1 00:00:04,480 --> 00:00:12,440 Speaker 1: Welcome to tech Stuff, a production from iHeartRadio. Hey there, 2 00:00:12,480 --> 00:00:16,400 Speaker 1: and welcome to tech Stuff. I'm your host Jonathan Strickland. 3 00:00:16,440 --> 00:00:19,919 Speaker 1: I'm an executive producer with iHeart Podcasts and how the 4 00:00:20,000 --> 00:00:22,959 Speaker 1: tech are you? You know. A couple of years ago, 5 00:00:23,520 --> 00:00:28,040 Speaker 1: it seemed like AI had just exploded into the mainstream 6 00:00:28,520 --> 00:00:32,120 Speaker 1: out of nowhere. Now, in reality, thousands of people have 7 00:00:32,200 --> 00:00:36,839 Speaker 1: been working for decades to bring various AI implementations to 8 00:00:36,880 --> 00:00:40,160 Speaker 1: a point where they could be deployed in the real world, 9 00:00:40,440 --> 00:00:43,279 Speaker 1: not just you know, R and D projects. I think 10 00:00:43,320 --> 00:00:47,080 Speaker 1: a lot of us are hyper aware of generative AI 11 00:00:47,159 --> 00:00:50,559 Speaker 1: in particular. That's the kind of application that's easy for 12 00:00:50,600 --> 00:00:54,080 Speaker 1: the average person to have experienced and perceived, you know, 13 00:00:54,200 --> 00:00:57,920 Speaker 1: first hand. But there's a lot more going on in 14 00:00:58,040 --> 00:01:01,960 Speaker 1: AI than just chat by and image generators, that kind 15 00:01:02,000 --> 00:01:07,040 Speaker 1: of thing. There's widespread agreement that AI is definitely going 16 00:01:07,080 --> 00:01:12,000 Speaker 1: to have an enormous impact on pretty much everything. There's 17 00:01:12,080 --> 00:01:16,479 Speaker 1: less agreement about how this transformation is going to manifest 18 00:01:16,800 --> 00:01:20,400 Speaker 1: or in what time frame it's going to happen. We've 19 00:01:20,400 --> 00:01:24,000 Speaker 1: got a lot of companies that have made some fairly 20 00:01:24,520 --> 00:01:28,880 Speaker 1: aggressive moves into the space, either developing AI or trying 21 00:01:28,920 --> 00:01:33,920 Speaker 1: to implement AI solutions, and it hasn't always worked out. 22 00:01:34,400 --> 00:01:38,880 Speaker 1: In the meanwhile, you've got this incredible environment in which entrepreneurs, 23 00:01:39,200 --> 00:01:43,520 Speaker 1: computer scientists, venture capital investors, and more are all trying 24 00:01:43,560 --> 00:01:49,480 Speaker 1: to leverage the moment. Like that whole idea of opportunity knocking. 25 00:01:49,520 --> 00:01:52,360 Speaker 1: You've got to seize on that opportunity because who knows 26 00:01:52,400 --> 00:01:55,120 Speaker 1: when it's going to come around again, Even if they're 27 00:01:55,160 --> 00:02:01,480 Speaker 1: not actually prepared to execute upon that opportunity. Sometimes leveraging 28 00:02:01,520 --> 00:02:05,040 Speaker 1: opportunity just doesn't go the way you want it to. Now, 29 00:02:05,080 --> 00:02:08,880 Speaker 1: I've already published tech Stuff episodes in the past about 30 00:02:08,960 --> 00:02:13,280 Speaker 1: failed tech startups. Some of those startups reached really high 31 00:02:13,360 --> 00:02:18,320 Speaker 1: levels of valuation, like topping a billion dollars, which catapults 32 00:02:18,360 --> 00:02:22,680 Speaker 1: them into the so called unicorn status. A unicorn is 33 00:02:22,760 --> 00:02:25,480 Speaker 1: a startup doesn't have to be tech. It's a startup 34 00:02:25,520 --> 00:02:29,320 Speaker 1: business that hits evaluation of a billion dollars at some 35 00:02:29,480 --> 00:02:34,359 Speaker 1: point or another. But even with that kind of high valuation, 36 00:02:34,800 --> 00:02:38,760 Speaker 1: and even with the initial excitement and support of investors 37 00:02:38,760 --> 00:02:43,800 Speaker 1: who have really deep pockets, startups sometimes collapse for one 38 00:02:43,880 --> 00:02:46,880 Speaker 1: reason or another. So today I thought we should talk 39 00:02:46,919 --> 00:02:51,680 Speaker 1: about a couple of digital health AI startups that have 40 00:02:52,040 --> 00:02:57,280 Speaker 1: followed this particular path. While AI enthusiasm has built to 41 00:02:57,360 --> 00:03:00,600 Speaker 1: what I would call a frenzy, in more recent months 42 00:03:00,880 --> 00:03:05,720 Speaker 1: that enthusiasm has waned somewhat. Investors have kind of backed 43 00:03:05,760 --> 00:03:08,120 Speaker 1: off a little bit from AI. There have been a 44 00:03:08,200 --> 00:03:11,480 Speaker 1: lot of articles that have been published that say any 45 00:03:11,639 --> 00:03:15,520 Speaker 1: real gains from AI are probably years down the road, 46 00:03:15,880 --> 00:03:19,680 Speaker 1: at least for most implementations. This is not necessarily universal 47 00:03:19,720 --> 00:03:22,280 Speaker 1: for all AI, which is an important thing to remember. 48 00:03:22,360 --> 00:03:27,519 Speaker 1: Not all AI is the same, but for many implementations 49 00:03:27,800 --> 00:03:32,240 Speaker 1: it's just too early to think of AI making an 50 00:03:32,400 --> 00:03:38,240 Speaker 1: enormous impact on business objectives. And so again, investments have 51 00:03:38,360 --> 00:03:41,400 Speaker 1: started to taper off quite a bit. People are being 52 00:03:41,520 --> 00:03:46,160 Speaker 1: more particular with their investments for lots of reasons, but 53 00:03:46,240 --> 00:03:48,680 Speaker 1: a big one of those is the perception that AI 54 00:03:48,840 --> 00:03:52,040 Speaker 1: is not really ready for prime time to totally transform 55 00:03:52,080 --> 00:03:56,960 Speaker 1: everything right now. So a lot of AI companies or 56 00:03:57,000 --> 00:04:00,600 Speaker 1: companies that have, you know, kind of leveraged AI to 57 00:04:01,320 --> 00:04:04,040 Speaker 1: be their sales pitch have kind of found themselves in 58 00:04:04,120 --> 00:04:08,200 Speaker 1: dire straits. The condition of dire straits, not the band. 59 00:04:09,080 --> 00:04:12,840 Speaker 1: Dire straits, not the sultans of swing. Now, before I 60 00:04:12,880 --> 00:04:18,080 Speaker 1: get to specifics, let's just establish some general factors, all right. 61 00:04:18,200 --> 00:04:23,520 Speaker 1: So artificial intelligence is an expensive discipline. It is a 62 00:04:23,720 --> 00:04:29,279 Speaker 1: resource hungry computer application. So you have to either own 63 00:04:29,640 --> 00:04:33,920 Speaker 1: or have access to really powerful data centers. For any 64 00:04:34,080 --> 00:04:38,200 Speaker 1: AI application that's going beyond just a proof of concept. 65 00:04:38,240 --> 00:04:41,479 Speaker 1: If you plan on launching something that is meant to 66 00:04:41,480 --> 00:04:44,960 Speaker 1: be a product or service that is in part or 67 00:04:45,080 --> 00:04:48,839 Speaker 1: wholly dependent upon AI, you have to have access to 68 00:04:48,880 --> 00:04:52,560 Speaker 1: that compute power. Now, on top of that, AI applications 69 00:04:52,600 --> 00:04:58,679 Speaker 1: typically require specific types of powerful parallel processing capabilities, which 70 00:04:58,720 --> 00:05:02,000 Speaker 1: means that you need a particular kind of computer chip. 71 00:05:02,080 --> 00:05:04,640 Speaker 1: You can't just get any computer chip, even a really 72 00:05:04,640 --> 00:05:07,760 Speaker 1: fast one. You need one that's really good at handling 73 00:05:07,800 --> 00:05:11,159 Speaker 1: parallel processing. So, like in the early days of AI, 74 00:05:11,320 --> 00:05:15,599 Speaker 1: GPUs were a really big part of AI processing because 75 00:05:15,640 --> 00:05:21,360 Speaker 1: GPUs graphics processing units typically are designed to be parallel processors. 76 00:05:21,360 --> 00:05:24,239 Speaker 1: They have multiple cores, and they can do multi threading 77 00:05:24,240 --> 00:05:27,239 Speaker 1: and work on a lot of different problems all simultaneously. 78 00:05:27,520 --> 00:05:29,919 Speaker 1: AI needs that capability, or at least a lot of 79 00:05:30,000 --> 00:05:35,000 Speaker 1: machine learning processes require that kind of computational power. So 80 00:05:35,080 --> 00:05:38,280 Speaker 1: the chips best suited to do that are not always 81 00:05:38,440 --> 00:05:42,039 Speaker 1: in plentiful supply. So that means we've got this big 82 00:05:42,160 --> 00:05:46,000 Speaker 1: pool of AI companies. Some of them are part of 83 00:05:46,160 --> 00:05:51,000 Speaker 1: much larger organizations like Microsoft or Google, but all of 84 00:05:51,040 --> 00:05:54,479 Speaker 1: them are competing for a limited supply of processors that 85 00:05:54,560 --> 00:05:59,240 Speaker 1: are suitable for handling AI computational loads. Not everyone is 86 00:05:59,279 --> 00:06:01,520 Speaker 1: going to come out of winner in that kind of competition. 87 00:06:01,960 --> 00:06:05,920 Speaker 1: You know, big companies obviously have a huge advantage over 88 00:06:06,120 --> 00:06:10,480 Speaker 1: smaller startups that are dependent upon rounds of fundraising from investors. 89 00:06:10,520 --> 00:06:13,920 Speaker 1: They're going to venture capitalists to get a influx of 90 00:06:14,000 --> 00:06:16,640 Speaker 1: cash in order to be able to do business. Meanwhile, 91 00:06:16,640 --> 00:06:20,560 Speaker 1: you have these monoliths like Microsoft and Google that have 92 00:06:21,279 --> 00:06:26,160 Speaker 1: decades of wealth that have been generated and they lean 93 00:06:26,200 --> 00:06:28,880 Speaker 1: on that in order to get an advantage in the market. 94 00:06:29,279 --> 00:06:33,800 Speaker 1: So that's one thing. Another is that scaling up is 95 00:06:33,960 --> 00:06:38,479 Speaker 1: particularly challenging for AI companies. This is hard for any startup. Right, 96 00:06:38,680 --> 00:06:42,400 Speaker 1: a startup can come up with a brilliant idea and 97 00:06:42,880 --> 00:06:45,839 Speaker 1: have an idea that truly has a good place in 98 00:06:45,880 --> 00:06:48,360 Speaker 1: the market. Once you get to a level of scale 99 00:06:48,560 --> 00:06:52,880 Speaker 1: where you can make this a revenue generating business, but 100 00:06:53,000 --> 00:06:56,200 Speaker 1: getting there is hard. Right, if it's manufacturing, then you 101 00:06:56,240 --> 00:06:57,880 Speaker 1: have to figure out, well, how are we going to 102 00:06:58,480 --> 00:07:02,760 Speaker 1: afford to manufac facture in the bulk we need in 103 00:07:02,839 --> 00:07:05,760 Speaker 1: order to make this a viable business for services, How 104 00:07:05,760 --> 00:07:09,400 Speaker 1: do we make sure that our business can provide the 105 00:07:09,480 --> 00:07:13,200 Speaker 1: services to the customer base we're going to need in 106 00:07:13,320 --> 00:07:17,520 Speaker 1: order to have a viable business. These are non trivial challenges. 107 00:07:18,000 --> 00:07:20,440 Speaker 1: You have to figure out how do you actually meet 108 00:07:20,480 --> 00:07:24,320 Speaker 1: the demand that you're hoping to create. Now, first, there 109 00:07:24,360 --> 00:07:25,920 Speaker 1: has to be a demand there in the first place. 110 00:07:25,960 --> 00:07:28,280 Speaker 1: And if there's no existing demand, you have to create 111 00:07:28,280 --> 00:07:30,480 Speaker 1: that demand. Then you have to be able to deliver 112 00:07:31,360 --> 00:07:35,800 Speaker 1: value for that demand. These are really hard problems, so 113 00:07:35,880 --> 00:07:38,160 Speaker 1: a lot of startups do fail, like whether they're in 114 00:07:38,200 --> 00:07:41,880 Speaker 1: the tech sector or not, and for AI companies it 115 00:07:41,960 --> 00:07:46,640 Speaker 1: is particularly difficult. Scientists might develop a truly intriguing use 116 00:07:46,680 --> 00:07:50,280 Speaker 1: for artificial intelligence. The work's great on a small scale, 117 00:07:50,600 --> 00:07:53,360 Speaker 1: like yeah, they can prove it works for a small 118 00:07:54,000 --> 00:07:58,200 Speaker 1: test group or maybe a region or a very specific industry. 119 00:07:58,400 --> 00:08:00,920 Speaker 1: But then to grow that so that you can meet 120 00:08:00,960 --> 00:08:04,120 Speaker 1: the needs of customers around the world that could require 121 00:08:04,320 --> 00:08:08,120 Speaker 1: way more resources than you can actually afford even as 122 00:08:08,120 --> 00:08:12,120 Speaker 1: a unicorn. So AI startups can end up burning through 123 00:08:12,240 --> 00:08:17,520 Speaker 1: cash really quickly, not through terrible mismanagement, though of course 124 00:08:17,600 --> 00:08:22,160 Speaker 1: that can happen too, but just because HEYI is expensive. 125 00:08:22,520 --> 00:08:24,360 Speaker 1: So I wanted to clear that up at the start 126 00:08:24,400 --> 00:08:26,760 Speaker 1: because I don't want to get the impression that the 127 00:08:26,800 --> 00:08:30,679 Speaker 1: folks that are behind these businesses necessarily did something wrong. 128 00:08:30,840 --> 00:08:34,040 Speaker 1: Although in one case we'll see that there is at 129 00:08:34,120 --> 00:08:38,040 Speaker 1: least one news outlet that very much feels like the 130 00:08:38,160 --> 00:08:41,200 Speaker 1: founder of a company did many things wrong. I don't 131 00:08:41,200 --> 00:08:44,040 Speaker 1: want to say that they were necessarily bad at managing money. 132 00:08:44,640 --> 00:08:47,080 Speaker 1: That could be a factor as well, but I'm trying 133 00:08:47,080 --> 00:08:51,280 Speaker 1: to separate this from the dot com bubble of the 134 00:08:51,360 --> 00:08:55,160 Speaker 1: late nineties. So with the dot com bubble, you had 135 00:08:55,200 --> 00:09:00,280 Speaker 1: all these startups that got enormous investments in cash, and 136 00:09:00,360 --> 00:09:02,400 Speaker 1: you know, some of them went public on the stock 137 00:09:02,440 --> 00:09:06,160 Speaker 1: market and their stocks inflated to ridiculous levels. And you 138 00:09:06,240 --> 00:09:10,400 Speaker 1: had these companies that didn't have fully baked business plans 139 00:09:10,440 --> 00:09:15,000 Speaker 1: in place. They were absolutely swimming in cash, and a 140 00:09:15,000 --> 00:09:19,640 Speaker 1: lot of times people were making extravagant purchases like crazy 141 00:09:19,760 --> 00:09:23,920 Speaker 1: office things like like a full bar or whatever, without 142 00:09:24,080 --> 00:09:27,720 Speaker 1: actually being able to put those assets to work to 143 00:09:27,840 --> 00:09:30,560 Speaker 1: create a business that could stand on its own. And 144 00:09:30,679 --> 00:09:36,400 Speaker 1: ultimately the bubble burst and the entire industry collapsed. You know, 145 00:09:36,520 --> 00:09:39,760 Speaker 1: some companies survived, but a lot of them didn't. Well, 146 00:09:39,760 --> 00:09:43,160 Speaker 1: I want to draw a line between those and what 147 00:09:43,400 --> 00:09:46,360 Speaker 1: might be an AI tech bubble. I think it's fair 148 00:09:46,400 --> 00:09:51,160 Speaker 1: to call it an AI tech bubble because one almost 149 00:09:51,240 --> 00:09:56,000 Speaker 1: universal issue for all the AI startups is this challenge 150 00:09:56,000 --> 00:10:01,800 Speaker 1: that artificial intelligence is inherently expensive. They could also fall 151 00:10:01,920 --> 00:10:05,240 Speaker 1: victim to the same problems that we solve with dot 152 00:10:05,280 --> 00:10:08,719 Speaker 1: com businesses. That's still a possibility. I'm not saying that 153 00:10:08,960 --> 00:10:13,559 Speaker 1: AI companies are somehow immune to human frailties of going 154 00:10:13,640 --> 00:10:17,200 Speaker 1: overboard and like everybody gets a new car or whatever 155 00:10:17,240 --> 00:10:21,080 Speaker 1: it might be, but that the very nature of AI 156 00:10:21,120 --> 00:10:25,520 Speaker 1: itself becomes a massive risk as far as seeing a 157 00:10:25,559 --> 00:10:30,560 Speaker 1: return on investment. So with all that set, we're ready 158 00:10:30,600 --> 00:10:34,840 Speaker 1: to start diving into discussions of a pair of different 159 00:10:35,040 --> 00:10:40,520 Speaker 1: digital health companies that were largely centered around the idea 160 00:10:40,640 --> 00:10:46,239 Speaker 1: of artificial intelligence revolutionizing the way we do certain things 161 00:10:46,240 --> 00:10:50,680 Speaker 1: in healthcare. Whether or not artificial intelligence was actually playing 162 00:10:50,720 --> 00:10:53,440 Speaker 1: a part in that, that's more of an open question. 163 00:10:53,679 --> 00:10:55,960 Speaker 1: We're going to get into that in just a moment. 164 00:10:56,080 --> 00:10:58,840 Speaker 1: Before we do that, let's take a quick break to 165 00:10:58,920 --> 00:11:11,560 Speaker 1: thank our sponsors. Okay, we're back. Let's talk about our 166 00:11:11,679 --> 00:11:17,160 Speaker 1: first digital health company and what happened. And in these cases, 167 00:11:17,200 --> 00:11:19,439 Speaker 1: I think it's safe to say that the AI component 168 00:11:19,679 --> 00:11:23,520 Speaker 1: was really just one contributing factor to how these two 169 00:11:24,240 --> 00:11:27,920 Speaker 1: health company startups failed over time. I think it was 170 00:11:27,960 --> 00:11:32,079 Speaker 1: a major contributing factor, but just one of multiple factors. 171 00:11:32,320 --> 00:11:35,359 Speaker 1: But another is that there was an understandable but arguably 172 00:11:35,480 --> 00:11:41,120 Speaker 1: foolish rush of cash influence in the health space following 173 00:11:41,160 --> 00:11:45,440 Speaker 1: the twenty twenty COVID outbreak. Now that rush of cash 174 00:11:45,640 --> 00:11:50,000 Speaker 1: was understandable because obviously the pandemic had an enormous impact 175 00:11:50,040 --> 00:11:54,040 Speaker 1: on people all around the world. There were tons of regions, 176 00:11:54,280 --> 00:11:59,720 Speaker 1: entire countries that were operating under lockdown conditions, sometimes for 177 00:11:59,800 --> 00:12:03,600 Speaker 1: so several months at a stretch. Now, obviously that would 178 00:12:03,679 --> 00:12:07,160 Speaker 1: change how we do pretty much everything from how we 179 00:12:07,240 --> 00:12:12,640 Speaker 1: work to how students were attending lessons, to how you 180 00:12:12,920 --> 00:12:15,800 Speaker 1: actually got to see a physician if you needed one, 181 00:12:16,120 --> 00:12:19,880 Speaker 1: and so out of necessity, health companies new and old 182 00:12:20,120 --> 00:12:23,880 Speaker 1: attempted to adapt to this new reality. Now, it could 183 00:12:23,920 --> 00:12:28,120 Speaker 1: be really hard for large established companies to adapt to 184 00:12:28,280 --> 00:12:33,040 Speaker 1: rapidly changing conditions. Being nimble isn't exactly a common trait 185 00:12:33,280 --> 00:12:38,640 Speaker 1: for legacy organizations. That opened up opportunities for younger startups 186 00:12:38,640 --> 00:12:41,959 Speaker 1: to innovate in the space and to attempt to serve 187 00:12:42,080 --> 00:12:47,320 Speaker 1: customers in ways that larger organizations simply couldn't replicate. And 188 00:12:47,440 --> 00:12:53,240 Speaker 1: one of those companies was called Babylon. Now, Babylon was 189 00:12:53,320 --> 00:12:56,840 Speaker 1: founded years before the pandemic. It was founded way back 190 00:12:56,880 --> 00:13:00,800 Speaker 1: in twenty thirteen. Technically it's one year year younger than 191 00:13:00,840 --> 00:13:03,679 Speaker 1: the other digital health company we'll talk about in a 192 00:13:03,679 --> 00:13:08,439 Speaker 1: little bit. Babylon launched in the United Kingdom. Ultimately it 193 00:13:08,440 --> 00:13:12,040 Speaker 1: would extend services to other parts of the world, primarily 194 00:13:12,160 --> 00:13:15,000 Speaker 1: in Asia, but also the United States and a couple 195 00:13:15,040 --> 00:13:19,319 Speaker 1: of places in Africa. It was a subscription based healthcare 196 00:13:19,400 --> 00:13:23,120 Speaker 1: services company, and initially it was one that had a 197 00:13:23,160 --> 00:13:27,640 Speaker 1: fairly simple approach. Customers or patients in other words, would 198 00:13:27,640 --> 00:13:33,000 Speaker 1: communicate with healthcare professionals via text messages and video conferences. 199 00:13:33,240 --> 00:13:38,520 Speaker 1: So it was a telehealth solutions company, which again not 200 00:13:38,720 --> 00:13:43,160 Speaker 1: that groundbreaking, right, there were other telehealth solutions out there, 201 00:13:43,559 --> 00:13:46,800 Speaker 1: and at this stage there was no real AI component. 202 00:13:47,240 --> 00:13:51,559 Speaker 1: This was all let's put patients in touch with doctors 203 00:13:52,000 --> 00:13:54,200 Speaker 1: and do it in a way where the patient doesn't 204 00:13:54,240 --> 00:13:56,440 Speaker 1: necessarily have to take time out of his or her 205 00:13:56,640 --> 00:14:00,880 Speaker 1: or their day to go and meet with physician. They 206 00:14:00,880 --> 00:14:04,319 Speaker 1: could do it through this app. So you might say, well, 207 00:14:04,320 --> 00:14:07,240 Speaker 1: where is the AI component if this episode is about 208 00:14:07,320 --> 00:14:10,880 Speaker 1: AI health startups. Well, that took the form of a 209 00:14:11,040 --> 00:14:14,600 Speaker 1: chat bot that was developed later on in Babylon. It 210 00:14:14,640 --> 00:14:18,040 Speaker 1: was an idea that I think was present from the beginning, 211 00:14:18,120 --> 00:14:21,080 Speaker 1: Like this was a goal early on was to develop 212 00:14:21,160 --> 00:14:24,520 Speaker 1: a chat bot that would be able to interact with patients, 213 00:14:24,960 --> 00:14:27,680 Speaker 1: and the chatbot would be able to answer questions, and 214 00:14:27,880 --> 00:14:31,440 Speaker 1: ultimately Babylon claimed it would be able to do things 215 00:14:31,480 --> 00:14:35,960 Speaker 1: like diagnose patients. So not just like answer questions about 216 00:14:36,120 --> 00:14:41,080 Speaker 1: physician availability or simple questions that might lead to a 217 00:14:41,120 --> 00:14:45,440 Speaker 1: way to alleviate symptoms that are acutely bothering a patient, 218 00:14:45,640 --> 00:14:50,400 Speaker 1: but actually diagnosing the underlying cause of those symptoms. That 219 00:14:50,520 --> 00:14:54,480 Speaker 1: is a huge claim, I mean as a remarkable claim, 220 00:14:54,520 --> 00:14:58,360 Speaker 1: and it requires remarkable evidence to support it. When you're 221 00:14:58,400 --> 00:15:02,240 Speaker 1: talking about healthcare, there's a high bar you need to meet, 222 00:15:02,600 --> 00:15:05,960 Speaker 1: a high bar of confidence. If you do not meet that, 223 00:15:06,440 --> 00:15:09,920 Speaker 1: then that means you probably should not not Probably you 224 00:15:10,000 --> 00:15:13,760 Speaker 1: should not offer these services to patients who were talking 225 00:15:13,920 --> 00:15:19,200 Speaker 1: literally matters of life and death. So understandably, a lot 226 00:15:19,240 --> 00:15:22,840 Speaker 1: of critics were raising concerns about how reliable this chatbot 227 00:15:23,040 --> 00:15:26,520 Speaker 1: actually was and whether or not it was ethical to 228 00:15:26,640 --> 00:15:32,480 Speaker 1: even suggest that a chatbot could accurately diagnose someone's ailments 229 00:15:32,560 --> 00:15:35,480 Speaker 1: through interacting with that patient. And again, all of this 230 00:15:35,640 --> 00:15:39,720 Speaker 1: was happening before the pandemic and well before Open Ai 231 00:15:39,880 --> 00:15:43,760 Speaker 1: really opened the floodgates with chat GPT in twenty twenty two, 232 00:15:44,280 --> 00:15:49,520 Speaker 1: So a very aggressive approach toward positioning AI as a 233 00:15:49,560 --> 00:15:53,440 Speaker 1: solution to a complicated problem. Now, in the past, I've 234 00:15:53,480 --> 00:15:58,040 Speaker 1: talked about how AI generated answers can sometimes be incomplete 235 00:15:58,360 --> 00:16:02,680 Speaker 1: or unreliable. That's you using today's AI chatbots, which are 236 00:16:03,320 --> 00:16:06,040 Speaker 1: miles ahead of the stuff that was being developed in 237 00:16:06,120 --> 00:16:08,720 Speaker 1: the twenty tens. So imagine putting your life in the 238 00:16:08,800 --> 00:16:15,680 Speaker 1: virtual hands of a fallible chatbot back in twenty eighteen. Wolf. However, 239 00:16:16,000 --> 00:16:19,680 Speaker 1: according to a great piece that appeared in Wired, it 240 00:16:19,720 --> 00:16:23,360 Speaker 1: was written by Grace Brown. It's titled The Fall of 241 00:16:23,440 --> 00:16:27,680 Speaker 1: Babylon is a Warning for AI Unicorns. Well, according to 242 00:16:27,680 --> 00:16:31,640 Speaker 1: that piece, the AI bit of Babylon might have been 243 00:16:31,640 --> 00:16:33,520 Speaker 1: a stretch in the first place, it might have been 244 00:16:33,520 --> 00:16:38,640 Speaker 1: disingenuous to reference this as artificial intelligence. So Brown cites 245 00:16:38,680 --> 00:16:41,840 Speaker 1: a consulting doctor by the name of Hugh Harvey, who 246 00:16:41,920 --> 00:16:45,560 Speaker 1: at one time worked for Babylon. Now. According to Harvey, 247 00:16:45,840 --> 00:16:49,600 Speaker 1: the AI decision tree that it would follow when interacting 248 00:16:49,640 --> 00:16:53,800 Speaker 1: with patients was essentially an Excel spreadsheet that correlated to 249 00:16:53,880 --> 00:16:57,320 Speaker 1: different parts of the human body. So a patient using 250 00:16:57,440 --> 00:17:01,480 Speaker 1: Babylon's app would indicate, you know, where their symptoms were 251 00:17:01,520 --> 00:17:04,600 Speaker 1: affecting them, like oh, my leg is itching or something 252 00:17:04,640 --> 00:17:07,960 Speaker 1: like that. The app would essentially hone in on possible 253 00:17:08,000 --> 00:17:13,440 Speaker 1: diagnoses by eliminating all the stuff that wasn't a potential 254 00:17:13,520 --> 00:17:17,360 Speaker 1: candidate for an explanation, which is not a very sophisticated 255 00:17:17,480 --> 00:17:22,520 Speaker 1: method of determining what the underlying cause is. And as 256 00:17:22,640 --> 00:17:27,240 Speaker 1: Harvey told Brown, quote, I was like, well, this isn't 257 00:17:27,359 --> 00:17:31,520 Speaker 1: really artificial intelligence, end quote. But whether we should classify 258 00:17:31,560 --> 00:17:35,159 Speaker 1: the inner workings of Babylon as AI or not AI 259 00:17:35,400 --> 00:17:39,280 Speaker 1: was definitely part of the company's messaging to investors. So 260 00:17:39,680 --> 00:17:42,719 Speaker 1: you could say, well, this isn't really artificial intelligence. This 261 00:17:42,800 --> 00:17:46,639 Speaker 1: is a very simplistic decision tree. There's no artificial intelligence 262 00:17:46,680 --> 00:17:49,919 Speaker 1: going on here. There's no decision making, but that's not 263 00:17:50,000 --> 00:17:55,000 Speaker 1: how the company was marketing their capabilities to potential customers. 264 00:17:55,359 --> 00:17:59,040 Speaker 1: Babylon was saying, we use artificial intelligence to help treat 265 00:17:59,119 --> 00:18:02,280 Speaker 1: patients to die noose and treat them. So, whether you 266 00:18:02,320 --> 00:18:04,760 Speaker 1: want to argue that AI was happening or not, the 267 00:18:04,800 --> 00:18:08,080 Speaker 1: company certainly was claiming that to be the case. And 268 00:18:08,160 --> 00:18:11,560 Speaker 1: Babylon initially did pretty well when it came to raising investments. 269 00:18:11,600 --> 00:18:15,520 Speaker 1: So by twenty nineteen, before the pandemic, Babylon had raised 270 00:18:15,520 --> 00:18:19,119 Speaker 1: more than half a billion dollars in funding over the years. 271 00:18:19,359 --> 00:18:23,000 Speaker 1: So remember it was founded in twenty thirteen. By twenty nineteen, 272 00:18:23,200 --> 00:18:26,320 Speaker 1: more than half a billion dollars in various investment rounds. 273 00:18:26,640 --> 00:18:29,679 Speaker 1: In twenty twenty one, the company played the risky maneuver 274 00:18:29,800 --> 00:18:32,920 Speaker 1: of going public through the use of a special purpose 275 00:18:33,000 --> 00:18:39,080 Speaker 1: Acquisition company or SPAC SPAC aka a blank check company. 276 00:18:39,359 --> 00:18:42,879 Speaker 1: Now I have talked about spacks before, but let's have 277 00:18:42,960 --> 00:18:47,760 Speaker 1: a quick refresher. Typically, when a private company is preparing 278 00:18:47,760 --> 00:18:52,440 Speaker 1: to transition into a publicly traded company where the average 279 00:18:52,440 --> 00:18:56,240 Speaker 1: citizen can buy stock in the company, it first has 280 00:18:56,280 --> 00:18:59,240 Speaker 1: to go through an extensive set of steps in order 281 00:18:59,359 --> 00:19:03,159 Speaker 1: to get to the IPO or initial public offering. This 282 00:19:03,280 --> 00:19:06,440 Speaker 1: involves a ton of scrutiny from regulators. Here in the 283 00:19:06,560 --> 00:19:10,840 Speaker 1: United States, it's the Securities and Exchange Commission, or SEC. 284 00:19:11,400 --> 00:19:16,720 Speaker 1: As Kate Ashford wrote in Forbes Advisor quote, going public 285 00:19:17,119 --> 00:19:21,240 Speaker 1: is a challenging, time consuming process that's difficult for most 286 00:19:21,240 --> 00:19:25,400 Speaker 1: companies to navigate alone. A private company planning an IPO 287 00:19:25,600 --> 00:19:29,280 Speaker 1: needs not only to prepare itself for an exponential increase 288 00:19:29,280 --> 00:19:32,280 Speaker 1: in public scrutiny, but it also has to file a 289 00:19:32,440 --> 00:19:36,640 Speaker 1: ton of paperwork and financial disclosures to meet the requirements 290 00:19:36,680 --> 00:19:41,040 Speaker 1: of the Securities and Exchange Commission SEC, which oversees public 291 00:19:41,200 --> 00:19:44,639 Speaker 1: companies end quote. So if a startup is looking to 292 00:19:44,640 --> 00:19:47,760 Speaker 1: get access to a ton of cash through going public 293 00:19:47,880 --> 00:19:51,439 Speaker 1: and it's a bit strapped for time, an alternative to 294 00:19:51,600 --> 00:19:56,480 Speaker 1: the IPO is the SPAC. So with a SPACK you 295 00:19:56,600 --> 00:20:00,199 Speaker 1: have a holding company. So this company doesn't really make 296 00:20:00,359 --> 00:20:04,679 Speaker 1: or do anything. It's kind of like an empty envelope. 297 00:20:04,960 --> 00:20:07,120 Speaker 1: So the one thing it can do is it can 298 00:20:07,160 --> 00:20:10,520 Speaker 1: go through all the regulatory processes required to go public 299 00:20:10,640 --> 00:20:13,520 Speaker 1: and become a publicly traded company. So now you've got 300 00:20:13,560 --> 00:20:16,520 Speaker 1: a publicly traded company that doesn't actually do anything else. 301 00:20:16,840 --> 00:20:20,320 Speaker 1: So using this empty shell of a company, you then 302 00:20:20,480 --> 00:20:24,920 Speaker 1: can acquire a private company that's just itching to go public, 303 00:20:25,240 --> 00:20:28,000 Speaker 1: but it doesn't have the time or the ability to 304 00:20:28,040 --> 00:20:30,840 Speaker 1: do this through the IPO method, or if they did 305 00:20:30,880 --> 00:20:34,679 Speaker 1: do an IPO, the value of stock that would be 306 00:20:34,680 --> 00:20:37,040 Speaker 1: determined through that process would be much lower than what 307 00:20:37,080 --> 00:20:40,600 Speaker 1: they actually want it to be. So your SPAC, your 308 00:20:40,640 --> 00:20:46,640 Speaker 1: SPAC acquires this private startup. Now through the transitive property 309 00:20:46,640 --> 00:20:50,119 Speaker 1: of ownership, that startup is a publicly traded company, or 310 00:20:50,119 --> 00:20:53,399 Speaker 1: at least it's part of a publicly traded company, And 311 00:20:53,440 --> 00:20:55,600 Speaker 1: it's like the startup got a chance to skip all 312 00:20:55,600 --> 00:20:58,440 Speaker 1: that boring paperwork and get straight to the part where 313 00:20:58,440 --> 00:21:02,120 Speaker 1: people throw money at it. However, if it turns out 314 00:21:02,160 --> 00:21:05,119 Speaker 1: the startup doesn't have the ability to succeed in the 315 00:21:05,160 --> 00:21:08,960 Speaker 1: public marketplace, while all of this can then come crashing down. 316 00:21:09,080 --> 00:21:12,480 Speaker 1: Shareholders can lose confidence in the company, They can sell 317 00:21:12,480 --> 00:21:16,320 Speaker 1: off their shares, share prices can fall, that big old 318 00:21:16,400 --> 00:21:19,120 Speaker 1: pile of money can start to shrink, and it's almost 319 00:21:19,160 --> 00:21:21,879 Speaker 1: like skipping all those steps that are intended to make 320 00:21:21,920 --> 00:21:24,600 Speaker 1: sure that companies can make the transition from private to 321 00:21:24,640 --> 00:21:28,080 Speaker 1: public in a sustainable way might be a bad idea. 322 00:21:28,480 --> 00:21:31,320 Speaker 1: That's what happened with Babylon, at least to some extent. 323 00:21:31,640 --> 00:21:36,640 Speaker 1: A SPAC called al Kourie Global Acquisition Corporation acquired Babylon 324 00:21:36,760 --> 00:21:40,040 Speaker 1: in October twenty twenty one. The SPAC, in turn had 325 00:21:40,080 --> 00:21:45,160 Speaker 1: the backing of palanteer Is, Peter Thiel's big data analytics company. 326 00:21:45,480 --> 00:21:48,600 Speaker 1: It was one of many investors that were part of this. 327 00:21:49,080 --> 00:21:54,439 Speaker 1: Babylon's valuation was estimated at four point two billion with 328 00:21:54,560 --> 00:21:58,800 Speaker 1: a B dollars, and it is wild to think that 329 00:21:59,080 --> 00:22:02,879 Speaker 1: just two years later Babylon would get sold off for 330 00:22:03,040 --> 00:22:06,800 Speaker 1: parts as the value of the company had totally collapsed. 331 00:22:06,920 --> 00:22:10,880 Speaker 1: And by collapsed, I mean that eighteen months after being 332 00:22:10,920 --> 00:22:13,919 Speaker 1: listed in the New York Stock Exchange, the stock price 333 00:22:14,119 --> 00:22:17,879 Speaker 1: was ninety nine percent lower than where it had started off. 334 00:22:18,400 --> 00:22:21,640 Speaker 1: How did that happen? I'll explain more, but first let's 335 00:22:21,640 --> 00:22:35,320 Speaker 1: take another quick break. Okay, we have Babylon, a startup 336 00:22:35,359 --> 00:22:39,920 Speaker 1: from twenty thirteen that reaches incredible heights through this reverse 337 00:22:40,000 --> 00:22:44,720 Speaker 1: merger process with a SPACK and is worth or value 338 00:22:44,800 --> 00:22:47,960 Speaker 1: that I should say, four point two billion dollars. Well, 339 00:22:47,960 --> 00:22:51,720 Speaker 1: according to Brown's Piece and Wired, Babylon was actually already 340 00:22:51,760 --> 00:22:54,520 Speaker 1: in trouble. By the time it joined the stock exchange, 341 00:22:54,680 --> 00:22:58,280 Speaker 1: the company was running through cash very quickly in an 342 00:22:58,359 --> 00:23:02,199 Speaker 1: attempt to scale the business, to grow it beyond what 343 00:23:02,320 --> 00:23:05,119 Speaker 1: it already was. Now. As I mentioned at the top, 344 00:23:05,320 --> 00:23:09,560 Speaker 1: AI businesses in particular are really costly to scale, So 345 00:23:09,720 --> 00:23:12,679 Speaker 1: unless you've got really deep pockets, like the pockets of 346 00:23:12,680 --> 00:23:16,479 Speaker 1: one of the big five tech companies Microsoft, Google, Meta, Apple, 347 00:23:16,600 --> 00:23:20,359 Speaker 1: or Amazon, well, you'll likely find challenges in making your 348 00:23:20,400 --> 00:23:23,560 Speaker 1: money last long enough for you to scale properly and 349 00:23:24,000 --> 00:23:28,000 Speaker 1: be self sufficient. Babylon was spending way more money than 350 00:23:28,040 --> 00:23:30,560 Speaker 1: it was bringing in, so it was losing money year 351 00:23:30,600 --> 00:23:33,960 Speaker 1: over year, and as a publicly traded company, Babylon had 352 00:23:34,000 --> 00:23:36,560 Speaker 1: to share this information with the sec. You know, if 353 00:23:36,600 --> 00:23:39,280 Speaker 1: you're a privately held company, you don't have to talk 354 00:23:39,320 --> 00:23:43,200 Speaker 1: about how much money you lost. The public remains uninformed, 355 00:23:43,240 --> 00:23:47,679 Speaker 1: But with publicly traded ones, that information gets filed and 356 00:23:47,720 --> 00:23:51,000 Speaker 1: it becomes available to the public, so you can see 357 00:23:51,040 --> 00:23:55,440 Speaker 1: how much money the company is losing year over year. Well, clearly, 358 00:23:55,560 --> 00:23:59,600 Speaker 1: shareholders lost confidence. The stock price crashed, and just a 359 00:23:59,600 --> 00:24:03,000 Speaker 1: couple of years after having going public with that SPAC transaction, 360 00:24:03,200 --> 00:24:07,480 Speaker 1: Babylon went into administration. In the UK and bankruptcy in 361 00:24:07,520 --> 00:24:10,480 Speaker 1: the US. So administration in the UK is kind of 362 00:24:10,520 --> 00:24:13,800 Speaker 1: similar to bankruptcy here in the United States. They're not identical, 363 00:24:14,040 --> 00:24:18,680 Speaker 1: but they are similar processes. It's meant to try and 364 00:24:19,040 --> 00:24:22,240 Speaker 1: return as much value to investors as possible while a 365 00:24:22,560 --> 00:24:27,800 Speaker 1: business effectively shuts down. So Brown's piece gives more details 366 00:24:27,960 --> 00:24:32,159 Speaker 1: about what was actually going on within Babylon, but in general, 367 00:24:32,400 --> 00:24:34,679 Speaker 1: it was a case of a company spending money it 368 00:24:34,760 --> 00:24:37,560 Speaker 1: had not yet raised in the hopes of hitting that 369 00:24:37,680 --> 00:24:41,359 Speaker 1: sweet spot and delivering upon the company's value proposition. The 370 00:24:41,400 --> 00:24:44,320 Speaker 1: stories of Babylon sound kind of similar to what I 371 00:24:44,440 --> 00:24:48,600 Speaker 1: heard about Aharranos Now Farrannose was that infamous high tech 372 00:24:49,000 --> 00:24:53,639 Speaker 1: health company that absolutely imploded after an expose revealed that 373 00:24:53,680 --> 00:24:57,080 Speaker 1: the company's flagship product, a device that was meant to 374 00:24:57,400 --> 00:25:00,960 Speaker 1: analyze a tiny micro drop of blood and potentially run 375 00:25:01,160 --> 00:25:04,240 Speaker 1: hundreds of different medical tests on It turned out that 376 00:25:04,280 --> 00:25:07,760 Speaker 1: product just did not work as advertised, and in fact, 377 00:25:08,280 --> 00:25:10,639 Speaker 1: it might not ever work at all, at least not 378 00:25:10,800 --> 00:25:13,600 Speaker 1: to the extent that was being promised by the company. 379 00:25:14,040 --> 00:25:15,960 Speaker 1: No matter how much effort was put into it, it 380 00:25:16,000 --> 00:25:19,159 Speaker 1: was going to run up against some fundamental limitations that 381 00:25:19,280 --> 00:25:23,160 Speaker 1: meant it just could not work the way it was envisioned, 382 00:25:23,359 --> 00:25:25,840 Speaker 1: and that the whole company was essentially a house of 383 00:25:25,880 --> 00:25:29,480 Speaker 1: cards built upon this belief that ultimately tech can do 384 00:25:29,600 --> 00:25:32,520 Speaker 1: anything if you just work at it hard enough, and 385 00:25:32,880 --> 00:25:35,800 Speaker 1: it turned out that just wasn't true. Well, it sounds 386 00:25:35,800 --> 00:25:40,080 Speaker 1: like Babylon suffered a kind of similar fate. Now check 387 00:25:40,080 --> 00:25:42,480 Speaker 1: out Grace Brown's article on Wired to read a more 388 00:25:42,520 --> 00:25:45,399 Speaker 1: detailed story about that. But we need to move on 389 00:25:45,600 --> 00:25:48,960 Speaker 1: at this point. So Babylon ultimately goes out of business, 390 00:25:49,240 --> 00:25:53,360 Speaker 1: sells its various business divisions and assets off to other 391 00:25:53,400 --> 00:25:56,240 Speaker 1: companies to return as much value to investors as possible, 392 00:25:56,359 --> 00:25:58,760 Speaker 1: and goes by by Now we're going to talk about 393 00:25:58,800 --> 00:26:03,080 Speaker 1: a different digital health company that had AI aspirations, this 394 00:26:03,119 --> 00:26:07,760 Speaker 1: one called Olive, sometimes called Olive AI. So for this bit, 395 00:26:08,000 --> 00:26:12,040 Speaker 1: I'm referencing a few different articles that I found particularly 396 00:26:12,119 --> 00:26:15,000 Speaker 1: helpful while reading up on the company. One of those 397 00:26:15,400 --> 00:26:19,359 Speaker 1: is an article by Emily Olsen in healthcare drive dot com. 398 00:26:19,400 --> 00:26:21,840 Speaker 1: It was written back in November twenty twenty three. It 399 00:26:21,920 --> 00:26:26,280 Speaker 1: is titled health AI startup Olive to shut down. So 400 00:26:26,400 --> 00:26:29,640 Speaker 1: spoiler alert there, except you know, that's what this episode's 401 00:26:29,640 --> 00:26:32,520 Speaker 1: all about. So maybe not so much a spoiler. I 402 00:26:32,600 --> 00:26:36,199 Speaker 1: also used another article by Giles Bruce. This one was 403 00:26:36,200 --> 00:26:39,520 Speaker 1: for Becker Hospital Review. It was titled the Rise and 404 00:26:39,560 --> 00:26:43,199 Speaker 1: Fall of Olive AI, a timeline that gave some you know, 405 00:26:43,520 --> 00:26:46,800 Speaker 1: simple little moments in time of what was going on 406 00:26:46,920 --> 00:26:49,560 Speaker 1: within Olive. And there were others as well. There was 407 00:26:49,600 --> 00:26:54,040 Speaker 1: an article by Free Press staff of Free Press Columbus 408 00:26:54,119 --> 00:26:58,119 Speaker 1: is in Columbus, Ohio that was very useful and also 409 00:26:58,440 --> 00:27:03,040 Speaker 1: not at all unbiased. Let's I'll talk about so like Babylon, 410 00:27:03,560 --> 00:27:07,439 Speaker 1: Olive actually got its start well before the current AI craze, 411 00:27:07,560 --> 00:27:09,919 Speaker 1: not to mention before the pandemic. It launched back in 412 00:27:09,960 --> 00:27:14,600 Speaker 1: twenty twelve in Columbus, Ohio a guy named Sean Lane, 413 00:27:15,160 --> 00:27:19,200 Speaker 1: whom the Columbus Free Press said, developed quote shadowy and 414 00:27:19,280 --> 00:27:23,159 Speaker 1: shady AI software which promised to cut administrative costs for 415 00:27:23,240 --> 00:27:27,080 Speaker 1: healthcare providers in quote, led Olive AI for a little 416 00:27:27,080 --> 00:27:30,679 Speaker 1: more than a decade before the company totally collapsed. The 417 00:27:30,720 --> 00:27:33,399 Speaker 1: Free Press has a lot of things to say about 418 00:27:33,400 --> 00:27:38,359 Speaker 1: Sean Lane, and they are pretty darn critical. They pull 419 00:27:38,520 --> 00:27:42,399 Speaker 1: no punches in their take. For example, that piece points 420 00:27:42,400 --> 00:27:46,040 Speaker 1: out that Shawn Lane incorporated a new company on the 421 00:27:46,160 --> 00:27:48,640 Speaker 1: very same day that olive Ai announced it was going 422 00:27:48,680 --> 00:27:51,640 Speaker 1: on a business So they said, well, that doesn't sit 423 00:27:51,720 --> 00:27:54,240 Speaker 1: well with us. Like your company that you led for 424 00:27:54,280 --> 00:27:58,440 Speaker 1: more than a decade spectacularly fails, and on that same 425 00:27:58,520 --> 00:28:01,520 Speaker 1: day that it shuts down, you announce or not announced, 426 00:28:01,560 --> 00:28:05,560 Speaker 1: but you incorporate a new company. That seems kind of questionable. 427 00:28:05,960 --> 00:28:07,960 Speaker 1: That's what the Free Press was saying. So if you 428 00:28:08,000 --> 00:28:10,840 Speaker 1: want to read some serious shade directed at Lane and 429 00:28:10,920 --> 00:28:14,240 Speaker 1: olive Ai, check out the article in Free Press Columbus. 430 00:28:14,359 --> 00:28:18,720 Speaker 1: It's titled out of control venture capitalysts throw more millions 431 00:28:18,760 --> 00:28:22,760 Speaker 1: at disgraced Columbus CEO. But again, note that there might 432 00:28:22,800 --> 00:28:25,320 Speaker 1: be a teenc bit of bias in that reporting. I'm 433 00:28:25,320 --> 00:28:28,960 Speaker 1: not saying it's misplaced bias, but it's there, all right. 434 00:28:29,080 --> 00:28:32,600 Speaker 1: So oli Ai, let's talk about what the company's sales 435 00:28:32,680 --> 00:28:35,399 Speaker 1: pitch was. So this was a B to B kind 436 00:28:35,480 --> 00:28:40,080 Speaker 1: of company, meaning it would count other businesses as its customers. 437 00:28:40,240 --> 00:28:43,160 Speaker 1: It's a business to business company. It didn't interface with 438 00:28:43,240 --> 00:28:46,680 Speaker 1: private citizens or anything like that. And the company's main 439 00:28:46,760 --> 00:28:49,640 Speaker 1: product was a software package that was meant to help 440 00:28:49,800 --> 00:28:54,680 Speaker 1: healthcare companies automate certain processes such as keeping tabs on 441 00:28:55,160 --> 00:28:58,520 Speaker 1: patients insurance coverage, making sure that you know their insurance 442 00:28:58,560 --> 00:29:03,560 Speaker 1: is still active that thing, or processing authorization requests through 443 00:29:03,800 --> 00:29:07,360 Speaker 1: an automated system. So essentially, the idea was to streamline 444 00:29:07,400 --> 00:29:12,040 Speaker 1: the numerous and repetitive tasks that are involved in healthcare administration. 445 00:29:12,440 --> 00:29:14,520 Speaker 1: And this was a pitch that a lot of investors 446 00:29:14,520 --> 00:29:17,600 Speaker 1: loved because it suggested that healthcare companies would be able 447 00:29:17,680 --> 00:29:22,720 Speaker 1: to significantly decrease their costs and increase their efficiency while 448 00:29:22,760 --> 00:29:26,720 Speaker 1: passing savings on to customers. Oh no, wait, sorry, no, 449 00:29:26,880 --> 00:29:30,360 Speaker 1: I forget that last part. I actually meant while generating 450 00:29:30,440 --> 00:29:35,480 Speaker 1: massive profits that mean huge shareholder returns, customers the patients, 451 00:29:35,480 --> 00:29:38,120 Speaker 1: they would still see the same costs because, after all, 452 00:29:38,160 --> 00:29:41,520 Speaker 1: here in the United States, it's usually an insurance company 453 00:29:41,520 --> 00:29:44,800 Speaker 1: that's actually paying up. No one cares if an insurance 454 00:29:44,840 --> 00:29:47,479 Speaker 1: company has to pay the same amount for services that 455 00:29:47,600 --> 00:29:51,760 Speaker 1: are actually costing less because the hospital or other healthcare 456 00:29:52,000 --> 00:29:55,200 Speaker 1: service provider has found a more efficient way of doing business. 457 00:29:55,320 --> 00:29:57,520 Speaker 1: They don't care if the insurance company is still paying 458 00:29:57,520 --> 00:30:01,840 Speaker 1: the same amount, even if the services themselves technically cost less. 459 00:30:02,120 --> 00:30:07,440 Speaker 1: Of course, insurance companies might care, and then therefore insurance 460 00:30:07,480 --> 00:30:10,720 Speaker 1: customers are going to care because ultimately those providers are 461 00:30:10,720 --> 00:30:15,520 Speaker 1: going to pass those costs down to the insurance customers 462 00:30:15,800 --> 00:30:18,000 Speaker 1: and they're gonna the customers are going to see higher 463 00:30:18,000 --> 00:30:21,040 Speaker 1: deductibles and higher premiums that kind of thing. But never 464 00:30:21,120 --> 00:30:24,240 Speaker 1: mind all that, that doesn't matter to investors, right. So 465 00:30:24,320 --> 00:30:29,080 Speaker 1: the earliest version of ol of debuted back in twenty seventeen, 466 00:30:29,160 --> 00:30:30,959 Speaker 1: so this is like five years after the company has 467 00:30:30,960 --> 00:30:35,280 Speaker 1: been founded, and the company enjoyed support from investors throughout 468 00:30:35,320 --> 00:30:39,240 Speaker 1: its early years. But like Babylon and countless other digital 469 00:30:39,280 --> 00:30:42,960 Speaker 1: health companies, it was the pandemic that would send the 470 00:30:42,960 --> 00:30:46,160 Speaker 1: company's fortunes to the moon. That's when investors were just 471 00:30:46,360 --> 00:30:49,680 Speaker 1: pouring huge amounts of money into these digital health companies. 472 00:30:49,800 --> 00:30:53,240 Speaker 1: So in twenty twenty, all of Ai raised nearly a 473 00:30:53,600 --> 00:30:58,040 Speaker 1: billion dollars in funding. That's just in one year, and 474 00:30:58,080 --> 00:31:00,400 Speaker 1: this is after the company had been incorporated for nearly 475 00:31:00,480 --> 00:31:04,560 Speaker 1: a decade. So in twenty twenty one, Olive acquired another 476 00:31:04,720 --> 00:31:09,160 Speaker 1: AI focused healthcare company called Empiric Health, which itself was 477 00:31:09,200 --> 00:31:12,560 Speaker 1: a spinoff from yet another healthcare company called inter Mountain 478 00:31:12,600 --> 00:31:16,000 Speaker 1: Health out of Salt Lake City, Utah. Stuff gets really complicated, 479 00:31:16,040 --> 00:31:20,320 Speaker 1: not just from a technical perspective, So empiric Health focused 480 00:31:20,360 --> 00:31:25,200 Speaker 1: on clinical analytics and used artificial intelligence to identify potential 481 00:31:25,240 --> 00:31:30,080 Speaker 1: irregularities in clinical procedures. So essentially, the tool was meant 482 00:31:30,120 --> 00:31:35,080 Speaker 1: to isolate instances of unwanted clinical variation so that healthcare 483 00:31:35,120 --> 00:31:38,760 Speaker 1: companies could address any problems early on before they become 484 00:31:38,800 --> 00:31:42,880 Speaker 1: bigger issues. By the summer of twenty twenty two, Olive 485 00:31:43,200 --> 00:31:47,360 Speaker 1: AI was in a totally different financial position because the 486 00:31:47,400 --> 00:31:52,360 Speaker 1: economy was no longer booming. Olive had potentially over extended itself, 487 00:31:52,600 --> 00:31:55,040 Speaker 1: so the company did what countless others did in the 488 00:31:55,040 --> 00:31:58,840 Speaker 1: summer of twenty twenty two, it held extensive layoffs, so 489 00:31:58,920 --> 00:32:02,440 Speaker 1: around four hundred fifs if the employees at Olive were 490 00:32:02,520 --> 00:32:06,400 Speaker 1: let go. Things however, did not improve, and so like Babylon, 491 00:32:06,800 --> 00:32:10,320 Speaker 1: Olive ultimately would begin selling off components of its own 492 00:32:10,400 --> 00:32:13,040 Speaker 1: business to other companies it was so it was essentially 493 00:32:13,080 --> 00:32:15,920 Speaker 1: getting broken down for parts, and this might be one 494 00:32:15,920 --> 00:32:18,680 Speaker 1: of the reasons that the Free Press of Columbus is 495 00:32:18,760 --> 00:32:23,800 Speaker 1: so critical of CEO Sean Lane, because the layoffs affected 496 00:32:23,880 --> 00:32:27,080 Speaker 1: many people in Ohio, and a lot of people likely 497 00:32:27,120 --> 00:32:30,680 Speaker 1: felt that leaders like Sean Lane were exploiting the products 498 00:32:30,680 --> 00:32:33,840 Speaker 1: of labor so that they and other investors could hit 499 00:32:33,880 --> 00:32:37,000 Speaker 1: the eject button while avoiding the worst of the consequences. 500 00:32:37,120 --> 00:32:39,200 Speaker 1: You know, if you actually play your cards right, you 501 00:32:39,240 --> 00:32:40,880 Speaker 1: could end up better off than you did when you 502 00:32:40,920 --> 00:32:43,160 Speaker 1: started the whole thing. And sure, a whole bunch of 503 00:32:43,200 --> 00:32:45,960 Speaker 1: employees and former customers might not be able to say 504 00:32:46,040 --> 00:32:48,760 Speaker 1: the same, but you got yours. Gush darn it. At 505 00:32:48,840 --> 00:32:50,800 Speaker 1: least that's the feeling I get when reading the Free 506 00:32:50,840 --> 00:32:53,960 Speaker 1: Press article, which I could be projecting here. I'm sure 507 00:32:54,040 --> 00:32:57,000 Speaker 1: the truth of the matter is far less cynical than that. 508 00:32:57,400 --> 00:33:01,440 Speaker 1: By how much I don't know, But like Babylon, critics, 509 00:33:01,440 --> 00:33:05,280 Speaker 1: including former Olive employees, argued that a lot of the 510 00:33:05,320 --> 00:33:10,320 Speaker 1: AI powered components weren't really true AI when you got 511 00:33:10,360 --> 00:33:15,600 Speaker 1: down to it, or they were extremely simplistic automated processes that, 512 00:33:15,680 --> 00:33:19,400 Speaker 1: depending upon your perspective, don't actually meet the threshold to 513 00:33:19,520 --> 00:33:22,720 Speaker 1: be called artificial intelligence. Now, I would say that's a 514 00:33:22,760 --> 00:33:28,560 Speaker 1: slippery slope, because defining artificial intelligence is deceivingly difficult. Heck, 515 00:33:28,640 --> 00:33:32,640 Speaker 1: for that matter, defining human intelligence is actually really tricky. 516 00:33:33,040 --> 00:33:37,560 Speaker 1: So is an automated algorithm artificial intelligence? And if not, 517 00:33:37,920 --> 00:33:40,960 Speaker 1: how complicated does the system need to be in order 518 00:33:40,960 --> 00:33:44,480 Speaker 1: to qualify as AI does there need to be some 519 00:33:44,520 --> 00:33:47,880 Speaker 1: sort of decision making component to it in order to 520 00:33:47,920 --> 00:33:51,320 Speaker 1: be AI. I don't actually have the answers to these questions. 521 00:33:51,360 --> 00:33:54,080 Speaker 1: You know, what's AI to one person might not be 522 00:33:54,240 --> 00:33:57,080 Speaker 1: AI to someone else, Which is kind of like the 523 00:33:57,360 --> 00:34:00,920 Speaker 1: legal definition of pornography in someplace is where it said 524 00:34:01,440 --> 00:34:02,920 Speaker 1: I can't tell you what it is, but I know 525 00:34:02,960 --> 00:34:04,920 Speaker 1: it when I see it. It's kind of that similar 526 00:34:04,960 --> 00:34:09,520 Speaker 1: situation anyway. With Oli, the problem was that once the 527 00:34:09,560 --> 00:34:13,280 Speaker 1: post pandemic boom had settled, the company was facing high 528 00:34:13,280 --> 00:34:17,320 Speaker 1: costs of business and revenue just wasn't keeping up. Hiring 529 00:34:17,400 --> 00:34:20,480 Speaker 1: freezes and layoffs in twenty twenty two were followed by 530 00:34:20,520 --> 00:34:23,960 Speaker 1: some high profile departures from the company. The chief financial 531 00:34:23,960 --> 00:34:26,879 Speaker 1: officer and the chief product officer both left by the 532 00:34:26,920 --> 00:34:29,960 Speaker 1: fall of twenty twenty two. Olive also saw its client 533 00:34:30,040 --> 00:34:33,840 Speaker 1: base Diminish providers began to shop around to some of 534 00:34:33,960 --> 00:34:38,200 Speaker 1: Olive's competition, so the company began to lose customers and 535 00:34:38,320 --> 00:34:42,040 Speaker 1: the ending was not yet set in stone. As late 536 00:34:42,120 --> 00:34:46,160 Speaker 1: as March twenty twenty three, Olive continued to raise hundreds 537 00:34:46,320 --> 00:34:51,000 Speaker 1: of millions of dollars collectively, the company raised more venture 538 00:34:51,080 --> 00:34:56,160 Speaker 1: capital funding than any other health tech startup in history, 539 00:34:56,480 --> 00:34:58,800 Speaker 1: but that was not enough to make the business model 540 00:34:58,880 --> 00:35:02,359 Speaker 1: actually work, so Olive sold off different parts of its 541 00:35:02,400 --> 00:35:06,160 Speaker 1: business to various companies, and also faced a lawsuit from 542 00:35:06,160 --> 00:35:11,120 Speaker 1: Ohio's state Economic Development Department because the company had failed 543 00:35:11,120 --> 00:35:13,480 Speaker 1: to live up to an obligation it had made in 544 00:35:13,560 --> 00:35:16,440 Speaker 1: order to provide a certain number of jobs in return 545 00:35:16,520 --> 00:35:20,640 Speaker 1: for the considerable tax incentives that it had enjoyed. So 546 00:35:20,760 --> 00:35:26,120 Speaker 1: on Halloween twenty twenty three, all of AI shut down. Now, 547 00:35:26,160 --> 00:35:29,440 Speaker 1: these are just two examples of Heck, it's just two 548 00:35:29,480 --> 00:35:33,440 Speaker 1: examples of digital health companies with AI components to it 549 00:35:33,719 --> 00:35:39,560 Speaker 1: that shut down despite the huge boom and AI investment. 550 00:35:39,920 --> 00:35:43,520 Speaker 1: If we extend that to AI startups in general, there 551 00:35:43,600 --> 00:35:47,959 Speaker 1: are tons of examples of AI startups that have had 552 00:35:48,000 --> 00:35:51,480 Speaker 1: to shut down over the last year or so. And again, 553 00:35:51,880 --> 00:35:56,120 Speaker 1: that's not necessarily an indication that the business itself was 554 00:35:56,160 --> 00:35:59,160 Speaker 1: a bad idea, or that the service or product they 555 00:35:59,200 --> 00:36:03,279 Speaker 1: planned to provide just had no place. That might not 556 00:36:03,400 --> 00:36:08,880 Speaker 1: be the case. AI is inherently a difficult discipline to 557 00:36:08,920 --> 00:36:12,200 Speaker 1: get into and make it work from a business perspective, 558 00:36:12,400 --> 00:36:14,520 Speaker 1: It needs to work really well. It needs to be 559 00:36:14,560 --> 00:36:18,319 Speaker 1: dependable and replicable, Like you need to make sure you 560 00:36:18,320 --> 00:36:21,479 Speaker 1: can rely on the results and that if you ask 561 00:36:21,600 --> 00:36:23,560 Speaker 1: the thing twenty times, you're going to get the right 562 00:36:23,600 --> 00:36:27,000 Speaker 1: answer all twenty times. That's hard to do from a 563 00:36:27,040 --> 00:36:30,680 Speaker 1: technical level. But also, as I mentioned multiple times, just 564 00:36:30,760 --> 00:36:36,080 Speaker 1: the expense of running an AI centric business is so 565 00:36:36,400 --> 00:36:40,399 Speaker 1: high that in order to make enough money to cover 566 00:36:40,480 --> 00:36:43,400 Speaker 1: all the costs of operation and then make profit on 567 00:36:43,440 --> 00:36:46,440 Speaker 1: top of that, it's really hard. You either have to 568 00:36:46,480 --> 00:36:49,520 Speaker 1: scale up super fast so that you're able to meet 569 00:36:49,840 --> 00:36:53,759 Speaker 1: an enormous number of customers around the world, or you 570 00:36:53,840 --> 00:36:55,680 Speaker 1: have to price yourself at a level where you're going 571 00:36:55,760 --> 00:36:57,960 Speaker 1: to see a return, but then you run the risk 572 00:36:58,000 --> 00:37:00,520 Speaker 1: of no one buying your product or service because it's 573 00:37:00,600 --> 00:37:03,640 Speaker 1: way too expensive. Yeah, it might be AI powered, but 574 00:37:03,719 --> 00:37:06,680 Speaker 1: why do I want to spend ten times more than 575 00:37:06,760 --> 00:37:09,200 Speaker 1: I would if I go with a human powered company. 576 00:37:09,320 --> 00:37:12,040 Speaker 1: It's going to get me reliable results, it just won't 577 00:37:12,080 --> 00:37:15,480 Speaker 1: be AI driven results. So yeah, we're still in this 578 00:37:15,560 --> 00:37:21,000 Speaker 1: world where AI it's got incredible potential, Like I can't 579 00:37:21,080 --> 00:37:26,440 Speaker 1: even begin to imagine the potential AI has to transform 580 00:37:26,719 --> 00:37:30,799 Speaker 1: how we do everything. If it's applied properly. But the 581 00:37:30,920 --> 00:37:34,600 Speaker 1: challenges of getting there are considerable, and they're not going 582 00:37:34,640 --> 00:37:38,600 Speaker 1: to be solved overnight. And it doesn't matter how flashy 583 00:37:38,719 --> 00:37:42,360 Speaker 1: an AI company is or how excited investors are to 584 00:37:42,920 --> 00:37:47,000 Speaker 1: try and get in on that particular gold rush. It's 585 00:37:47,160 --> 00:37:50,359 Speaker 1: not going to make the AI powered future get here 586 00:37:50,480 --> 00:37:55,520 Speaker 1: any quicker. It might actually slow things down. So, as always, 587 00:37:55,760 --> 00:38:00,840 Speaker 1: I recommend employing critical thinking whenever you encounter anything, honestly, 588 00:38:00,920 --> 00:38:05,400 Speaker 1: but particularly when you encounter information or news about artificial intelligence. 589 00:38:05,760 --> 00:38:09,080 Speaker 1: Use critical thinking because again, I do believe there are 590 00:38:09,160 --> 00:38:12,680 Speaker 1: ways where AI is going to make a positive difference 591 00:38:12,800 --> 00:38:17,320 Speaker 1: in how we go about doing different tasks. But slapping 592 00:38:17,360 --> 00:38:21,120 Speaker 1: AI onto something does not automatically make it better, just 593 00:38:21,160 --> 00:38:24,000 Speaker 1: as I would tell the hosts of the podcast The 594 00:38:24,000 --> 00:38:27,760 Speaker 1: Besties that throwing the adjective super in a video games 595 00:38:27,800 --> 00:38:32,040 Speaker 1: title does not automatically make it better. That's just a 596 00:38:32,120 --> 00:38:36,440 Speaker 1: general joke at The Besties. I listened to an episode 597 00:38:36,719 --> 00:38:40,359 Speaker 1: recently where they were jokeingally suggesting that if you have 598 00:38:40,440 --> 00:38:42,359 Speaker 1: super in the title, it must mean that the game 599 00:38:42,440 --> 00:38:44,840 Speaker 1: is better. So great show. By the way, I have 600 00:38:44,920 --> 00:38:47,440 Speaker 1: no connection to the Besties. I don't even know any 601 00:38:47,480 --> 00:38:50,080 Speaker 1: of the people who are the hosts of that show. 602 00:38:50,239 --> 00:38:52,960 Speaker 1: But if you like video game discussions, you should definitely 603 00:38:53,040 --> 00:38:55,640 Speaker 1: check it out. That's just a free plug from yours truly, 604 00:38:55,960 --> 00:38:58,440 Speaker 1: and again I have no connection to them. They're not 605 00:38:58,520 --> 00:39:01,279 Speaker 1: an iHeart podcast than like that. It's just a show 606 00:39:01,280 --> 00:39:04,440 Speaker 1: I enjoy. That's it for this episode. I'll probably do 607 00:39:04,520 --> 00:39:07,400 Speaker 1: more episodes about AI startups and kind of talk about 608 00:39:07,680 --> 00:39:09,839 Speaker 1: the challenges they face, because I really do think there 609 00:39:09,840 --> 00:39:13,440 Speaker 1: are some startups out there, including in the digital health space, 610 00:39:13,800 --> 00:39:18,200 Speaker 1: that are trying to do really interesting, important work. But 611 00:39:18,320 --> 00:39:21,640 Speaker 1: in many cases, I think the folks who perhaps are 612 00:39:22,120 --> 00:39:26,279 Speaker 1: the leaders behind those companies may not have a full 613 00:39:26,400 --> 00:39:29,279 Speaker 1: understanding or appreciation of how hard it's going to be, 614 00:39:29,640 --> 00:39:33,440 Speaker 1: and that ends up falling on the actual experts in 615 00:39:33,480 --> 00:39:37,680 Speaker 1: the field, the computer scientists, etc. To try and realize 616 00:39:37,719 --> 00:39:44,640 Speaker 1: a vision that is inherently extremely difficult to accomplish, not impossible, necessarily, 617 00:39:45,000 --> 00:39:48,759 Speaker 1: but very challenging. So I'll probably do some more of 618 00:39:48,800 --> 00:39:51,120 Speaker 1: these in the future, not so that I could just say, Haha, 619 00:39:51,200 --> 00:39:53,640 Speaker 1: look at these companies that didn't make it, but to 620 00:39:53,960 --> 00:39:57,600 Speaker 1: get a deeper understanding of why didn't they make it, 621 00:39:57,760 --> 00:40:00,239 Speaker 1: or for the ones that do make it, what's set 622 00:40:00,280 --> 00:40:03,040 Speaker 1: them apart, because I think there's some valuable lessons to 623 00:40:03,080 --> 00:40:05,920 Speaker 1: be learned there. In the meantime, I hope all of 624 00:40:05,960 --> 00:40:08,799 Speaker 1: you out there are doing well, and I will talk 625 00:40:08,840 --> 00:40:19,240 Speaker 1: to you again really soon. Tech Stuff is an iHeartRadio production. 626 00:40:19,520 --> 00:40:24,560 Speaker 1: For more podcasts from iHeartRadio, visit the iHeartRadio app, Apple Podcasts, 627 00:40:24,680 --> 00:40:30,240 Speaker 1: or wherever you listen to your favorite shows.