20.829 --> 00:18:23.190 you. Everyone would say it's broken in some way, 00:18:23.390 --> 00:18:26.049 shape or form. And there's enough things you 00:18:26.049 --> 00:18:28.730 control. that if you want to focus on what you 00:18:28.730 --> 00:18:30.769 can control, just accepting that the environment 00:18:30.769 --> 00:18:33.329 sucks, you can do enough to get healthy today. 00:18:33.559 --> 00:18:36.579 John, this reminds me of the conversation that 00:18:36.579 --> 00:18:38.420 you and I had a few months ago. We were having 00:18:38.420 --> 00:18:41.200 dinner together and you were mentioning again, 00:18:41.279 --> 00:18:42.980 a conversation that you're having with your children, 00:18:42.980 --> 00:18:47.420 right? About AI. And I don't want to steal your 00:18:47.420 --> 00:18:49.380 thunder because this was really your idea, but 00:18:49.380 --> 00:18:52.420 maybe you can share a little bit of the insights 00:18:52.420 --> 00:18:54.299 that you shared with me about how you're talking 00:18:54.299 --> 00:18:56.720 to your kids about how AI is going to be impacting 00:18:56.720 --> 00:18:58.900 them in the work world. Yeah, you're talking 00:18:58.900 --> 00:19:00.539 about the conversation we had about knowledge 00:19:00.539 --> 00:19:05.099 and wisdom, right? Yeah, exactly. Yeah, so what 00:19:05.099 --> 00:19:07.480 computers do really well is they handle knowledge 00:19:07.480 --> 00:19:08.960 really, really well. And when I say knowledge, 00:19:08.960 --> 00:19:10.799 I'm talking about things that our brain handles 00:19:10.799 --> 00:19:14.700 kind of in the cerebral cortex and how we process 00:19:14.700 --> 00:19:17.420 information. And Josiah is going to have to kind 00:19:17.420 --> 00:19:20.000 of make sure that he's keeping me saying the 00:19:20.000 --> 00:19:22.200 words correctly here. But broadly speaking, the 00:19:22.200 --> 00:19:24.019 way that a computer processes information is 00:19:24.019 --> 00:19:26.460 really similar to our cerebral cortex. And that 00:19:26.460 --> 00:19:28.259 means that a lot of the work we do there today 00:19:28.259 --> 00:19:30.079 is going to be done by computers later on because 00:19:30.079 --> 00:19:33.440 they're going to be faster, better, on for 24 00:19:33.440 --> 00:19:34.920 hours, like they're just, and they're going to 00:19:34.920 --> 00:19:36.519 make less mistakes. Like that's just the way 00:19:36.519 --> 00:19:38.539 that computers are. However, there's two other 00:19:38.539 --> 00:19:40.460 parts of our brain, the amygdala and the limbic 00:19:40.460 --> 00:19:43.119 system that process information in a much different 00:19:43.119 --> 00:19:45.700 way. It's not just different processors. It's 00:19:45.700 --> 00:19:47.599 actually different biological architecture where 00:19:47.599 --> 00:19:49.740 I'm not the expert at that. But what I do know 00:19:49.740 --> 00:19:52.140 is that information in information out is different 00:19:52.140 --> 00:19:54.539 in those three parts of the brain. And the computer 00:19:54.539 --> 00:19:57.819 has the first one like nailed. And today, I've 00:19:57.819 --> 00:20:00.319 not heard a lot of people talking about making 00:20:00.319 --> 00:20:02.359 different architecture that solves for two and 00:20:02.359 --> 00:20:05.339 three. I've heard lots of emulation conversations. 00:20:05.740 --> 00:20:08.660 How do we emulate emotions? And that would be 00:20:08.660 --> 00:20:12.880 kind of similar to someone who has clinical autism 00:20:12.880 --> 00:20:15.480 level one, and they emulate emotions because 00:20:15.480 --> 00:20:17.240 they've observed it and they can think about 00:20:17.240 --> 00:20:19.619 it, but they're not really feeling in that same 00:20:19.619 --> 00:20:23.019 way that you or I might feel. And computers can 00:20:23.019 --> 00:20:25.539 do that, but we're trying to actually write ways 00:20:25.539 --> 00:20:27.730 to get deeper and better emulators and try to 00:20:27.730 --> 00:20:30.349 re -architect the software so that we can actually 00:20:30.349 --> 00:20:33.430 have computers be more thoughtful about the way 00:20:33.430 --> 00:20:35.190 they're interacting with the human. And then 00:20:35.190 --> 00:20:38.890 teaching my kids, getting human experience, those 00:20:38.890 --> 00:20:40.630 are the things that computers aren't going to 00:20:40.630 --> 00:20:42.690 be able to take the place of in the workforce 00:20:42.690 --> 00:20:47.269 of the future. So how do you find skills that 00:20:47.269 --> 00:20:50.400 are going to be valuable? 20 years now, 30 years 00:20:50.400 --> 00:20:52.319 now, and how do you build toward those? You can 00:20:52.319 --> 00:20:54.839 create value today and create value in the future. 00:20:54.900 --> 00:20:56.839 And then how do you learn how to learn? Because 00:20:56.839 --> 00:20:59.380 you're going to have to be pliable as we go through 00:20:59.380 --> 00:21:01.859 the next, the coming decades. When we were having 00:21:01.859 --> 00:21:04.180 this conversation, you used this phrase that 00:21:04.180 --> 00:21:06.240 I just loved. And so I'm going to, I'm going 00:21:06.240 --> 00:21:08.359 to make sure that you get credit for this phrase. 00:21:08.519 --> 00:21:12.299 You said it's all about wisdom work, right? So, 00:21:12.380 --> 00:21:14.059 and, and, and I thought that that was such a 00:21:14.059 --> 00:21:17.720 great way of explaining in a very shorthand way. 00:21:17.819 --> 00:21:21.240 What you're talking about about these human experiences 00:21:21.240 --> 00:21:25.500 that are not going to be about information and 00:21:25.500 --> 00:21:28.420 information processing, but it's about something 00:21:28.420 --> 00:21:30.599 fundamentally qualitatively different. I mean, 00:21:30.680 --> 00:21:32.500 can you maybe tell us a little bit more about 00:21:32.500 --> 00:21:35.289 like What does that even mean to you? I think 00:21:35.289 --> 00:21:37.690 it's a great question. In wisdom work, I mean, 00:21:37.809 --> 00:21:40.430 it's certainly not a technical or clinical term, 00:21:40.529 --> 00:21:42.269 right? This is a shorthand way for me to talk 00:21:42.269 --> 00:21:44.269 to my kids, right? You made up this phrase, but 00:21:44.269 --> 00:21:46.369 I was like, this is beautiful. I love it. Yeah. 00:21:46.650 --> 00:21:48.650 And what it really is, it's saying like all the 00:21:48.650 --> 00:21:50.569 stuff that computers aren't going to be able 00:21:50.569 --> 00:21:53.990 to handle well, right? So it's how do you do 00:21:53.990 --> 00:21:56.250 the things that aren't going to be done well 00:21:56.250 --> 00:21:58.109 by a computer and how do you lean into that? 00:21:58.170 --> 00:22:00.170 And so that's going to be about, you know, dealing 00:22:00.170 --> 00:22:02.309 with people. I think that's going to be a really 00:22:02.309 --> 00:22:04.329 important part of the future. That could be managing 00:22:04.329 --> 00:22:07.349 people. That could be clinical therapy or clinical 00:22:07.349 --> 00:22:10.309 intervention a lot of different ways. My understanding 00:22:10.309 --> 00:22:12.650 of the brain is that when you go to make decisions 00:22:12.650 --> 00:22:16.900 in a rational, healthy kind of kind of non -abnormal 00:22:16.900 --> 00:22:19.180 psychology kind of brain, that when you make 00:22:19.180 --> 00:22:22.259 a decision, it's generally made emotionally first, 00:22:22.259 --> 00:22:24.180 and then you justify that decision logically. 00:22:24.359 --> 00:22:26.759 And so if someone makes a decision emotionally 00:22:26.759 --> 00:22:29.119 and can't find the logic to back it up, we sometimes 00:22:29.119 --> 00:22:32.700 describe that as buyer's remorse, right? And 00:22:32.700 --> 00:22:35.180 then when somebody makes a decision logically, 00:22:35.640 --> 00:22:38.359 but doesn't have the emotional will to kind of 00:22:38.359 --> 00:22:41.279 commit to it, we have analytical paralysis, right? 00:22:41.559 --> 00:22:43.519 And if you have neither, you just don't make 00:22:43.519 --> 00:22:46.750 the decision. But if you have the decision made 00:22:46.750 --> 00:22:49.029 emotionally and then you justify it with the 00:22:49.029 --> 00:22:51.210 logic, that's where you get people that make 00:22:51.210 --> 00:22:53.569 long -term decisions that are good. And hopefully 00:22:53.569 --> 00:22:55.529 for my kids, they can learn how to do things 00:22:55.529 --> 00:22:58.029 that have kind of delayed gratification, which 00:22:58.029 --> 00:23:00.170 is an important skill set next to everything 00:23:00.170 --> 00:23:01.549 else we're talking about, right? And that's a 00:23:01.549 --> 00:23:04.569 whole other topic. But wisdom work is about understanding 00:23:04.569 --> 00:23:07.059 how the brain works. outside of the cerebral 00:23:07.059 --> 00:23:09.619 cortex and not just the black and white logic, 00:23:09.980 --> 00:23:12.079 but the feelings behind it and what feelings 00:23:12.079 --> 00:23:14.680 are driving that and getting really honest about 00:23:14.680 --> 00:23:17.259 those feelings. And I tell my kids like, you 00:23:17.259 --> 00:23:19.420 know, you can have your own feelings and those 00:23:19.420 --> 00:23:21.880 are going to be valid based on your lived experiences. 00:23:22.119 --> 00:23:24.319 You can have your own opinions. And I'm going 00:23:24.319 --> 00:23:25.940 to have a different level of respect for your 00:23:25.940 --> 00:23:27.920 opinions based on how you formed them. So if 00:23:27.920 --> 00:23:29.599 you've had opinions that are formed because you've 00:23:29.599 --> 00:23:31.920 taken both sides of the argument, you've know 00:23:31.920 --> 00:23:33.960 how to passionately argue both sides. You understand 00:23:33.960 --> 00:23:36.119 all the logic behind both sides. and then you 00:23:36.119 --> 00:23:38.140 form your opinion based on that, I'll have deep 00:23:38.140 --> 00:23:39.920 respect for your opinion. Even if I disagree 00:23:39.920 --> 00:23:42.019 with it, I'll have a deep respect for you. Now, 00:23:42.140 --> 00:23:43.680 if you have an opinion because you read something 00:23:43.680 --> 00:23:45.240 on the internet and that's your opinion now, 00:23:45.440 --> 00:23:47.079 it's still your opinion. I'm just not gonna respect 00:23:47.079 --> 00:23:48.759 it as much because you haven't formed it well. 00:23:49.019 --> 00:23:51.539 The facts we can't change though. Like you don't 00:23:51.539 --> 00:23:54.480 get two sets of facts. And you as a scientist 00:23:54.480 --> 00:23:57.039 know this, like data is the data is the data. 00:23:57.220 --> 00:23:59.480 I just saw something the other day, a marketing 00:23:59.480 --> 00:24:01.460 company marketed something based on a study. 00:24:01.500 --> 00:24:03.599 And I went back and read the study and the study 00:24:03.599 --> 00:24:06.299 was clear. No conclusive evidence. This is true. 00:24:06.519 --> 00:24:08.240 And the marketer was like, this thing studied 00:24:08.240 --> 00:24:09.960 this thing. And it's like, that should be against 00:24:09.960 --> 00:24:12.220 the law. Like it should, you should not, like 00:24:12.220 --> 00:24:15.279 it is so misleading. I was, I was angry. I can 00:24:15.279 --> 00:24:17.579 name the feeling I was angry. Well, those things, 00:24:17.599 --> 00:24:20.079 those things infuriate me as well, too. And I 00:24:20.079 --> 00:24:22.700 think that you're also tapping into something 00:24:22.700 --> 00:24:26.079 really important that I think as scientists we 00:24:26.079 --> 00:24:27.640 sometimes have a little bit of trouble with, 00:24:27.740 --> 00:24:31.339 which is the idea that emotion and intuition. 00:24:31.859 --> 00:24:34.700 These are a part of the human experience that 00:24:34.700 --> 00:24:39.000 are important in how we make decisions and important 00:24:39.000 --> 00:24:44.640 and that we use facts. Hopefully, we use logic, 00:24:45.180 --> 00:24:50.299 but we also use intuition and feeling and that 00:24:50.299 --> 00:24:54.859 is not going to be, at least as I can understand 00:24:54.859 --> 00:24:59.279 it now, easily replaced by even the best of AIs. 00:24:59.819 --> 00:25:01.420 And I know, Josiah, you weren't there with us 00:25:01.420 --> 00:25:04.190 when we were having this conversation, but Certainly 00:25:04.190 --> 00:25:07.450 you must be thinking about how AI is changing 00:25:07.450 --> 00:25:09.730 what people are doing and what people are not 00:25:09.730 --> 00:25:12.289 doing at work because you're actively trying 00:25:12.289 --> 00:25:15.130 to do that. So I mean, what, when you're thinking 00:25:15.130 --> 00:25:18.089 about at Prasena, what do you see as the great 00:25:18.089 --> 00:25:20.990 benefits of using AI and maybe even some of the 00:25:20.990 --> 00:25:23.930 challenges of where this could come in? Using 00:25:23.930 --> 00:25:26.890 AI is going to unlock things that we don't know. 00:25:27.009 --> 00:25:29.890 As humans, we are bad at predicting the future. 00:25:30.130 --> 00:25:32.910 I'm fascinated by the drawings I've seen from 00:25:32.910 --> 00:25:37.109 the 1930s and 40s of the optimistic 2020s and 00:25:37.109 --> 00:25:39.609 the people riding around on the waves with their 00:25:39.609 --> 00:25:41.250 video phones hanging out in front of them talking 00:25:41.250 --> 00:25:43.789 to the people. You know, we can laugh at that 00:25:43.789 --> 00:25:46.089 and see how cute it is and draw analogies between 00:25:46.089 --> 00:25:48.410 that and the iPhones and FaceTime. But let's 00:25:48.410 --> 00:25:50.809 face that, we're not very good at predicting 00:25:50.809 --> 00:25:52.809 what's going to happen and the effects of that 00:25:52.809 --> 00:25:56.250 technology on our interactions. AI in the short 00:25:56.250 --> 00:25:59.490 term has really made a difference in how I work. 00:25:59.809 --> 00:26:03.170 And I committed to making the AI that I make 00:26:03.170 --> 00:26:05.869 have a positive impact, putting safeguard rails 00:26:05.869 --> 00:26:09.089 around it, thinking about how people will feel 00:26:09.089 --> 00:26:12.049 when they use my technology. And yes, I do use 00:26:12.049 --> 00:26:14.809 that intentionally as a technologist, thinking 00:26:14.809 --> 00:26:17.589 about people's feelings. We've done a disservice 00:26:17.589 --> 00:26:21.109 in society and not emphasizing people's need 00:26:21.109 --> 00:26:23.630 to have strong and positive labels for their 00:26:23.630 --> 00:26:25.670 feelings, understanding the nuance of their feelings 00:26:25.670 --> 00:26:28.170 and their emotions. My research has led me to 00:26:28.170 --> 00:26:31.029 conclude that memories are formed first from 00:26:31.029 --> 00:26:34.509 our limbic system. The limbic system encodes 00:26:34.509 --> 00:26:36.990 our memories of how we feel about situations, 00:26:37.430 --> 00:26:40.950 and then our minds go back to backfill those 00:26:40.950 --> 00:26:43.710 memories with facts as best as we can fit them. 00:26:44.269 --> 00:26:47.150 If we don't have a rational capacity for labeling 00:26:47.150 --> 00:26:49.509 our feelings, we're going to have difficulties 00:26:49.509 --> 00:26:51.710 processing those feelings as memories of the 00:26:51.710 --> 00:26:54.190 future. But the point is, is that if I can focus 00:26:54.190 --> 00:26:56.990 on how the technology that I make helps people 00:26:56.990 --> 00:26:59.230 feel, makes people feel, I can make a better, 00:26:59.450 --> 00:27:01.710 more grounded experience for them. So practically 00:27:01.710 --> 00:27:04.390 speaking, that looks like when I'm making a patient 00:27:04.390 --> 00:27:07.730 advocate app that helps them advocate for themselves, 00:27:07.890 --> 00:27:10.309 know what questions to ask and respond to the 00:27:10.309 --> 00:27:13.089 patient about their medical situation. I'm not 00:27:13.089 --> 00:27:15.529 going to just throw in a large language model, 00:27:15.809 --> 00:27:18.289 five page prompt, call it a day, you often have 00:27:18.289 --> 00:27:20.960 a cocktail. I'm going to take some serious work 00:27:20.960 --> 00:27:23.299 in putting guardrails around it, deconstructing 00:27:23.299 --> 00:27:26.559 the experience, thinking about, can I take emotional 00:27:26.559 --> 00:27:29.619 cues? Can I factor in their sociological background, 00:27:29.960 --> 00:27:33.700 their family history? Can I put in place rules 00:27:33.700 --> 00:27:36.660 around what the AI can say and can't say when 00:27:36.660 --> 00:27:38.880 it can defer to the human provider? I'm going 00:27:38.880 --> 00:27:41.470 to put a very nuanced... thought to the experience 00:27:41.470 --> 00:27:43.630 so that their feelings are considered. And at 00:27:43.630 --> 00:27:45.410 the end of the day, it's going to be a very simple 00:27:45.410 --> 00:27:47.710 experience. It can be a chat window or a phone 00:27:47.710 --> 00:27:49.589 call, but there is going to be much more work 00:27:49.589 --> 00:27:51.650 put into considering the human experience rather 00:27:51.650 --> 00:27:53.930 than just the technology or how easy or hard 00:27:53.930 --> 00:27:57.099 it is for me to create. That's great. Go ahead, 00:27:57.200 --> 00:27:59.539 John, please. My hope for this is that if we 00:27:59.539 --> 00:28:01.519 get really good at prompting AI, it's going to 00:28:01.519 --> 00:28:03.339 teach us to ask better questions of one another, 00:28:03.380 --> 00:28:06.740 not just of the computer. You give a bad prompt 00:28:06.740 --> 00:28:09.140 to an AI, you get a bad set of data out. And 00:28:09.140 --> 00:28:11.279 if you're humble enough to recognize that, hey, 00:28:11.319 --> 00:28:13.140 that's because of me, not because of the computer, 00:28:13.599 --> 00:28:15.420 then you start to learn to ask better questions. 00:28:15.519 --> 00:28:17.980 And so when you look at intent analysis or sentiment 00:28:17.980 --> 00:28:21.079 analysis in computers, what it's doing is it's 00:28:21.079 --> 00:28:24.670 actually looking at the surface level. great 00:28:24.670 --> 00:28:26.490 therapist would say the problem the patient brings 00:28:26.490 --> 00:28:28.950 you isn't the problem. It's usually several layers 00:28:28.950 --> 00:28:31.109 deeper than that. And that's work that we can 00:28:31.109 --> 00:28:33.809 maybe teach computers to do over time. But I 00:28:33.809 --> 00:28:36.529 think better better serve would be having a computer, 00:28:36.529 --> 00:28:38.930 you know, psychologically code the conversation 00:28:38.930 --> 00:28:40.750 and then indicate to the human like, hey, there's 00:28:40.750 --> 00:28:42.789 something deeper here. I can't get it. That's 00:28:42.789 --> 00:28:44.910 your job. That's the wisdom work. Right. And 00:28:44.910 --> 00:28:47.470 so that's, you know, in this yesterday show up 00:28:47.470 --> 00:28:48.950 very clearly in a conversation where someone 00:28:48.950 --> 00:28:51.089 was telling me about another person. And they're 00:28:51.089 --> 00:28:52.609 like, you know, they're motivated by money. And 00:28:52.609 --> 00:28:55.000 I said, I'm sorry. you know, respectfully and 00:28:55.000 --> 00:28:56.180 expertly, I'm going to tell you that they're 00:28:56.180 --> 00:28:58.299 not motivated by money. In fact, no one's motivated 00:28:58.299 --> 00:29:00.400 by money. They're motivated by what that money 00:29:00.400 --> 00:29:02.660 does for them or what it means to them. So maybe 00:29:02.660 --> 00:29:04.440 it's ego. Maybe it's a scorecard. Maybe it's 00:29:04.440 --> 00:29:06.440 what they can buy. Maybe it's security. Maybe 00:29:06.440 --> 00:29:08.460 it's something else. But money is always the 00:29:08.460 --> 00:29:11.259 secondary thing. And so if you don't understand 00:29:11.259 --> 00:29:12.940 the underpinning motivation, you're going to 00:29:12.940 --> 00:29:14.480 have a hard time connecting with that person 00:29:14.480 --> 00:29:16.650 the way that you want to. That's an amazing point. 00:29:16.970 --> 00:29:19.769 I think we are running to the end of our time 00:29:19.769 --> 00:29:21.710 together, although I could keep talking with 00:29:21.710 --> 00:29:24.130 the two of you about how you're using AI and 00:29:24.130 --> 00:29:28.609 your thoughts about AI for hours, given my own 00:29:28.609 --> 00:29:31.630 interests in this area run really deep and also 00:29:31.630 --> 00:29:34.269 you're just both a lot of fun to talk to. I want 00:29:34.269 --> 00:29:36.990 to give you each a chance to sort of say if there's 00:29:36.990 --> 00:29:39.509 anything that you want to end with, especially 00:29:39.509 --> 00:29:41.609 about the work that you're doing at Prasena and 00:29:41.609 --> 00:29:45.720 how AI is shaping that. Maybe you'll turn to 00:29:45.720 --> 00:29:47.920 Josiah and then let John kind of close us out. 00:29:48.339 --> 00:29:50.976 My view on AI, if I could say anything is... 00:29:50.970 --> 00:29:53.029 see through the hype and look at it as a tool. 00:29:53.369 --> 00:29:55.369 Inherently, it's about how we use the tool. I 00:29:55.369 --> 00:29:57.690 would look like an idiot if I showed up to a 00:29:57.690 --> 00:29:59.609 wedding with a hammer to cut a cake. I'd look 00:29:59.609 --> 00:30:01.430 like an idiot if I showed up to a construction 00:30:01.430 --> 00:30:04.329 site with a cake cutter to hang drywall. AI is 00:30:04.329 --> 00:30:07.509 a tool, learn what it is, learn how it can assist 00:30:07.509 --> 00:30:09.329 you and use the right tool for the right job. 00:30:09.509 --> 00:30:11.970 And that's what I try to do with AI and tools 00:30:11.970 --> 00:30:14.869 in Prasena to help our patients and help our 00:30:14.869 --> 00:30:17.809 providers. I'll end up with our latest research 00:30:17.809 --> 00:30:19.410 that hasn't yet been published, but our data 00:30:19.410 --> 00:30:21.890 has been collected. We had 79 patients with an 00:30:21.890 --> 00:30:25.650 average A1C of 11 .06 coming out of the hospital. 00:30:25.990 --> 00:30:28.349 And in the first 90 days, dropped them to 7 .8, 00:30:28.349 --> 00:30:31.849 and 180 days to 7 .2, with 30 % coming off of 00:30:31.849 --> 00:30:34.589 insulin at the six -month mark. The clinical 00:30:34.589 --> 00:30:36.470 results we're having, because we've leaned into 00:30:36.470 --> 00:30:39.930 the tools, are really fun to talk about. And 00:30:39.930 --> 00:30:42.809 it took us 10 years to get here. We've been playing 00:30:42.809 --> 00:30:45.819 with AI for a decade. I think it's really, it's 00:30:45.819 --> 00:30:47.779 like we're another overnight success 10 years 00:30:47.779 --> 00:30:50.500 in the making. And so, you know, I think, I think 00:30:50.500 --> 00:30:52.640 this is, it's a thing that's here to stay. And 00:30:52.640 --> 00:30:54.099 it's something that's going to take time for 00:30:54.099 --> 00:30:56.680 people to learn. And it's going to be frustrating 00:30:56.680 --> 00:30:58.440 in moments. Certainly we talk about all these 00:30:58.440 --> 00:31:00.240 great things we're doing when we're on podcasts 00:31:00.240 --> 00:31:02.380 like this, but you know, we also, we don't talk 00:31:02.380 --> 00:31:04.500 about the, all the, the long weekends that, you 00:31:04.500 --> 00:31:06.539 know, Josiah, Dustin and I spent 16 hours on 00:31:06.539 --> 00:31:08.819 a Saturday and Sunday banging our head into a 00:31:08.819 --> 00:31:11.359 wall to get it wrong 32 times. And then we finally 00:31:11.359 --> 00:31:13.779 get something that, that makes sense and is helpful. 00:31:14.039 --> 00:31:16.210 So. it's a tool and it's a tool that's being 00:31:16.210 --> 00:31:18.269 shaped and developed right now. So it's not a 00:31:18.269 --> 00:31:20.910 mass generated hammer or a video game that's 00:31:20.910 --> 00:31:22.710 been worked on for 10 years that has no bugs. 00:31:22.890 --> 00:31:25.150 Like AI is this thing that's constantly being 00:31:25.150 --> 00:31:27.329 relearned and every time a new model comes out 00:31:27.329 --> 00:31:30.049 from any of the major producers, we have to evaluate 00:31:30.049 --> 00:31:31.970 the whole model in light of what we know and 00:31:31.970 --> 00:31:34.029 then maybe even unplug something and replug something. 00:31:34.089 --> 00:31:36.390 So I'm constantly pushing on Josiah for more 00:31:36.390 --> 00:31:39.009 modular, flexible development and to be able 00:31:39.009 --> 00:31:41.529 to be more nimble more as we grow. And he of 00:31:41.529 --> 00:31:43.869 course is like constantly pushing. back on me 00:31:43.869 --> 00:31:45.670 about the unreasonable demands that I'm making 00:31:45.670 --> 00:31:48.789 about technology. But the reality is what it's 00:31:48.789 --> 00:31:50.910 going to lead to is better patient agency, better 00:31:50.910 --> 00:31:53.450 healthcare outcomes, and people being able to 00:31:53.450 --> 00:31:56.130 spend more time focusing on human connection 00:31:56.130 --> 00:31:59.190 with each other if we want it to. And so my hope 00:31:59.190 --> 00:32:00.930 is that we have a more connected world, not a 00:32:00.930 --> 00:32:03.470 less connected world because of AI, if we choose 00:32:03.470 --> 00:32:05.710 to use the tool that way. And I think that, you 00:32:05.710 --> 00:32:07.730 know, in our, we're seeing that happen for real 00:32:07.730 --> 00:32:10.130 inside of our practice, with our patients, with 00:32:10.130 --> 00:32:12.910 our team, with each other. We're a much more 00:32:12.910 --> 00:32:15.190 connected community than we were even a year 00:32:15.190 --> 00:32:17.289 ago because we have more time. more space, more 00:32:17.289 --> 00:32:19.809 energy, more to talk about. And so for anybody 00:32:19.809 --> 00:32:21.430 out there, I'd really encourage you like use 00:32:21.430 --> 00:32:24.950 AI to drive forward human connection, not to 00:32:24.950 --> 00:32:26.869 isolate yourself, which is already too big a 00:32:26.869 --> 00:32:29.529 problem, not to become more polar in the conversation, 00:32:29.549 --> 00:32:31.809 but to come back and meet people where they are. 00:32:31.829 --> 00:32:33.589 Like go meet your neighbor right where they are. 00:32:33.789 --> 00:32:36.250 It is the most amazing human experience to do 00:32:36.250 --> 00:32:38.289 that. Like that's what we're really excited about. 00:32:38.329 --> 00:32:40.589 Like how do we do this for a billion people in 00:32:40.589 --> 00:32:43.210 the next 50 years? That's our goal. Amazing. 00:32:43.410 --> 00:32:46.059 Thank you, John and Josiah for joining us. on 00:32:46.059 --> 00:32:47.900 such a great discussion. If you want to learn 00:32:47.900 --> 00:32:50.900 more about the work of Prasena Health, you can 00:32:50.900 --> 00:32:54.059 go to their website, prasena .com, P -R -E -C 00:32:54.059 --> 00:32:57.000 -I -N -A .com. And if you have questions for 00:32:57.000 --> 00:32:58.619 the guests on our show or you want to support 00:32:58.619 --> 00:33:01.059 the transformative research and programs here 00:33:01.059 --> 00:33:05.400 at USC, you can email us at listenuppeople at 00:33:05.400 --> 00:33:08.759 usc .edu. And just one more time, thank you both 00:33:08.759 --> 00:33:10.839 so much for joining me. Thanks, Eric. We'll be 00:33:10.839 --> 00:33:12.420 back anytime. We'd love to geek out about this 00:33:12.420 --> 00:33:15.299 stuff. 100%. you