1 00:00:00,000 --> 00:00:01,440 Hey everyone and welcome back. 2 00:00:01,440 --> 00:00:04,200 Today we're going to be looking at AI doctors. 3 00:00:04,200 --> 00:00:05,400 AI doctors. 4 00:00:05,400 --> 00:00:06,240 Well, not really doctors, 5 00:00:06,240 --> 00:00:08,600 but AI helping doctors with diagnoses. 6 00:00:08,600 --> 00:00:09,440 Right. 7 00:00:09,440 --> 00:00:10,980 We found this super interesting research paper 8 00:00:10,980 --> 00:00:14,040 from Jaminetwork open all about it. 9 00:00:14,040 --> 00:00:15,640 Yeah, it's a really cool study. 10 00:00:15,640 --> 00:00:18,000 They actually use chat GPT plus. 11 00:00:18,000 --> 00:00:19,680 Oh wow, the GPT-4 version. 12 00:00:19,680 --> 00:00:23,600 Yeah, GPT-4 to see if it could help doctors, 13 00:00:23,600 --> 00:00:26,240 you know, make better diagnoses. 14 00:00:26,240 --> 00:00:28,920 So we're talking real doctors using AI. 15 00:00:28,920 --> 00:00:30,880 Real doctors, real cases. 16 00:00:30,880 --> 00:00:33,040 So try and figure out what's wrong with their patients. 17 00:00:33,040 --> 00:00:33,880 Yeah. 18 00:00:33,880 --> 00:00:35,040 Like a digital medical detective. 19 00:00:35,040 --> 00:00:35,880 Exactly. 20 00:00:35,880 --> 00:00:38,360 And the question is, does having access to something 21 00:00:38,360 --> 00:00:41,240 like chat GPT actually change how doctors think? 22 00:00:41,240 --> 00:00:42,360 Change how they diagnose. 23 00:00:42,360 --> 00:00:44,400 Yeah, can it actually be useful 24 00:00:44,400 --> 00:00:45,800 or is it just a distraction? 25 00:00:45,800 --> 00:00:46,640 Right. 26 00:00:46,640 --> 00:00:47,480 So to figure that out. 27 00:00:47,480 --> 00:00:48,640 They got a bunch of doctors together, right? 28 00:00:48,640 --> 00:00:49,840 Yeah, they have 50 doctors. 29 00:00:49,840 --> 00:00:50,680 50 doctors. 30 00:00:50,680 --> 00:00:52,080 Across a few different specialties. 31 00:00:52,080 --> 00:00:52,920 Oh, like. 32 00:00:52,920 --> 00:00:54,320 Family medicine, internal medicine. 33 00:00:54,320 --> 00:00:55,360 Emergency medicine. 34 00:00:55,360 --> 00:00:56,760 Yeah, so a good mix. 35 00:00:56,760 --> 00:00:57,840 To see how they diagnose. 36 00:00:57,840 --> 00:01:00,560 Right, and they gave them these complex medical cases. 37 00:01:00,560 --> 00:01:02,040 Like little medical mysteries. 38 00:01:02,040 --> 00:01:02,880 Yeah, exactly. 39 00:01:02,880 --> 00:01:04,080 They had to put all the clues together. 40 00:01:04,080 --> 00:01:06,960 Symptoms, test results, you know, patient history. 41 00:01:06,960 --> 00:01:07,920 The whole thing. 42 00:01:07,920 --> 00:01:09,160 And figure out what's going on. 43 00:01:09,160 --> 00:01:10,760 And they had to explain how they got there. 44 00:01:10,760 --> 00:01:13,120 Yeah, so it wasn't just multiple choice. 45 00:01:13,120 --> 00:01:13,960 Oh, interesting. 46 00:01:13,960 --> 00:01:15,800 They had to actually write out their reasoning 47 00:01:15,800 --> 00:01:16,680 step by step. 48 00:01:16,680 --> 00:01:19,720 Like they were making a note in the patient's chart. 49 00:01:19,720 --> 00:01:21,160 Exactly. 50 00:01:21,160 --> 00:01:23,480 They called it structured reflection. 51 00:01:23,480 --> 00:01:24,560 Structured reflection. 52 00:01:24,560 --> 00:01:27,280 Which basically means show your work. 53 00:01:27,280 --> 00:01:28,120 I like that. 54 00:01:28,120 --> 00:01:29,560 So you can really see how they're thinking. 55 00:01:29,560 --> 00:01:30,920 So they've got these doctors. 56 00:01:30,920 --> 00:01:32,680 They've got these mystery cases. 57 00:01:32,680 --> 00:01:35,600 And then they throw chat GPT-4 into the mix. 58 00:01:35,600 --> 00:01:36,440 Exactly. 59 00:01:36,440 --> 00:01:39,720 What happened when the AI had to go at these same cases? 60 00:01:39,720 --> 00:01:41,240 Well, this is the really interesting part. 61 00:01:41,240 --> 00:01:42,080 OK. 62 00:01:42,080 --> 00:01:44,880 The AI on its own actually scored higher 63 00:01:44,880 --> 00:01:46,120 than any of the doctors. 64 00:01:46,120 --> 00:01:47,120 Wait, really? 65 00:01:47,120 --> 00:01:48,840 Yeah, it was significantly better. 66 00:01:48,840 --> 00:01:52,200 The AI was better than all these expert doctors. 67 00:01:52,200 --> 00:01:53,880 At diagnosing these cases. 68 00:01:53,880 --> 00:01:54,920 That's kind of freaky. 69 00:01:54,920 --> 00:01:55,920 It is a little bit. 70 00:01:55,920 --> 00:01:57,880 Does that mean robots are going to be our doctors now? 71 00:01:57,880 --> 00:01:59,120 Well, not so fast. 72 00:01:59,120 --> 00:01:59,880 OK. 73 00:01:59,880 --> 00:02:01,000 Here's the thing. 74 00:02:01,000 --> 00:02:03,280 When the doctors were allowed to use GPT-4, 75 00:02:03,280 --> 00:02:05,840 along with their usual resources, 76 00:02:05,840 --> 00:02:08,280 their scores didn't actually get that much better. 77 00:02:08,280 --> 00:02:10,200 So it's good on its own. 78 00:02:10,200 --> 00:02:11,040 It seems that way. 79 00:02:11,040 --> 00:02:12,320 But not good at working with a team. 80 00:02:12,320 --> 00:02:14,680 Yeah, it's like it's a great soloist, 81 00:02:14,680 --> 00:02:15,880 but not a team player. 82 00:02:15,880 --> 00:02:17,520 So what's going on there? 83 00:02:17,520 --> 00:02:19,360 Well, that's where things get really interesting. 84 00:02:19,360 --> 00:02:20,520 OK, tell me more. 85 00:02:20,520 --> 00:02:23,480 It might not just be about how smart the AI is. 86 00:02:23,480 --> 00:02:24,600 That's something else. 87 00:02:24,600 --> 00:02:27,920 But about how we as humans interact with it. 88 00:02:27,920 --> 00:02:32,800 So you can't just say, hey AI, what's wrong with this person? 89 00:02:32,800 --> 00:02:33,640 Not quite. 90 00:02:33,640 --> 00:02:34,480 There's more to it. 91 00:02:34,480 --> 00:02:37,000 It's like having a really powerful search engine. 92 00:02:37,000 --> 00:02:37,280 Right. 93 00:02:37,280 --> 00:02:39,160 But if you don't know how to use it, 94 00:02:39,160 --> 00:02:40,480 you're not going to find what you need. 95 00:02:40,480 --> 00:02:42,520 So you have to know how to ask the right questions. 96 00:02:42,520 --> 00:02:43,280 Exactly. 97 00:02:43,280 --> 00:02:45,800 And that's something that this study highlighted. 98 00:02:45,800 --> 00:02:46,280 Oh, hells in the air. 99 00:02:46,280 --> 00:02:50,720 This idea of prompting how you phrase your questions to the AI. 100 00:02:50,720 --> 00:02:51,320 I see. 101 00:02:51,320 --> 00:02:55,160 The doctors in the study weren't given any specific training 102 00:02:55,160 --> 00:02:56,760 on how to use GPT-4. 103 00:02:56,760 --> 00:02:57,000 Right. 104 00:02:57,000 --> 00:02:59,560 So they might not have been using it to its full potential. 105 00:02:59,560 --> 00:03:03,600 Like they got a fancy new gadget, but no instructions. 106 00:03:03,600 --> 00:03:04,640 Exactly. 107 00:03:04,640 --> 00:03:05,560 So that's a big take away. 108 00:03:05,560 --> 00:03:08,440 Yeah, it's not enough to just introduce AI into health care. 109 00:03:08,440 --> 00:03:08,960 Right. 110 00:03:08,960 --> 00:03:11,360 We have to figure out how to integrate it properly. 111 00:03:11,360 --> 00:03:12,200 How to use it, right? 112 00:03:12,200 --> 00:03:13,560 How to train people. 113 00:03:13,560 --> 00:03:15,760 So it's not about AI versus doctors. 114 00:03:15,760 --> 00:03:18,640 No, it's more about how they can work together. 115 00:03:18,640 --> 00:03:20,680 Both bringing something to the table. 116 00:03:20,680 --> 00:03:21,680 Exactly. 117 00:03:21,680 --> 00:03:24,320 The doctor with their experience in human touch. 118 00:03:24,320 --> 00:03:29,080 And the AI with all this information and new perspectives. 119 00:03:29,080 --> 00:03:31,680 It could be a really powerful combination. 120 00:03:31,680 --> 00:03:32,720 It's pretty exciting. 121 00:03:32,720 --> 00:03:34,400 But we have to be realistic. 122 00:03:34,400 --> 00:03:35,320 Absolutely. 123 00:03:35,320 --> 00:03:37,520 This was just one study with one AI. 124 00:03:37,520 --> 00:03:38,120 Right. 125 00:03:38,120 --> 00:03:39,200 There's a lot more out there. 126 00:03:39,200 --> 00:03:42,040 It's like saying we tried one type of electric car. 127 00:03:42,040 --> 00:03:43,280 And it didn't work so well. 128 00:03:43,280 --> 00:03:45,200 And then saying all electric cars are bad. 129 00:03:45,200 --> 00:03:46,080 We can't do that. 130 00:03:46,080 --> 00:03:48,040 We need more research, more data. 131 00:03:48,040 --> 00:03:49,440 And who knows what's coming next. 132 00:03:49,440 --> 00:03:51,280 Exactly. 133 00:03:51,280 --> 00:03:52,040 AI is constantly evolving. 134 00:03:52,040 --> 00:03:52,920 So this study. 135 00:03:52,920 --> 00:03:54,480 It's a snapshot in time. 136 00:03:54,480 --> 00:03:56,640 Oh, look at what's possible right now. 137 00:03:56,640 --> 00:03:58,360 But there's so much more to come. 138 00:03:58,360 --> 00:04:00,480 It's a really interesting start, though. 139 00:04:00,480 --> 00:04:00,960 Yeah. 140 00:04:00,960 --> 00:04:03,800 It shows the potential of AI in health care. 141 00:04:03,800 --> 00:04:05,720 And some of the challenges we need to think about. 142 00:04:05,720 --> 00:04:06,440 Exactly. 143 00:04:06,440 --> 00:04:08,920 And that's what makes this study so important. 144 00:04:08,920 --> 00:04:13,040 Gives us a lot to think about for the future of AI in health care. 145 00:04:13,040 --> 00:04:13,520 Yeah. 146 00:04:13,520 --> 00:04:16,200 How do we actually train doctors to use these tools? 147 00:04:16,200 --> 00:04:16,440 Why? 148 00:04:16,440 --> 00:04:18,880 It's like giving someone a supercomputer. 149 00:04:18,880 --> 00:04:20,040 But they have no idea how to code. 150 00:04:20,040 --> 00:04:20,600 Exactly. 151 00:04:20,600 --> 00:04:23,600 You wouldn't expect them to be an expert overnight. 152 00:04:23,600 --> 00:04:25,760 So we need AI training for doctors. 153 00:04:25,760 --> 00:04:26,160 Yeah. 154 00:04:26,160 --> 00:04:28,880 Imagine if medical school started incorporating that 155 00:04:28,880 --> 00:04:30,200 into the curriculum. 156 00:04:30,200 --> 00:04:31,200 That's a cool idea. 157 00:04:31,200 --> 00:04:34,320 Doctors could learn how to really use these systems. 158 00:04:34,320 --> 00:04:35,640 How to ask the right questions. 159 00:04:35,640 --> 00:04:37,600 Get the most useful information. 160 00:04:37,600 --> 00:04:39,360 Like it could really enhance their skills. 161 00:04:39,360 --> 00:04:41,200 It would be like giving them a superpower. 162 00:04:41,200 --> 00:04:42,400 A medical superpower. 163 00:04:42,400 --> 00:04:44,000 Tapping into all this knowledge. 164 00:04:44,000 --> 00:04:45,880 And it's not just diagnosis either, right? 165 00:04:45,880 --> 00:04:46,360 No. 166 00:04:46,360 --> 00:04:49,760 There's so much more AI could do in health care. 167 00:04:49,760 --> 00:04:50,400 Like what? 168 00:04:50,400 --> 00:04:53,200 Think about personalized treatment plans. 169 00:04:53,200 --> 00:04:54,720 Tailored to each patient. 170 00:04:54,720 --> 00:04:58,040 Based on their genes, their lifestyle, their medical history. 171 00:04:58,040 --> 00:04:58,360 Whoa. 172 00:04:58,360 --> 00:04:59,200 That's next level. 173 00:04:59,200 --> 00:05:01,160 Or what about drug discovery? 174 00:05:01,160 --> 00:05:03,760 Using AI to find new medicines. 175 00:05:03,760 --> 00:05:04,160 Yes. 176 00:05:04,160 --> 00:05:05,560 Sifting through all that data. 177 00:05:05,560 --> 00:05:07,240 Finding the most promising leads. 178 00:05:07,240 --> 00:05:08,600 The possibilities are huge. 179 00:05:08,600 --> 00:05:09,640 It's pretty exciting. 180 00:05:09,640 --> 00:05:10,920 But we can't get carried away. 181 00:05:10,920 --> 00:05:11,960 I have to be realistic. 182 00:05:11,960 --> 00:05:13,800 There are challenges too. 183 00:05:13,800 --> 00:05:14,560 Like what? 184 00:05:14,560 --> 00:05:16,280 Well, first of all, we need to make sure 185 00:05:16,280 --> 00:05:18,560 these systems are reliable. 186 00:05:18,560 --> 00:05:19,520 They have to be accurate. 187 00:05:19,520 --> 00:05:21,160 And safe for patients. 188 00:05:21,160 --> 00:05:22,840 So the information they give is good. 189 00:05:22,840 --> 00:05:25,000 And their decisions make sense. 190 00:05:25,000 --> 00:05:26,080 What about bias? 191 00:05:26,080 --> 00:05:27,200 That's a big one. 192 00:05:27,200 --> 00:05:29,040 AI can be biased, can it? 193 00:05:29,040 --> 00:05:29,360 Yeah. 194 00:05:29,360 --> 00:05:32,640 If the data it's trained on is biased, the AI can be too. 195 00:05:32,640 --> 00:05:34,480 And that could be a problem for patients. 196 00:05:34,480 --> 00:05:34,840 Yeah. 197 00:05:34,840 --> 00:05:37,760 We need to make sure these systems are fair for everyone. 198 00:05:37,760 --> 00:05:39,120 And then there's privacy. 199 00:05:39,120 --> 00:05:39,720 Of course. 200 00:05:39,720 --> 00:05:41,000 All that patient data. 201 00:05:41,000 --> 00:05:42,800 It's really sensitive information. 202 00:05:42,800 --> 00:05:44,000 We have to keep it safe. 203 00:05:44,000 --> 00:05:44,640 Absolutely. 204 00:05:44,640 --> 00:05:46,880 Data security is crucial. 205 00:05:46,880 --> 00:05:48,200 So there's a lot to figure out. 206 00:05:48,200 --> 00:05:48,480 Yeah. 207 00:05:48,480 --> 00:05:49,720 It's a complex landscape. 208 00:05:49,720 --> 00:05:51,360 But it's also really hopeful. 209 00:05:51,360 --> 00:05:52,040 I agree. 210 00:05:52,040 --> 00:05:55,040 AI has the potential to change health care for the better. 211 00:05:55,040 --> 00:05:56,640 We just have to be smart about it. 212 00:05:56,640 --> 00:06:01,200 We need doctors, ethicists, data scientists, 213 00:06:01,200 --> 00:06:02,240 everyone working together. 214 00:06:02,240 --> 00:06:04,320 To figure out the best way forward. 215 00:06:04,320 --> 00:06:07,320 It's a conversation we all need to be part of. 216 00:06:07,320 --> 00:06:08,960 Because it's going to affect all of us. 217 00:06:08,960 --> 00:06:11,280 The future of health care is being shaped right now. 218 00:06:11,280 --> 00:06:13,200 And we have a chance to make it amazing. 219 00:06:13,200 --> 00:06:16,480 I feel like we're just at the beginning of this whole AI 220 00:06:16,480 --> 00:06:17,520 and health care thing. 221 00:06:17,520 --> 00:06:18,120 Oh, yeah. 222 00:06:18,120 --> 00:06:18,640 Definitely. 223 00:06:18,640 --> 00:06:20,240 This is just scratching the surface. 224 00:06:20,240 --> 00:06:21,680 This study was really interesting. 225 00:06:21,680 --> 00:06:22,160 Yeah. 226 00:06:22,160 --> 00:06:24,600 But it definitely raises more questions than it answers. 227 00:06:24,600 --> 00:06:26,680 I think that's a good thing, though. 228 00:06:26,680 --> 00:06:27,280 How so? 229 00:06:27,280 --> 00:06:28,920 It means people are engaged. 230 00:06:28,920 --> 00:06:30,120 They're thinking about it. 231 00:06:30,120 --> 00:06:31,960 Thinking about AI and their own health care. 232 00:06:31,960 --> 00:06:32,800 Exactly. 233 00:06:32,800 --> 00:06:33,800 And that's important. 234 00:06:33,800 --> 00:06:36,120 So let's flip the script for a second. 235 00:06:36,120 --> 00:06:36,680 OK. 236 00:06:36,680 --> 00:06:38,240 And ask our listeners a question. 237 00:06:38,240 --> 00:06:39,320 What do you think? 238 00:06:39,320 --> 00:06:42,120 We saw how AI can help with diagnosis. 239 00:06:42,120 --> 00:06:42,760 Right. 240 00:06:42,760 --> 00:06:44,320 But what else could it do? 241 00:06:44,320 --> 00:06:44,680 Yeah. 242 00:06:44,680 --> 00:06:47,160 What other ways could AI be used in health care? 243 00:06:47,160 --> 00:06:49,080 To help patients and doctors. 244 00:06:49,080 --> 00:06:51,840 Maybe it's something like scheduling appointments. 245 00:06:51,840 --> 00:06:54,000 Or analyzing medical images. 246 00:06:54,000 --> 00:06:55,840 Maybe even creating those personalized health 247 00:06:55,840 --> 00:06:56,840 plans we talked about. 248 00:06:56,840 --> 00:06:59,200 Based on your genes and lifestyle. 249 00:06:59,200 --> 00:07:01,040 There are so many possibilities. 250 00:07:01,040 --> 00:07:02,440 And what about your concerns? 251 00:07:02,440 --> 00:07:02,640 Yeah. 252 00:07:02,640 --> 00:07:04,960 What would make you feel comfortable with AI 253 00:07:04,960 --> 00:07:06,760 being involved in your health care? 254 00:07:06,760 --> 00:07:08,040 Is it transparency? 255 00:07:08,040 --> 00:07:10,800 Knowing how the AI is making decisions. 256 00:07:10,800 --> 00:07:12,000 What about data privacy? 257 00:07:12,000 --> 00:07:14,440 Making sure your information is safe. 258 00:07:14,440 --> 00:07:17,640 Or is it about making sure the human element is still there? 259 00:07:17,640 --> 00:07:19,360 That doctor-patient connection? 260 00:07:19,360 --> 00:07:19,760 Yeah. 261 00:07:19,760 --> 00:07:20,400 That's important. 262 00:07:20,400 --> 00:07:22,240 These are all really important questions. 263 00:07:22,240 --> 00:07:23,680 And there's no easy answers. 264 00:07:23,680 --> 00:07:25,680 But we need to be having these conversations. 265 00:07:25,680 --> 00:07:27,800 Thinking about the ethical side of things. 266 00:07:27,800 --> 00:07:31,080 Because AI in health care is here to stay. 267 00:07:31,080 --> 00:07:32,000 It's the future. 268 00:07:32,000 --> 00:07:35,560 And we all have a role to play in shaping that future. 269 00:07:35,560 --> 00:07:37,120 Making sure it benefits everyone. 270 00:07:37,120 --> 00:07:37,920 Exactly. 271 00:07:37,920 --> 00:07:39,680 Well, this has been an awesome deep dive. 272 00:07:39,680 --> 00:07:40,080 Yeah. 273 00:07:40,080 --> 00:07:40,800 Really fascinating. 274 00:07:40,800 --> 00:07:42,880 We've learned a lot about AI in health care. 275 00:07:42,880 --> 00:07:45,480 And there's still so much more to explore. 276 00:07:45,480 --> 00:07:47,880 But for now, we're going to wrap things up. 277 00:07:47,880 --> 00:07:49,200 Thanks for joining us. 278 00:07:49,200 --> 00:07:52,880 And be sure to tune in next time for another deep dive. 279 00:07:52,880 --> 00:08:14,400 Until then, stay curious.