0 00:00:00,010 --> 00:00:07,177 SUB BY : DENI AUROR@ https://aurorarental.blogspot.com/ 1 00:00:07,206 --> 00:00:08,341 You know how they say 2 00:00:08,374 --> 00:00:10,743 there are two certainties in life, right? 3 00:00:10,777 --> 00:00:12,445 Death and taxes. 4 00:00:12,478 --> 00:00:15,014 Can't we get rid of one of those? 5 00:00:15,048 --> 00:00:17,584 See, 100 years ago, life expectancy was only 45, 6 00:00:17,617 --> 00:00:18,718 can you believe that? 7 00:00:18,752 --> 00:00:20,886 Then by the 1950s, it was up to 65, 8 00:00:20,920 --> 00:00:22,355 and today, it's almost 80. 9 00:00:22,388 --> 00:00:25,792 Tomorrow, who knows? Right? 10 00:00:25,825 --> 00:00:27,527 Healthcare has made huge progress. 11 00:00:27,560 --> 00:00:30,529 We've eradicated epidemics that used to kill millions, 12 00:00:30,563 --> 00:00:32,465 but life is fragile. 13 00:00:32,498 --> 00:00:34,934 People still get sick, or pass away 14 00:00:34,967 --> 00:00:36,903 for reasons that maybe 15 00:00:36,936 --> 00:00:39,606 should be, someday curable. 16 00:00:41,274 --> 00:00:44,344 What if we could improve diagnosis? 17 00:00:44,377 --> 00:00:47,446 Innovate to predict illness instead of just react to it? 18 00:00:47,480 --> 00:00:49,015 In this episode, 19 00:00:49,048 --> 00:00:51,183 we'll see how machine learning is combating 20 00:00:51,217 --> 00:00:53,352 one of the leading causes of blindness, 21 00:00:53,386 --> 00:00:56,055 and enabling a son with a neurological disease 22 00:00:56,089 --> 00:00:58,123 to communicate with his family. 23 00:00:58,157 --> 00:01:00,259 AI is changing the way we think 24 00:01:00,293 --> 00:01:01,861 about mind and body, 25 00:01:01,894 --> 00:01:03,162 life and death, 26 00:01:03,195 --> 00:01:05,098 and what we value most, 27 00:01:05,131 --> 00:01:07,100 our human experience. 28 00:01:09,936 --> 00:01:14,006 ...and our other co-captain, Number 8! Tim Shaw! 29 00:01:18,144 --> 00:01:20,546 We've lived with his football dream, 30 00:01:20,580 --> 00:01:22,248 All the way back to sixth grade 31 00:01:22,281 --> 00:01:23,817 when his coach said, 32 00:01:23,850 --> 00:01:25,784 "This kid is gonna go a long way." 33 00:01:25,818 --> 00:01:27,286 From that point on, 34 00:01:27,320 --> 00:01:29,556 Tim was doing pushups in his bedroom at night, 35 00:01:29,589 --> 00:01:32,158 Tim was the first one at practice. 36 00:01:32,191 --> 00:01:33,993 Tim took it seriously. 37 00:01:44,303 --> 00:01:47,173 Number 8, Tim Shaw! 38 00:01:49,108 --> 00:01:51,210 I don't know what they're doing out there, 39 00:01:51,244 --> 00:01:53,179 and I don't know who they comin' to! 40 00:01:53,212 --> 00:01:55,648 For as long as he can remember, 41 00:01:55,682 --> 00:01:58,351 Tim Shaw dreamed of three letters... 42 00:01:58,384 --> 00:02:00,119 N-F-L. 43 00:02:01,254 --> 00:02:03,089 He was a natural from the beginning. 44 00:02:03,122 --> 00:02:05,458 As a kid, he was fast and athletic. 45 00:02:05,491 --> 00:02:09,495 He grew into 235 pounds of pure muscle, 46 00:02:09,528 --> 00:02:13,266 and at 23, he was drafted to the pros. 47 00:02:13,299 --> 00:02:16,068 His dream was real. 48 00:02:16,101 --> 00:02:19,505 He was playing professional football. 49 00:02:19,538 --> 00:02:21,441 Hello, I'm with Tim Shaw. 50 00:02:21,474 --> 00:02:24,276 You get to start this season right. What's it feel like? 51 00:02:24,310 --> 00:02:26,278 It's that amazing pre-game electricity, 52 00:02:26,312 --> 00:02:30,917 the butterflies are there, and I'm ready to hit somebody. You might wanna look out. 53 00:02:30,950 --> 00:02:32,885 Hey, Titans fans, it's Tim Shaw here, 54 00:02:32,919 --> 00:02:35,855 linebacker and special teams animal. 55 00:02:35,888 --> 00:02:37,190 He loves what he does. 56 00:02:37,223 --> 00:02:39,525 He says, "They pay me to hit people!" 57 00:02:42,328 --> 00:02:44,664 I'm here to bring you some truth, 58 00:02:44,697 --> 00:02:46,232 a little bit of truth, 59 00:02:46,265 --> 00:02:47,900 and so we'll call it T-Shaw's truth, 60 00:02:47,934 --> 00:02:50,737 'cause it's not all the way true, but it's my truth. 61 00:03:27,140 --> 00:03:29,308 In 2012, 62 00:03:29,341 --> 00:03:31,777 my body started to do things it hadn't done before. 63 00:03:31,810 --> 00:03:34,647 My muscles were twitching, I was stumbling, 64 00:03:34,680 --> 00:03:38,985 or I was not making a play I would have always made. 65 00:03:39,018 --> 00:03:40,553 I just wasn't the same athlete, 66 00:03:40,586 --> 00:03:44,591 I wasn't the same football player that I'd always been. 67 00:04:03,443 --> 00:04:07,313 The three letters that had defined Tim's life up to that point 68 00:04:07,347 --> 00:04:10,817 were not the three letters that the doctor told him that day. 69 00:04:10,850 --> 00:04:11,985 A-L-S. 70 00:04:37,609 --> 00:04:39,178 Okay... 71 00:04:39,211 --> 00:04:43,349 A-L-S, which stands for "amyotrophic lateral sclerosis," 72 00:04:43,382 --> 00:04:45,785 is also known as Lou Gehrig's Disease. 73 00:04:45,818 --> 00:04:49,488 It causes the death of neurons controlling voluntary muscles. 74 00:04:49,521 --> 00:04:52,258 He can't even scratch his head... 75 00:04:52,291 --> 00:04:53,559 Better yet? 76 00:04:53,592 --> 00:04:55,627 ...none of those physical things 77 00:04:55,661 --> 00:04:58,397 that were so easy for him before. 78 00:04:58,430 --> 00:05:01,233 He has to think about every step he takes. 79 00:05:01,267 --> 00:05:03,469 So Tim's food comes in this little container. 80 00:05:03,502 --> 00:05:05,071 We're gonna mix it with water. 81 00:05:17,149 --> 00:05:19,152 As the disease progresses, 82 00:05:19,185 --> 00:05:20,719 muscles weaken. 83 00:05:20,753 --> 00:05:24,423 Simple everyday actions, like walking, talking, and eating, 84 00:05:24,456 --> 00:05:26,859 take tremendous effort. 85 00:05:49,915 --> 00:05:52,952 Tim used to call me on the phone in the night, 86 00:05:52,985 --> 00:05:54,420 and he had voice recognition, 87 00:05:54,453 --> 00:05:56,221 and he would speak to the phone, 88 00:05:56,255 --> 00:05:57,957 and say, "Call Dad." 89 00:05:57,990 --> 00:06:02,228 His phone didn't recognize the word "Dad." 90 00:06:02,261 --> 00:06:05,531 So, he had said to me... 91 00:06:05,564 --> 00:06:07,866 "Dad, I've changed your name. 92 00:06:07,900 --> 00:06:11,304 I'm calling... I now call you "Yo-yo." 93 00:06:11,337 --> 00:06:14,907 So he would say into his phone, "Call Yo-yo." 94 00:06:16,742 --> 00:06:20,879 Tim has stopped a lot of his communication. 95 00:06:20,912 --> 00:06:23,315 He just doesn't talk as much as he used to, 96 00:06:23,349 --> 00:06:25,384 and I, I miss that. 97 00:06:25,418 --> 00:06:26,752 I miss it. 98 00:06:26,785 --> 00:06:29,555 - What do you think about my red beard? - No opinion. 99 00:06:29,589 --> 00:06:31,290 That means he likes it, 100 00:06:31,324 --> 00:06:32,858 just doesn't wanna say on camera. 101 00:06:32,892 --> 00:06:35,862 Now, my favorite was when you had the handlebar moustache. 102 00:06:35,895 --> 00:06:38,998 Language, the ability to communicate with one another. 103 00:06:39,031 --> 00:06:42,734 It's something that makes us uniquely human, 104 00:06:42,768 --> 00:06:47,139 making communication an impactful application for AI. 105 00:06:47,172 --> 00:06:49,842 Yeah, that'll be fun. 106 00:06:57,516 --> 00:06:59,318 My name is Julie. 107 00:06:59,351 --> 00:07:01,120 I'm a product manager here at Google. 108 00:07:01,153 --> 00:07:04,490 For the past year or so, I've been working on Project Euphonia. 109 00:07:04,523 --> 00:07:06,759 Project Euphonia has two different goals. 110 00:07:06,792 --> 00:07:09,529 One is to improve speech recognition 111 00:07:09,562 --> 00:07:12,531 for people who have a variety of medical conditions. 112 00:07:12,564 --> 00:07:15,268 The second goal is to give people their voice back, 113 00:07:15,301 --> 00:07:18,571 which means actually recreating the way they used to sound 114 00:07:18,604 --> 00:07:20,573 before they were diagnosed. 115 00:07:20,606 --> 00:07:22,341 If you think about communication, 116 00:07:22,374 --> 00:07:24,476 it starts with understanding someone, 117 00:07:24,510 --> 00:07:26,178 and then being understood, 118 00:07:26,211 --> 00:07:27,980 and for a lot of people, 119 00:07:28,014 --> 00:07:31,884 their voice is like their identity. 120 00:07:31,917 --> 00:07:34,920 In the US alone, roughly one in ten people 121 00:07:34,954 --> 00:07:36,789 suffer acquired speech impairments, 122 00:07:36,822 --> 00:07:39,592 which can be caused by anything from ALS, 123 00:07:39,625 --> 00:07:43,029 to strokes, to Parkinson's, to brain injuries. 124 00:07:43,062 --> 00:07:45,397 Solving it is a big challenge, 125 00:07:45,431 --> 00:07:49,735 which is why Julie partnered with a big thinker to help. 126 00:08:07,319 --> 00:08:11,524 Dimitri is a world-class research scientist and inventor. 127 00:08:11,557 --> 00:08:14,326 He's worked at IBM, Princeton, and now Google, 128 00:08:14,359 --> 00:08:18,197 and holds over 150 patents. 129 00:08:18,230 --> 00:08:19,431 Accomplishments aside, 130 00:08:19,464 --> 00:08:21,800 communication is very personal to him. 131 00:08:21,833 --> 00:08:24,770 Dimitri has a pretty strong Russian accent, 132 00:08:24,803 --> 00:08:27,973 and also he learned English when he was already deaf, 133 00:08:28,007 --> 00:08:30,209 so he never heard himself speak English. 134 00:08:32,144 --> 00:08:34,079 Oh, you do? Oh, okay. 135 00:08:34,113 --> 00:08:37,049 Technology can't yet help him hear his own voice. 136 00:08:37,082 --> 00:08:40,419 He uses AI-powered Live Transcribe 137 00:08:40,453 --> 00:08:42,321 to help him communicate. 138 00:08:42,355 --> 00:08:43,789 Okay, that's awesome. 139 00:08:43,823 --> 00:08:47,260 So we partnered up with Dimitri to train a recognizer 140 00:08:47,293 --> 00:08:50,663 that did a much better job at recognizing his voice. 141 00:08:50,696 --> 00:08:53,832 The model that you're using right now for recognition, 142 00:08:53,865 --> 00:08:57,069 what data did you train it on? 143 00:09:01,307 --> 00:09:04,277 So, how does speech recognition work? 144 00:09:07,312 --> 00:09:10,683 First, the sound of our voice is converted into a waveform, 145 00:09:10,716 --> 00:09:13,118 which is really just a picture of the sound. 146 00:09:13,152 --> 00:09:16,288 Waveforms are then matched to transcriptions, 147 00:09:16,321 --> 00:09:18,357 or "labels" for each word. 148 00:09:18,391 --> 00:09:21,726 These maps exist for most words in the English language. 149 00:09:21,760 --> 00:09:24,530 This is where machine learning takes over. 150 00:09:24,563 --> 00:09:26,865 Using millions of voice samples, 151 00:09:26,898 --> 00:09:28,700 a deep learning model is trained 152 00:09:28,734 --> 00:09:31,704 to map input sounds to output words. 153 00:09:31,737 --> 00:09:35,341 Then the algorithm uses rules, such as grammar and syntax, 154 00:09:35,374 --> 00:09:37,176 to predict each word in a sentence. 155 00:09:37,209 --> 00:09:39,244 This is how AI can tell the difference 156 00:09:39,278 --> 00:09:42,581 between "there," "their," and "they're." 157 00:09:42,614 --> 00:09:45,584 The speech recognition model that Google uses 158 00:09:45,618 --> 00:09:47,619 works very well for people 159 00:09:47,652 --> 00:09:50,590 who have a voice that sounds similar 160 00:09:50,623 --> 00:09:53,225 to the examples that were used to train this model. 161 00:09:53,258 --> 00:09:56,462 In 90% of cases, it will recognize what you want to say. 162 00:09:56,495 --> 00:09:58,797 Dimitri's not in that 90%. 163 00:09:58,830 --> 00:10:01,033 For someone like him, it doesn't work at all. 164 00:10:01,066 --> 00:10:04,870 So he created a model based on a sample of one. 165 00:10:23,489 --> 00:10:25,291 But making a new unique model 166 00:10:25,324 --> 00:10:27,893 with unique data for every new and unique person 167 00:10:27,927 --> 00:10:29,795 is slow and inefficient. 168 00:10:29,828 --> 00:10:32,097 Tim calls his dad "Yo-yo." 169 00:10:32,131 --> 00:10:35,700 Others with ALS may call their dads something else. 170 00:10:35,734 --> 00:10:37,502 Can we build one machine 171 00:10:37,536 --> 00:10:40,539 that recognizes many different people, 172 00:10:40,572 --> 00:10:42,374 and how can we do it fast? 173 00:10:42,408 --> 00:10:45,277 So this data doesn't really exist. 174 00:10:45,310 --> 00:10:46,878 We have to actually collect it. 175 00:10:46,912 --> 00:10:50,549 So we started this partnership with ALS TDI in Boston. 176 00:10:50,582 --> 00:10:52,818 They helped us collect voice samples 177 00:10:52,851 --> 00:10:54,286 from people who have ALS. 178 00:10:54,319 --> 00:10:55,588 This is for you, T. Shaw. 179 00:10:55,621 --> 00:10:59,025 One, two, three! 180 00:11:02,928 --> 00:11:07,265 I hereby accept your ALS ice bucket challenge. 181 00:11:09,501 --> 00:11:11,870 When the ice bucket challenge went viral, 182 00:11:11,904 --> 00:11:17,443 millions joined the fight, and raised over $220 million for ALS research. 183 00:11:17,476 --> 00:11:20,779 There really is a straight line from the ice bucket challenge 184 00:11:20,812 --> 00:11:22,448 to the Euphonia Project. 185 00:11:29,187 --> 00:11:31,957 ALS Therapy Development Institute is an organization 186 00:11:31,990 --> 00:11:35,227 that's dedicated to finding treatments and cures for ALS. 187 00:11:35,260 --> 00:11:37,029 We are life-focused. 188 00:11:37,062 --> 00:11:40,299 How can we use technologies we have 189 00:11:40,332 --> 00:11:42,634 to help these people right away? 190 00:11:42,667 --> 00:11:45,604 Yeah, they're actually noisier. 191 00:11:45,638 --> 00:11:47,406 That's a good point. 192 00:11:47,439 --> 00:11:49,107 I met Tim a few years ago 193 00:11:49,140 --> 00:11:50,976 shortly after he had been diagnosed. 194 00:11:51,009 --> 00:11:53,645 Very difficult to go public, 195 00:11:53,678 --> 00:11:55,681 but it was made very clear to me 196 00:11:55,714 --> 00:11:57,116 that the time was right. 197 00:11:57,149 --> 00:12:00,486 He was trying to understand what to expect in his life, 198 00:12:00,519 --> 00:12:02,688 but he was also trying to figure out, 199 00:12:02,721 --> 00:12:04,690 "All right, what part can I play?" 200 00:12:04,723 --> 00:12:07,526 All the ice bucket challenges and the awareness 201 00:12:07,559 --> 00:12:09,461 have really inspired me also. 202 00:12:09,494 --> 00:12:10,763 If we can just step back, 203 00:12:10,796 --> 00:12:13,064 and say, "Where can I shine a light?" 204 00:12:13,098 --> 00:12:15,067 or "Where can I give a hand?" 205 00:12:15,100 --> 00:12:17,436 When the ice bucket challenge happened, 206 00:12:17,469 --> 00:12:21,640 we had this huge influx of resources of cash, 207 00:12:21,673 --> 00:12:23,542 and that gave us the ability 208 00:12:23,576 --> 00:12:26,678 to reach out to people with ALS who are in our programs 209 00:12:26,711 --> 00:12:28,714 to share their data with us. 210 00:12:28,747 --> 00:12:31,583 That's what got us the big enough data sets 211 00:12:31,616 --> 00:12:33,786 to really attract Google. 212 00:12:36,688 --> 00:12:38,790 Fernando didn't initially set out 213 00:12:38,823 --> 00:12:40,893 to make speech recognition work better, 214 00:12:40,926 --> 00:12:44,263 but in the process of better understanding the disease, 215 00:12:44,296 --> 00:12:47,332 he built a huge database of ALS voices, 216 00:12:47,366 --> 00:12:51,437 which may help Tim and many others. 217 00:12:54,873 --> 00:12:56,742 It automatically uploaded it. 218 00:12:56,775 --> 00:12:57,977 Oh. 219 00:13:59,571 --> 00:14:01,574 How many have you done, Tim? 220 00:14:07,379 --> 00:14:08,880 2066? 221 00:14:08,914 --> 00:14:12,217 Tim, he wants to find every way that he can help. 222 00:14:12,251 --> 00:14:16,388 It's inspiring to see his level of enthusiasm, 223 00:14:16,421 --> 00:14:20,626 and his willingness to record lots and lots of voice samples. 224 00:14:20,659 --> 00:14:23,128 To turn all this data into real help, 225 00:14:23,161 --> 00:14:25,197 Fernando partnered with one of the people 226 00:14:25,230 --> 00:14:28,199 who started the Euphonia Project, Michael Brenner... 227 00:14:28,233 --> 00:14:30,335 - Hey, Fernando. - Hey, how are you doing? 228 00:14:30,368 --> 00:14:32,337 ...a Google research scientist 229 00:14:32,371 --> 00:14:34,105 and Harvard-trained mathematician 230 00:14:34,138 --> 00:14:35,507 who's using machine learning 231 00:14:35,540 --> 00:14:38,343 to solve scientific Hail Marys, like this one. 232 00:14:38,377 --> 00:14:42,381 Tim Shaw has recorded almost 2,000 utterances, 233 00:14:42,414 --> 00:14:44,616 and so we decided to apply our technology 234 00:14:44,649 --> 00:14:48,553 to see if we could build a recognizer that understood him. 235 00:14:51,156 --> 00:14:54,292 The goal, right, for Tim, is to get it so that it works 236 00:14:54,326 --> 00:14:56,829 outside of the things that he recorded. 237 00:14:56,862 --> 00:14:58,963 The problem is that we have no idea 238 00:14:58,997 --> 00:15:00,799 how big of a set that this will work on. 239 00:15:00,832 --> 00:15:04,102 Dimitri had recorded upwards of 15,000 sentences, 240 00:15:04,136 --> 00:15:07,105 which is just an incredible amount of data. 241 00:15:07,138 --> 00:15:09,674 We couldn't possibly expect anyone else 242 00:15:09,707 --> 00:15:11,376 to record so many sentences, 243 00:15:11,410 --> 00:15:14,012 so we know that we have to be able to do this 244 00:15:14,046 --> 00:15:16,147 with much less recordings from a person. 245 00:15:16,181 --> 00:15:18,283 So it's not clear it will work. 246 00:15:22,287 --> 00:15:24,189 - That didn't work at all. - Not at all. 247 00:15:24,223 --> 00:15:26,191 He said, "I go the opposite way," 248 00:15:26,224 --> 00:15:28,060 and it says, "I know that was." 249 00:15:28,093 --> 00:15:29,728 When it doesn't recognize, 250 00:15:29,761 --> 00:15:32,864 we jiggle around the parameters of the speech recognizer, 251 00:15:32,897 --> 00:15:34,432 then we give it another sentence, 252 00:15:34,466 --> 00:15:37,069 and the idea is that you'll get it to understand. 253 00:15:37,102 --> 00:15:39,338 Can we go to the beach? 254 00:15:42,140 --> 00:15:43,742 - Yes! Got it. - Got it. 255 00:15:43,776 --> 00:15:45,844 That's so cool. Okay, let's try another. 256 00:15:45,878 --> 00:15:47,980 If Tim Shaw gets his voice back, 257 00:15:48,013 --> 00:15:50,716 he may no longer feel that he is defined, 258 00:15:50,749 --> 00:15:53,585 or constrained, by three letters, 259 00:15:53,618 --> 00:15:55,187 but that's a big "if." 260 00:15:55,220 --> 00:15:58,289 While Michael and team Euphonia work away, 261 00:15:58,323 --> 00:16:01,560 let's take a moment and imagine what else is possible 262 00:16:01,593 --> 00:16:03,228 in the realm of the senses. 263 00:16:03,261 --> 00:16:05,396 Speech. 264 00:16:05,430 --> 00:16:07,499 Hearing. 265 00:16:07,532 --> 00:16:08,934 Sight. 266 00:16:08,967 --> 00:16:10,869 Can AI predict blindness? 267 00:16:12,037 --> 00:16:13,405 Or even prevent it? 268 00:16:50,509 --> 00:16:53,178 Santhi does not have an easy life. 269 00:16:53,211 --> 00:16:56,348 It's made more difficult because she has diabetes, 270 00:16:56,381 --> 00:16:58,717 which is affecting her vision. 271 00:17:05,223 --> 00:17:10,229 If Santhi doesn't get medical help soon, she may go blind. 272 00:17:12,664 --> 00:17:15,067 Complications of diabetes 273 00:17:15,100 --> 00:17:16,502 include heart disease, 274 00:17:16,535 --> 00:17:17,669 kidney disease, 275 00:17:17,702 --> 00:17:20,506 but one of the really important complications 276 00:17:20,539 --> 00:17:21,807 is diabetic retinopathy. 277 00:17:21,840 --> 00:17:24,309 The reason it's so important is that 278 00:17:24,342 --> 00:17:26,979 it's one of the lead causes of blindness worldwide. 279 00:17:27,012 --> 00:17:29,681 This is particularly true in India. 280 00:17:53,171 --> 00:17:55,239 In the early stages, it's symptomless, 281 00:17:55,273 --> 00:17:57,276 but that's when it's treatable, 282 00:17:57,309 --> 00:17:59,377 so you want to screen them early on, 283 00:17:59,410 --> 00:18:01,346 before they actually lose vision. 284 00:18:05,150 --> 00:18:06,418 In the early stages, 285 00:18:06,451 --> 00:18:08,353 if a doctor is examining the eye, 286 00:18:08,387 --> 00:18:10,622 or you take a picture of the back of the eye, 287 00:18:10,656 --> 00:18:14,493 you will see lots of those bleeding spots in the retina. 288 00:18:21,533 --> 00:18:25,136 Today, the doctors are not enough to do the screening. 289 00:18:25,169 --> 00:18:27,071 We are very limited ophthalmologists, 290 00:18:27,105 --> 00:18:28,673 so there should be other ways 291 00:18:28,706 --> 00:18:31,043 where you can screen the diabetic patients 292 00:18:31,076 --> 00:18:32,644 for diabetic complications. 293 00:18:32,678 --> 00:18:33,845 In the US, 294 00:18:33,878 --> 00:18:37,148 there are about 74 eye doctors for every million people. 295 00:18:37,181 --> 00:18:39,751 In India, there are only 11. 296 00:18:39,784 --> 00:18:42,587 So just keeping up with the sheer number of patients, 297 00:18:42,620 --> 00:18:45,423 let alone giving them the attention and care they need, 298 00:18:45,457 --> 00:18:48,360 is overwhelming, if not impossible. 299 00:18:48,393 --> 00:18:52,264 We probably see about 2,000 to 2,500 patients 300 00:18:52,297 --> 00:18:53,765 every single day. 301 00:18:53,798 --> 00:18:56,201 The interesting thing with diabetic retinopathy 302 00:18:56,234 --> 00:18:59,338 is there are ways to screen and get ahead of the problem. 303 00:18:59,371 --> 00:19:02,340 The challenge is that not enough patients undergo screening. 304 00:19:02,373 --> 00:19:05,310 Like Tim Shaw's ALS speech recognizer, 305 00:19:05,343 --> 00:19:08,012 this problem is also about data, 306 00:19:08,046 --> 00:19:09,648 or lack of it. 307 00:19:09,681 --> 00:19:13,118 To prevent more people from experiencing vision loss, 308 00:19:13,152 --> 00:19:16,221 Dr. Kim wanted to get ahead of the problem. 309 00:19:19,958 --> 00:19:21,660 So there's a hemorrhage. 310 00:19:21,693 --> 00:19:24,662 All these are exudates. 311 00:19:24,696 --> 00:19:27,499 Dr. Kim called up a team at Google. 312 00:19:27,532 --> 00:19:29,267 Made up of doctors and engineers, 313 00:19:29,301 --> 00:19:31,536 they're exploring ways to use machine learning 314 00:19:31,570 --> 00:19:35,373 to solve some of the world's leading healthcare problems. 315 00:19:35,407 --> 00:19:39,110 So we started with could we train an AI model 316 00:19:39,144 --> 00:19:41,780 that can somehow help read these images, 317 00:19:41,813 --> 00:19:45,850 that can decrease the number of doctors required 318 00:19:45,883 --> 00:19:47,018 to do this task. 319 00:19:47,051 --> 00:19:48,587 So this is the normal view. 320 00:19:48,620 --> 00:19:50,522 When you start looking more deeply, 321 00:19:50,555 --> 00:19:53,225 then this can be a microaneurysm, right? 322 00:19:53,258 --> 00:19:55,460 - This one here? - Could be. 323 00:19:55,493 --> 00:19:58,296 The team uses the same kind of machine learning 324 00:19:58,329 --> 00:20:00,665 that allows us to organize our photos 325 00:20:00,699 --> 00:20:02,667 or tag friends on social media, 326 00:20:02,701 --> 00:20:05,437 image recognition. 327 00:20:05,470 --> 00:20:07,005 First, models are trained 328 00:20:07,038 --> 00:20:10,442 using tagged images of things like cats or dogs. 329 00:20:10,475 --> 00:20:12,543 After looking at thousands of examples, 330 00:20:12,577 --> 00:20:15,747 the algorithm learns to identify new images 331 00:20:15,780 --> 00:20:17,382 without any human help. 332 00:20:17,415 --> 00:20:19,785 For the retinopathy project, 333 00:20:19,818 --> 00:20:21,886 over 100,000 eye scans 334 00:20:21,920 --> 00:20:23,421 were graded by eye doctors 335 00:20:23,455 --> 00:20:26,958 who rated each eye scan on a scale from one to five, 336 00:20:26,991 --> 00:20:28,560 from healthy to diseased. 337 00:20:28,593 --> 00:20:30,228 These images were then used 338 00:20:30,261 --> 00:20:32,564 to train a machine learning algorithm. 339 00:20:32,597 --> 00:20:35,199 Over time, the AI learned to predict 340 00:20:35,233 --> 00:20:37,835 which eyes showed signs of disease. 341 00:20:37,869 --> 00:20:40,238 This is the assistant's view 342 00:20:40,272 --> 00:20:41,706 where the model's predictions 343 00:20:41,740 --> 00:20:45,243 are actually projected on the original image, 344 00:20:45,277 --> 00:20:49,047 and it's picking up the pathologies very nicely. 345 00:20:49,080 --> 00:20:51,649 To get help implementing the technology, 346 00:20:51,682 --> 00:20:53,718 Lily's team reached out to Verily, 347 00:20:53,751 --> 00:20:56,287 the life sciences unit at Alphabet. 348 00:20:56,320 --> 00:20:58,089 So, how was India? 349 00:20:58,123 --> 00:20:59,357 Oh, amazing! 350 00:20:59,391 --> 00:21:01,226 Verily came out of Google X, 351 00:21:01,259 --> 00:21:05,296 and we sit at the intersection of technology, life science, and healthcare. 352 00:21:05,330 --> 00:21:07,532 What we try to do is think about big problems 353 00:21:07,565 --> 00:21:09,300 that are affecting many patients, 354 00:21:09,333 --> 00:21:11,436 and how can we bring the best tools 355 00:21:11,469 --> 00:21:12,603 and best technologies 356 00:21:12,637 --> 00:21:14,205 to get ahead of the problems. 357 00:21:14,239 --> 00:21:17,709 The technical pieces are so important, and so is the methodology. 358 00:21:17,743 --> 00:21:19,644 How do you capture the right image, 359 00:21:19,677 --> 00:21:21,412 and how does the algorithm work, 360 00:21:21,446 --> 00:21:24,015 and how do you deploy these tools not only here, 361 00:21:24,048 --> 00:21:25,550 but in rural conditions? 362 00:21:25,583 --> 00:21:28,186 If we can speed up this diagnosis process 363 00:21:28,219 --> 00:21:30,088 and augment the clinical care, 364 00:21:30,121 --> 00:21:31,990 then we can prevent blindness. 365 00:21:32,023 --> 00:21:36,128 There aren't many other bigger problems that affect more patients. 366 00:21:36,161 --> 00:21:39,030 Diabetes affects 400 million worldwide, 367 00:21:39,063 --> 00:21:41,433 70 million in India alone, 368 00:21:41,466 --> 00:21:44,502 which is why Jessica and Lily's teams 369 00:21:44,536 --> 00:21:47,405 began testing AI-enabled eye scanners there, 370 00:21:47,439 --> 00:21:49,374 in its most rural areas, 371 00:21:49,407 --> 00:21:52,276 like Dr. Kim's Aravind Eye Clinics. 372 00:21:52,310 --> 00:21:55,146 - Is the camera on? - Now it's on. 373 00:21:55,179 --> 00:21:56,147 Yeah. 374 00:21:56,180 --> 00:21:57,415 So once the camera is up, 375 00:21:57,449 --> 00:21:59,484 we need to check network connectivity. 376 00:21:59,517 --> 00:22:01,419 The patient comes in. 377 00:22:01,452 --> 00:22:03,521 They get pictures of the back of the eye. 378 00:22:03,554 --> 00:22:05,389 One for the left eye, and right eye. 379 00:22:05,423 --> 00:22:07,992 The images are uploaded to this algorithm, 380 00:22:08,026 --> 00:22:10,561 and once the algorithm performs its analysis, 381 00:22:10,595 --> 00:22:13,364 it sends the results back to the system, 382 00:22:13,397 --> 00:22:15,734 along with a referral recommendation. 383 00:22:15,767 --> 00:22:17,936 It's good. It's up and running. 384 00:22:17,969 --> 00:22:20,705 Because the algorithm works in real time, 385 00:22:20,739 --> 00:22:23,442 you can get a real-time answer to a doctor, 386 00:22:23,475 --> 00:22:26,445 and that real-time answer comes back to the patient. 387 00:22:30,382 --> 00:22:32,250 Once you have the algorithm, 388 00:22:32,284 --> 00:22:34,853 it's like taking your weight measurement. 389 00:22:34,886 --> 00:22:36,054 Within a few seconds, 390 00:22:36,087 --> 00:22:39,358 the system tells you whether you have retinopathy or not. 391 00:22:40,391 --> 00:22:41,793 In the past, 392 00:22:41,826 --> 00:22:44,462 Santhi's condition could've taken months to diagnose, 393 00:22:44,495 --> 00:22:45,864 if diagnosed at all. 394 00:23:20,131 --> 00:23:23,501 By the time an eye doctor would've been able to see her, 395 00:23:23,535 --> 00:23:26,671 Santhi's diabetes might have caused her to go blind. 396 00:23:27,705 --> 00:23:29,841 Now, with the help of new technology, 397 00:23:29,874 --> 00:23:31,376 it's immediate, 398 00:23:31,409 --> 00:23:33,378 and she can take the hour-long bus ride 399 00:23:33,411 --> 00:23:35,580 to Dr. Kim's clinic in Madurai 400 00:23:35,613 --> 00:23:37,149 for same-day treatment. 401 00:23:49,560 --> 00:23:51,163 Now thousands of patients 402 00:23:51,196 --> 00:23:53,631 who may have waited weeks or months to be seen 403 00:23:53,664 --> 00:23:56,601 can get the help they need before it's too late. 404 00:24:11,115 --> 00:24:12,951 Thank you, sir. 405 00:24:27,598 --> 00:24:29,033 Retinopathy 406 00:24:29,067 --> 00:24:32,003 is when high blood sugar damages the retina. 407 00:24:32,037 --> 00:24:34,639 Blood leaks, and the laser treatment 408 00:24:34,672 --> 00:24:36,541 basically "welds" the blood vessels 409 00:24:36,574 --> 00:24:38,777 to stop the leakage. 410 00:24:38,810 --> 00:24:41,746 Routine eye exams can spot the problem early. 411 00:24:41,779 --> 00:24:44,048 In rural or remote areas, like here, 412 00:24:44,082 --> 00:24:48,753 AI can step in and be that early detection system. 413 00:24:57,662 --> 00:25:01,466 I think one of the most important applications of AI. 414 00:25:01,499 --> 00:25:04,736 is in places where doctors are scarce. 415 00:25:04,769 --> 00:25:08,038 In a way, what AI does is make intelligence cheap, 416 00:25:08,072 --> 00:25:11,509 and now imagine what you can do when you make intelligence cheap. 417 00:25:11,543 --> 00:25:13,978 People can go to doctors they couldn't before. 418 00:25:14,012 --> 00:25:16,848 It may not be the impact that catches the most headlines, 419 00:25:16,881 --> 00:25:20,218 but in many ways it'll be the most important impact. 420 00:25:40,238 --> 00:25:42,774 AI now is this next generation of tools 421 00:25:42,807 --> 00:25:45,510 that we can apply to clinically meaningful problems, 422 00:25:45,543 --> 00:25:49,314 so AI really starts to democratize healthcare. 423 00:25:56,521 --> 00:25:59,090 The work with diabetic retinopathy 424 00:25:59,123 --> 00:26:02,260 is opening our eyes to so much potential. 425 00:26:03,394 --> 00:26:05,230 Even within these images, 426 00:26:05,263 --> 00:26:07,565 we're starting to see some interesting signals 427 00:26:07,598 --> 00:26:11,002 that might tell us about someone's risk factors for heart disease. 428 00:26:11,035 --> 00:26:13,070 And from there, you start to think about 429 00:26:13,104 --> 00:26:16,274 all of the images that we collect in medicine. 430 00:26:16,307 --> 00:26:18,609 Can you use AI or an algorithm 431 00:26:18,642 --> 00:26:20,745 to help patients and doctors 432 00:26:20,778 --> 00:26:23,214 get ahead of a given diagnosis? 433 00:26:23,248 --> 00:26:26,750 Take cancer as an example of how AI can help save lives. 434 00:26:26,784 --> 00:26:29,120 We could take a sample of somebody's blood 435 00:26:29,153 --> 00:26:31,489 and look for the minuscule amounts of cancer DNA 436 00:26:31,523 --> 00:26:36,561 or tumor DNA in that blood. This is a great application for machine learning. 437 00:26:36,594 --> 00:26:38,363 And why stop there? 438 00:26:38,396 --> 00:26:39,998 Could AI accomplish 439 00:26:40,031 --> 00:26:42,767 what human researchers have not yet been able to? 440 00:26:42,800 --> 00:26:45,136 Figuring out how cells work well enough 441 00:26:45,170 --> 00:26:48,506 that you can understand why a tumor grows and how to stop it 442 00:26:48,540 --> 00:26:50,475 without hurting the surrounding cells. 443 00:26:50,508 --> 00:26:52,677 And if cancer could be cured, 444 00:26:52,710 --> 00:26:54,445 maybe mental health disorders, 445 00:26:54,478 --> 00:26:56,081 like depression, or anxiety. 446 00:26:56,114 --> 00:26:58,682 There are facial and vocal biomarkers 447 00:26:58,716 --> 00:27:00,585 of these mental health disorders. 448 00:27:00,619 --> 00:27:03,521 People check their phones 15 times an hour. 449 00:27:03,555 --> 00:27:05,089 So that's an opportunity 450 00:27:05,123 --> 00:27:08,093 to almost do, like, a well-being checkpoint. 451 00:27:08,126 --> 00:27:10,462 You can flag that to the individual, 452 00:27:10,495 --> 00:27:11,562 to a loved one, 453 00:27:11,595 --> 00:27:14,365 or in some cases even to a doctor. 454 00:27:14,399 --> 00:27:17,368 If you look at the overall field of medicine, 455 00:27:17,401 --> 00:27:21,939 how do you do a great job of diagnosing illness? 456 00:27:21,972 --> 00:27:24,575 Having artificial intelligence, 457 00:27:24,608 --> 00:27:27,946 the world's greatest diagnostician, helps. 458 00:27:31,616 --> 00:27:33,050 At Google, 459 00:27:33,084 --> 00:27:35,386 Julie and the Euphonia team have been working for months 460 00:27:35,419 --> 00:27:38,323 trying to find a way for former NFL star Tim Shaw 461 00:27:38,356 --> 00:27:39,958 to get his voice back. 462 00:27:47,665 --> 00:27:50,167 Yes! So Zach's team, the DeepMind team, 463 00:27:50,201 --> 00:27:53,738 has built a model that can imitate your voice. 464 00:27:53,771 --> 00:27:55,073 For Tim, we were lucky, 465 00:27:55,106 --> 00:27:59,043 because, you know, Tim has a career of NFL player, 466 00:27:59,077 --> 00:28:02,613 so he did multiple radio interviews and TV interviews, 467 00:28:02,647 --> 00:28:04,349 so he sent us this footage. 468 00:28:04,382 --> 00:28:07,685 Hey, this is Tim Shaw, special teams animal. 469 00:28:07,719 --> 00:28:10,721 Christmas is coming, so we need to find out 470 00:28:10,754 --> 00:28:12,356 what the Titans players are doing. 471 00:28:12,390 --> 00:28:14,859 If you gotta hesitate, that's probably a "no." 472 00:28:14,892 --> 00:28:17,261 Tim will be able to type what he wants, 473 00:28:17,294 --> 00:28:21,565 and the prototype will say it in Tim's voice. 474 00:28:21,598 --> 00:28:22,967 I've always loved attention. 475 00:28:23,000 --> 00:28:24,936 Don't know if you know that about me. 476 00:28:24,969 --> 00:28:26,904 She's gonna shave it for you. 477 00:28:26,937 --> 00:28:29,073 Interpreting speech is one thing, 478 00:28:29,106 --> 00:28:31,575 but re-creating the way a real person sounds 479 00:28:31,608 --> 00:28:33,311 is an order of magnitude harder. 480 00:28:33,344 --> 00:28:36,280 Playing Tecmo Bowl, eating Christmas cookies, and turkey. 481 00:28:36,313 --> 00:28:40,151 Voice imitation is also known as voice synthesis, 482 00:28:40,185 --> 00:28:43,220 which is basically speech recognition in reverse. 483 00:28:43,253 --> 00:28:46,925 First, machine learning converts text back into waveforms. 484 00:28:46,958 --> 00:28:50,061 These waveforms are then used to create sound. 485 00:28:50,094 --> 00:28:53,965 This is how Alexa and Google Home are able to talk to us. 486 00:28:53,998 --> 00:28:56,734 Now the teams from DeepMind and Google AI 487 00:28:56,768 --> 00:28:58,336 are working to create a model 488 00:28:58,369 --> 00:29:01,739 to imitate the unique sound of Tim's voice. 489 00:29:01,773 --> 00:29:03,841 Looks like it's computing. 490 00:29:03,874 --> 00:29:05,943 But it worked this morning? 491 00:29:05,976 --> 00:29:08,012 We have to set expectations quite low. 492 00:29:08,045 --> 00:29:11,115 I don't know how our model is going to perform. 493 00:29:11,149 --> 00:29:13,851 I hope that Tim will understand 494 00:29:13,884 --> 00:29:16,287 and actually see the technology for what it is, 495 00:29:16,320 --> 00:29:20,692 which is a work in progress and a research project. 496 00:29:20,725 --> 00:29:25,997 After six months of waiting, Tim Shaw is about to find out. 497 00:29:26,030 --> 00:29:28,365 The team working on his speech recognition model 498 00:29:28,399 --> 00:29:31,102 is coming to his house for a practice run. 499 00:29:39,009 --> 00:29:41,079 Good girl, come say hello. 500 00:29:41,913 --> 00:29:43,548 - Hi! - Oh, hi! 501 00:29:43,581 --> 00:29:44,749 Welcome. 502 00:29:44,782 --> 00:29:45,850 - Hi! - Come in. 503 00:29:45,883 --> 00:29:47,752 Thanks for having us. 504 00:29:47,785 --> 00:29:49,921 He's met some of you before, right? 505 00:29:49,954 --> 00:29:51,823 How are you doing, Tim? 506 00:29:51,856 --> 00:29:53,591 - Hi, Tim. - Good to see you. 507 00:29:53,625 --> 00:29:55,259 - Hello. - Hello. 508 00:29:55,293 --> 00:29:56,327 Hi. 509 00:29:56,361 --> 00:29:59,097 Hi, I'm Julie. We saw each other on the camera. 510 00:30:01,065 --> 00:30:03,401 It's warmer here than it is in Boston. 511 00:30:03,434 --> 00:30:05,202 As it should be. 512 00:30:09,340 --> 00:30:10,408 Okay. 513 00:30:11,408 --> 00:30:13,377 Lead the way, Tim. 514 00:30:13,410 --> 00:30:16,714 I'm excited to share with Tim and his parents 515 00:30:16,747 --> 00:30:18,249 what we've been working on. 516 00:30:18,283 --> 00:30:21,018 I'm a little bit nervous. I don't know if the app 517 00:30:21,052 --> 00:30:24,155 is going to behave the way we hope it will behave, 518 00:30:24,188 --> 00:30:27,792 but I'm also very excited, to learn new things 519 00:30:27,825 --> 00:30:29,260 and to hear Tim's feedback. 520 00:30:29,293 --> 00:30:32,063 So I brought two versions with me. 521 00:30:32,096 --> 00:30:34,899 I was supposed to pick, but I decided to just bring both 522 00:30:34,932 --> 00:30:37,335 just in case one is better than the other, 523 00:30:37,368 --> 00:30:38,803 and, just so you know, 524 00:30:38,836 --> 00:30:41,639 this one here was trained 525 00:30:41,672 --> 00:30:44,075 only using recordings of your voice, 526 00:30:44,108 --> 00:30:47,445 and this one here was trained using recordings of your voice, 527 00:30:47,478 --> 00:30:51,215 and also from other participants from ALS TDI 528 00:30:51,248 --> 00:30:54,685 who went through the same exercise of... 529 00:30:54,718 --> 00:30:57,388 So, okay. 530 00:30:57,422 --> 00:30:59,924 I was hoping we could give them a try. 531 00:30:59,958 --> 00:31:02,260 Are we ready? 532 00:31:03,995 --> 00:31:06,965 Who are you talking about? 533 00:31:15,206 --> 00:31:16,307 It got it. 534 00:31:16,340 --> 00:31:18,243 It got it. 535 00:31:24,582 --> 00:31:27,451 Is he coming? 536 00:31:29,554 --> 00:31:30,889 Yes. 537 00:31:33,991 --> 00:31:36,961 Are you working today? 538 00:31:46,037 --> 00:31:48,873 It's wonderful. 539 00:31:48,906 --> 00:31:49,907 Cool. 540 00:31:49,941 --> 00:31:51,909 Thank you for trying this. 541 00:31:51,943 --> 00:31:53,044 - Wow! - It's fabulous. 542 00:31:53,077 --> 00:31:55,279 What I love, it made mistakes, 543 00:31:55,312 --> 00:31:57,749 - and then it corrected itself. - Yeah. 544 00:31:57,782 --> 00:31:59,650 I was watching it like, "That's not it," 545 00:31:59,683 --> 00:32:02,686 - and then it went... - Then it does it right. 546 00:32:02,720 --> 00:32:04,088 These were phrases, 547 00:32:04,122 --> 00:32:06,890 part of the 70% that we actually used 548 00:32:06,924 --> 00:32:08,192 to train the model, 549 00:32:08,225 --> 00:32:11,795 but we also set aside 30% of the phrases, 550 00:32:11,828 --> 00:32:14,431 so this might not do as well, 551 00:32:14,465 --> 00:32:18,102 but I was hoping that we could try some of these too. 552 00:32:18,136 --> 00:32:20,070 So what we've already done 553 00:32:20,104 --> 00:32:23,474 is him using phrases that were used to train the app. 554 00:32:23,508 --> 00:32:24,608 That's right. 555 00:32:24,641 --> 00:32:27,578 Now we're trying to see if it can recognize phrases 556 00:32:27,611 --> 00:32:30,915 - that weren't part of that. - Yes, that's right. 557 00:32:30,948 --> 00:32:32,383 So let's give it a try? 558 00:32:33,917 --> 00:32:36,387 Do you want me to? 559 00:32:39,656 --> 00:32:43,061 Do you have the time to play? 560 00:32:47,965 --> 00:32:50,501 What happens afterwards? 561 00:32:53,270 --> 00:32:55,840 Huh. So, on the last one, 562 00:32:55,873 --> 00:32:58,242 this one got it, and this one didn't. 563 00:32:58,275 --> 00:33:02,546 - We'll pause it. So... - I love the first one, where it says, 564 00:33:02,580 --> 00:33:04,782 - "Can you help me take a shower?" 565 00:33:04,815 --> 00:33:07,919 - That's not at all what he said. - -I know, 566 00:33:07,952 --> 00:33:10,487 you've gotta be really careful what you ask for. 567 00:33:12,156 --> 00:33:15,559 So if, when it's interpreting his voice, 568 00:33:15,592 --> 00:33:17,295 and it makes some errors, 569 00:33:17,328 --> 00:33:19,263 is there a way we can correct it? 570 00:33:19,296 --> 00:33:22,066 Yeah. We want to add the option 571 00:33:22,099 --> 00:33:24,335 for you guys to fix the recordings, 572 00:33:24,368 --> 00:33:27,104 but as of today, because this is the very first time 573 00:33:27,137 --> 00:33:28,706 we actually tried this, 574 00:33:28,739 --> 00:33:30,474 we don't have it yet. 575 00:33:30,508 --> 00:33:32,910 This is still a work in progress. 576 00:33:32,943 --> 00:33:34,811 We have a speech recognition model 577 00:33:34,845 --> 00:33:36,680 that works for Tim Shaw, 578 00:33:36,714 --> 00:33:38,449 which is, you know, one person, 579 00:33:38,482 --> 00:33:41,185 and we're really hoping that, you know, 580 00:33:41,218 --> 00:33:43,254 this technology can work for many people. 581 00:33:43,287 --> 00:33:46,090 There's something else I want you to try, 582 00:33:46,123 --> 00:33:47,525 if that's okay? 583 00:33:47,558 --> 00:33:50,761 We're working with another team at Google called DeepMind. 584 00:33:50,794 --> 00:33:55,366 They're specialized in voice imitation and synthesis. 585 00:33:56,500 --> 00:33:57,468 In 2019, 586 00:33:57,501 --> 00:34:00,605 Tim wrote a letter to his younger self. 587 00:34:02,540 --> 00:34:05,776 They are words written by a 34-year-old man with ALS 588 00:34:05,810 --> 00:34:08,145 who has trouble communicating 589 00:34:08,179 --> 00:34:09,513 sent back in time 590 00:34:09,547 --> 00:34:15,252 to a 22-year-old on the cusp of NFL greatness. 591 00:34:15,285 --> 00:34:18,389 So let me give this a try. 592 00:34:18,422 --> 00:34:21,892 I just like using this letter because it's just so beautiful, 593 00:34:21,925 --> 00:34:24,729 so let me see if this is gonna work. 594 00:34:27,231 --> 00:34:30,567 So, I've decided to write you this letter 595 00:34:30,601 --> 00:34:32,636 'cause I have so much to tell you. 596 00:34:32,670 --> 00:34:33,905 I want to explain to you 597 00:34:33,938 --> 00:34:35,973 why it's so difficult for me to speak, 598 00:34:36,006 --> 00:34:37,974 the diagnosis, all of it, 599 00:34:38,008 --> 00:34:39,443 and what my life is like now, 600 00:34:39,476 --> 00:34:41,712 'cause one day, you will be in my shoes, 601 00:34:41,745 --> 00:34:44,348 living with the same struggles. 602 00:34:57,961 --> 00:35:01,599 It's his voice, that I'd forgotten. 603 00:35:14,945 --> 00:35:16,347 We do. 604 00:36:05,062 --> 00:36:06,998 We're so happy to be working with you. 605 00:36:07,031 --> 00:36:08,466 It's really an honor. 606 00:36:12,536 --> 00:36:14,938 The thought that one day, 607 00:36:14,971 --> 00:36:18,409 that can be linked with this, 608 00:36:18,442 --> 00:36:20,310 and when you speak as you are now, 609 00:36:20,344 --> 00:36:22,947 it will sound like that, is... 610 00:36:29,319 --> 00:36:31,689 It's okay. We'll wait. 611 00:36:33,057 --> 00:36:34,758 There is a lot of unknown 612 00:36:34,791 --> 00:36:37,861 and still a lot of research to be conducted. 613 00:36:37,894 --> 00:36:41,198 We're really trying to have a proof of concept first, 614 00:36:41,232 --> 00:36:44,268 and then expand to not only people who have ALS, 615 00:36:44,302 --> 00:36:45,669 but people who had a stroke, 616 00:36:45,702 --> 00:36:48,539 or a traumatic brain injury, multiple sclerosis, 617 00:36:48,572 --> 00:36:51,008 any types of neurologic conditions. 618 00:36:51,041 --> 00:36:52,776 Maybe other languages, too, you know? 619 00:36:52,809 --> 00:36:56,413 I would really like this to work for French, for example. 620 00:36:56,446 --> 00:36:58,683 Wouldn't it be a wonderful opportunity 621 00:36:58,716 --> 00:37:01,518 to bring technology to problems that we're solving 622 00:37:01,551 --> 00:37:03,253 in life science and healthcare, 623 00:37:03,286 --> 00:37:05,222 and in fact, it's a missed opportunity 624 00:37:05,256 --> 00:37:07,892 if we don't try to bring the best technologies 625 00:37:07,925 --> 00:37:09,126 to help people. 626 00:37:09,159 --> 00:37:10,728 This is really just the beginning. 627 00:37:10,761 --> 00:37:13,097 Just the beginning indeed. 628 00:37:13,130 --> 00:37:15,332 Imagine the possibilities. 629 00:37:15,365 --> 00:37:19,169 I think in the imaginable future for AI and healthcare 630 00:37:19,202 --> 00:37:22,139 is that there is no healthcare anymore, 631 00:37:22,173 --> 00:37:24,008 because nobody needs it. 632 00:37:24,041 --> 00:37:27,744 You could have an AI that is directly talking to your immune system, 633 00:37:27,778 --> 00:37:30,480 and is actually preemptively creating the antibodies 634 00:37:30,514 --> 00:37:32,917 for the epidemics that are coming your way, 635 00:37:32,950 --> 00:37:34,184 and will not be stopped. 636 00:37:34,218 --> 00:37:37,254 This will not happen tomorrow, but it's the long-term goal 637 00:37:37,288 --> 00:37:38,756 that we can point towards. 638 00:37:43,827 --> 00:37:46,163 Tim had never heard his own words 639 00:37:46,196 --> 00:37:48,799 read out loud before today. 640 00:37:48,832 --> 00:37:51,568 Neither had his parents. 641 00:37:51,602 --> 00:37:54,939 Every single day is a struggle for me. 642 00:37:54,972 --> 00:37:58,075 I can barely move my arms. 643 00:37:58,109 --> 00:38:00,410 Have fun. 644 00:38:00,443 --> 00:38:02,346 I can't walk on my own, 645 00:38:02,379 --> 00:38:06,550 so I recently started using a wheelchair. 646 00:38:07,551 --> 00:38:10,454 I have trouble chewing and swallowing. 647 00:38:10,487 --> 00:38:12,857 I'd kill for a good pork chop. 648 00:38:12,890 --> 00:38:16,794 Yes, my body is failing, but my mind is not giving up. 649 00:38:18,262 --> 00:38:20,430 Find what's most important in your life, 650 00:38:20,464 --> 00:38:22,566 and live for that. 651 00:38:24,034 --> 00:38:26,937 Don't let three letters, NFL, 652 00:38:26,970 --> 00:38:28,372 define you... 653 00:38:31,809 --> 00:38:36,514 ...the same way I refuse to let three letters define me. 654 00:38:43,254 --> 00:38:45,455 One of the things Tim has taught us, 655 00:38:45,489 --> 00:38:48,925 and I think it's a lesson for everyone... 656 00:38:48,959 --> 00:38:51,962 Medically speaking, Tim's life has an end to it. 657 00:38:51,995 --> 00:38:55,966 In fact, five years ago we were told he only had two to five years left. 658 00:38:55,999 --> 00:38:57,368 We're already past that. 659 00:38:58,735 --> 00:39:01,638 He has learned very quickly 660 00:39:01,672 --> 00:39:05,442 that today is the day that we have, 661 00:39:05,475 --> 00:39:08,579 and we can ruin today 662 00:39:08,612 --> 00:39:10,314 by thinking about yesterday 663 00:39:10,347 --> 00:39:12,082 and how wonderful it used to be, 664 00:39:12,116 --> 00:39:16,053 and, "Oh, woe is me," and "I wish it was like that." 665 00:39:16,086 --> 00:39:17,521 We can also ruin today 666 00:39:17,554 --> 00:39:19,222 by looking into the future, 667 00:39:19,256 --> 00:39:20,290 and in Tim's case, 668 00:39:20,323 --> 00:39:22,425 how horrible this is going to be. 669 00:39:22,459 --> 00:39:23,794 "This is going to happen," 670 00:39:23,827 --> 00:39:25,996 "I won't be able to do this anymore." 671 00:39:26,029 --> 00:39:28,665 So if we go either of those directions, 672 00:39:28,698 --> 00:39:31,068 it spoils us from being present today. 673 00:39:31,101 --> 00:39:32,569 That's a lesson for all of us. 674 00:39:32,602 --> 00:39:35,406 Whether we have an ALS diagnosis or not, 675 00:39:35,439 --> 00:39:38,241 try to see the good and the blessing of every day. 676 00:39:38,275 --> 00:39:40,644 You're here with us today. 677 00:39:40,677 --> 00:39:42,747 It's going to be a good day.