Latest / Get Real with Jim Weiss / Brian Anderson and CHAI Are Making Sure AI Stays Honest

Transcript

Brian Anderson and CHAI Are Making Sure AI Stays Honest · Get Real with Jim Weiss

• 0:00
  Speaker 1: Okay, So two things came up in the episode, and
• 0:02
  Speaker 1: I would love to ask your take on them because
• 0:04
  Speaker 1: you asked Bryan's what do you think is the definition
• 0:08
  Speaker 1: of good and responsible AI?
• 0:11
  Speaker 2: You know, my view of it is the AI serving
• 0:15
  Speaker 2: the purpose you intended to serve. I think that to
• 0:18
  Speaker 2: me is the number one point. And in serving it,
• 0:21
  Speaker 2: do you serve it at high quality, at the highest
• 0:24
  Speaker 2: quality that you possibly can? And reliability and trustworthiness. So
• 0:29
  Speaker 2: I think you just got to know that I use
• 0:32
  Speaker 2: AI tools all the time and sometimes they spit back
• 0:35
  Speaker 2: some things that I question, and sometimes it feels right
• 0:40
  Speaker 2: because sometimes you just may be asking the question to
• 0:43
  Speaker 2: get the answer you want. That's where we got to
• 0:46
  Speaker 2: take some of that bias out of it all. That's
• 0:50
  Speaker 2: where AI, if it's really going to help us be smarter.
• 0:54
  Speaker 2: You know, it has to be objective and it has
• 0:57
  Speaker 2: to be administered at the highest quality level, and that
• 1:01
  Speaker 2: means good quality data, because that was something Brian talks
• 1:05
  Speaker 2: about in the piece that that's the biggest challenge, having
• 1:08
  Speaker 2: high quality data that informed the AI model. Good to
• 1:14
  Speaker 2: see you.
• 1:14
  Speaker 3: Man, Thanks for having me.
• 1:16
  Speaker 2: We've got a lot to get real about. I mean,
• 1:18
  Speaker 2: AI couldn't before it's like, well, I wanted to ask,
• 1:22
  Speaker 2: so why now, why you? And what was missing at
• 1:25
  Speaker 2: the intersection of AI and healthcare in particular that motivated
• 1:29
  Speaker 2: you to co found CHAI and twenty.
• 1:31
  Speaker 3: One as the pandemic was happening, if you recalled, during
• 1:35
  Speaker 3: that time, Operation Warp Speed was stood up. I was
• 1:37
  Speaker 3: asked by the Trump administration and then subsequently the Biden
• 1:41
  Speaker 3: administration to help lead data analytics for Operation Warp Speed.
• 1:44
  Speaker 3: And you had private sector technology companies and health systems
• 1:48
  Speaker 3: working alongside the federal government and state and local governments.
• 1:52
  Speaker 3: And we appreciated the impact we were having. When you
• 1:55
  Speaker 3: have organizations that are inherently competitive working collectively together, real impact,
• 2:00
  Speaker 3: real progress can happen, as we saw in the pandemic.
• 2:03
  Speaker 3: We appreciated that, and we asked a very basic question,
• 2:05
  Speaker 3: which is, you know, is there something more that we
• 2:07
  Speaker 3: can be doing in this moment? And we looked at AI.
• 2:09
  Speaker 3: I mean, this is before chat GPT and all of
• 2:12
  Speaker 3: the JENAI buzz happened, and we asked a basic question,
• 2:16
  Speaker 3: which was do we have consensus at a technically specific
• 2:20
  Speaker 3: level on what responsible AI should look like in health
• 2:24
  Speaker 3: And the answer we quickly came to I still remember this.
• 2:26
  Speaker 3: We had one of our colleagues from Stanford who looked
• 2:29
  Speaker 3: across you know, literature and said, you know, there are
• 2:31
  Speaker 3: two hundred and twenty plus competing different approaches to responsible AI.
• 2:36
  Speaker 3: So so no, there is no technical consensus, certainly because
• 2:39
  Speaker 3: of the pandemic how we saw AI was being used
• 2:41
  Speaker 3: that while in a consequential sector like health, we probably
• 2:45
  Speaker 3: should have consensus. And that has been our north star.
• 2:48
  Speaker 3: Is it bringing together a group of clinicians from health
• 2:51
  Speaker 3: systems and the builders from the innovation space to build
• 2:55
  Speaker 3: consensus on what good AI looks like in healthcare and
• 2:58
  Speaker 3: building that trust. I think that was the other thing
• 3:00
  Speaker 3: that I certainly appreciated in the pandemic was the lack
• 3:04
  Speaker 3: of trust that we have in innovative digital technologies in
• 3:09
  Speaker 3: healthcare and AI. Think there are many polls and surveys
• 3:12
  Speaker 3: out there that show the majority of the population doesn't
• 3:15
  Speaker 3: trust AI, trusts AI even less when it's consequential use
• 3:19
  Speaker 3: cases like my own personal health. And so that's been
• 3:21
  Speaker 3: the mission of CHI building those technically specific industry standards
• 3:26
  Speaker 3: in partnership with doctors and nurses and patient advocates and
• 3:29
  Speaker 3: the technology companies that build the tools that we use.
• 3:32
  Speaker 2: The question is you're working with the current administration on
• 3:35
  Speaker 2: any of this? I assume so, And how is CHI
• 3:38
  Speaker 2: developing consensus driven guidelines for responsible AI and health, Like,
• 3:43
  Speaker 2: what's the process you're undertaking.
• 3:46
  Speaker 3: When we started CHAI, right, it started again looking at
• 3:50
  Speaker 3: Operation warp Speed. It was a public private partnership and
• 3:52
  Speaker 3: I think we appreciated that there is power when you
• 3:56
  Speaker 3: bring together the innovation community and the public section, and
• 4:00
  Speaker 3: particularly an AI where you have a incredibly fast moving technology.
• 4:06
  Speaker 3: It's almost breathlessly developing new capabilities month by month. That
• 4:11
  Speaker 3: you want to inform the policy makers the regulators on
• 4:15
  Speaker 3: what those new capabilities are and what those new uses are,
• 4:19
  Speaker 3: because that's going to inform appropriate informed regulation and where
• 4:25
  Speaker 3: we need the kinds of incentives to unlock that next
• 4:28
  Speaker 3: set of use cases or capabilities that can improve you know,
• 4:32
  Speaker 3: in the case of health, healthcare and outcomes for our patients.
• 4:34
  Speaker 3: And so that vision of establishing a public private partnership
• 4:38
  Speaker 3: has been central to CHY. And so how do we
• 4:40
  Speaker 3: do that. We do that fundamentally through a number of
• 4:43
  Speaker 3: working groups, right, and so I'd mentioned we build these
• 4:45
  Speaker 3: technically specific industry standards. It sounds I admit very geeky.
• 4:49
  Speaker 3: But what we do is we build these working groups
• 4:52
  Speaker 3: that bring together experts from startup companies, from health systems
• 4:57
  Speaker 3: with doctors and nurses and put them in a room,
• 5:00
  Speaker 3: virtual room most of the time, and they'll go use
• 5:02
  Speaker 3: case by use case. So use case might be how
• 5:05
  Speaker 3: do we use AI to support doctors with ambient listening,
• 5:10
  Speaker 3: which is patients aren't familiar with that, or your listeners
• 5:12
  Speaker 3: aren't familiar with that. That's where doctors essentially are having
• 5:15
  Speaker 3: a device in their examine that's listening into the patient
• 5:17
  Speaker 3: and provider talking and helping the provider to construct a
• 5:21
  Speaker 3: note for their medical record. And it doesn't require the
• 5:24
  Speaker 3: provider to like stare at the computer screen. And that's
• 5:27
  Speaker 3: a really exciting thing for doctors because we get into
• 5:30
  Speaker 3: medicine because we want to connect with our patients, not
• 5:32
  Speaker 3: because we want to stare at a computer screen. And
• 5:34
  Speaker 3: so when you bring together those innovators and those doctors
• 5:37
  Speaker 3: and nurses, and then you bring in leaders from the
• 5:40
  Speaker 3: FDA or from other departments within health and Human services
• 5:45
  Speaker 3: at the federal level, and you tell them how doctors
• 5:48
  Speaker 3: are using this right, what patients are concerned about. That
• 5:51
  Speaker 3: then informs them so that when they go and make
• 5:55
  Speaker 3: regulation or go and create policy, they can do that
• 5:59
  Speaker 3: in a way that is informed. And that is essentially
• 6:01
  Speaker 3: what Chai is trying to do is we're trying to
• 6:04
  Speaker 3: create a space, a safe space, where we can inform
• 6:08
  Speaker 3: our public officials on how to best incentivize and create
• 6:12
  Speaker 3: the safe and appropriate regulations that aren't going to stiple innovation.
• 6:16
  Speaker 3: You know, I'll be candid with you and say that
• 6:17
  Speaker 3: relationship got off to a fantastic start with the current administration,
• 6:22
  Speaker 3: literally from a few days after President Trump was elected
• 6:25
  Speaker 3: back in November of twenty twenty four, started talking to
• 6:28
  Speaker 3: the transition team, and then fast forward to January, we
• 6:32
  Speaker 3: really got a strong engagement in a number of our
• 6:36
  Speaker 3: working groups, from leaders from NIH to CMS, to the
• 6:40
  Speaker 3: Office of National Coordinator to the Veterans Administration, and so
• 6:44
  Speaker 3: it was really exciting. And then somewhere along the way,
• 6:48
  Speaker 3: I think there was a pretty significant misunderstanding about what
• 6:51
  Speaker 3: Chai is. People started calling us a gatekeeper or artel,
• 6:56
  Speaker 3: and these couldn't be things further from the truth. Chai
• 6:59
  Speaker 3: tries to create a space where we can incentivize and
• 7:03
  Speaker 3: accelerate innovation by coming up with a common definition of
• 7:06
  Speaker 3: what good is right. If you think back to econ
• 7:10
  Speaker 3: one oh one in college or in high school, one
• 7:12
  Speaker 3: of the most important lessons for a builder is to
• 7:15
  Speaker 3: know your customer right, to understand what your customer wants.
• 7:18
  Speaker 3: And we're creating that space where the builders and their customers,
• 7:21
  Speaker 3: health systems, payers, life science companies can essentially co create
• 7:26
  Speaker 3: what they want. What does the definition of good responsible
• 7:29
  Speaker 3: AI look like? And with that common definition accelerate the
• 7:32
  Speaker 3: kind of sales and growth of companies that we need
• 7:35
  Speaker 3: to innovate in the space we're trying to work through that.
• 7:38
  Speaker 3: I've gone on the record multiple times letting folks in
• 7:41
  Speaker 3: the current administration know that we would love to engage
• 7:44
  Speaker 3: them and understand what their concerns are, and again to
• 7:47
  Speaker 3: create that space where we can be a resource to
• 7:51
  Speaker 3: educate our elected officials on what the private sector, innovation community,
• 7:55
  Speaker 3: and the leaders of healthcare are dealing with AI.
• 7:59
  Speaker 2: Well, is it Congress that you need to interact with
• 8:02
  Speaker 2: more or is it the governement?
• 8:04
  Speaker 3: Yeah, I mean so we speak a lot with congressional officials.
• 8:07
  Speaker 3: I think certainly because AI is moving so quickly, it's
• 8:09
  Speaker 3: hard to keep up with the technology. I have the
• 8:12
  Speaker 3: privilege of having a full time job that allows me
• 8:14
  Speaker 3: to do that and my team too. We try to
• 8:18
  Speaker 3: support our elected officials by being an educational resource, a
• 8:21
  Speaker 3: bipartisan or nonpartisan educational resource on what responsible AI is.
• 8:25
  Speaker 3: And so we speak to Congress a lot, and then
• 8:27
  Speaker 3: we speak to the executive branch officials when they're thinking about, well,
• 8:31
  Speaker 3: if there's a concern about a need for a regulation,
• 8:34
  Speaker 3: what are the potential downstream consequences if we create a
• 8:38
  Speaker 3: very specific regulation, or if we create too broad of
• 8:41
  Speaker 3: a regulation, or if we create a regulation in the
• 8:44
  Speaker 3: transparency space, And again, trying to offer CHAI as an
• 8:48
  Speaker 3: educational resource to support their regulations. What CHAI is not
• 8:52
  Speaker 3: is a regulatory entity. What we want to be is
• 8:54
  Speaker 3: an educational resource.
• 8:56
  Speaker 2: Right, And that's I guess where I was thinking. Maybe
• 9:00
  Speaker 2: or an administration doesn't want a lot of regulations, right,
• 9:02
  Speaker 2: I think that's their goal. So how do you help
• 9:07
  Speaker 2: walk because it's a real line to walk, right. I
• 9:09
  Speaker 2: don't know how you balance those two.
• 9:12
  Speaker 3: There is no easy answer. I'll be, you know, the
• 9:14
  Speaker 3: first to admit that this is something that you have
• 9:16
  Speaker 3: to have a great deal of humility when you're talking
• 9:19
  Speaker 3: about AI because the space is moving so quickly, and
• 9:22
  Speaker 3: there are so many things that we don't know yet.
• 9:24
  Speaker 3: I think some lessons that I've learned, and this certainly
• 9:27
  Speaker 3: isn't an exhaustive list. Creating regulations that are too technically
• 9:31
  Speaker 3: specific is generally a bad idea. There are multiple examples
• 9:35
  Speaker 3: of the previous administration where regulations were created that I
• 9:40
  Speaker 3: think we're well intended, but we're so focused on the
• 9:44
  Speaker 3: capabilities of the models at that moment right twenty twenty
• 9:47
  Speaker 3: two to twenty twenty three, that if you create a regulation,
• 9:50
  Speaker 3: for example, that an AI model of a certain size
• 9:54
  Speaker 3: needs to be regulated if it's above that line in
• 9:56
  Speaker 3: the sand, right, because any models that back in twenty
• 9:59
  Speaker 3: twenty two twenty twenty three that were above that size, like, wow,
• 10:02
  Speaker 3: they were dangerous, that they could do bad things. But
• 10:04
  Speaker 3: then fast forward to now, for example, we have models
• 10:07
  Speaker 3: that are much smaller than that line in the sand
• 10:10
  Speaker 3: that was drawn in twenty twenty two that are conceivably
• 10:13
  Speaker 3: even more dangerous because they're more capable. And so I
• 10:16
  Speaker 3: think it's a balance between being specific enough that the
• 10:20
  Speaker 3: regulation is actually going to have some appropriate intended consequence
• 10:25
  Speaker 3: that you want, but not so specific and rigid that
• 10:28
  Speaker 3: it's going to be out of date in six months.
• 10:31
  Speaker 3: I think another lesson I've learned is creating a framework
• 10:36
  Speaker 3: for the industry to fill in the details is generally,
• 10:40
  Speaker 3: I think a good idea. I mean, we started try
• 10:42
  Speaker 3: because there were no industry standards in this space right
• 10:44
  Speaker 3: There's not a technical level of specificity that we have
• 10:47
  Speaker 3: in other more mature areas of the economy. And for
• 10:50
  Speaker 3: the government to come in and define before the industry
• 10:55
  Speaker 3: has defined what those standards are, I don't think that's
• 10:59
  Speaker 3: a good idea. But I do think it's a good
• 11:01
  Speaker 3: idea to say, here's a general framework about how we
• 11:05
  Speaker 3: think about safety or responsible AI, and you, industry, you
• 11:09
  Speaker 3: fill in the details. But here's how we think about transparency,
• 11:12
  Speaker 3: or here's how we think about farns or accountability. But industry,
• 11:16
  Speaker 3: we want you to build the consensus to fill into details.
• 11:19
  Speaker 3: And so I think those are some examples of some
• 11:20
  Speaker 3: lessons I've learned on what might make good regulations, But
• 11:24
  Speaker 3: again really defer to the regulatory community in the state
• 11:28
  Speaker 3: and federal government to actually do that well.
• 11:30
  Speaker 2: On your website, there are timelines for major rollouts and
• 11:34
  Speaker 2: use case work groups and cross cutting work groups for
• 11:37
  Speaker 2: mental health, chatbots, clinical trials, ambient AI, agentic, AI payment, integrity,
• 11:43
  Speaker 2: risk categorization, AI governance, and so much more. Let's hear
• 11:48
  Speaker 2: about which ones are the most mission critical? Are there
• 11:51
  Speaker 2: some that are more important than others, and that probably
• 11:55
  Speaker 2: some are not so controversial. It's okay, everybody agrees, and
• 12:00
  Speaker 2: then there's a few areas we really need to hammer out.
• 12:03
  Speaker 3: I'll give you three and a fourth one that I
• 12:06
  Speaker 3: think are really important. The first one is governance. This
• 12:08
  Speaker 3: is something that we have heard really loudly from our
• 12:12
  Speaker 3: community of health systems and payers and health plans, the
• 12:17
  Speaker 3: need to understand what are the best practices on how
• 12:21
  Speaker 3: do we manage these tools. A health system or a
• 12:25
  Speaker 3: pair may have gone through a procurement process and spent
• 12:28
  Speaker 3: millions of dollars to purchase this tool, this AI tool,
• 12:31
  Speaker 3: and now that they have it deployed, well, how the
• 12:33
  Speaker 3: heck do we manage it and maintain it? Because like
• 12:36
  Speaker 3: any tool, like any hammer or tool that we might have, all.
• 12:40
  Speaker 2: The tech and software is only is epic is a
• 12:43
  Speaker 2: good example, right, is how you use it?
• 12:45
  Speaker 3: That's right? That's right, and you need to have that
• 12:47
  Speaker 3: kind of insight into how it's performing. What are the
• 12:50
  Speaker 3: metrics that we should be monitoring for its performance and
• 12:53
  Speaker 3: its safety, and how frequently should we be monitoring it?
• 12:56
  Speaker 3: And what is the threshold for performance degradation before we
• 12:59
  Speaker 3: pull it off the production line and need to retune
• 13:01
  Speaker 3: it or retrain it. These are all questions that we're
• 13:04
  Speaker 3: still figuring out as a community, and so AI governance
• 13:07
  Speaker 3: is perhaps I think the single most important one that
• 13:10
  Speaker 3: we're focused on at a working group level in terms
• 13:12
  Speaker 3: of building out those playbooks for health systems and prepayers
• 13:16
  Speaker 3: and health plans. The second one is in the agentic space.
• 13:19
  Speaker 3: I believe that one of the most transformative moments in
• 13:23
  Speaker 3: healthcare is going to come when semi autonomous autonomous AI
• 13:29
  Speaker 3: tools move from a tool where I as a clinician
• 13:32
  Speaker 3: and using a tool with my patient, from tool to
• 13:36
  Speaker 3: care team member, meaning a member of a care team
• 13:38
  Speaker 3: that's able to operate in a semi autonomous or an
• 13:40
  Speaker 3: autonomous way. You know, there's a great story that just
• 13:42
  Speaker 3: came out yesterday about an AI company, Doctronic. They're a
• 13:46
  Speaker 3: member of CHAI and they're building an a gentic AI
• 13:49
  Speaker 3: tool that solves a lot of the pressing problems we
• 13:53
  Speaker 3: have in healthcare today. Namely the scarcity of access to
• 13:57
  Speaker 3: providers to help adjust a metiction in a timely manner.
• 14:01
  Speaker 3: And so this concept of management of chronic disease in
• 14:05
  Speaker 3: resource impoverished areas like rural communities, where I might have
• 14:10
  Speaker 3: to drive one hundred miles or wait three weeks to
• 14:12
  Speaker 3: see a provider. Now I can see an agentic AI
• 14:14
  Speaker 3: doctor that might help manage my diabetes or my hypertension meds,
• 14:19
  Speaker 3: and I don't have to wait three weeks. I can
• 14:21
  Speaker 3: get online or on my phone and talk to the
• 14:23
  Speaker 3: agentic doctor. That I think is going to be a
• 14:26
  Speaker 3: really transformative moment when that becomes more widely.
• 14:29
  Speaker 2: Wait, does that mean nobody has to go to med
• 14:31
  Speaker 2: school anymore?
• 14:33
  Speaker 3: Well, I mean it brings up an interesting question. I
• 14:35
  Speaker 3: was talking to an MIT professor earlier today about this
• 14:38
  Speaker 3: exact question. I think it means that we need to
• 14:41
  Speaker 3: rethink how we train future physicians. This concept of governance
• 14:46
  Speaker 3: and how we manage and monitor these tools is going
• 14:50
  Speaker 3: to be really important. No one has developed a training
• 14:52
  Speaker 3: curriculum on AI governance, but I think that's one of
• 14:54
  Speaker 3: the most pressing things that medical students doctors of the
• 14:58
  Speaker 3: future are going to need to do because we're as physicians,
• 15:00
  Speaker 3: we're going to be monitoring and maintaining these agentic tools
• 15:03
  Speaker 3: across the board. If you had to ask me, you know, Jim,
• 15:05
  Speaker 3: I think what is that keeps me up at night
• 15:08
  Speaker 3: with AI? It's the alignment problem we have with it,
• 15:11
  Speaker 3: which basically comes down to how do we ensure that
• 15:15
  Speaker 3: these models are aligned to our values? And in a
• 15:20
  Speaker 3: sector like pharaoh space or autonomous driving, it's fairly straightforward, right,
• 15:25
  Speaker 3: don't kill people, right, it's an easier thing to align to.
• 15:28
  Speaker 3: But when it comes to human health, our values are
• 15:31
  Speaker 3: very complex, right. They're intertwined with our faith, with our family,
• 15:36
  Speaker 3: with our upbringing. The different values that people have about
• 15:39
  Speaker 3: their health can get quite complex. And as we move
• 15:42
  Speaker 3: into this agentic future, ensuring that these models, right, these
• 15:47
  Speaker 3: future AI doctors and nurses are aligned to our values
• 15:51
  Speaker 3: and can understand the difference between my values as Brian
• 15:55
  Speaker 3: and your values as Jim, is going to be really
• 15:57
  Speaker 3: important when it comes to making shared decision making.
• 16:01
  Speaker 2: So what is the membership structure and how do companies
• 16:05
  Speaker 2: and stakeholders work with you?
• 16:07
  Speaker 3: Yeah, so CHAI is a member driven community, as you're mentioning,
• 16:10
  Speaker 3: so a coalition of three thousand organizations participate. Membership is
• 16:15
  Speaker 3: membership dues based, and so right to be able to
• 16:17
  Speaker 3: fund and support and facilitate the working groups. The writing,
• 16:21
  Speaker 3: the development of the tools and products that we offer
• 16:24
  Speaker 3: to our members are membership dues associated with that, and
• 16:27
  Speaker 3: those dues are based on the size of the organization,
• 16:30
  Speaker 3: So the larger the organization, the higher the dues. It's
• 16:34
  Speaker 3: really affordable for startups to join. The broader community is
• 16:37
  Speaker 3: entirely free. I have a strong belief that this is
• 16:40
  Speaker 3: something that should be open and available to everyone that
• 16:44
  Speaker 3: wants to participate. So we have quite an active community.
• 16:46
  Speaker 3: When we launch any of the products of the working
• 16:49
  Speaker 3: groups you listed earlier, we share that first with our
• 16:52
  Speaker 3: community and we ask for feedback and whether or not
• 16:54
  Speaker 3: you were part of the working group that developed it
• 16:56
  Speaker 3: or just part of the community that is actively involved
• 16:59
  Speaker 3: in using this stuff, we want their feedback.
• 17:01
  Speaker 2: How does CHAI have teeth, like, how do you ensure
• 17:05
  Speaker 2: members use these models thro're developing or is that all voluntary?
• 17:09
  Speaker 3: Great question. I have done a lot of work in
• 17:13
  Speaker 3: the standard space better part of a decade, and what
• 17:15
  Speaker 3: I don't want CHI to be is a group that
• 17:18
  Speaker 3: just publishes white papers or standards that you know, set
• 17:21
  Speaker 3: up on bookshelf and gather dust and no one adopts
• 17:24
  Speaker 3: or uses. And so we made a commitment that, you know,
• 17:28
  Speaker 3: we would have a very robust feedback process one and
• 17:33
  Speaker 3: so anything that we publish, we are committed to iterating
• 17:36
  Speaker 3: on that on an annual basis with our community. So
• 17:39
  Speaker 3: particularly because AIS movies so quickly.
• 17:41
  Speaker 2: Yeah, annual seems like too long.
• 17:44
  Speaker 3: In some ways it might be, and so open to
• 17:46
  Speaker 3: revisiting that. So there's an iterative kind of feedback loop
• 17:49
  Speaker 3: to the content we're developing to make sure that it
• 17:51
  Speaker 3: doesn't go stale. The second part, and this I think
• 17:54
  Speaker 3: is the more interesting part, is one of the things
• 17:57
  Speaker 3: I think we appreciated in this work is when you
• 18:00
  Speaker 3: come up with a common definition of what good in
• 18:03
  Speaker 3: this case, good responsible AI looks like at a technically
• 18:06
  Speaker 3: specific level, it's much easier to develop a set of
• 18:10
  Speaker 3: objective metrics evaluation criteria to be able to objectively say,
• 18:17
  Speaker 3: how does this model perform according to this technical spec
• 18:20
  Speaker 3: Is it performing well, is it performing poorly? And really
• 18:24
  Speaker 3: having a commitment to transparency around how models are performing
• 18:29
  Speaker 3: against said definition of responsible AI, and so what we've
• 18:33
  Speaker 3: committed to do in HI is create this registry that
• 18:37
  Speaker 3: I had shared earlier that we're going to be announcing
• 18:39
  Speaker 3: and launching in April and opening that up to health
• 18:43
  Speaker 3: systems that want to begin sharing how models deployed locally
• 18:49
  Speaker 3: at their health system are performing. This is something that's
• 18:52
  Speaker 3: never been done before, and honestly, it is harder than
• 18:55
  Speaker 3: I thought it would be because when you get down
• 18:57
  Speaker 3: to the nitty gritty about how a model is performing
• 19:00
  Speaker 3: and what its accuracy is, it becomes easier to compare
• 19:03
  Speaker 3: apples to apples, like how is one ambient listening product
• 19:08
  Speaker 3: compare against another ambient listening product? How does one sepsist
• 19:11
  Speaker 3: clinical decision support tool compare against another sepsist clinical decision
• 19:15
  Speaker 3: support tool. But that kind of transparency and objectiveness around
• 19:20
  Speaker 3: the performance of these models is how we build trust
• 19:23
  Speaker 3: and how we hold all of us accountable. Like if
• 19:27
  Speaker 3: the metrics suck, then well we need to go do
• 19:29
  Speaker 3: better and develop better metrics. But if there's a product
• 19:32
  Speaker 3: out there that's, by golly, it's like it's scoring in
• 19:35
  Speaker 3: the stratosphere, and then there are these other products that
• 19:38
  Speaker 3: are doing really poorly well, that should beg a few
• 19:42
  Speaker 3: questions about why is one performing so poorly while the
• 19:45
  Speaker 3: other's performing so highly. And I think we're going to
• 19:47
  Speaker 3: get to a better place when we have that kind
• 19:49
  Speaker 3: of transparency and that kind of accountability. And so I
• 19:52
  Speaker 3: think that's how we have relevance and accountability and a
• 19:57
  Speaker 3: desire I think to move this field forward with that
• 20:00
  Speaker 3: kind of radical transparency.
• 20:01
  Speaker 2: How did working at Athena shape of goals you're now
• 20:04
  Speaker 2: implementing at Chai.
• 20:05
  Speaker 3: Yeah, no, I mean Athena was a fantastic place to
• 20:08
  Speaker 3: get started in the AI space. This was, you know,
• 20:11
  Speaker 3: fifteen or so years ago. I remember when I first
• 20:13
  Speaker 3: started at Athena. This is again rewinding the tape a
• 20:16
  Speaker 3: little bit. Athena at the time when I joined was
• 20:20
  Speaker 3: I think the first major EHR vendor on the Internet, right,
• 20:24
  Speaker 3: that was a big deal to have an EHR vendor
• 20:26
  Speaker 3: totally SaaS based on the Internet. You didn't kind of
• 20:29
  Speaker 3: deploy it and install it in your hospital basement. And
• 20:32
  Speaker 3: because of that, we had a single instance that had
• 20:38
  Speaker 3: all the data being captured in a similar way with
• 20:42
  Speaker 3: you know, various kind of small permutations in the configurations
• 20:45
  Speaker 3: for a local clinic, but the data was essentially captured
• 20:49
  Speaker 3: in a homogeneous way that allowed us to have incredible
• 20:53
  Speaker 3: power when it came to developing data sets to train
• 20:57
  Speaker 3: models on and to develop clinical support tools on. And
• 21:01
  Speaker 3: so I think that was like one lesson that I
• 21:04
  Speaker 3: learned is when you have thirty plus thousand providers all
• 21:08
  Speaker 3: using the same EHR instance capturing data in the same way, gosh,
• 21:13
  Speaker 3: you can find things out really quickly to support those providers.
• 21:17
  Speaker 3: And so, you know, when we started rolling out AI services,
• 21:21
  Speaker 3: I think one of the first areas that we focused
• 21:23
  Speaker 3: on was new guidelines come out all the time, Right.
• 21:26
  Speaker 2: Then is CHAI helping them with the guidelines?
• 21:29
  Speaker 3: I assume, yeah, So actually, yeah, I'll get back to that.
• 21:32
  Speaker 2: That's the thing for that point.
• 21:34
  Speaker 3: But at Athena, what we appreciated was because doctors don't
• 21:37
  Speaker 3: have time to read these we can develop clinical decision
• 21:40
  Speaker 3: support or CDs tools, some of which were powered by AI,
• 21:44
  Speaker 3: to identify patients that were potentially at high risk for
• 21:48
  Speaker 3: whatever the outcome that this guideline was trying to stop,
• 21:52
  Speaker 3: and then intervene on the provider's behalf to reach out
• 21:56
  Speaker 3: to that patient, to get that patient to come in
• 21:58
  Speaker 3: to be seen by that provider to get the appropriate care.
• 22:01
  Speaker 3: So I think that was one really exciting space where
• 22:03
  Speaker 3: CHAI is working with the specialty societies. Now, if you
• 22:06
  Speaker 3: ask any provider who do they trust for the clinical
• 22:10
  Speaker 3: best practices of their specialty, they go to the specialty society.
• 22:14
  Speaker 3: They turn to AAP for pediatrics, or ACP for internal
• 22:19
  Speaker 3: medicine docs or ASCO if I'm an oncologists and with AI,
• 22:24
  Speaker 3: it is a tool JIM, and a tool to be
• 22:27
  Speaker 3: used in clinical guidelines or clinical workflows. And so what
• 22:31
  Speaker 3: we're doing because we want to build trust with the
• 22:34
  Speaker 3: end users physicians and nurses and patients. But in this
• 22:38
  Speaker 3: case for nurses and physicians, for clinicians, we want to
• 22:41
  Speaker 3: partner with those societies that they trust to help build
• 22:44
  Speaker 3: out what these new guidelines for the use of AI
• 22:48
  Speaker 3: management of hypertension, or management of diabetes or management of
• 22:52
  Speaker 3: you fill in the blank disease. AI is a tool
• 22:55
  Speaker 3: to be used in that and we want to build
• 22:57
  Speaker 3: that kind of trust, and so we have a strategic
• 23:00
  Speaker 3: effort in partnering with those societies to build that educational
• 23:03
  Speaker 3: content out.
• 23:04
  Speaker 2: I feel like there's momentum in the industry to move
• 23:08
  Speaker 2: this way. Everyone's questioning it, but I don't feel resistance
• 23:12
  Speaker 2: to it, like the way you might think, right, there's
• 23:15
  Speaker 2: some fear.
• 23:16
  Speaker 3: Well, there's on both sides. Right, there's fear that it's
• 23:18
  Speaker 3: going to harm us, and then there's fear that if
• 23:20
  Speaker 3: we do it too conservatively, we're going to be left
• 23:23
  Speaker 3: behind and we're going to lose the race, or we're
• 23:25
  Speaker 3: going to miss out on lost opportunities. And I get
• 23:29
  Speaker 3: all that, but I think to your point, Jim, right,
• 23:31
  Speaker 3: if you look at the history of humans inventing tools,
• 23:34
  Speaker 3: if we want to be the author of that next
• 23:36
  Speaker 3: chapter and be the deciders of how these tools are used,
• 23:41
  Speaker 3: we need to come up with that definition of what
• 23:44
  Speaker 3: is the responsible way to use it, what is the
• 23:46
  Speaker 3: way that leads to human flourishing, to use it where
• 23:48
  Speaker 3: these tools are aligned to our values, because if they're not,
• 23:50
  Speaker 3: who's going to use it? And so I get the
• 23:53
  Speaker 3: tension where we want to move fast. I'm a product
• 23:57
  Speaker 3: guy myself, right, I come from Athena, and so I
• 23:59
  Speaker 3: understand and developing these things in as fast as a
• 24:02
  Speaker 3: way as possible. But in health there's that different set
• 24:06
  Speaker 3: of equities of people's lives are really at stake, and
• 24:08
  Speaker 3: so I do believe that principles like transparency and fairness
• 24:14
  Speaker 3: are in the long arc of history for AI are
• 24:17
  Speaker 3: going to be the ones that are going to win out,
• 24:19
  Speaker 3: because if you look at the history of other consequential
• 24:21
  Speaker 3: inventions in human society, we have made a collective decision
• 24:27
  Speaker 3: that we want these things to lead to human flourishing,
• 24:30
  Speaker 3: and we have the power to decide what are those
• 24:33
  Speaker 3: guardrails for fairness or transparency or robustness. So I'm excited
• 24:38
  Speaker 3: to be on that journey.
• 24:39
  Speaker 2: Well, isn't it's set up that way? I think the openness,
• 24:42
  Speaker 2: I mean, that's what all this seems to be. It's
• 24:44
  Speaker 2: like an open shared platform more than other texts that
• 24:49
  Speaker 2: I've seen in the past year. You know, Like so
• 24:51
  Speaker 2: when you think of Epic or even a Thing or
• 24:54
  Speaker 2: any of those things, it wasn't as transparent, And maybe
• 24:58
  Speaker 2: that was one of the issues. But because because there's
• 25:00
  Speaker 2: so many interested parties in this, you would think has
• 25:04
  Speaker 2: to be wide open.
• 25:06
  Speaker 3: Yes, I think you're right, but I don't think that
• 25:08
  Speaker 3: the through line to that next phase in this is
• 25:13
  Speaker 3: as straight as we might hope it to be.
• 25:16
  Speaker 2: Well, nothing ever is right, and you know, the question
• 25:19
  Speaker 2: will be, you know, will AI create greater access and
• 25:22
  Speaker 2: ultimately drive down costs, which was the goal of a
• 25:27
  Speaker 2: lot of these earlier texts, right, but you know now
• 25:30
  Speaker 2: the accessibility of all this to even just a regular
• 25:34
  Speaker 2: patient who has enough interest in it. Because I had
• 25:37
  Speaker 2: a podcast where we were talking about a guy who's
• 25:41
  Speaker 2: created his own cancer therapy, right, and now he's opened
• 25:46
  Speaker 2: that up to challenge others to develop these individualized cancer
• 25:50
  Speaker 2: therapies that you know, this was for a sarcoma And
• 25:53
  Speaker 2: maybe you're aware of this example, but I had Andrew
• 25:56
  Speaker 2: Hesselon and you know he was talking to me about it.
• 25:59
  Speaker 2: He mentioned that example, and I just feel like all
• 26:03
  Speaker 2: this transparency will lead to a lot more access and
• 26:09
  Speaker 2: ultimately drive down the costs absolutely.
• 26:13
  Speaker 3: I mean AI. I think, unlike all of the other
• 26:16
  Speaker 3: digital technology innovations in the health space that have come
• 26:20
  Speaker 3: before it, AI has the power to address the single
• 26:25
  Speaker 3: biggest driver of cost in health, which is the labor
• 26:28
  Speaker 3: cost of physicians and nurses. And simultaneously, to your point,
• 26:33
  Speaker 3: as the opportunity to address the scarcity problem that so
• 26:36
  Speaker 3: many communities struggle with access and if we're able to
• 26:40
  Speaker 3: drive down the cost because we have now agentic tools
• 26:43
  Speaker 3: that can help with shared decision making for diabetes management
• 26:48
  Speaker 3: or hypertension management, or helping patients get their medications. If
• 26:51
  Speaker 3: we're talking about the future with physical AI or robots
• 26:55
  Speaker 3: and at the same time being able to have these
• 26:58
  Speaker 3: technologies deploy across rural communities and the global South, I'm
• 27:04
  Speaker 3: really excited about that future. One of the challenges, Jim,
• 27:07
  Speaker 3: that I honestly don't have the answer to, is how
• 27:09
  Speaker 3: do we upscale all of the nurses and doctors out
• 27:14
  Speaker 3: there that are going to have to be navigating this
• 27:18
  Speaker 3: future collectively? How do we manage these tools when none
• 27:22
  Speaker 3: of us have had that AI governance training that we need.
• 27:25
  Speaker 2: Well, that's why CHA is important and not necessary, and
• 27:29
  Speaker 2: we got to get the word out about it so
• 27:31
  Speaker 2: that academic centers join your coalition. I assume they already are,
• 27:36
  Speaker 2: but yeah, I mean the bottom line, you know, look,
• 27:39
  Speaker 2: patients are already off doing this, right, I know, I
• 27:42
  Speaker 2: do it. I manage getting my own GLP one and
• 27:45
  Speaker 2: whatever it is. It's like, this thing is well on
• 27:49
  Speaker 2: its way and we just want to make sure it's done,
• 27:52
  Speaker 2: and I think the coolest way possible.
• 27:55
  Speaker 3: So agreed. If you ask any of the major technology
• 27:58
  Speaker 3: companies that have one of these AI models, what is
• 28:00
  Speaker 3: the biggest search category for patients like you and me.
• 28:03
  Speaker 3: It's health, every single one of them. Health is the
• 28:06
  Speaker 3: biggest category.
• 28:07
  Speaker 2: Yeah, well, this is the category everybody's been saying it.
• 28:10
  Speaker 2: I don't think there's a more important zone, right, I mean,
• 28:14
  Speaker 2: I guess there's security, But at.
• 28:16
  Speaker 3: A personal level, that's the one that hits home the most. Health.
• 28:19
  Speaker 2: Yeah, you're building a national network of certified independent assurance labs, right,
• 28:24
  Speaker 2: So what are these labs and what will they do?
• 28:27
  Speaker 3: Yeah? Now, this is I appreciate this question, Jim. This
• 28:30
  Speaker 3: is something where a lot of ink has been spilt
• 28:32
  Speaker 3: in the media about this. When we think about assurance,
• 28:36
  Speaker 3: let me just kind of back up for a second,
• 28:38
  Speaker 3: and if I'm a product developer at a technology company,
• 28:41
  Speaker 3: every technology company worth their salt is going to have
• 28:44
  Speaker 3: a quality assurance internal quality assurance process. If you don't,
• 28:48
  Speaker 3: you shouldn't be working at that company or buying what
• 28:50
  Speaker 3: they're selling. So internal qa QC checks are the standard
• 28:55
  Speaker 3: thing that you do when you're developing the technology. And
• 28:58
  Speaker 3: it's no different when you think about AI. And so
• 29:01
  Speaker 3: if we think about AI, and we think about AI
• 29:04
  Speaker 3: and its life cycle, and I'll kind of boil it
• 29:06
  Speaker 3: down to just three basic phases, right, there's the development phase,
• 29:09
  Speaker 3: there's the deployment phase, and there's the maintenance, monitoring or
• 29:14
  Speaker 3: governance phase. So development, deployment, governance. In each one of
• 29:18
  Speaker 3: those three phases, there are quality assurance equities and important
• 29:23
  Speaker 3: things to consider. If I'm in the development phase, it
• 29:25
  Speaker 3: goes to what I just said, right, every company before
• 29:28
  Speaker 3: they commercially sell something wants to have an internal QAQC process.
• 29:32
  Speaker 3: So at HI, what we've developed is a set of partners.
• 29:35
  Speaker 3: You can call them labs, I'm gonna call them partners,
• 29:38
  Speaker 3: partners that have a set of capabilities that are going
• 29:42
  Speaker 3: to enable those technology companies to rapidly train and test
• 29:47
  Speaker 3: their models before they sell it, to make sure that
• 29:49
  Speaker 3: it's up to snuff, that it's actually going to perform well.
• 29:51
  Speaker 3: Every company wants that. If you ask any health tech
• 29:54
  Speaker 3: company with the biggest problem they have, it is access
• 29:57
  Speaker 3: to high quality data. These surance partners for that developer
• 30:02
  Speaker 3: community have access to high quality do Let's look at
• 30:05
  Speaker 3: the deployment phase. You look at the deployment community. Let's
• 30:07
  Speaker 3: talk about health systems. Health systems get inundated by technology
• 30:12
  Speaker 3: companies wanting to sell them. When you ask any health system,
• 30:14
  Speaker 3: they'll probably tell you at least ten a day. And
• 30:17
  Speaker 3: then when they have a procurement process for an AI
• 30:20
  Speaker 3: tool that they're trying to solicit and choose from, they
• 30:22
  Speaker 3: may get hundreds of respondents and they want to be
• 30:25
  Speaker 3: able to separate the wheat from the shaft and kind
• 30:28
  Speaker 3: of down select to whoever the finalists are. And one
• 30:31
  Speaker 3: of the best ways to do that is understand how
• 30:33
  Speaker 3: those models perform and how those models test, and so
• 30:37
  Speaker 3: having a set of independent entities that those health systems
• 30:41
  Speaker 3: can turn to for quality assurance testing is huge. Sometimes
• 30:45
  Speaker 3: that quality is if you're a big health system, you
• 30:47
  Speaker 3: could do that QA testing yourself. If you're a small one,
• 30:49
  Speaker 3: like one of these safety Night clinics, you might turn
• 30:52
  Speaker 3: to NACK itself to do that testing for you. In
• 30:55
  Speaker 3: the governance phase, it comes down to doing that monitoring,
• 30:58
  Speaker 3: that performance monitoring right that I'd mentioned, That is one
• 31:02
  Speaker 3: of I think the most important things for all health
• 31:04
  Speaker 3: systems that they need to be able to stand up
• 31:05
  Speaker 3: those set of technologies, and so we have a set
• 31:07
  Speaker 3: of partners to be able to do that. So that
• 31:10
  Speaker 3: kind of quality assurance service is being offered for the
• 31:13
  Speaker 3: developer community, the deployer community, and in general for the
• 31:16
  Speaker 3: governance monitoring needs that both communities.
• 31:20
  Speaker 2: Have, right and are you helping the supplements industry and
• 31:24
  Speaker 2: the pharma industries too as they do their R and
• 31:28
  Speaker 2: D are you getting tapped into by those organizations.
• 31:31
  Speaker 3: So we're working in the clinical trial space, so agnostic,
• 31:35
  Speaker 3: I would say, we're agnostic to what the specific drug,
• 31:39
  Speaker 3: be it a supplement or a medication that's regulated by
• 31:42
  Speaker 3: the FDA. Is I think we all believe that AI
• 31:46
  Speaker 3: is going to accelerate clinical trials. I mean, there's so
• 31:49
  Speaker 3: many different ways that AI can do, from designing new
• 31:52
  Speaker 3: molecules that no humans ever thought of, to dentifying patients
• 31:56
  Speaker 3: and engaging patients to join a trial that they didn't
• 31:59
  Speaker 3: even know that they qualified for, to be able to
• 32:02
  Speaker 3: do the data ascertainment that I'd mentioned earlier, and kind
• 32:05
  Speaker 3: of teasing out the cause and effect in ways that
• 32:07
  Speaker 3: a statistician can't do without AI. And so we have
• 32:12
  Speaker 3: an effort focused on building out those best practices in
• 32:16
  Speaker 3: industry standards for clinical trials in general. Does it involve pharma, yes,
• 32:22
  Speaker 3: Does it include you know, the supplement industry, yes. Are
• 32:25
  Speaker 3: we focused on one specific use case within that No?
• 32:28
  Speaker 2: No, And there's going to be common threads that they
• 32:31
  Speaker 2: all follow as I was a little more curious about diagnostics,
• 32:36
  Speaker 2: molecular diagnostics and that early detections zone and area, which
• 32:41
  Speaker 2: will also potentially keep people out of the system. And
• 32:45
  Speaker 2: I guess if we can make it cost effective to
• 32:47
  Speaker 2: scan and thraw a blood out of everybody, that should theoretically,
• 32:53
  Speaker 2: again emphasis on prevention versus treatment. You got it.
• 32:57
  Speaker 3: I mean think that's one of the great powers of
• 32:59
  Speaker 3: AI is being able to separate that weak signal from
• 33:03
  Speaker 3: the noise that a overwhelmed primary care provider is going
• 33:07
  Speaker 3: to miss ninety nine times out of one hundred, but
• 33:09
  Speaker 3: the model won't. And so I think there's a series
• 33:13
  Speaker 3: of exciting use cases in that space that we're going
• 33:15
  Speaker 3: to be focusing on as well.
• 33:16
  Speaker 2: Well. I think we're going to have to do a
• 33:17
  Speaker 2: part two or three or four. Thanks so much for this.
• 33:21
  Speaker 2: I like to get to three questions in the end,
• 33:24
  Speaker 2: But before I do that, I always like a good
• 33:26
  Speaker 2: origin story. I mean, how did you end up here?
• 33:29
  Speaker 2: What in your background led you to this point. It's
• 33:33
  Speaker 2: a good question, Jim, because there can't be an easy job, right,
• 33:36
  Speaker 2: So there's that simple point, and you know you've taken
• 33:40
  Speaker 2: the slings and arrows. So why.
• 33:43
  Speaker 3: I grew up in California and in Colorado. So I'm
• 33:45
  Speaker 3: Native American, and I grew up on the Southern New Reservation,
• 33:49
  Speaker 3: which is where all my extended family still lives in
• 33:52
  Speaker 3: southwestern Colorado. And if you know anything about tribal nations,
• 33:55
  Speaker 3: you know that the very rural you oftentimes have challenges
• 33:59
  Speaker 3: with getting access to high quality healthcare. And with that
• 34:04
  Speaker 3: as a personal history, I wanted to be a doctor
• 34:08
  Speaker 3: that could support my community that I lived in, and
• 34:12
  Speaker 3: so for the longest time, I wanted to be a
• 34:14
  Speaker 3: pediatrician that practiced clinically that served my community. And as
• 34:19
  Speaker 3: I went through college and then went to medical school
• 34:21
  Speaker 3: and then started practicing, and holy smokes, look at this
• 34:24
  Speaker 3: digital technology and how broken it is. I got really frustrated,
• 34:29
  Speaker 3: deeply frustrated with the digital tools that I was using clinically.
• 34:33
  Speaker 3: It was a challenging moment. For the sake of time,
• 34:36
  Speaker 3: I'll boil it down to you know, I became so
• 34:38
  Speaker 3: frustrated with the tools that I thought that I could
• 34:40
  Speaker 3: better make an impact if I started trying to solve
• 34:44
  Speaker 3: that problem. So I got into the digital technology digital
• 34:47
  Speaker 3: health space. I started working at as Athena, hit it
• 34:50
  Speaker 3: off with Jonathan and Ed and the team there, and
• 34:54
  Speaker 3: really found a calling in serving the patients that the
• 34:57
  Speaker 3: virtual patients that I had through the providers that I
• 35:00
  Speaker 3: worked with in building better technologies. And I still believe
• 35:03
  Speaker 3: that to the stage that I can have a profound
• 35:06
  Speaker 3: impact in delivering higher quality care to the patients of
• 35:11
  Speaker 3: the United States by leading an effort to build a
• 35:16
  Speaker 3: consensus driven set of definitions about what good responsible AI
• 35:20
  Speaker 3: looks like for our frontline clinicians, for our doctors and nurses,
• 35:23
  Speaker 3: and our patients. It's hard, though, I mean, all the
• 35:25
  Speaker 3: gray hairs that Zoom is airbrushing out that you don't see.
• 35:29
  Speaker 3: There are a lot of arrows being thrown itvy.
• 35:31
  Speaker 2: I mean, look at this, bro, Come on, you're good.
• 35:34
  Speaker 2: You look fine. You got that right? No, it's good.
• 35:37
  Speaker 2: I think it's always good to know the why behind
• 35:40
  Speaker 2: the mission and how you got here, and it's I
• 35:44
  Speaker 2: think good for our audience to know that. And how
• 35:47
  Speaker 2: I think, truly sincere you are about partnering with all
• 35:51
  Speaker 2: parts of the system to really get us to the
• 35:54
  Speaker 2: best place. I you know, I'm highly optimistic about all
• 35:57
  Speaker 2: of it. That's why I'm going to get the word
• 35:59
  Speaker 2: out that there's an urgency to get these things going.
• 36:03
  Speaker 2: And what I tell everyone and get reals, prevent, prevent,
• 36:07
  Speaker 2: focus on the hat. Yeah. Yeah, we're on the same
• 36:10
  Speaker 2: page there. Let's make it happen, all right. Three questions.
• 36:13
  Speaker 2: I like to end with a night I'm ready. What's
• 36:15
  Speaker 2: the best vacation you've ever taken? Oo?
• 36:19
  Speaker 3: Our family's happy places? Barbados. We love going to Barbados.
• 36:23
  Speaker 3: I have a wife and three kids and we go
• 36:26
  Speaker 3: there each February. So I'm here in Boston where it's
• 36:29
  Speaker 3: super cold and we enjoy going during our February break
• 36:32
  Speaker 3: down there to the beach and sun.
• 36:34
  Speaker 2: Yeah. Here, that's awesome. Barbados is what? That's Caribbean? Right, yep, Yepane.
• 36:39
  Speaker 2: So if there was an album that's the soundtrack of
• 36:42
  Speaker 2: your life, what's that album?
• 36:44
  Speaker 3: That is a great question. So I'm gonna I'm gonna
• 36:49
  Speaker 3: get spiritual here for a second. Gym, forgive me for that. So,
• 36:52
  Speaker 3: I mean I am active in my church. I'm a Christian.
• 36:56
  Speaker 3: There is a album or song, Graves to Garden, which
• 37:00
  Speaker 3: tells a story of the Christian faith of death and
• 37:03
  Speaker 3: resurrection and kind of what all that means and that
• 37:06
  Speaker 3: really resonates with me.
• 37:07
  Speaker 2: Awesome, that's great. Music is so important and healing, and
• 37:12
  Speaker 2: it is doing all the things we needed, and certainly
• 37:15
  Speaker 2: in religion, it's one of the things I like the
• 37:17
  Speaker 2: best about religion is the music part.
• 37:20
  Speaker 3: I amen to that.
• 37:22
  Speaker 2: Yeah, what's something you want to get real honest about.
• 37:26
  Speaker 3: We have a crisis of trust about AI. We have
• 37:31
  Speaker 3: so many people talking past one another. Right, We talked
• 37:34
  Speaker 3: about this tension about we want to not lose this
• 37:39
  Speaker 3: race and AI, but we also want to reap the
• 37:42
  Speaker 3: benefits and the promise of AI solving all these maladies
• 37:46
  Speaker 3: and challenges we have in hell. I think you can
• 37:49
  Speaker 3: have both, right, It's not a binary choose one or
• 37:53
  Speaker 3: the other. Right, you can develop guardrails that are thoughtful
• 37:57
  Speaker 3: and appropriate, that build trust and unleash American dynamism and
• 38:03
  Speaker 3: innovation and do both of those things. And I feel
• 38:07
  Speaker 3: like it is not accurate to say that we can
• 38:10
  Speaker 3: only have one or the other. I would love to
• 38:12
  Speaker 3: have a space where we could get real and have
• 38:15
  Speaker 3: honest conversation about what walking and chewing gum looks like
• 38:18
  Speaker 3: at the same time.
• 38:19
  Speaker 2: Well, that's the mission to try, and I think let's
• 38:22
  Speaker 2: get more people on board with it, and I like
• 38:25
  Speaker 2: the consumer reports analogy. You know, the more you can
• 38:29
  Speaker 2: show that value, the better it'll go. I think you're
• 38:33
  Speaker 2: just at the beginning of it, you know. I hear
• 38:35
  Speaker 2: what you're saying, but I think that will even out
• 38:38
  Speaker 2: and smooth out too over time. You're just in the
• 38:42
  Speaker 2: fire right now.
• 38:43
  Speaker 3: I hope you're right, Jim, you're feeling the heat.
• 38:46
  Speaker 2: We are. Yeah, all right, Well, thanks for taking it,
• 38:50
  Speaker 2: and thanks for doing this podcast.
• 38:52
  Speaker 3: Well, it's an honor to be on with you, Jim,
• 38:54
  Speaker 3: and thanks for.
• 38:57
  Speaker 2: Hey everyone, thanks for joining me on the Get Real podcast.
• 39:00
  Speaker 2: You know, I hope you feel a little healthier hanging
• 39:02
  Speaker 2: out with me. You know, it's going to be a
• 39:04
  Speaker 2: lot of fun. Passing this along to others. Please subscribe below,
• 39:09
  Speaker 2: Please share. I'm excited not just to talk about healthcare,
• 39:13
  Speaker 2: but a lot of other fun topics. It's about making
• 39:17
  Speaker 2: healthy talk about healthcare. It is not a lecture. This
• 39:21
  Speaker 2: is supposed to be fun. And look forward to spending
• 39:23
  Speaker 2: more time with you in the future.