January 22, 2018
- Management & Marketing
Episode 4: Artificial Intelligence, Machine Learning and the Health Benefits Industry
And now for something completely indifferent
Episode 4 Transcript
[0:00:15.4] SM: Hello and welcome to another episode of And Now For Something Completely Indifferent, a Canadian health benefits industry podcast brought to you from your friends at Green Shield Canada.
I am Sarah Murphy, and in this podcast in our number of episodes, we’ll be discussing the hottest topics and trends in Canadian health benefits. We’re looking at this podcast as if it’s the podcast that no one asked for, but we’re doing it anyway.
Before we get started with today’s topic, we’d like to remind our listeners that the views expressed in this podcast are those of the individual speaking and not necessarily the views of GSC. We will likely talk about sensitive and maybe even controversial subjects from time to time and we therefore reserve the right to potentially offend, and we are apologizing for it upfront.
You can download this podcast from our website at greenshield.ca/podcast, or you can subscribe to it from wherever you get your podcasts. We also encourage you to read our publications, the inside story and follow the script, which you can also download from our website, and please be sure to follow the conversation on Twitter and Linkedin.
That’s it. Let’s get started with today’s topic, and I will introduce our host for the day, David Willows.
[0:01:33.4] DW: Hi, Sarah.
[0:01:34.4] SM: Hello.
[0:01:35.7] DW: Happy New Year to you.
[0:01:36.9] SM: Same to you.
[0:01:37.9] DW: This is our first podcast of 2018 after three in short succession at the end of 2017. We have crossed over the New Year. It’s still alive. The podcast lives. I know you were away over the break and in very, very cold Florida. That’s an odd sentence to say, isn’t it? But the whole world knows how cold it was in Florida when you were there. What were you doing?
[0:02:02.5] SM: Well, my family and I, we went to Disneyworld. Magical.
[0:02:05.8] DW: You had a 7-day package for Disneyworld.
[0:02:07.5] SM: Mm-hmm.
[0:02:08.4] DW: What was the highest temperature that was achieved in Orlando that week?
[0:02:11.3] SM: Okay. I have to say, I can’t be too dramatic. The first day we had a high of 20, our first afternoon there. That was great. After that, it was about a high of 8.
[0:02:20.0] DW: Okay, and you bought toques in Florida.
[0:02:22.0] SM: Yeah.
[0:02:22.6] DW: Canadians are buying toques in Florida.
[0:02:23.9] SM: We left our 4 toques in our car when we got to the airport, and then as soon as we got to Magic Kingdom, we bought 4 toques in U.S. dollars. I was angry.
[0:02:33.0] DW: That’s a terribly sad story. We have a couple of guests with us today, and there was something you told me about Disney that I want us to bring up later, because today we’re going to talk about the world of artificial intelligence and machine learning, and I think both you and I are curious in whether some of the Disney experiences had an underpinning of some of these ideas. So we are going to test our guests on that.
So we have two guests in the podcast studio today. They are not GSCers, they work at a company called Memotext, and sitting across from me is Amos Adler, and I was going to say your first name is one of the probably most mispronounced names that you can come close.
[0:03:16.8] AA: That was better. That was not bad. Emphasis with me would have been a little better upfront, but it was pretty good.
[0:03:25.7] DW: Okay, because I’m sure you get Amos.
[0:03:27.4] AA: I do.
[0:03:28.2] DW: You get Amos.
[0:03:29.7] AA: I do. Yes.
[0:03:31.3] DW: Your partner in crime here I think has the least mispronounced name. How do you say it?
[0:03:37.4] BS: Well, today it sounds like Batman, but it’s actually Bill.
[0:03:40.3] DW: It is Bill. Bill Simpson.
[0:03:41.5] SM: Bill. Wow!
[0:03:43.0] BS: The most Anglo-Saxon combination of names you ever had.
[0:03:46.4] DW: Yes. We nailed that, I think. Yes, Bill, is a trooper for being in here today. He is doing Christian Bale as Batman. He does not normally sound like that, but I think it adds certain level of malevolence to the podcast or something. You’re going to talk about AI, that’s going to seem kind of even more dangerous than it really is.
[0:04:02.8] BS: It’s almost like we’re out of Terminator here.
[0:04:04.7] DW: Yeah. Okay. Good. Good. Thank you. Even if you’re faking it, thank you. It’s adding a lot to the podcast. I want the listeners to have some context for who these people are, because you’re not normal insurance company type people. I don’t think Sarah and I — Or either, but I know. That’s another story.
[0:04:21.7] SM: Careful.
[0:04:23.8] DW: Give us your bios.
[0:04:24.7] AA: Our bios. I am not Batman, but we all talk like these at Memotext. My bio personally is really sort of a background in bringing together technology process, health data and really try to figure out how it is that we can make you use out of all the awesome tools that are out there now really to help patients, power patients, and at the same time kind of help the people that we work with, i.e., GSC, makes money or at least break even on things like patient engagement and helping people out. That’s sort of where I come from.
I like to call myself, and actually Bill helped me coin this term, a scientician, which is really kind of like a business guy that pretends he’s a scientist at the same time, and —
[0:05:16.4] DW: If you give you any technical answers today, we just should know you’re pretending.
[0:05:19.5] AA: Yes, mostly. Absolutely.
[0:05:22.1] DW: Okay.
[0:05:22.6] AA: Yeah, that’s sort of where I come from, business guy/science guy on TV that has built a career in digital health over the last close to a decade.
[0:05:32.4] DW: Okay. Great. Welcome. Beside you, Batman, you are a science guy, aren’t you? Tell us about your background.
[0:05:39.6] BS: I am. In contrast to Amos’ businetist moniker or — Yeah, scientistian. I call myself a businetist.
[0:05:48.5] DW: Okay. Do you guys do any work in between sort or coming up with quirky new words?
[0:05:52.5] BS: There’s a lot of miscellaneous word definition in Memotext.
[0:05:55.7] AA: We have a lot of memes too. There’s a lot of memes involved in what we do.
[0:05:58.8] BS: Basically what that means is that we can stop laughing at my voice anytime. Basically, that means that I have a science background. I worked in neuroscience for years. Spent nearly 10 years working in mental health, and then I sort of branched out into the business world and took on a more data-centric, data science type of approach.
My background is really in clinical data analysis, management of that, and then getting the most out of it. In psychiatry in particular, which is where I came from, nothing is very specific. Everything is very subjectively oriented and there’s not a lot of objectivity in there. The work that we do at Memotext essentially brings objectivity into sort of clinical interactions. That’s really how I got interested in this whole thing, and then leveraging all of the emerging technology and analytics platforms that are out there.
[0:06:53.0] DW: Okay. That’s very interesting. You’ve worked also on a couple of projects, and I know you’ve at least spent some time with other organizations in our sphere, other carriers, etc. I’m not asking you to obviously pitch your company and stuff like that, but what is it that Memotext does that probably has traditionally been missing from the health-carrier world? What are you able to offer, people like us that may not be readily available? Obviously, I think Bill is the only neuroscientist that’s ever been in our building. That’s an obvious first, but what is it that you’ve seen about our industry where you think you can fill in some gaps?
[0:07:31.3] AA: It sort of aligns with our mission, which is ultimately to make health data useful. So really, in terms of insurance in the carrier world, both in Canada and the U.S., there is this vast store of health data that hasn’t really been leveraged to the extent that it could be used. You have the ability more and more to identify risk through health data, to identify opportunity, but in sort of in a pretty rudimentary kind of way.
We are trying to bring that to the next level to basically be able to use health data. When I say health data, health data can mean everything from actual medical claims data to the fact that you clicked on something to the fact that you are walking 3,000 steps a day, instead of 10,000 steps a day and potentially even stuff like your Facebook usage or the fact that you visited Tim Hortons five times this week.
All those things can be construed as health data and really it’s just a different view of how you can approach the data that insurance in particular has in order to then sort of achieve the goals that I’ve mentioned briefly before, which are how do you power people to be healthier? When I say be healthier, achieve their healthcare goals, which is a very specific individual patient goal. At the same time, try to help the bottom line of the people that we’re working with, which in insurance and in Canada is a particularly challenging fete to try to attempt.
[0:09:08.5] DW: You say that more so than in the states?
[0:09:10.7] AA: From an insurance perspective, yes.
[0:09:12.5] DW: Okay. Tell us about that.
[0:09:14.4] AA: The business of health insurance in the U.S. with regulation to utilization of healthcare services is much more clear, because health plans and insurance companies pay for all health services, unless you’re within Medicare, Medicaid kind of sort of world. Whereas in Canada, there is a certain point where health insurers pass the bill over to the government. That makes the business case vastly different in the two countries.
[0:09:45.3] DW: It means the data is fractured as well. It’s not in one —
[0:09:48.6] AA: It is completely fractured, so the value proposition is very different. Ultimately, the objective that we’re trying to achieve is the same. We’re trying to make people — Tell people live healthier lives through their health data almost despite themselves. We use health data really to understand individuals so that we can engage them in the way that will be most effective for them. That means using health data to drive things like digital patient engagement programs, apps, mobile programs, adaptive content delivery. That’s kind of — Yeah, that’s all over the place there.
[0:10:28.1] DW: I think we’ve experienced that in the work that we’ve done with you and sort of the very interesting ways that you have been able to use our — And let me make this clear, unidentified data.
The other thing that we’ve attached to our work with Memotext is the concept that we’re working with these experts in AI, artificial intelligence and ML, machine learning. For our listeners, I did want to get from sort of the mouths of experts some set of clear but simple definitions of what those things are and maybe how insurance companies can be using these concepts differently going forward.
Sarah, your Disney experience led you to believe that maybe there was some interesting AI happening there. I want you to confront Bill, the neuroscientist with your experience there and see if that qualifies as AI.
[0:11:21.0] SM: Okay. So when you are at Disneyworld, which is as we all know, the most magical place on earth. From certain conversations I’ve had with people who have been to Disney, I have heard that perhaps there’s this whole underworld underneath the property that is constantly monitoring the happiness…
[0:11:40.3] AA: You do know that they bought Star Wars, right?
[0:11:42.4] SM: Right. Exactly.
[0:11:42.4] AA: There is a whole dark side potential —
[0:11:45.9] BS: And a whole billion dollar cash flow every two Christmases that they just magically reinvest into the park.
[0:11:54.6] SM: They’re constantly monitoring the happiness and they’re doing this through crazy technologies and they are monitoring everybody on social media. You have a magic band on your arm that follows you everywhere I go and they know exactly where you are at all times and what you’re thinking and what you’re doing. My take on it is whenever they sense that at a park, that people aren’t happy, maybe the wait times are really long, the weather sucks, that all of a sudden they just read things that are happening and they bring out Mickey Mouse, or in my experience, when my kids were having a really nasty domestic, magically somebody appeared out of nowhere and offered us free ice cream coupons, because they didn’t want our children to be crying, because nobody cries at Disney. That to me says there are some crazy AI going on. They know everything that’s going on at all times. Is that true?
[0:12:37.5] AA: True.
[0:12:37.9] BS: I would love to say that is true. It definitely sounds like it would be AI for sure.
[0:12:44.3] DW: Let me turn this back to the world that we live in, and can you, for all the average folks, including Sarah and me that are in the room today, what is artificial intelligence and how is that different than machine learning?
[0:12:58.3] BS: The simplest definition of artificial intelligence is a computer program that can do something without being given explicit instructions to do so, and machine learning is one of the tools that’s used to create artificial intelligence.
You could have, for example, a facial recognition AI. So when you upload a picture to social media and it recognizes your face, that’s technically AI. You didn’t program the computer to find your face, but what powers that AI is a long sequence of facial recognition algorithms that used machine learning to find the key features to identify your face. Machine learning is used in the development of those technologies and then its deployment in an AI sense, if that makes sense.
[0:13:45.2] DW: Yes.
[0:13:47.0] AA: Machine learning, also sort of a way to understand it is you used to have to write a recipe for something for instructions. Now you can just feed a computer data and it will infer the recipe itself.
[0:14:01.4] DW: Okay. How does this come in to the work that you do? You’ve talked about sort of your very expansive use of data and you’re looking at it in different ways and probably our industry has in the past. Where do these concepts then come in to the work that you do?
[0:14:14.3] BS: I think — I like to sort of call it as data mining advanced, because you had whole departments in these sort of large data rich organizations before where you had people sitting there and trying to pull threads together, trying to pull the connections between different data points. The problem is, is that when you have so much data and you have so many complex things, particularly like human behavior, which is not necessarily rational all the time. You can’t wrap your head around it as an analyst. You just can’t. But what these advanced algorithms can do is they can find those patterns that you wouldn’t be able to find yourself by looking at possible directions by tons of transformations of the data. Sort of the most advance sense, you feed an input and you ask it for an output, and it does anywhere between five and 15 sort of transformations in between. You have no idea what it’s doing in there, and what it’s doing might not have any conceptual relationship to the outcome, but it eventually figures it out, and that’s kind of sort of the most advance case of this.
The way that we’re seeing that used in insurance, in particular in health data, is we’re talking about the determinants of health. Well, genetics and other sort of biological factors account for about 30% of health outcomes, but 70% is these sort of social determinants of health. How motivated are you to do well in managing your condition? How close do you live to your grocery store? How often do you do this and that and the other thing?
We need these advanced algorithms that don’t think in a linear way in order to find those meaningful connections that can help us predict potentially when somebody is going to be deteriorating in their condition or know what things to offer them in order to help incentive them to do better.
[0:16:13.8] AA: I’d like to give more of a businetist — business-like answer.
[0:16:18.9] DW: That’s why you’re here.
[0:16:21.2] AA: Using machine learning to identify populations that are ripe for change, identify populations that will be changing soon. So the possibility to predict when somebody may deteriorate in their health or move from one sort of classification of use of medications to another. That’s sort of using machine learning to identify opportunities within the health data and sort of putting together a puzzle as much as you can from the data that you have. That’s the first piece of it.
The second piece of it is from within the data, identifying the levers or the opportunities to make changes or to affect behavior or affect that — Or mitigate the risk that you might find from within the data itself.
[0:17:12.7] DW: In 2018, do we have examples in the Canadian healthcare scene where this is being done and we’re starting to see some success or some outcomes there or are we still too early?
[0:17:24.1] AA: I would venture to say that some of the work that we’re doing here is probably some of the earliest commercial work that’s been done in this space. There is obviously academic work that’s being done. I personally have spoken to a number of academics that are working on stuff that is similar to what we are doing, but actually without sounding sort of too — What’s the word, cocky? They’re not even close to what we’re trying to achieve using the data that we have access to.
I think there’s a lot of research that’s being done from the perspective of the government, things like — I don’t know if you’ve heard of ICES, not the terrorist guys.
[0:18:04.4] DW: No. That’s a branding problem here on Ontario.
[0:18:06.2] AA: Yes, absolutely.
[0:18:09.2] DW: Not their fault either. They were there first, I think.
[0:18:11.7] AA: They were there first. The work that’s being done in public health data by this group —
[0:18:16.8] DW: You should tell people who that group is, since we’re sort of joking about their name.
[0:18:19.8] AA: I don’t even know what it stands for? Do you know what it stands for?
[0:18:22.7] BS: I can’t recall.
[0:18:23.7] DW: ICES is a provincial body in Ontario.
[0:18:25.8] AA: Yeah, not the bad guys. What? ICES. It could be ICES.
[0:18:35.1] DW: Still too close for comfort.
[0:18:35.7] AA: Still bad branding.
[0:18:36.9] BS: It makes me think of Mr. Freeze in Batman Forever. Not good branding.
[0:18:41.0] AA: There’s definitely a Batman theme, sort of dark side AI theme going on. That’s not to say that AI is necessarily dark. There’s actually — The argument of ethical use of AI and machine learning, and that’s where they sort of in its infancy, but some of the things that I hear about it and feel and see within the world of machine learning and AI is not so much that there’s fear of Skynet coming over and taking over. It’s more about AI and machine learning augmenting our abilities, making a stronger, making a smarter, being able to take advantage of all of these data, because we don’t have — Part of the reason we haven’t been able to do this until now is because we didn’t have the vast amounts of data.
AI has been around for 60 years, machine learning. These are not new things. It’s only now that we have the computer power and the amounts of data that we can actually do this in a cost-effective way.
[0:19:39.6] BS: I do want to chime in there with one other actual clinical practical example of what’s going on. The world of cancer care has actually embraced AI quite a bit. There’s several medical device level technologies now that help pathologists diagnose different types of cancer based on image recognition and things like that, and you can argue that the use of genetic data, analyzing genetic data against the medicines that they used to treat cancer sort of personalize genomics in a way. That’s also sort of an AI big data problem, and that’s been put to use quite effectively in cancer care where they’ll genotype your tumor and then you have a very short list of medications that you know are targeted to that specific genetic makeup of that tumor.
That arm has actually embraced it quite strongly, but as I almost said, there’s a lot of other things that are at the fringes of commercialization that are in university research labs across the country that are just starting to make their way out.
[0:20:43.9] DW: Just to finish off. Maybe an unfair question, but can you, perhaps with your crystal ball, look 10 years ahead and guess at something that may happen in the healthcare system that might sort of freak us out at this point in time that we couldn’t fathom that this could be done?
[0:21:00.8] AA: 10 years from now.
[0:21:01.5] BS: 2027.
[0:21:03.5] DW: Specifically 2027.
[0:21:04.5] AA: I’m going to give my opinion on that while he forms his thought. The thing about it is the leaps in — We all like to complain about healthcare. It’s like our national pastime. We have it so good that we — Anyways, that’s a different story.
[0:21:18.9] DW: This is coming from a person — I know you do a lot of work south of the border, so I know that’s one of your common themes with me.
[0:21:23.4] AA: The leaps that we make in healthcare based on technology, we don’t really notice them, but if you think about cancer care, as Bill is mentioning, five years ago to today, the level of personalization, the level of precision medicine and how that is actually embedded within healthcare in Canada already, is an order of magnitude from where we were 5 years ago, but it’s one of these things you don’t really notice, because, “Oh, it’s not sort of celebrated with fireworks and that kind of thing.”
If you want to talk about sort of cool futuristic kind of things, and people are working on tri-quarters, where they can detect things using devices. I think we’re going to have personally far less human involvement in healthcare.
Right now already, you’re seeing a very, very quick, and in a healthcare quick evolution from a human decision making to machine decision making that receives a human checkmark at the end of it. Decision support, which will be eventually be overtaken ultimately by machine learning, AIs that make decisions, make diagnosis. We’re just a few years away from, I’m pretty sure, melanoma diagnosis being a commercial endeavor from AI without actually involving a dermatologist. Sorry. Dermatologists.
[0:22:49.2] BS: They’re already angry enough.
[0:22:50.9] AA: Yeah, that’s sort of — I see a very hand-off human position in healthcare in the future. That’s for sure.
[0:22:58.8] DW: Okay.
[0:22:59.6] BS: I would wager to say that it’s nice to try and think about all the flashy cool stuff that might be coming down with this technology, especially if you consider all the cool things you can do with the consumer grade, at the consumer level with your smartphone, stuff like this. But calling it back, I once saw a talk from a leading EMR president in the U.S. and he said that there are three things that technology is really good at; administrative automation, resource control and network building, and if you apply those three pillars of technology to healthcare, healthcare is none of them.
We are still sending people to different doctors via faxes. My specialist still doesn’t have my GPs notes and vice-versa. I can’t go to a doctor outside of my GPs personal network of friends that he likes to refer to.
[0:23:50.8] DW: Your friend here is complaining about the Canadian healthcare system.
[0:23:52.7] AA: I know.
[0:23:54.7] BS: But there is a point. What I’m saying is, is that right now we’re reaching a level of sort of digital infiltration of these systems at a level that we’ve never seen before and at a pace that is much faster that it has ever been.
In 10 years, I would love to say that everybody’s electronic health records will be online. You’ll be able to freely move your health information between institutions and the entire care experience from the patient perspective will be 15 times better than it is today, but that could be still be a longshot.
[0:24:27.0] AA: I agree with that. Two things, one, the use of things like blockchain to decentralize health data will be huge. That’s going to be a big driver going forward for portability of data. Then even pulling it back further, things like there’s — I’m not a futurist, but there’s a lot of work that’s happening in basically interchangeable human parts, like stuff that you’ll be able to 3D print, extremely precision medicine. It’s going to be very — Our children’s children will have a very different healthcare experience, and it’s already changing now.
We talk a lot about wellness, and there’s a whole generation of people that are — Wellness will be normal for them. Preventive medicine, like the next generations. That’s going to be part of their existence as supposed to the sick care that we sort of live in today.
[0:25:23.6] DW: Okay. That’s terrific, and that’s half an hour sort of flown by, and I want to give some sympathy to Bill’s voice.
[0:25:31.3] BS: I appreciate that.
[0:25:33.4] DW: I do think, if the Memotext team will agree, we’re going to have to try you back at some point, and I think we’re going to talk specifically at that time about some of the work you’re doing here, and we may have some AI/machine learning outcomes in flight at that moment, but I think you’ve given us sort of a good introductory lesson into what’s happening in our world broadly with some of these concepts and very specifically in the healthcare system. I thank you again, and we’ll talk to you next time.
[0:25:57.9] AA: Thank you.
[0:25:58.9] BS: Thanks, guys.
[0:25:59.1] DW: Okay.
[END OF INTERVIEW]
[0:26:06.4] SM: Thank you to our listeners for tuning in to another edition of And Now For Something Completely Indifferent, a Canadian health benefits podcast. To be sure to get future episodes of this podcast, please subscribe wherever you get your podcasts, or visit our website, greenshield.ca/podcast to download our episodes.
As a reminder, we talk about these issues consistently in our publications, which are available on our website. Specifically for today’s topics, you can check out our December issue of the inside story.
Thank you for listening, and we’ll talk again soon.