WEBVTT 00:04.079 --> 00:06.190 People imagine a robot sprinting into 00:06.190 --> 00:08.412 combat , pulling out wounded soldiers , 00:08.449 --> 00:10.616 or racing through a collapsed building 00:10.616 --> 00:12.838 to rescue civilians , but that's really 00:12.838 --> 00:15.060 in the far future . Today we're focused 00:15.060 --> 00:17.282 on the first step , teaching AI to find 00:17.282 --> 00:19.338 casualties and assess their status , 00:19.338 --> 00:21.560 always under meaningful human control . 00:21.560 --> 00:23.727 That's what the DARPA triage challenge 00:23.727 --> 00:25.671 is really about , not robots , but 00:25.671 --> 00:28.059 creating life saving tools that work in 00:28.059 --> 00:30.540 chaos . Communication is one of the 00:30.540 --> 00:32.919 hardest things in an emergency response . 00:33.139 --> 00:35.540 And if the robots are not able to 00:35.540 --> 00:38.500 communicate well and clearly to the 00:38.500 --> 00:41.060 medic , the information that they need . 00:41.430 --> 00:43.909 It will not help them at all . The 00:43.909 --> 00:46.020 technology being developed here , the 00:46.020 --> 00:48.131 AI that can rapidly assess injuries , 00:48.139 --> 00:50.229 prioritize patients , and support 00:50.229 --> 00:52.630 overwhelmed medics is already 00:52.630 --> 00:54.709 transformative . Instead of making 00:54.709 --> 00:56.653 those quick decisions on the fly , 00:56.653 --> 00:58.820 especially when you know , like if you 00:58.820 --> 00:58.750 have patients that are already deceased , 00:58.869 --> 01:00.980 you'll be able to do a quick check on 01:00.980 --> 01:02.980 them and be like , OK , yeah , that 01:02.980 --> 01:04.758 category is correct . The robot 01:04.758 --> 01:04.389 definitely pulled me out of tunnel 01:04.389 --> 01:06.611 vision at one point . It would call out 01:06.611 --> 01:08.722 to us like , medic , medic , I have a 01:08.722 --> 01:10.889 critically injured patient . So at one 01:10.889 --> 01:13.111 point , I'm treating and then I hear it 01:13.111 --> 01:15.167 say some words and I'll just look up 01:15.167 --> 01:14.470 real quick like I'm still working on 01:14.470 --> 01:16.692 this guy , but I have someone else that 01:16.692 --> 01:18.859 I'm going to have to go to immediately 01:18.859 --> 01:21.081 afterwards . They were very accurate in 01:21.081 --> 01:20.709 pinpointing where our patients were . 01:20.910 --> 01:23.190 They were very accurate in calling out 01:23.190 --> 01:25.790 like if this person needed help . Today , 01:26.010 --> 01:27.788 there's virtually no artificial 01:27.788 --> 01:29.788 intelligence designed for frontline 01:29.788 --> 01:31.889 medical triage . The DARPA triage 01:31.889 --> 01:33.790 challenge is closing that gap , 01:34.169 --> 01:36.489 building AI systems that can process 01:36.489 --> 01:39.199 vital signs , spot critical injuries , 01:39.370 --> 01:41.690 and give medics decision support when 01:41.690 --> 01:44.319 every second counts . So we really 01:44.319 --> 01:46.541 think that one of the benefits of using 01:46.541 --> 01:48.839 AI in an austere environment and you 01:48.839 --> 01:50.783 know , potentially a mass casualty 01:50.783 --> 01:52.950 incident is that it's not going to get 01:52.950 --> 01:54.950 overwhelmed and it'll give the same 01:55.160 --> 01:57.216 level of fidelity of prediction each 01:57.216 --> 01:59.216 time . And we hope that some of the 01:59.216 --> 02:01.438 interpretive tools we develop will help 02:01.438 --> 02:03.160 the medics more consistently , 02:03.160 --> 02:05.327 efficiently , and to save more lives . 02:05.327 --> 02:07.327 We're looking at , you know , being 02:07.327 --> 02:09.216 able to actually point the medics 02:09.216 --> 02:11.216 towards where they're most needed . 02:11.320 --> 02:14.619 We're looking at it from a very removed 02:14.619 --> 02:17.679 and mechanical way , uh , which is 02:17.679 --> 02:19.960 extraordinarily hard to do , uh , 02:20.119 --> 02:23.520 during a mass casualty event , and that 02:23.520 --> 02:27.300 kind of makes these systems perfect at 02:27.639 --> 02:30.199 this particular job . In healthcare 02:30.199 --> 02:32.639 today , AI is helping to read X-rays , 02:32.759 --> 02:34.639 pathology reports , and analyze 02:34.639 --> 02:36.695 hospital data . But when it comes to 02:36.695 --> 02:38.820 the chaotic mass casualty of triage , 02:39.240 --> 02:41.018 there's a void . We're building 02:41.018 --> 02:42.907 technologies for the world's most 02:42.907 --> 02:45.129 challenging casualty events , and those 02:45.129 --> 02:47.184 technologies are going to accelerate 02:47.184 --> 02:49.018 progress for the entire field of 02:49.018 --> 02:49.440 disaster medicine .