WEBVTT 00:00.533 --> 00:03.700 (audience applauding) 00:11.370 --> 00:13.500 - I hope everyone had a good lunch or is busy 00:13.500 --> 00:15.343 finishing up an excellent lunch. 00:18.800 --> 00:21.593 I'm joined with two close friends of mine, 00:22.670 --> 00:25.020 and I'm probably the only person who can say this 00:25.020 --> 00:26.370 in the entire world. 00:26.370 --> 00:28.786 I work with and for both of them. 00:28.786 --> 00:29.670 (audience laughing) 00:29.670 --> 00:32.540 So I want to make sure I disclose my conflict of interest 00:32.540 --> 00:34.300 to start with. 00:34.300 --> 00:35.810 So General Shanahan went 00:35.810 --> 00:38.500 to Michigan - Go Blue 00:38.500 --> 00:42.380 - Go Blue, ROTC, entered the service of our country 00:42.380 --> 00:43.453 in 1984. 00:45.010 --> 00:46.900 He's been promoted a gazillion times, 00:46.900 --> 00:48.120 he was in charge of a whole bunch 00:48.120 --> 00:49.360 of intelligence activities, 00:49.360 --> 00:51.430 a whole bunch of operational activities, 00:51.430 --> 00:55.540 and eventually we needed somebody operationally 00:55.540 --> 00:58.220 to implement AI in the entire DoD 00:58.220 --> 01:01.720 and he was the perfect choice. 01:01.720 --> 01:04.370 So I work with him in my role as Chairman of the DIB. 01:05.310 --> 01:09.133 Kent Walker was a federal prosecutor. 01:10.346 --> 01:13.910 Law and order federal prosecutor who then chose to come 01:13.910 --> 01:15.370 to Silicon Valley, actually I think worked 01:15.370 --> 01:17.100 at eBay for awhile. 01:17.100 --> 01:20.540 And then we snagged him, we being Google, maybe 15 years 01:20.540 --> 01:21.730 ago, - Yeah, I guess 01:21.730 --> 01:24.150 - 15 years every day together 01:25.070 --> 01:27.870 and during that time not only did he set up our legal 01:27.870 --> 01:31.490 function but is now in charge of all global policy, 01:31.490 --> 01:33.120 PR, all those sorts of things together 01:33.120 --> 01:35.630 so very, very significant players. 01:35.630 --> 01:39.100 And what I thought we should do since you all have heard 01:39.100 --> 01:44.100 from me plenty, is simply start and perhaps Kent, 01:44.460 --> 01:46.440 we should just have you make some comments 01:46.440 --> 01:48.730 about the world as you see it today. 01:48.730 --> 01:51.250 - Sure, so thank you very much Eric, 01:51.250 --> 01:53.000 General Shanahan, it's a pleasure to be with you today 01:53.000 --> 01:54.530 and with all of you. 01:54.530 --> 01:58.450 The topic of today's panel, Private Public Partnerships, 01:58.450 --> 02:00.510 is extraordinarily important to me. 02:00.510 --> 02:03.990 I grew up in this community, my father was in the service 02:03.990 --> 02:06.870 for 24 years, I was born and spent the first years 02:06.870 --> 02:09.350 of my life on U.S. military bases. 02:09.350 --> 02:11.860 My father finished his career at Lockheed, 02:11.860 --> 02:14.700 so I feel a profound commitment to getting this right, 02:14.700 --> 02:17.816 to making sure that the private sector, the defense sector, 02:17.816 --> 02:22.090 and universities can work together in the best possible way. 02:22.090 --> 02:25.410 Before we jump in to the thoughts on how we can accomplish 02:25.410 --> 02:29.180 that, I wanted to take on two issues up front. 02:29.180 --> 02:32.610 It's been frustrating to hear concerns around 02:32.610 --> 02:35.650 our commitments to the national security and defense 02:35.650 --> 02:38.880 and so I wanted to set the record straight on two issues. 02:38.880 --> 02:40.360 First on China. 02:40.360 --> 02:43.750 In 2010 you may remember that Google was public 02:43.750 --> 02:46.360 about an attack on our infrastructure that originated 02:46.360 --> 02:50.930 in China, sophisticated cyber security attack. 02:50.930 --> 02:53.770 We learned a lot from that experience and while a number 02:53.770 --> 02:56.600 of our peer companies have significant commercial 02:56.600 --> 02:59.610 and AI operations in China, we have chosen to scope 02:59.610 --> 03:02.260 our operations there very carefully. 03:02.260 --> 03:06.110 Our focus is on advertising and on work supporting 03:06.110 --> 03:07.310 an open source platform. 03:08.150 --> 03:10.670 Second, with regard to the more general question 03:10.670 --> 03:13.210 of national security and our engagement 03:13.210 --> 03:17.770 in the Maven Project, it is an area where it's right 03:17.770 --> 03:20.410 that we decided to press the reset button 03:20.410 --> 03:22.750 until we had an opportunity to develop our own set 03:22.750 --> 03:25.040 of AI principles, our own work with regard 03:25.040 --> 03:27.760 to internal standards and review processes. 03:27.760 --> 03:30.680 But that was the decision focused on a discreet contract, 03:30.680 --> 03:32.995 not a broader statement about our willingness 03:32.995 --> 03:36.060 or history of working with the Department of Defense 03:36.060 --> 03:38.290 and the National Security Administration. 03:38.290 --> 03:42.270 We continue to do that, we are committed to doing that, 03:42.270 --> 03:45.200 and that work builds on a long tradition of work 03:45.200 --> 03:49.860 throughout the Valley on national security generally. 03:49.860 --> 03:53.300 It's important to remember that the history of the Valley 03:53.300 --> 03:55.910 in large measure builds on government technologies 03:55.910 --> 04:00.910 from radar to the internet to GPS to some of the work 04:00.940 --> 04:04.160 on autonomous vehicles and personal assistants 04:04.160 --> 04:05.760 you're seeing now. 04:05.760 --> 04:08.560 Just in the last couple of weeks we had an extraordinary 04:08.560 --> 04:11.430 accomplishment with regard to quantum supremacy 04:11.430 --> 04:13.730 which moved forward the frontiers of science 04:13.730 --> 04:16.570 and technology, but that was not an achievement 04:16.570 --> 04:17.920 by Google alone. 04:17.920 --> 04:20.220 It built on research that had been done 04:20.220 --> 04:22.320 at the University of California in Santa Barbara, 04:22.320 --> 04:24.620 it benefited from extensive consultation 04:24.620 --> 04:27.190 with research scientists at NASA, 04:27.190 --> 04:30.070 it was carried out in many ways on super computers 04:30.070 --> 04:31.900 from the Department of Energy. 04:31.900 --> 04:35.900 So those kinds of exchanges and collaborations 04:35.900 --> 04:39.470 are really key to what has made America technological 04:39.470 --> 04:42.700 innovation as successful as it's been. 04:42.700 --> 04:45.190 And just as we feel as we're contributing 04:45.190 --> 04:48.260 to the defense community, national security community, 04:48.260 --> 04:51.570 a lot of that work, that community, is a part of Google. 04:51.570 --> 04:54.000 We have lots of vets who work at Google. 04:54.000 --> 04:56.110 We go above and beyond to make sure that reservists 04:56.110 --> 04:58.800 working at Google can complete their military service 04:58.800 --> 05:01.860 while having thriving careers, and even our tools 05:01.860 --> 05:04.580 we have tried to take steps to make sure that vets 05:04.580 --> 05:08.620 transitioning to civilian life can make the best use 05:08.620 --> 05:11.433 of their military skills in the private sector. 05:12.530 --> 05:15.880 As we do that we also are fully engaged in a wide 05:15.880 --> 05:18.840 variety of work with different agencies. 05:18.840 --> 05:21.340 With the JAIC we are working on a number of national 05:21.340 --> 05:25.750 mission initiatives from cyber security to health care 05:25.750 --> 05:27.470 to business automation. 05:27.470 --> 05:31.360 With DARPA we are working on a number of fundamental 05:31.360 --> 05:34.710 projects to ensure the robustness of AI, 05:34.710 --> 05:38.770 to identify deep fakes and to progress work on the end 05:38.770 --> 05:42.990 of Moore's Law and how to progress the operation 05:42.990 --> 05:46.010 of hardware and use software/hardware interfaces 05:46.010 --> 05:47.480 in better ways. 05:47.480 --> 05:50.650 So as we take on those kinds of things we're eager 05:50.650 --> 05:54.040 to do more, we are pursuing actively additional 05:54.040 --> 05:57.400 certifications that will allow us to more fully engage 05:57.400 --> 06:00.200 across a range of these different topic areas. 06:00.200 --> 06:02.410 And we think that's extremely important. 06:02.410 --> 06:04.430 At the same time we think there's a great partnership 06:04.430 --> 06:07.450 to be had on the work that the DIB has announced 06:07.450 --> 06:10.120 in the last week, their AI principles, 06:10.120 --> 06:12.860 which were I thought very well done, 06:12.860 --> 06:15.140 it's a lengthy document but a thoughtful document. 06:15.140 --> 06:17.380 And really continues to work the groundwork 06:17.380 --> 06:20.230 that was laid by the Department of Defense back 06:20.230 --> 06:23.940 in I think it was 2012 with directive three thousand 06:23.940 --> 06:27.030 zero nine which talked about the use of human judgment 06:28.293 --> 06:29.850 in the application of advanced technologies, 06:29.850 --> 06:32.880 the charter of the JAIC, the work the DoD has done 06:32.880 --> 06:34.230 with its own AI principles. 06:35.242 --> 06:37.270 In the private sector, we too have been trying 06:37.270 --> 06:38.360 to drive forward on this. 06:38.360 --> 06:42.090 We have not only put out principles in very common 06:42.090 --> 06:44.430 and overlapping areas, there's a lot in common 06:44.430 --> 06:45.680 in these questions. 06:45.680 --> 06:50.680 Safety, human judgment, accountability, explainability, 06:50.970 --> 06:54.740 fairness are all critical areas where different 06:56.090 --> 06:58.260 actors in the space, each have different things 06:58.260 --> 07:01.430 to contribute and I think that's critically important. 07:01.430 --> 07:04.550 This is a shared responsibility to get this right. 07:04.550 --> 07:07.420 As the DIB report notes, we need a global framework, 07:07.420 --> 07:10.110 we need a global approach to these issues endorsing 07:10.110 --> 07:12.240 the OECD framework around these issues, 07:12.240 --> 07:15.940 extremely important and something that we want to support, 07:15.940 --> 07:17.540 and we are working together to figure out 07:17.540 --> 07:19.660 where are the complementarities. 07:19.660 --> 07:21.130 Because at the end of the day we are a proud 07:21.130 --> 07:24.010 American company, we are committed to the defense 07:24.010 --> 07:26.590 of the United States, our allies and the safety 07:26.590 --> 07:29.140 and security of the world, we are eager to continue 07:29.140 --> 07:32.500 this work and think about places we can work together 07:32.500 --> 07:34.250 to build on each other's strengths. 07:35.230 --> 07:37.100 - Well thank you Kent. 07:37.100 --> 07:39.930 General, take us through what you're up to 07:39.930 --> 07:41.045 at the JAIC. 07:41.045 --> 07:42.500 - Well so first of all let me say thanks, 07:42.500 --> 07:45.050 it is great to be here and I thank both Eric 07:45.050 --> 07:46.650 and Kent for the opportunity to do this. 07:46.650 --> 07:49.870 Admittedly though, I will say I'm a poor substitute 07:49.870 --> 07:51.570 for the Chairman of the Joint Chiefs of Staff, 07:51.570 --> 07:54.090 General Milley, although I say it's a lower probability 07:54.090 --> 07:55.800 of any headline-grabbing soundbites, 07:55.800 --> 07:57.770 so you get that with it. 07:57.770 --> 08:00.960 And I also confess this is undoubtedly the first 08:00.960 --> 08:03.470 and last time I will serve as a warmup act 08:03.470 --> 08:05.150 for Dr. Henry Kissinger 08:05.150 --> 08:07.391 (audience laughing) 08:07.391 --> 08:10.090 but hang on for the main event, as I say. 08:10.090 --> 08:12.800 I not only welcome but I relish the opportunity 08:12.800 --> 08:14.500 to have this broader conversation 08:14.500 --> 08:17.320 about public-private partnerships. 08:17.320 --> 08:19.700 When you ask me to reflect back on my two years 08:19.700 --> 08:22.620 as the Director of Project Maven, and just about 08:22.620 --> 08:25.650 a year in the seat as the director of the JAIC, 08:25.650 --> 08:28.310 there is one over-arching thing that continues 08:28.310 --> 08:30.840 to resonate strongly with me. 08:30.840 --> 08:34.750 It's the importance, and I would say the necessity, 08:34.750 --> 08:37.260 of strengthening bonds between government, 08:37.260 --> 08:39.203 industry and academia. 08:40.130 --> 08:42.927 This was said this morning, you brought it up 08:42.927 --> 08:44.630 and others had also mentioned it, 08:44.630 --> 08:48.350 is this idea this relationship should be depicted 08:48.350 --> 08:49.410 as a triangle. 08:49.410 --> 08:51.870 And it actually should be in the form 08:51.870 --> 08:56.260 of an equilateral triangle: government, academia, industry. 08:56.260 --> 08:59.920 I would suggest that that's largely the form it did take 08:59.920 --> 09:02.960 beginning in the 1950's and largely lasting 09:02.960 --> 09:05.050 until the early part of this decade. 09:05.050 --> 09:07.810 Walter Isaacson writes about this very eloquently 09:07.810 --> 09:10.970 and powerfully in his book, The Innovators. 09:10.970 --> 09:14.850 It is what really drove to Silicon Valley today. 09:14.850 --> 09:16.920 It's not the case today. 09:16.920 --> 09:19.880 At best the sides of the triangle are no longer 09:19.880 --> 09:23.976 equidistant, you might even say they are distorted 09:23.976 --> 09:27.350 or a little frayed in addition to being 09:27.350 --> 09:28.860 of different lengths. 09:28.860 --> 09:32.160 The reasons for that are complex and they're multi-fold. 09:32.160 --> 09:36.700 Snowden, Apple encryption, mismatched operating tempo 09:36.700 --> 09:39.810 and agility, different business models, general mistrust 09:39.810 --> 09:41.680 between the government and industry. 09:41.680 --> 09:44.220 We started talking past each other instead 09:44.220 --> 09:45.800 of with each other. 09:45.800 --> 09:49.410 The task is made much more difficult today by the fact 09:49.410 --> 09:52.180 that industry is moving so much faster than the 09:52.180 --> 09:54.640 Department of Defense, in fact the rest of government, 09:54.640 --> 09:58.040 when it comes to the adoption and integration of AI. 09:58.040 --> 10:00.220 We're playing perpetual catch-up. 10:00.220 --> 10:03.460 And some employees in the tech industry see no 10:03.460 --> 10:06.510 compelling reason to work with the Department of Defense 10:06.510 --> 10:08.560 and even those who want to work with DoD, 10:08.560 --> 10:11.560 which I say is far more than sometimes is portrayed, 10:11.560 --> 10:14.170 I say put everybody in this room in that category, 10:14.170 --> 10:17.490 we don't make it easy for them. 10:17.490 --> 10:20.070 So I would just reinforce some of the themes 10:20.070 --> 10:23.120 that are in the Security Commission's report 10:23.120 --> 10:25.450 or the interim report, and that is this idea 10:25.450 --> 10:30.450 of a share sense of responsibility about our IA future. 10:30.640 --> 10:33.860 A shared vision about the importance of trust 10:33.860 --> 10:36.110 and transparency. 10:36.110 --> 10:39.340 Our national security depends on it. 10:39.340 --> 10:42.600 And even for those who for various reasons 10:42.600 --> 10:46.260 still view DoD with suspicion or who are reluctant 10:46.260 --> 10:48.890 to accept that we are in a strategic competition 10:48.890 --> 10:51.970 with China, I would hope they would still agree with us 10:51.970 --> 10:55.890 that AI is a critical component of our nation's prosperity, 10:55.890 --> 10:59.430 vitality and self-sufficiency. 10:59.430 --> 11:02.320 So in other words, no matter where you stand 11:02.320 --> 11:05.320 in respect to the government's future use of AI-enabling 11:05.320 --> 11:10.070 technologies, I submit that we can never attain 11:10.070 --> 11:13.510 the vision outlined in the Commission's interim report 11:13.510 --> 11:16.950 without industry and academia with us together 11:16.950 --> 11:18.600 in an equal partnership. 11:18.600 --> 11:21.730 There's too much at stake to do otherwise, 11:21.730 --> 11:23.360 we are in this together. 11:23.360 --> 11:27.700 Public-private partnerships are the very essence 11:27.700 --> 11:30.060 of America's success as a nation, 11:30.060 --> 11:32.950 not only in the Department of Defense but across 11:32.950 --> 11:34.900 the entire United States government. 11:34.900 --> 11:36.900 So the message we want to send today, 11:36.900 --> 11:40.343 we have to make this triangle back to what it used to be. 11:41.210 --> 11:42.623 - Well thank you General. 11:44.230 --> 11:46.950 I think I'm gonna ask a couple of questions to both of you 11:46.950 --> 11:48.810 and let's start with the same question to both. 11:48.810 --> 11:52.253 Kent, talk about Maven some more. 11:53.090 --> 11:54.482 - Sure. 11:54.482 --> 11:57.030 (audience laughing) 11:57.030 --> 11:59.810 Well I think it's no secret that we came up 11:59.810 --> 12:01.200 as a consumer company. 12:01.200 --> 12:03.310 We are quickly evolving into also becoming 12:03.310 --> 12:06.000 an enterprise company and putting a lot of resources 12:06.000 --> 12:08.090 into that, but there are different protocols 12:08.090 --> 12:11.040 and different ways of engaging, and as we go along 12:11.040 --> 12:14.400 that journey, I'd be lying to tell you that everybody, 12:14.400 --> 12:16.670 all our employees have an identical view 12:16.670 --> 12:19.040 on a lot of hard issues, they don't. 12:19.040 --> 12:22.110 But in some ways that debate and discussion 12:22.110 --> 12:25.600 is the positive as well as the negative. 12:25.600 --> 12:27.570 It's, in many ways it's in our DNA but it's 12:27.570 --> 12:29.020 in the DNA of America. 12:29.020 --> 12:31.820 You could argue that that kind of constructive debate 12:31.820 --> 12:34.660 is America's first innovation. 12:34.660 --> 12:36.470 You look at great research scientists 12:36.470 --> 12:39.280 like Richard Feynman who was one of the leading thinkers 12:39.280 --> 12:41.850 in quantum mechanics who was also notoriously 12:41.850 --> 12:44.370 iconoclastic, free-thinking guy. 12:44.370 --> 12:48.320 We think out of that comes incredible strength. 12:48.320 --> 12:51.990 If we work together well we can actually have a more robust, 12:51.990 --> 12:55.150 more resilient framework, a framework that helps 12:55.150 --> 12:57.870 build social trust as well as a framework that works 12:57.870 --> 12:59.480 for the world. 12:59.480 --> 13:01.880 So as we put forward our AI principles 13:01.880 --> 13:05.180 and our governing processes, because an important thing 13:05.180 --> 13:08.010 to note is that the principles in a sense are easy. 13:08.010 --> 13:11.970 As the DIB report notes, the report devotes a couple 13:11.970 --> 13:14.770 of pages to the principles and a long section 13:14.770 --> 13:17.690 to the implementation because you quickly discover 13:17.690 --> 13:20.400 that a lot of the hard problems are when the principles 13:20.400 --> 13:22.060 conflict and are challenging. 13:22.060 --> 13:24.270 We've had debates about whether or not to publish 13:24.270 --> 13:26.734 a paper on lip reading which has... 13:26.734 --> 13:28.160 - [Eric] Say that again. 13:28.160 --> 13:30.430 - We have had debates about whether to publish 13:30.430 --> 13:32.670 a paper on lip reading. 13:32.670 --> 13:34.940 Lip reading is a great benefit to people who are 13:34.940 --> 13:37.230 hard of hearing around the world, et cetera, 13:37.230 --> 13:38.920 but you could imagine it could be misused 13:38.920 --> 13:41.650 for surveillance and other kinds of purposes. 13:41.650 --> 13:44.200 After reviewing a particular technology, 13:44.200 --> 13:45.720 we determined that it was appropriate 13:45.720 --> 13:48.170 to publish because that particular technology 13:48.170 --> 13:51.120 was useful really only in one-to-one settings 13:51.120 --> 13:53.200 not for surveillance at a distance. 13:53.200 --> 13:55.260 But it's an example of the kinds of discussions 13:55.260 --> 13:57.780 we have around issues like lip reading 13:57.780 --> 14:00.900 or facial recognition or other challenging questions 14:00.900 --> 14:03.500 where we have to come to terms with the reality, 14:03.500 --> 14:05.080 the trade-offs that we're making. 14:05.080 --> 14:08.800 Very much the case in a lot of these issues as well 14:08.800 --> 14:10.150 but we think there's an awful lot of room 14:10.150 --> 14:13.260 for collaboration and coordination on cyber security, 14:13.260 --> 14:16.370 on logistics, on transportation, on health care, 14:16.370 --> 14:18.150 many more topics where we're already engaged 14:18.150 --> 14:19.710 with the military. 14:19.710 --> 14:20.870 - [Eric] General, same question. 14:20.870 --> 14:22.580 Tell us more about Maven. 14:22.580 --> 14:24.920 - Okay, so when we started Project Maven, 14:24.920 --> 14:27.610 our intent was to go after commercial industry. 14:27.610 --> 14:30.340 Eric and the DIB had told us this is where the solutions 14:30.340 --> 14:32.200 already exist, do not reinvent the wheel, 14:32.200 --> 14:33.230 it happens out there. 14:33.230 --> 14:35.520 And our approach was a simple one. 14:35.520 --> 14:38.740 We wanted everybody in the market that was a small 14:38.740 --> 14:41.730 startup of 15 people, which is one of the companies 14:41.730 --> 14:45.520 we got on contract, to the biggest internet data 14:45.520 --> 14:47.920 cyber cloud companies in the world. 14:47.920 --> 14:50.060 And one of those happened to be Project Google. 14:50.060 --> 14:52.960 Why did we go after Project, or why did we have 14:52.960 --> 14:54.320 to Google with Project Maven, 14:54.320 --> 14:58.310 because we wanted to take the best AI talent in the world 14:58.310 --> 15:01.120 and put it against our most wicked problem set, 15:01.120 --> 15:03.230 wide-area motion imagery, it's an extraordinary 15:03.230 --> 15:07.210 difficult problem to go after and we did a very successful 15:07.210 --> 15:09.590 collaboration with the Google team on this. 15:09.590 --> 15:12.420 What was happening internal to the company is how 15:12.420 --> 15:15.320 that played out is a little bit, a different story, 15:15.320 --> 15:17.910 but we got all the way to the end of the contract 15:17.910 --> 15:20.930 and we got products that we were very pleased with. 15:20.930 --> 15:22.640 Now it was unfortunate I think even for some 15:22.640 --> 15:24.800 of the software engineers on that project, 15:24.800 --> 15:26.450 they got to the point where they almost felt 15:26.450 --> 15:29.500 a little bit ostracized because others criticized 15:29.500 --> 15:31.410 them for working with the Department of Defense. 15:31.410 --> 15:33.460 But day to day, from the senior-most leader 15:33.460 --> 15:36.850 down to the people working on the Project Maven team 15:36.850 --> 15:39.720 we got tremendous support from Project, 15:39.720 --> 15:41.520 in Maven from Google. 15:41.520 --> 15:44.630 What we found though, and this is really the critique 15:44.630 --> 15:48.400 on both sides, is we lost the narrative very quickly. 15:48.400 --> 15:50.980 And part of this was about the company made a strategic 15:50.980 --> 15:53.230 decision really not to be public about what they wanted 15:53.230 --> 15:54.063 to do. 15:54.063 --> 15:55.840 Our approach in the Department of Defense is willing 15:55.840 --> 15:58.760 to talk as much as the company wanted us to talk about, 15:58.760 --> 16:00.170 we do whatever the market would bear, 16:00.170 --> 16:01.750 in very general terms, we didn't want to get 16:01.750 --> 16:03.240 into operational specifics. 16:03.240 --> 16:04.950 This was a project, was intelligence, 16:04.950 --> 16:07.380 surveillance, reconnaissance on a drone, 16:07.380 --> 16:09.610 remotely piloted aircraft, it had no weapons on it, 16:09.610 --> 16:11.650 it was not a weapons project, it is not a weapons 16:11.650 --> 16:14.430 project, but what happened is we started hearing 16:14.430 --> 16:16.390 these wild stories and assumptions 16:16.390 --> 16:19.530 about what Project Maven was and was not to the point where 16:19.530 --> 16:21.810 if you google today, no pun intended, 16:21.810 --> 16:22.810 you actually google, 16:23.690 --> 16:26.320 the adjective controversial has now been 16:26.320 --> 16:28.530 inserted permanently in front of Project Maven. 16:28.530 --> 16:30.800 It was not controversial to me, it was not controversial 16:30.800 --> 16:33.190 to the team, I would say it's not controversial 16:33.190 --> 16:35.640 to anybody right now beyond some people who just 16:35.640 --> 16:36.900 don't like what we're doing. 16:36.900 --> 16:39.262 So I guess what I bring it all the way full circle 16:39.262 --> 16:42.030 is this is an interesting point of, I've thought a lot, 16:42.030 --> 16:43.820 and I'm not sure everybody fully appreciates 16:43.820 --> 16:45.610 or agrees with me. 16:45.610 --> 16:47.770 I view what happened with Google and Maven as 16:47.770 --> 16:50.200 a little bit of a canary in a coal mine. 16:50.200 --> 16:52.630 The fact that it happened when it did as opposed 16:52.630 --> 16:54.500 on the verge of a conflict or a crisis where 16:54.500 --> 16:56.760 asking for their help, we've gotten some of that 16:56.760 --> 16:58.340 out of the way, you've heard Kent talking a little bit 16:58.340 --> 17:01.440 about of a reset here and how much the company 17:01.440 --> 17:04.000 and all the other companies that we deal with 17:04.000 --> 17:06.100 want to work with the Department of Defense. 17:06.100 --> 17:08.660 I think that narrative is an important narrative, 17:08.660 --> 17:11.810 it happened, it would have happened to somebody else 17:11.810 --> 17:14.630 at some point, but this idea of transparency 17:14.630 --> 17:17.080 and a willingness to talk about what each side 17:17.080 --> 17:19.520 is trying to achieve may be the biggest lessons 17:19.520 --> 17:21.368 of all that I took from it. 17:21.368 --> 17:23.740 - It's a real tragedy that we don't wear hats 17:23.740 --> 17:25.850 anymore because I could borrow three hats 17:25.850 --> 17:27.560 and figure out which hat I'm wearing. 17:27.560 --> 17:30.250 With my DIB hat on, I can tell you that when 17:30.250 --> 17:34.300 I met General Shanahan the real problem inside the military 17:34.300 --> 17:37.880 is that we take these exquisitely trained soldiers, 17:37.880 --> 17:40.160 airmen, so forth and so on, and we put them in front 17:40.160 --> 17:42.750 of mind-numbing observational tasks. 17:42.750 --> 17:46.040 They literally watch screens all day. 17:46.040 --> 17:49.000 And it's a terrible waste of the human asset 17:49.000 --> 17:50.820 that the military produces. 17:50.820 --> 17:53.650 And so there's a huge opportunity to try to sort 17:53.650 --> 17:56.060 of get them to work at a higher level position 17:56.060 --> 17:59.452 and that's why the DIB recommended the, 17:59.452 --> 18:01.240 made the procedure and indeed the creation 18:01.240 --> 18:03.760 of the Joint Center for AI which Kent, 18:03.760 --> 18:06.070 you both stood up and now head. 18:06.070 --> 18:08.394 Let's talk about another question for both of you 18:08.394 --> 18:11.780 which has to do with ethics. 18:11.780 --> 18:14.530 Now, in the middle of the kerfuffle that went on 18:14.530 --> 18:17.810 inside of Google, Kent had the good idea of having 18:17.810 --> 18:22.530 a formal AI ethics proposal, and he drove inside of Google 18:22.530 --> 18:26.070 an ethics process which produced a remarkable public 18:26.070 --> 18:30.330 document, now I have my Google hat on, which I think 18:30.330 --> 18:33.220 is really quite definitive, and I think maybe you could 18:33.220 --> 18:36.410 talk about that and then similarly, 18:36.410 --> 18:40.058 the DIB produced a proposal to the military 18:40.058 --> 18:43.200 and I believe you are the customer for the proposal 18:43.200 --> 18:46.093 that we wrote on military AI ethics. 18:47.870 --> 18:50.860 I assume both of you are in favor, since Kent wrote 18:50.860 --> 18:53.330 the first one and all the other industry companies have now 18:53.330 --> 18:56.450 largely copied, variance of your approach in one form 18:56.450 --> 18:58.870 or another, I assume you are in favor of this. 18:58.870 --> 19:01.180 What are the consequences of these ethics things, 19:01.180 --> 19:05.230 does it really work, does, for example, does Google 19:05.230 --> 19:09.370 prevent, does Google turn off things or stop doing things 19:09.370 --> 19:11.870 like in the last little while, I mean, how does it 19:11.870 --> 19:14.603 actually work, and same question for you General, 19:16.960 --> 19:19.180 there are people who claim that the military won't operate 19:19.180 --> 19:20.830 under ethics principles. 19:20.830 --> 19:23.780 In our report we cite the many rules of the military 19:23.780 --> 19:24.960 is required to operate under. 19:24.960 --> 19:26.040 Maybe you could talk about that. 19:26.040 --> 19:26.890 So Kent? 19:26.890 --> 19:27.800 - Sure. 19:27.800 --> 19:30.700 So I think as the General noted, having frameworks 19:30.700 --> 19:33.440 in place early on, both the set of principles 19:33.440 --> 19:37.047 but then also the review processes and escalation 19:37.047 --> 19:39.850 opportunities is a critical part of internal 19:39.850 --> 19:42.120 as well as external transparency. 19:42.120 --> 19:44.500 It's right, among our principles we've talked about 19:44.500 --> 19:47.560 surveillance being a concern, so we want to make sure 19:47.560 --> 19:51.210 that some of the recognition tools and the image 19:51.210 --> 19:53.510 tracking software that we're developing are deployed 19:53.510 --> 19:54.800 in appropriate ways. 19:54.800 --> 19:56.480 We want to be a good partner, we don't want to pull away 19:56.480 --> 19:59.410 support but we want to make sure we know the scope 19:59.410 --> 20:02.640 of the project that we're developing, and when we're 20:02.640 --> 20:04.920 licensing that for commercial uses have a sense 20:04.920 --> 20:06.400 of the direction of travel there. 20:06.400 --> 20:08.550 I think that's a valuable thing for both sides 20:08.550 --> 20:10.680 in terms of making sure that expectations are clear 20:10.680 --> 20:13.110 and in terms of building no only trust internally 20:13.110 --> 20:14.850 but trust across society. 20:14.850 --> 20:18.100 So another example would be when it comes to general 20:18.100 --> 20:21.450 purpose APIs for facial recognition where you don't know 20:21.450 --> 20:23.550 necessarily what use is gonna be made of them, 20:23.550 --> 20:26.320 we said until we develop more policy and more 20:26.320 --> 20:29.250 technological safeguards we're gonna be very cautious 20:29.250 --> 20:30.830 about proceeding in that area. 20:30.830 --> 20:33.060 Another example is when it comes to weapons, 20:33.060 --> 20:35.320 we have said this is a nascent technology, 20:35.320 --> 20:37.290 we want to be very careful about the application 20:37.290 --> 20:40.230 of AI in this area, so that's not an area 20:40.230 --> 20:42.560 that we're pursuing, given our background, 20:42.560 --> 20:45.080 we recognize the limits of our experience 20:45.080 --> 20:46.000 in that area. 20:46.000 --> 20:48.070 Obviously the military is gonna be deeper 20:48.070 --> 20:49.990 and have more understanding of safety implications 20:49.990 --> 20:50.823 and the like. 20:50.823 --> 20:52.350 So we're gonna continue to work through these 20:52.350 --> 20:53.183 different areas. 20:53.183 --> 20:55.470 I think there's a remarkable degree of convergence 20:55.470 --> 20:59.467 we see between the OECD, the DoD, the DIB, 21:00.500 --> 21:02.130 now internationally we're starting to see 21:02.130 --> 21:04.100 the European commissions say they're coming up 21:04.100 --> 21:06.300 with regulations for artificial intelligence 21:06.300 --> 21:07.670 in the next hundred days. 21:07.670 --> 21:09.500 I think this will be a very interesting exercise 21:09.500 --> 21:12.280 as we all pursue kind of a combination of how we 21:12.280 --> 21:15.180 build acceptance for this next generation of technologies. 21:16.040 --> 21:19.560 - And so looking at it through the DoD lens, 21:19.560 --> 21:22.080 this may be the best starting point, when you talk, 21:22.080 --> 21:24.730 Kent mentions this area of convergence between 21:24.730 --> 21:27.320 commercial industry, academia and the government, 21:27.320 --> 21:29.270 probably the AI ethics principles are as good 21:29.270 --> 21:31.580 as anything else to drive the stake in the ground, 21:31.580 --> 21:34.550 do we agree on all of these, some of these, 21:34.550 --> 21:36.480 and if we don't disagree let's get the conversation 21:36.480 --> 21:38.130 going, so it's a good starting point. 21:38.130 --> 21:39.810 Another part is, I need to state the obvious, 21:39.810 --> 21:44.260 that I can tell you with certainty 21:44.260 --> 21:49.260 that China and Russia did not embark on a 15-month process 21:49.730 --> 21:52.100 involving public hearings and discussion 21:52.100 --> 21:54.190 about the ethical, safe and lawful use 21:54.190 --> 21:56.310 of artificial intelligence, they're not doing it. 21:56.310 --> 21:58.800 And I don't expect they ever will do it. 21:58.800 --> 22:01.260 So people may question what the department is doing 22:01.260 --> 22:02.980 and why we're doing it, but I tell you what, 22:02.980 --> 22:07.050 we just embarked on this long process just to make sure 22:07.050 --> 22:09.680 we took into account all of the different voices 22:09.680 --> 22:12.340 on the ethical use of artificial intelligence 22:12.340 --> 22:14.210 and I would say the product that's been delivered 22:14.210 --> 22:16.940 is an excellent product shaped by a lot of people 22:16.940 --> 22:20.500 who spent time and attention against this. 22:20.500 --> 22:23.530 I've said this in other settings and over 35, 22:23.530 --> 22:25.860 pretty much 35 and a half years in uniform, 22:25.860 --> 22:28.400 I have never spent as much time on this question 22:28.400 --> 22:30.660 of the ethical use of given technology. 22:30.660 --> 22:33.240 The Department of Defense actually has a long 22:33.240 --> 22:35.680 and I would say a commendable history despite flaws 22:35.680 --> 22:37.830 along the way of looking at the ethical use 22:37.830 --> 22:40.700 of emerging technologies whatever they are. 22:40.700 --> 22:42.890 There are differences with artificial intelligence 22:42.890 --> 22:45.580 and what the DIB report does very well is start 22:45.580 --> 22:47.950 with here is what's similar to every other technology 22:47.950 --> 22:49.680 that ever been field in the department, 22:49.680 --> 22:51.380 here are some areas that may be different, 22:51.380 --> 22:54.140 we're not quite sure yet, and here are some substantive 22:54.140 --> 22:56.810 differences like systems that learn on their own. 22:56.810 --> 22:59.130 That's a pretty good framework for going after this. 22:59.130 --> 23:01.900 We have a way of looking at this, no matter if it's 23:01.900 --> 23:04.210 artificial intelligence or any other technology, 23:04.210 --> 23:07.440 our history, our processes, our approach and our training 23:07.440 --> 23:09.950 are in place to look at any emerging technology 23:09.950 --> 23:12.570 and how we bring it in from a pilot and a prototype 23:12.570 --> 23:13.580 into production. 23:13.580 --> 23:15.760 So now that this report has been presented 23:15.760 --> 23:17.970 to the Secretary of Defense, it is up, 23:17.970 --> 23:19.180 I get two questions now. 23:19.180 --> 23:20.610 One, what do you think about the report? 23:20.610 --> 23:23.440 It's an excellent report, provides the best possible 23:23.440 --> 23:25.000 starting point, and the number two is what are you 23:25.000 --> 23:26.010 going to do about it. 23:26.010 --> 23:27.780 This is where it gets really complicated. 23:27.780 --> 23:30.280 We have to come up with an implementation plan. 23:30.280 --> 23:32.740 It will not be a JAIC implementation plan, 23:32.740 --> 23:35.300 it will be a department-wide implementation plan 23:35.300 --> 23:37.780 taking these recommendations, putting something together 23:37.780 --> 23:40.740 through my boss, Mr. Dana Deasy as the Chief Information 23:40.740 --> 23:43.390 Officer of the department in making some recommendations 23:43.390 --> 23:45.960 on how we implement this for across the entire 23:45.960 --> 23:47.090 Department of Defense. 23:47.090 --> 23:49.820 That is not an overnight task, this is gonna take us 23:49.820 --> 23:52.590 a while to get this right but we now have an outstanding 23:52.590 --> 23:53.423 starting point. 23:54.310 --> 23:59.310 - So, that's a wonderful framing for where we are, 23:59.780 --> 24:02.410 I'd like to push a little bit on where this will go. 24:02.410 --> 24:06.280 Kent, let me give you an example. 24:06.280 --> 24:09.170 Open AI developed a technology which will allow 24:09.170 --> 24:12.870 arbitrary rewriting of text that was sufficiently good 24:12.870 --> 24:17.080 that they became concerned and they didn't release it 24:17.080 --> 24:19.580 and said they only released it in certain model ways 24:19.580 --> 24:21.040 to certain researchers. 24:21.040 --> 24:24.040 That's an example, and I asked them, and they said, 24:24.040 --> 24:26.770 I said did anyone put any pressure on you, 24:26.770 --> 24:30.030 and they said no, we just thought it was our good judgment. 24:30.030 --> 24:35.030 You famously, very early said on the face recognition 24:35.040 --> 24:38.220 thing, we're going to avoid that as a general purpose 24:38.220 --> 24:39.900 because of the dangers. 24:39.900 --> 24:43.530 Where will the industry end up in this sort of 24:43.530 --> 24:45.120 self restraint thing. 24:45.120 --> 24:48.030 Is it going to be a common set of principles, 24:48.030 --> 24:49.870 is this going to be, is the industry going to have 24:49.870 --> 24:53.650 to have an AI ethics common with respect to, 24:53.650 --> 24:55.590 you know, being careful? 24:55.590 --> 24:57.410 How will this play out in your view? 24:57.410 --> 24:59.390 - I think you already see some of that first to work 24:59.390 --> 25:01.890 across the industry with the partnership on artificial 25:01.890 --> 25:04.790 intelligence to exchange information on some of the work 25:04.790 --> 25:06.020 that's being done. 25:06.020 --> 25:08.610 It's going to be an evolving question as we develop 25:08.610 --> 25:10.550 more infrastructure, more of these frameworks 25:10.550 --> 25:13.130 about the appropriate limits of the use of artificial 25:13.130 --> 25:15.450 intelligence, the appropriate safeguards and checks 25:15.450 --> 25:18.660 and balances for a whole variety of different areas. 25:18.660 --> 25:22.070 But I think, I'm hopeful, that with a common groundwork, 25:22.070 --> 25:25.040 the way we've started to lay already, we're on the path 25:25.040 --> 25:28.090 to doing that, but this is true of any new technology. 25:28.090 --> 25:32.560 Any communications platform from the radio to television 25:32.560 --> 25:35.040 to the internet you've needed new regulatory 25:35.040 --> 25:37.440 infrastructures, new social conventions 25:37.440 --> 25:39.620 about how do you use these different tools. 25:39.620 --> 25:42.230 This is an extraordinarily powerful technology, 25:42.230 --> 25:45.190 we're at the early days, so I think it's understandable 25:45.190 --> 25:47.250 that you're seeing a variety of views come together 25:47.250 --> 25:49.080 but also notable that you're seeing the degree 25:49.080 --> 25:51.250 of convergence that you're seeing. 25:51.250 --> 25:56.150 - So General, you have talked inside the Pentagon 25:56.150 --> 25:59.713 about this notion of a new kind of warfare. 26:01.468 --> 26:03.130 And I think the term that you all use 26:03.130 --> 26:04.363 is algorithmic warfare. 26:06.630 --> 26:09.610 Take us through, in the same sense that Kent talked 26:09.610 --> 26:11.750 about how this new emergent thing sort of new 26:11.750 --> 26:15.370 and powerful, what's new and powerful about this 26:15.370 --> 26:17.930 technology in a military context. 26:17.930 --> 26:19.900 With your long experience and understanding, 26:19.900 --> 26:21.350 how the military frames it. 26:21.350 --> 26:23.740 What's the language, what's the positioning. 26:23.740 --> 26:26.640 - I go back to as we were formed and then 26:26.640 --> 26:29.840 Deputy Secretary of Defense Bob Work was in the room, 26:29.840 --> 26:32.110 I'll never forget it, it's like yesterday, 26:32.110 --> 26:35.310 designating us, okay, you are now formed as the team 26:35.310 --> 26:38.290 that's gonna figure out how you actually field AI, 26:38.290 --> 26:40.220 get away from the research piece of it, 26:40.220 --> 26:43.730 which was all happening wonderfully behind the scenes, 26:43.730 --> 26:46.000 but now we needed a team that was focusing on fielding 26:46.000 --> 26:48.820 to the warfighter, and the name that he gave us 26:48.820 --> 26:51.290 was the algorithmic warfare cross-functional team. 26:51.290 --> 26:52.550 It's not an accidental name. 26:52.550 --> 26:54.350 It's become Project Maven because it's much easier 26:54.350 --> 26:55.573 to say than aquifith, 26:57.458 --> 27:00.423 (audience laughing) 27:00.423 --> 27:02.590 - And, your acronyms are gonna kill me. 27:02.590 --> 27:04.750 Okay, let's - So let's just focus 27:04.750 --> 27:05.583 on algorithmic warfare. 27:05.583 --> 27:07.143 - Why don't you tell us what algorithmic warfare is. 27:07.143 --> 27:09.790 - So it's the idea we're going to face a fight in the 27:09.790 --> 27:12.120 future, we're used to fighting for 20 years in a certain 27:12.120 --> 27:15.100 type of fight, counter-terrorism, insurgencies, 27:15.100 --> 27:19.640 we are going to be shocked by the speed, the chaos, 27:19.640 --> 27:21.970 the bloodiness and the friction of the future fight 27:21.970 --> 27:24.260 in which this will be, maybe playing 27:24.260 --> 27:25.640 in microseconds at time. 27:25.640 --> 27:27.430 How do we envision that fight happening. 27:27.430 --> 27:29.880 It has to be algorithm against algorithm. 27:29.880 --> 27:33.220 It is a, as you described earlier, as we were talking 27:33.220 --> 27:36.450 about this, it's a Boydian OODA, how fast can we 27:36.450 --> 27:38.420 get inside somebody's decision cycle. 27:38.420 --> 27:40.467 - Remind people what the OODA is. 27:40.467 --> 27:43.800 - Colonel John Boyd, Air Force colonel who was 27:43.800 --> 27:47.550 sort of the author of the observe, orient, decide, act, 27:47.550 --> 27:50.840 which is how you get through the cycle of decision making 27:50.840 --> 27:52.770 which was really never about the decide or the act, 27:52.770 --> 27:54.460 it was more about the observe and orient, 27:54.460 --> 27:56.230 I think really about the orient phase. 27:56.230 --> 27:58.323 But in this future fight we're looking at, 27:58.323 --> 28:00.390 this would be happening so fast, if we're trying 28:00.390 --> 28:02.880 to do this by humans against machines 28:02.880 --> 28:05.310 and the other side has the machines and the algorithms 28:05.310 --> 28:08.180 that we don't, we're at an unacceptably high risk 28:08.180 --> 28:09.450 of losing that conflict. 28:09.450 --> 28:11.780 Now this is a challenging one because I think part 28:11.780 --> 28:14.950 of what you're getting at in that future scenario 28:14.950 --> 28:18.440 how are people going to be assured that our algorithms 28:18.440 --> 28:20.580 are going to work as intended and they don't take on 28:20.580 --> 28:22.200 a life of its own so to speak. 28:22.200 --> 28:24.590 What we will fall back on, I think this is a starting 28:24.590 --> 28:27.610 point for what the DIB principles gave us, 28:27.610 --> 28:30.380 is test, evaluation, validation and verification. 28:30.380 --> 28:33.000 We have to do a lot more work on the front end, 28:33.000 --> 28:35.650 by the time they field it we know it's being fielded. 28:35.650 --> 28:39.010 But I think we're really going to be at a disadvantage, 28:39.010 --> 28:42.760 if we think we're gonna be at a pure human against machine. 28:42.760 --> 28:45.350 It'll be human and machine on one side, 28:45.350 --> 28:48.200 human, machine on the other, but the temporal dimension, 28:48.200 --> 28:51.010 this fleeting superiority that you may be facing 28:51.010 --> 28:53.500 where decisions will be made that fast, it might be 28:53.500 --> 28:54.990 algorithm against algorithm. 28:54.990 --> 28:58.190 - Yeah, that was, to me the key question military matter 28:58.190 --> 29:01.650 is what happens when the whole scenario is faster 29:01.650 --> 29:02.853 than human decision making. 29:02.853 --> 29:05.080 Because I understand the way the military works 29:05.080 --> 29:08.210 is when there's a threat, in general people check 29:08.210 --> 29:10.740 with their superior, there's sort of a rule of engagement, 29:10.740 --> 29:13.200 there's human judgment, it's all built around some 29:13.200 --> 29:15.460 number of minutes, right? 29:15.460 --> 29:17.520 Not some number of nanoseconds. 29:17.520 --> 29:21.420 How will the military adjust its procedures 29:21.420 --> 29:23.440 to deal with this real possible threat? 29:23.440 --> 29:26.040 - It won't be driven by above. 29:26.040 --> 29:28.330 The innovation will happen at the lowest possible level. 29:28.330 --> 29:30.610 What we have to be able to do in places like JAIC 29:30.610 --> 29:33.520 or Maven is give people the policies and the authorities 29:33.520 --> 29:36.080 and the framework to do what they need to do. 29:36.080 --> 29:38.320 The innovation, the people that will say I have a 29:38.320 --> 29:40.780 solution to this, I'm going to write a code, 29:40.780 --> 29:42.660 I'm going to develop an algorithm and apply it to 29:42.660 --> 29:45.280 this problem set in the field if you've given me 29:45.280 --> 29:47.160 the data, the tools, the frameworks, the library, 29:47.160 --> 29:49.320 the standards, all those other things, we can do it. 29:49.320 --> 29:51.580 So if the idea in that fight, it will be more 29:51.580 --> 29:54.340 decentralized than a lot of people are comfortable 29:54.340 --> 29:56.960 with today and that brings risks with it. 29:56.960 --> 29:59.740 So we're talking about higher risks, higher consequence, 29:59.740 --> 30:02.620 but it's either that or risk losing the fight. 30:02.620 --> 30:05.890 So it's this idea of decentralized development, 30:05.890 --> 30:09.590 decentralized experimentation, decentralized innovation. 30:09.590 --> 30:12.030 The innovation as was described in one of the panels 30:12.030 --> 30:13.750 this morning happens at the bottom. 30:13.750 --> 30:15.930 We've got to give them the push from above 30:15.930 --> 30:18.120 to make it succeed. - And also add 30:18.120 --> 30:19.342 - Kent 30:19.342 --> 30:20.175 - in addition to the tempo component, 30:20.175 --> 30:23.840 there's new fronts in the cyber security, cyber defense. 30:23.840 --> 30:27.946 We're seeing already sort of efforts to destabilize 30:27.946 --> 30:29.460 with disinformation campaigns and the like. 30:29.460 --> 30:31.420 So the more we can work together to recognize 30:31.420 --> 30:34.320 those patterns across a wider battlefield if you will, 30:34.320 --> 30:35.570 the better for everybody. 30:36.580 --> 30:40.580 - Kent, do you have a model for how the industry, 30:40.580 --> 30:43.130 one of the themes of our whole conference is the industry 30:43.130 --> 30:45.720 and the government need to work together broadly, 30:45.720 --> 30:48.210 and obviously we have a senior general here, 30:48.210 --> 30:49.990 but I'm really referring to the government as a whole, 30:49.990 --> 30:52.050 and there's a whole lot more than just the DoD 30:52.050 --> 30:53.390 that needs AI. 30:53.390 --> 30:55.930 Do you have a model for how the industry should work 30:55.930 --> 30:58.830 with the federal government, the state governments, 30:58.830 --> 31:00.770 the DoDs and so forth? 31:00.770 --> 31:05.570 - Well, I've only talked about two important elements 31:05.570 --> 31:08.460 the DIB report talks, touches on as well. 31:08.460 --> 31:11.180 The first is this notion of trying to build broad trust 31:11.180 --> 31:13.060 in the application of new technologies, 31:13.060 --> 31:15.250 the second is the need for a global framework 31:15.250 --> 31:17.000 which helps with that process. 31:17.000 --> 31:19.110 The third, as General Shanahan alluded to, 31:19.110 --> 31:22.500 is a more operational administrative question 31:22.500 --> 31:26.700 of how do we make it as easy as possible for new companies 31:26.700 --> 31:29.110 to enter into these kinds of partnerships. 31:29.110 --> 31:31.110 So a lot of the innovation, a lot of the cutting edge 31:31.110 --> 31:33.880 research being done in Silicon Valley is not being done 31:33.880 --> 31:36.260 by large companies, it's being done by small companies. 31:36.260 --> 31:39.820 It's a rich ecosystem of innovation and it's challenging 31:39.820 --> 31:43.070 even for a company Google's size to start to get more 31:43.070 --> 31:44.960 involved in that environment. 31:44.960 --> 31:47.710 It's doubly difficulty for some of these smaller companies. 31:47.710 --> 31:51.850 So as we look at modernizing procurement from the military 31:51.850 --> 31:55.010 side and working with Congress as well to make it as quick, 31:55.010 --> 31:59.440 as nimble, as flexible as possible, responsive to new needs. 31:59.440 --> 32:02.420 Looking at increasing R and D funding across the board 32:02.420 --> 32:05.130 because that's traditionally been a really fertile 32:05.130 --> 32:07.960 ground for a lot of these collaborative enterprises 32:07.960 --> 32:09.400 to move forward. 32:09.400 --> 32:13.000 Looking at human resources exchanges, there's a lot 32:13.000 --> 32:16.350 of authorities out there which authorize private sector 32:16.350 --> 32:19.270 people to come into the government, but in practice 32:19.270 --> 32:20.850 it's harder than you would think. 32:20.850 --> 32:23.070 So a lot of that hard work on the ground I think 32:23.070 --> 32:25.190 is important to making this a success. 32:25.190 --> 32:27.450 - For both of you, because we're gonna be making 32:27.450 --> 32:30.338 recommendations that ideally will end up in legislation 32:30.338 --> 32:33.973 a year from now, plus or minus, are there specific things 32:33.973 --> 32:36.590 that we could do that would promote private-public 32:36.590 --> 32:39.960 partnerships, for example as you know the DoD has 32:39.960 --> 32:43.890 DIUX, SCO and a number of other groups that work 32:43.890 --> 32:46.927 very closely, of Incutel, obviously the extraordinary 32:46.927 --> 32:49.650 contribution that DARPA has played to our industry 32:49.650 --> 32:52.790 and to technology and to me personally, so forth and so on. 32:52.790 --> 32:55.857 So the sum of all of that, do you have a model, 32:55.857 --> 32:58.770 and I'll ask Kent the same question, do you have specific 32:58.770 --> 33:01.040 things that would be helpful that would decrease 33:01.040 --> 33:04.700 the friction and increase the cohesion between small 33:04.700 --> 33:07.190 companies, large companies, the federal government, 33:07.190 --> 33:08.860 procurement, the DoD? 33:08.860 --> 33:09.830 - A couple of different thoughts on that. 33:09.830 --> 33:12.030 First of all, there's so much that has started to happen 33:12.030 --> 33:13.810 over the last couple of years with places like 33:13.810 --> 33:16.810 Defense Digital Service, DIU, Kessel Run, Compile 33:16.810 --> 33:20.180 to Combat, all what I call these beginning insurgencies 33:20.180 --> 33:21.130 to get things moving. 33:21.130 --> 33:23.070 - And we should pause and say that these are each 33:23.070 --> 33:25.410 small teams of software people inside the DoD 33:25.410 --> 33:28.030 that have had an outside impact in changing 33:28.030 --> 33:30.740 the procedures in important aspects in the Air Force 33:30.740 --> 33:33.510 for example, in some of the K-Ox and things that like. 33:33.510 --> 33:35.340 - So that all got it started but we have to do 33:35.340 --> 33:37.130 is figure out how to institutionalize, how to make 33:37.130 --> 33:39.490 that systemic change across the Department of Defense 33:39.490 --> 33:41.100 which is the next hard part to do. 33:41.100 --> 33:42.910 And you ask about what are the ways we do that. 33:42.910 --> 33:44.790 I'll tell you one of the biggest ones is just talent, 33:44.790 --> 33:47.530 bringing in talent from the outside, from academia, 33:47.530 --> 33:48.363 from industry. 33:48.363 --> 33:50.710 Our chief scientist, Jo Crisman, who is here today, 33:50.710 --> 33:53.480 has time in the government in ARPA working for a startup 33:53.480 --> 33:56.160 in her last job, Nand Mulchandani who is our chief 33:56.160 --> 33:58.310 technical officer, 25 years in the Valley, 33:58.310 --> 34:00.920 he comes in and within 24 hours takes a different 34:00.920 --> 34:02.450 view of what we're trying to do. 34:02.450 --> 34:04.470 We need a lot more of that, we need sabbaticals, 34:04.470 --> 34:07.200 people coming in from academia for a year or two, 34:07.200 --> 34:08.570 going back out to the outside. 34:08.570 --> 34:10.560 Us putting people in education with industry, 34:10.560 --> 34:12.738 Secretary of Defense corporate fellowship. 34:12.738 --> 34:14.550 That's all beginning to happen. 34:14.550 --> 34:17.450 It needs to scale to the next level to really start 34:17.450 --> 34:20.000 to understand what we're each talking about. 34:20.000 --> 34:21.730 Me going out to the Valley and talking to the 34:21.730 --> 34:24.190 Suite, C-Suite, only gets so far. 34:24.190 --> 34:26.513 It's the peer-to-peer relationships and discussion 34:26.513 --> 34:28.250 that I think are gonna be more important than 34:28.250 --> 34:29.083 anything else. 34:29.083 --> 34:31.190 - And I very much agree with that, I think you've seen 34:31.190 --> 34:35.100 examples of it with the JAIC, with DARPA, 34:35.100 --> 34:36.890 we're priming the pump on a variety of really 34:36.890 --> 34:41.300 important areas whether that's training or in models 34:41.300 --> 34:43.277 and simulation, they're helpful in that, recruiting, 34:43.277 --> 34:45.320 a number of different areas. 34:45.320 --> 34:48.790 Another important component of this is the IT 34:48.790 --> 34:52.301 modernization because in many ways, the AI kernel 34:52.301 --> 34:56.110 is critical but it comes embedded within a larger 34:56.110 --> 34:59.190 environment of software that's oftentimes very difficult 34:59.190 --> 35:01.390 because you have to get security clearances 35:01.390 --> 35:03.710 and appropriate certification for all the elements 35:03.710 --> 35:05.700 of that piece, so there's that combination 35:05.700 --> 35:08.780 of successful individual experiments and trial runs 35:08.780 --> 35:11.710 to build the familiarity at the peer-to-peer level 35:11.710 --> 35:14.190 but also the systemic change to make it easier 35:14.190 --> 35:17.103 to have wider adoption of the technology more broadly. 35:18.900 --> 35:21.440 - It's time for us to finish up. 35:21.440 --> 35:24.810 My objective in this panel was to put to bed 35:24.810 --> 35:27.700 this notion that somehow Silicon Valley wouldn't 35:27.700 --> 35:30.720 work with the military and I think we've clearly seen 35:30.720 --> 35:32.390 examples, small companies, large companies, 35:32.390 --> 35:34.860 strong statement from Kent on those, and we just sort 35:34.860 --> 35:36.880 of move forward and build this collective 35:36.880 --> 35:40.840 between the private and the public partnerships. 35:40.840 --> 35:43.780 Kent, can you sort of summarize sort of the key 35:43.780 --> 35:46.690 takeaway that you want to offer us, the key message, 35:46.690 --> 35:48.193 the key word, the key, 35:49.970 --> 35:52.740 why are you here and why did you make a special trip 35:52.740 --> 35:54.023 just to make this point? 35:54.950 --> 35:57.930 - I want to be clear, and I'll restate what I said 35:57.930 --> 35:58.763 at the beginning. 35:58.763 --> 36:00.720 We are a proud American company, we are committed 36:00.720 --> 36:04.900 to the cause of national defense for the United States 36:04.900 --> 36:08.000 of America, for our allies and for peace and safety 36:08.000 --> 36:09.470 and security in the world. 36:09.470 --> 36:12.630 We approach that task thoughtfully as we do using 36:12.630 --> 36:15.690 approaching a variety of advanced technologies. 36:15.690 --> 36:17.870 We want to be thoughtful and make sure we have clear 36:17.870 --> 36:20.320 frameworks and transparency and understanding 36:20.320 --> 36:21.530 as we move forward. 36:21.530 --> 36:23.440 I think that it's a mission that the military 36:23.440 --> 36:25.810 and the U.S. government share, and I'm looking forward, 36:25.810 --> 36:27.760 we're looking forward to working more closely 36:27.760 --> 36:29.720 together in the future. 36:29.720 --> 36:32.740 - So General, you never like these things, 36:32.740 --> 36:35.490 but you're sort of the top, you're the tip, 36:35.490 --> 36:39.510 you're the fellow who's going to make this change happen 36:39.510 --> 36:43.670 across 3.2 million people, $660 billion, 36:43.670 --> 36:45.630 an enormous bureaucracy. 36:45.630 --> 36:47.480 How are you going to pull this off? 36:47.480 --> 36:49.029 - One person at a time. 36:49.029 --> 36:50.460 (audience laughing) 36:50.460 --> 36:53.610 It has to be a combination of top down. 36:53.610 --> 36:57.310 It was said on the previous panel, you must have 36:57.310 --> 37:00.260 the full support of leadership from the very top 37:00.260 --> 37:02.610 to show that it's a priority for the department. 37:02.610 --> 37:04.990 That's critical but also insufficient. 37:04.990 --> 37:06.740 You have to have the bottom up innovation, 37:06.740 --> 37:07.950 the people pushing from below. 37:07.950 --> 37:09.670 It's there today, there's no question, 37:09.670 --> 37:11.380 some of them are represented in this room. 37:11.380 --> 37:13.360 They already knows what that future needs to look like, 37:13.360 --> 37:15.580 how do we meet that in the middle, and give them 37:15.580 --> 37:18.210 the resources, the tools to succeed. 37:18.210 --> 37:20.610 The last thing I'll say is this is intimidating, 37:20.610 --> 37:22.100 this is a daunting task. 37:22.100 --> 37:23.060 No way around it. 37:23.060 --> 37:25.690 It is a multi-generational problem, it's going to 37:25.690 --> 37:28.030 require a multi-generational solution. 37:28.030 --> 37:29.890 I'm not gonna wake up tomorrow and suddenly realize 37:29.890 --> 37:30.800 we've got this all right. 37:30.800 --> 37:33.380 We're gonna have some fits and starts, some successes, 37:33.380 --> 37:35.820 some drawbacks, but just keep plowing ahead, 37:35.820 --> 37:37.910 and with the resources and the commitment 37:37.910 --> 37:40.940 of the department behind us I know we'll get there. 37:40.940 --> 37:41.773 - Well thank you. 37:41.773 --> 37:44.900 I think it's worth saying, I've worked with Kent 37:44.900 --> 37:47.920 for 15 years and with my Google hat on, I'll tell you 37:47.920 --> 37:51.180 I cannot be more proud of the impact that he's had 37:51.180 --> 37:55.470 on our society, the scale and the reach of our corporation. 37:55.470 --> 37:57.590 I think you can see this today. 37:57.590 --> 38:00.910 And General, I don't think Bob Work could have chosen 38:00.910 --> 38:03.100 a better person to lead this. 38:03.100 --> 38:05.960 Our partnership with you over the last three years, 38:05.960 --> 38:09.730 you really have moved the resources, gotten the money, 38:09.730 --> 38:12.420 gotten the attention and delivered, and there was 38:12.420 --> 38:13.580 no one before you. 38:13.580 --> 38:16.830 You are that person, so thank you both very much, 38:16.830 --> 38:17.873 thank you all. 38:17.873 --> 38:21.123 (audience applauding)