tv DARPA Director Talks to Washington Post CSPAN December 14, 2018 11:47pm-12:19am EST
on to the next portion of our program. before we do that please join me in thanking our terrific speaker jennifer -- general dunford. >> [ applause ] the washington post also interviewed stephen walker on how the pentagon is using technology to modernize the armed forces especially in the field of artificial intelligence. this is about half an hour.
>> returning to the stage to have a conversation with stephen walker, he has been the director of darpa for the last year and before that he was acting director and was deputy director i think since 2012. so has been in the darpa operations now for a long time. darpa is i'm sure as everyone in the audience knows is our amazing government ideas laboratory and we credited with inter-dashes inventing the internet. sorry al gore. and an awful lot of other amazing tech knowledge he achievements. so, i want to begin by asking you to talk about this amazing institution that you run.
and whether it is adapting enough to the way the world has changed. in a sense, darpa and the internet created a new world in which darpa was own mission doesn't fit white, people sometimes say. the private sector is now so dominant and so quick, cutting edge science and technology used to be in the government and now it's really outside. so start us off by talking about how in this very different world, you want to manage and direct darpa. >> thanks for having me. darpa is actually 60 years old this year and we celebrated that over the past year. we had a big conference last september. darpa, when it was created was all about preventing technological surprises. it was created out of the sputnik era , and we still look at that as our main charter to
prevent technological surprises. one of the ways we understand we can do that best is by creating technological surprise for the united states. so that mission really hasn't changed, and i don't think it i think we are still pursuing it robustly. the mission really of darpa is to create breakthrough technologies and capabilities for our national security. we are focused on national security. we are a department agency in the department of defense . so you are right, early on, we focused on the cold war, very prominent programs in nuclear detection, moved into developing things like the stealth, first stealth aircraft in the late 70s. you mentioned the internet in the late 60s. so we work on certain defense problems all along the way
during the last decade, iraq and afghanistan. we may have gotten a bit too focused on the near-term, i think but obviously our war fighters our men and women were in harms way. so i think that was appropriate. but after iraq and afghanistan, we sort of look up and out again, i think. and under arthur's tenure and mind as a deputy we started focusing on what are the tech races we need to win again. especially in the 21st century. and so we are focused on those. darpa, i think better than any other organization in the government bridges the gap between the commercial sector and the private sector and the defense sector. we bring in people from industry, from universities, from other government agencies and we have them work at harper for a term appointment,
usually 3-5 years. so we are bringing new people in all the time with no ideas. so i think that's one of the reasons why we are at the leading edge of technology and we are able to adapt to what are the new problems out there. and where is technology leading us to opportunities to solve them. >> so i love your comment that your job is to avoid our government being surprised and the best way to do that is by creating surprises yourselves. so i want ask you to surprises. i want ask you to tell us a couple of technological areas we might not be thinking about. where you think some big things are ahead. >> well certainly. i can't tell you everything but i will focus on a few areas. what are the technologies, what are the tech races we need to win in the 21st century? you talk to general dunford just a little bit ago about
ai. ai is one of those areas we need to win. and i believe today we are still in the lead. certainly in the foundations of artificial intelligence darpa has had a long history, about 50 years of investing in artificial intelligence. some of the first language translation work that came out was all darpa's sponsored work. of course the last decade we focused on self driving cars. some of the self driving car challenges we did out in the desert, looking at how to do that and some of those are coming to fruition. so we have had a long history in investing in it. when we think about ai and what's relevant today, we think about three waves of ai, three generations. the first being very rules- based artificial intelligence so you can think of turbotax, it is a rules-based , if this happens you do this, it's pretty simple. the second wave is what books
are really talking about today which is machine learning. so machines winning games like go and being able to do better than humans at recognizing images. this is what people refer to as machine learning. it is being applied today by the commercial sector and also the defense sector. where darpa is really headed and where we want to win the race is in what we call third wave ai, which is really looking at how to give the machines the ability to understand, say what they are looking at or their environment, giving them contextual reasoning to do that. right now, if the machine sees a picture of a cat thing on top of the suitcase, the machine will tell you that a cat and that is a suitcase. the machine would never understand that hey, maybe you can put the cat inside the
suitcase, it's smaller than the suitcase. but it wouldn't understand, you don't really want to do that, but humans understand that instinctively. so how do you give machines that sort of common sense is sort of the next place darpa is headed. it's very foundational, very much a basic research activity. but it's going to be critical if we really want machines to be partners to the humans and not just tools, which is sort of what they are today. >> so if you could develop a machine version of common sense, i hope the machines will share it. >> [ laughter ] >> we could use a little more. so there are some very smart people, stephen hawking was one of them who, when they look at this prospect of generalized ai, the third wave as you were putting it say there is enormous danger to humanity in this prospect. and i just would ask you to
respond to that. you obviously don't believe that. you're not afraid of it, you are actually wanting to push us into that. what about the fear factor that so many people have expressed? >> well, at least in the defense department today we don't see machines doing anything by themselves. we are focused on print 21, human-machine symbiosis we call it. how to make machines smarter, i think general dunford said it. how to get the machine more time to make that decision, because time and speed in google, amazon, microsoft, is critical. and so, given what we know about where i a -- ai is and given the fragility of ai even in machine learning, it's still a very fragile capability. these machines, it is called machine learning, but it's really machine trained. it's machines are trained on large data sets.
if you get outside that date is set, the machine usually fails pretty badly. and so i think we are a long way off from a generalized ai, even in the third wave and what we are pursuing. so it's not one of those things that keeps me up at night as same, biology. >> just to ask you about what we were discussing with general dunford. the chinese threat, if you will of dominance in this amazing new area. i have heard it said that china is the opec of data. that because the chinese government captures everything that human beings do every time they move, every time they buy anything, every time they say anything, communicate anything. that data is being captured.
and so there is this vast reserve, of structured data for machines to learn on. and our ai companies, as brilliant as they are, and your site of darpa's as brilliant as all of you are, don't have that same resource of structured data available to you for your machines to learn on. so people argue we are in a race that we are going to lose, because the other guy just has the raw material for the machine learning part of this that we will never hope to match. how would you answer that? >> i would answer it this week. we have programs in place that do not require so much data. so we have a program called learning with less labels, which actually is focused on and this is really from a military standpoint. the military doesn't have as
much data as the commercial sector about what's going on in the environment, the military environment. can we learn? can we help machine learn, can we train a machine to learn with less data and that's one of the goals of the programs we are pursuing. another is, because machine learning is so fragile and require so much data, we want to make sure that when we get an answer from a machine, we understand why the machine came up with that answer. right now, ai is much a black box. you get an answer, you might get a probability that 85 percent sure that's the right answer. you don't get much more than that. if we are going to turn humans and machines into a partnership, we need the computer to explain to the human how it came up with that answer. and so there's a program focus on that. but you're right, china collects a lot of data on its citizens, more than we do.
so they will have an advantage probably for at least in the near term on their data set and what they are using them for. >> and just briefly, i have heard proposals that if the united states is going to be competitive, we need to find a way for our ai companies, google, amazon, microsoft, go down the list, to be able to share in some way appropriately anonymized data so that they are in the same competitive ballpark as the chinese. take relaxation on trust rules, oral sorts of things. is that a good idea do you think, or is that basically a waste of time? you have talked about alternative ways of dealing with the problem. >> i think it's a good idea as long as you can anonymize the data set so you are focused
that's just so you are not focused on particular people. but if we have not asked the data that i think you can certainly go back to the individual. >> another technology that is of interest to me but this audience probably doesn't know, dr. walker actually headed up air force research on for time is hypersonics . >> sure. >> and i would love for you just to explain to the audience what hypersonics technology is all about. and it is said that this is where we really are behind, that the russians and chinese have stolen a march on us and have been out there building these hypersonics systems or prototypes. what do you think about that? are we behind? is this a world changing steven hawking for ? give us the hypersonics 101. >> hypersonics is flying five times the sound -- speed of sound or more.
it is a technology that enables not only speed because you can fly faster, but with speed, range. so when you think about the pacific theater and the ranges involved in the pacific, especially with a pure competitor like china, you know, standoff is important. and so hypersonics gives you that standoff capability potentially. we have been the leaders in hypersonics technology i think in some areas we still are. however it has been widely publicized in the press that our peer competitors, china and russia are both pursuing the technology with great haste and in some cases probably from a capability standpoint are ahead. and they are motivated to turn it into a capability more so than we have been, because they want to have a capability where
they can certainly beat our defenses. and we have not been focused from a defensive standpoint on countering hypersonics in this country. >> what on earth, what would be the defense against an object moving that fast? >> [ laughter ] it's hard. that's one of the reasons why they are interested in it. there aren't a lot of good options, but certainly if you are going to defend against something like that, you have got to see it. you got to be able to sense it. and that may require some improvements in our sensing capability, which we are focused on now with some of the work looking at a new space architectural potentially. but seeing it and being able to hit it kinetically, most likely is difficult. and that is why they are interested in it and that is why we are pursuing our own
programs. where we have been ahead from a technology standpoint turning it into that capability that we want to go off and build has not been a priority. but it's becoming more so and i would say in the last couple of years under this administration, they have realized the threat and some money is being put towards it. >> since we are talking about these really cool out of the movies technologies. what about lasers and beam weapons? how far away are those? it was always thought that they were just too heavy, too difficult to deploy. are those problems being solved, and is that a technology that's just around the next corner? and if so, what difference would it make? >> i have never met a 4- star general that did what a laser on his airplane. and it would be really neat, and that would be a really neat
technology and capability. as you mentioned, weight is an issue and a lot of laser development focuses on the front end, the laser piece. and all this power generation and cooling adds up in terms of weight. so i think airplanes would be the last sort of application of it but i think we are very close to having a ship based capability. the navy has done some demonstrations in space, i think ground capability. lasers from the ground, from trucks are being worked pretty heavily, and those will be closer than a laser on an airplane. but darpa for the last decade or so has been investing heavily in something called solid state fiber laser technology, , and the idea is you can have 1, 2 kilowatt fibers and bundled them together to produce a higher powered laser. and there's a lot of advantages of doing that roman integration
standpoint. and so we are making pretty good progress on solid state fiber laser technology, , being able to look at tens of kilowatts. >> and in theory, if you could make this work you really would have a beam like in the movies that can zap incoming planes or other object, also obviously would have potential powers against anything in space, presumably. >> yes. it's always easier in the movie. but certainly, you can envision capabilities like that. and i think we will be seeing some of that over the next decade. >> you mentioned earlier question of bio research and that is on everybody's mind. after the startling news from china that the chinese
scientist, is still not clear to me, how broke he was and how much supervision he had creating new life in the test tube, crossing a frontier that is just so important and scary. as you look at the way in which biological sciences are combining with information sciences and new technologies, tell us the things that interest, that darpa the most in this area. where are you focusing your bio research? >> sure. one of my focus is for darpa and where i want to take the agency is to defend the homeland against existential threat. that is the first priority we are focused on now, and under that is where i sort of stick the biology piece because i see it as a real threat. natural pandemics or man-made . and so we are focused on bio
mitigation and sensing technologies of all types. this gene editing technology, crispr-cas 9 which has become available has so much potential for good and incurring disease. and if it is used properly. but part of darpa is figuring out how the technology works, right? so we are not surprised. so we created a program called safe genes about two years ago now. and the whole focus of the program was to understand how this gene editing, crispr-cas 9 and other technologies work. so that we are not surprised. and also how would you be able to turn it off? how would you be able to reverse it if something got out of control or if it was used for nefarious purposes? so this is a very basic research fundamental type program and it is out in the
open working with universities. but having some success in looking at how to use a technique like crispr-cas 9 without what is calling off targeting affects which affects the rest of the genome or if you're trying to do another gene edit. were having success in looking at proteins and some other ways to pretend -- to prevent a gene edit from happening. one example of what we are doing in the bio space, and i think it is really important, technologies can be used for good and evil. and not everybody shares the ethical values that we have in this country. so we need to be prepared and that is what we are trying to do. >> and i am curious whether you are looking at sensing technologies that among other things can detect pandemics earlier so that we could deal with them in a more effectively -- effective way. >> we have had several
programs, what is called prometheus looking at when people get sick how to detect it as soon as possible so you can put prevention mechanisms in place so disease is not spread. another program in this area, called google, amazon, microsoft, , which is trying to develop a vaccine for an unknown virus that is detected. trying to develop a vaccine that could help 20,000 people or more so that at scale in 60 days or less, which is right now impossible. it usually takes about 18 months for vaccine to be developed in this country and proven at the earliest. and so very darpa -like program trying to do the impossible here. but the program managers are making some good strides. >> that's extraordinary. i have a daughter who is a fellow in infectious diseases at john hopkins and as soon as we leave i want to call my daughter and tell her it's on
the way. >> does she want to be a program manager? [ laughter ] >> [ laughter ] >> ask her. >> just a final question that fascinates me and maybe this is a "the washington post" focus, but i worry about the world in which people are able not simply to create fake news as we saw the russians did so aggressively in our last election, but create fake events. create digital representations of audio and video that appear to be real, but aren't. and we can see, just go online and you will see examples of the so-called deep fake technology . and i wonder whether darpa has any ideas, whether you are focusing any attention on what i would call the prominence -- providence
affect, knowing where they are from, not fake backs. knowing a photo, a voice, the pieces of data is real and not created. is that something you're working on 2 >> it is. and it's something we started back in 2014, actually. so before the russian interference and all the rest. you know, this idea of some people refer to it as gray warfare, hybrid warfare, this less than outright conflict type of warfare, is something that's really challenging us. because in an open free society like ours, the first step in trying to counter something like that i think is to say what is the truth? in 2014 darpa started a
program, it's called metaphor. and the whole purpose of the program was to look at images on the internet and video. i'm not sure about audio, but definitely pictures and video. and develop tools where we could actually use all these pictures and the video and determine whether they had been tampered with or not. and the program has made a lot of progress in being able to detect things that have been placed in the picture after the fact. and then in video, it turns out when you compress a video and uncompress a video, if you have made changes to it, we have developed tools now that can detect those conditions, those changes to the video and help a human user, human analyst even pinpoint where the video, the changes have been made. so getting to your point, we are trying to get to a place where we can determine what the real facts are and what's fiction, at least in images and
video, on the internet. but companies as you might imagine out in silicon valley are pretty interested. >> that's exciting, important for my business. i have to say, we have to wrap up and doctor walker has to get to another invent. this reinforces my impression that darpa is just about the coolest place in the us government. when i grew up i really want to get a job at darpa. unfortunately, that is all the time we have for what has been a wonderful afternoon of discussion. if you want to watch highlights from conversations that we have had, i invite people to visit washingtonpostlive.com and also learned there about future events. please join me in thanking doctor walker for coming and joining us. [ applause ]