[0:00:24] This hearing will come to order. Good afternoon. Welcome to each of the witnesses. Thank you for joining us. Thank you for everyone for attending this hearing. Throughout history, mankind has refused to accept the complacency of the status quo and has instead looked to harness creativity and imagination to reshape the world through innovation and disruption. The Industrial Revolution, Henry Ford's moving assembly line.
[0:00:54] The invention of flight and commercial aviation, and more recently, the creation of the Internet, have all acted as disruptive forces that have not only changed the way we live, but have been engines for commerce that have offered consumers enormous freedom. Today, we're on the verge of a new technological revolution. Thanks to the rapid advances in processing power, the rise of big data, cloud computing,
[0:01:24] Mobility due to wireless capability and advanced algorithms. Many believe that there may not be a single technology that will shape our world more in the next 50 years than artificial intelligence. In fact, some have observed that as powerful and transformative as the internet has been, it may be best remembered as the predicate for artificial intelligence and machine learning. Artificial intelligence is at an inflection point.
[0:01:56] While the concept of artificial intelligence has been around for at least 60 years, more recent breakthroughs, such as IBM's chess-playing deep blue victory over world champion Garry Kasparov, advancements in speech recognition, the emergence of self-driving cars, and IBM's computer Watson's victory in the TV game show Jeopardy, have brought artificial intelligence from mere concept
[0:02:27] Whether we recognize it or not, artificial intelligence is already seeping into our daily lives. In the healthcare sector, artificial intelligence is increasingly being used to predict diseases at an earlier stage, thereby allowing the use of preventative treatment which can help lead to better patient outcomes, faster healing, and lower costs. In transportation,
[0:02:55] Artificial intelligence is not only being used in smarter traffic management applications to reduce traffic, but is also set to disrupt the automotive industry through the emergence of self-driving vehicles. Consumers can harness the power of artificial intelligence through online search engines and virtual personal assistance via smart devices, such as Microsoft's Cortana, Apple's Siri,
[0:03:26] Amazon's Alexa and Google Home. Artificial intelligence also has the potential to contribute to economic growth in both the near and long term. A 2016 Accenture report predicted that artificial intelligence could double annual economic growth rates by 2035 and boost labor productivity by up to 40%. Furthermore, market research firm Forrester recently predicted
[0:03:56] that there will be a greater than 300% increase in investment in artificial intelligence in 2017 compared to 2016. While the emergence of artificial intelligence has the opportunity to improve our lives, it will also have vast implications for our country and the American people that Congress will need to consider moving forward. Workplaces will encounter new opportunities thanks to productivity enhancements.
[0:04:25] As artificial intelligence becomes more pervasive, Congress will need to consider its privacy implications. There is also a growing interest in this technology from foreign governments who are looking to harness this technology to give their countries a competitive advantage on the world stage. Today, the United States is the preeminent leader in developing artificial intelligence. But that could soon change. According to the Wall Street Journal, quote,
[0:04:55] The biggest buzz in China's internet industry isn't about besting global tech giants by better adapting existing business models for the Chinese market. Rather, it's about competing head-to-head with the U.S. and other tech powerhouses in the hottest area of technological innovation, artificial intelligence. Seeding leadership in developing artificial intelligence to China, Russia, and other foreign governments.
[0:05:23] We'll not only place the United States at a technological disadvantage, but it could have grave implications for national security. We are living in the dawn of artificial intelligence. And it is incumbent that Congress and this subcommittee begin to learn about the vast implications of this emerging technology to ensure that the United States remains a global leader throughout the 21st century. This is the first congressional hearing.
[0:05:54] And I am confident it will not be the last as this growing technology raises opportunities and potential threats at the same time. I look forward to hearing from our distinguished panel of experts today. And at this point, I'll yield to our subcommittee's ranking member, Senator Peters, to give an opening statement.
[0:06:16] Well, thank you, Chairman Cruz, for calling this very important meeting, and I'd like to thank the witnesses for taking the time as well, and look forward to hearing your expertise on this exciting area. Certainly, our nation's history has always been defined, in my mind, by our ongoing search for the next innovation, the next big thing that's going to advance our economy and our society.
[0:06:41] Our mastery of manufacturing and automation helped the United States establish its industrial and military might. Looking to the future, we must continue to harness this American drive to discover, to create the next big advancement, to keep our nation on the cutting edge. Over the last few decades, technology has changed the way we make and how we make it. Today, we boast an ever-evolving innovation ecosystem that's rooted in robotics and machine learning.
[0:07:10] And the subject of today's hearing, artificial intelligence. AI products like our smartphones intelligent personal assistant or our banks fraud detection technology are already improving the day-to-day lives of Americans. New advances in computer processing and cloud computing are driving rapid expansion in AI development in industries as diverse as healthcare, transportation, education, and security. And they've all gone through enormous change, as all of you know.
[0:07:39] We as a society need to help foster this broader ecosystem so we can capitalize on its benefits, including cleaner energy, new economic growth, improved safety and health, and greater accessibility for the elderly and disabled populations. Being from Michigan, I've had the opportunity to experience firsthand the development of self-driving cars.
[0:08:00] That are made possible by a combination of advanced automatic braking and lane changing systems, cameras, sensors, high performance computing, deep learning systems, 3D high definition maps, and artificial intelligence. Just last year, over 35,000 people died in motor vehicle crashes. But research suggests that about 94% of those accidents were the result of human error.
[0:08:27] With safe deployment, AI-fueled automated vehicles could significantly decrease this number, saving countless thousands of lives. In addition, the World Economic Forum estimates that the digital transformation of the automotive industry alone, just the automotive industry alone, will generate $67 billion in value for the sector and yield $3.1 trillion in societal benefits before 2025.
[0:08:53] That's billions injected into our economy to boost our competitiveness and billions saved due to major reductions in auto accident injuries and deaths, environmental degradation caused by traffic congestion and air pollution. More broadly, U.S. technology companies spent $8.5 billion on AI in 2015, more than four times the amount spent in 2010, and experts predict incredible growth in the coming years in healthcare, marketing, and finance sectors.
[0:09:23] A critical component of this subcommittee's work is promoting and preserving American competitiveness. And while the U.S. leads the world in investment and discoveries in AI, as we all know and as Chairman mentioned, China is quickly catching up. In 2014, Chinese scientists overtook U.S. scientists in terms of new papers published and citations of published papers in the area of deep learning, a cutting edge of AI research.
[0:09:50] Like computer processing, nanotechnology, and biotechnology, AI has the power to dramatically change and grow the economy, but we must invest in research and in STEM education to maintain our competitive advantage. Analysts by the Council of Economic Advisers shows that doubling or tripling all research investment, something that I fully support, will allow us to continue to grow. However, targeting increases in areas of high economic productivity like AI
[0:10:19] May offer benefits with a much smaller budgetary impact. In fact, a recent report from the White House recommends that current government spending on AI research should be doubled or even quadrupled to achieve optimal economic growth. Certainly look forward to hearing your comments on that. We've already seen enormous gains from federal investment in this area through NASA's work to develop AI applications for use in robotic spacecraft missions. These applications include planning, spacecraft autonomy, image processing,
[0:10:49] and Rover autonomy. NASA also utilizes AI and Earth observing satellites to optimize observations of natural disasters like volcanic eruptions. And I look forward to hearing more about this pioneering work by NASA and the ways in which other industries have benefited from date. And finally, while we must strive to optimize the full economic potential of AI, we must also address its potential impacts on the workforce that we know today.
[0:11:18] While new jobs will be created because of AI, we also have to think critically about the steps we can take today and in coming years to make sure that American workers are not left behind. The subcommittee plays a unique role in studying emerging technologies and examining ways to promote and harness scientific advancement for the greater good. And I appreciate the chairman's interest in this important issue. I look forward to working with him.
[0:11:44] And continuing to advance development at AI for the greater good. That's why I look with great anticipation to the testimony from each of you. Thank you. Thank you, Senator Peters. We'll now recognize the ranking member of the full committee, Senator Nelson, for an opening statement. Thank you, Mr. Chairman. Indeed, AI has helped the space program quite a bit. I'm delighted that a representative from JPL is here. JPL are the whiz kids.
[0:12:15] The rock stars, and in fact, they are rocket scientists. And so that's a part. I want to pick up, there's another part about AI, and that is the replacement of jobs. And we've got to prepare for that. For example, Elon Musk recently said that...
[0:12:45] A dilemma recently in an interview with CNBC. He predicted robots could eventually take over many jobs away from folks that they would have to depend on the government in order to have a living. Elon used the example of truck drivers who could be displaced in the future by autonomous vehicles.
[0:13:15] And those kind of advancements that would allow trucks to drive themselves. And yet, if a whole occupation is suddenly displaced, what do we do? We just came through an election where the loss of jobs was a big topic. And maybe truck drivers don't want to be trained to go on a computer.
[0:13:46] What are we going to do for the future? And so this is just another challenge that we faced as technology advances. And that's why we're here today. So thanks. I'm looking forward to it. Thank you, Senator Nelson. I'd now like to introduce our witnesses. The first witness is Dr. Eric Horvitz, who is the interim co-chair of the Partnership on Artificial Intelligence.
[0:14:14] and serves as the managing director of the Microsoft Research Redmond Lab. Dr. Horvitz's research contributions span theoretical and practical challenges with computing systems that learn from data and that can perceive, reason, and decide. His efforts have helped to bring multiple systems and services into the world, including innovations in transportation, healthcare, aerospace, e-commerce, online services, and operating systems.
[0:14:43] Dr. Andrew Moore is the Dean of the School of Computer Science at Carnegie Mellon University. Dr. Moore's background is in statistical machine learning, artificial intelligence, robotics, and statistical computation of large volumes of data, including decision and control algorithms. Mr. Gary Brockman is the co-founder and chief technology officer of OpenAI, a nonprofit artificial intelligence research company.
[0:15:12] Prior to OpenAI, Mr. Brockman was the CTO of Stipe, a financial technology company that builds tools enabling web commerce. And Dr. Steve Chen is the technical group supervisor of the artificial intelligence group and the senior research scientist in the mission planning and execution section at NASA's Jet Propulsion Laboratory. I would also note that Dr. Andrew Futrell was set to testify at our hearing today.
[0:15:43] But unfortunately, due to weather issues, he was unable to travel to D.C. today. Dr. Futrell is the chair of the Department of Genomic Medicine at the University of Texas MD Anderson Cancer Center in my hometown of Houston. We thank him for his previously submitted written testimony, and if there are no objections, I would like to submit Dr. Futrell's testimony into the hearing record. With that, we will move to the testimony, although I will note for each of you that your...
[0:16:11] Experience and wisdom in this topic will be welcomed and sorely needed. And intelligence, artificial or otherwise, is not something we deal with often because this is the United States Congress. And with that, Dr. Horvitz, you may give your testimony.
[0:16:31] Thank you Chairman Kruse, Nike member of Peters, and members of the subcommittee, and good afternoon and thank you for hosting this discussion on AI. It's fabulous. To start, AI is not one thing. AI is a constellation of disciplines including computer vision, machine learning, language understanding, reasoning and planning, and robotics. But they're all aimed at a shared aspiration, the scientific understanding of thought and intelligent behavior, and developing computing systems based on these understandings.
[0:17:00] Many advances over the 60-year history of AI are now actually part of our daily lives. Just consider the AI route planning algorithms we use daily in our GPS systems. And while we've seen many advances over time, it's clear that we're now at an inflection point. We're seeing an acceleration of AI competencies in many areas, and the inflection is driven by a confluence of several factors, and these include the unprecedented quantities of data.
[0:17:28] that have come available with the widespread digitization of our lives, increases in computing power over time, and the recent jumps in the prowess of our algorithms, the methods we use, particularly machine learning methods that learn to predict and to diagnose from data. The advances are putting unprecedented technologies in the hands of people, including real-time speech-to-speech translation among languages now available for Microsoft's Skype services and computer vision for assisting drivers.
[0:17:58] AI technology that is already available today could save thousands of lives and many billions of dollars if properly translated into practice. And the key opportunities before us include health care, transportation, education, as well as agriculture, manufacturing, and increasing accessibility for those with special needs. Other directions that are critical
[0:18:22] Include using AI advances to enhance the resilience and capacity of critical infrastructure like our electrical power grid and road network. So let's take as an example healthcare to start. AI is a veritable sleeping giant for healthcare. AI technologies will be extremely valuable for handling acute as well as chronic illnesses and for making our hospitals safer. As an example, we've built systems
[0:18:46] That can predict patient outcomes and that make patient specific recommendations to allocate scarce resources. And these prototypes have been applied to address such challenges as chronic diseases, hospital readmissions, hospital associated infections, and catching preventable errors in hospitals responsible for over a quarter of a million deaths per year in the United States.
[0:19:09] And while on the topic of saving lives, advances in pattern recognition systems enable us to develop effective automated braking and control systems that will keep us safer. We could take a big cut out of those 30,000 deaths that we've become accustomed to tolerating every year in our country, and we don't often think about the 300,000 incapacitating injuries on our roads every year.
[0:19:33] Numerous other AI innovations can help with safe driving. This week, our team just published results that show how we can leverage data from Minneapolis to build a routing system that identifies the safest routes to take, considering a multitude of factors at any moment, even how the sun is shining and the glare that it produces. Moving forward, key research directions in AI include focusing on human-AI collaboration, ensuring the robustness, safety, and security of AI systems,
[0:20:01] And identifying and addressing the ethical, legal, economic, and broader societal influences of AI. As an example of important research, we need to endow systems with the ability to explain their reasoning to people. People need to trust systems to use them effectively. And such trust requires transparency. There's also important work to be done with developing robust and resilient AI systems.
[0:20:25] Especially when these systems are used for high stakes, safety critical applications amidst the complexities of the open world. We must also be aware that AI systems can present new kinds of attack surfaces that can be disrupted by cyber attacks. So it's important to address rising cyber vulnerabilities as we develop and field these more capable AI systems. Another area of importance is the influence of AI on our nation's workforce and economy.
[0:20:53] And while estimates vary, the economic influence of AI will likely be in multiples of trillions of dollars. However, along with the expected benefits come concerns with how AI will affect jobs and income disparities. And these are important issues we must closely reflect, monitor, and plan to ensure a smooth transition to a world where AI plays a more important role. One thing for sure is that we urgently need to prioritize the training and retraining of the US workforce.
[0:21:22] So that our workforce skills are aligned with needs, we also need to double down on investments in STEM education and training. And moving forward, continued strong and public and private sector support of research and studies on the scientific and socioeconomic challenges of AI will be critical to ensuring that people in society get the most out of AI advances. So in summary, we expect AI advances to raise our quality of life, to empower citizens in new ways, but with
[0:21:50] As with any technical advance, we need to invest efforts to study and address potential challenges, concerns, inequities that may come along with the benefits that we expect from AI. Thanks very much. Thank you, Dr. Horvitz. Dr. Moore. Thank you, Chairman Cruz. Thank you, Chairman Cruz, Ranking Member Peters, and members of the subcommittee for convening this really important meeting.
[0:22:16] What I want to do with this testimony is three takeaways. First, what AI is and what it isn't right now. The artificial in artificial intelligence is there for a reason. Secondly, why things are changing so quickly right now. And thirdly, the thing that keeps me awake at night regarding US competitiveness in AI. So what is AI?
[0:22:45] When one of my students decides to build an AI, they always end up doing two things. One, they make sure the computer can perceive and understand the world through computer vision and through big data. Second, they do massive search, which means in all these examples, such as a car finding a route, the computer searches to find the best of billions and sometimes quintillions of possibilities.
[0:23:14] And does that very quickly and efficiently. And that's happening all the time. When you ask Siri a question, what's happening up there in the cloud is it is essentially trying out about 10 billion possible answers to your question and searching over them to score each one as to how likely is it to make you satisfied. If a car, autonomous car, is about to hit a deer and it's about
[0:23:41] 0.2 of a second to impact. It can spend the first 20th of a second gaming out millions of possibilities of what it can do, what the deer is going to do to maximize the chance that people are going to survive. So that's what's going on in AI. It is perception through big data and search. What's really changed in the last couple of years is
[0:24:07] Through the efforts of the United States in big data, we now have computers which can perceive enough for this to be useful. And this means that there's now a big land grab going on for researchers and entrepreneurs and students for finding all the places where we can use this. For example, this picture here is one of Sebastian LePage.
[0:24:32] who is VP of operations at Canova, a robotics arm company, working with Carnegie Mellon faculty on the question of if you are unable to use anything below your neck, but you want to indicate by nodding at something that you'd like to pick it up, perhaps nod again to say you want to bring it to your mouth, searching over the billions of possible things the arm can do safely to find the one which is most likely to make you, as the user, happy.
[0:24:59] This is an unambiguously good use of technology applicable to tens of thousands of our veterans, for example. But what excites me and the reason I'm in this business right now is that there's thousands of stories like this happening. And my students and faculty and folks all across the country who know, who have the skills, are exploring the ways of doing this. For example...
[0:25:25] One of our undergraduate students, working by herself, using OpenAI tools, managed to quickly come up with a system for checking open internet records to detect sex traffickers. And our algorithm has saved a couple of hundred young women from sex trafficking. That's one person who had the skills able to do this. Another example.
[0:25:50] This wasn't possible 12 months ago. Now we have drones zooming quickly through forests, able to dodge trees as they're going because they can now afford to plan so fast to avoid trees in sub-second time. Many other examples, so many that I could keep us going for half an hour. They all just make me so excited. But the thing that keeps me awake at night on all of this is the talent war.
[0:26:18] I really, really beseech that together we can get a million of the current middle school students in the country to become AI experts over the next five to ten years. At the moment, we have a tiny fraction, I would say less than 1% of the people available to work in this area who could work in this area. If you duplicated these four panel members 100 times each, we still would have too much to do.
[0:26:46] When it comes to taking advantage of all these opportunities. I estimate that when an internet company hires one of our students, they're making $5 to $10 million per student just by having that person on their payroll. And so the bidding wars for this talent are huge. And our students, instead of necessarily moving to work on AI for the Veterans Administration,
[0:27:13] AI for helping protect our warfighter. The majority of them are simply going off to internet companies, which is fine, but I want them in all the other sections of our economy as well. Similarly, if you look at every one of the big advances in artificial intelligence that are now active, they came from university professors, the majority of them from US university professors.
[0:27:37] We are beginning to lose many of our university professors to industry, and that is damaging our seed corn. Our university professors really are looking, the AI experts among them, are looking for sustained, stable funding, not necessarily lots of funding, so that they can realize their dreams of doing things like this. So I'm very excited. I'm very grateful for this subcommittee.
[0:28:02] Shining a light on this important issue. And I think the future is bright, but it really is an AI race at the moment. Thank you, Dr. Moore. Mr. Brockman. Thank you, Chairman. Thank you, Chairman Cruz, Ranking Member Peters, distinguished members of the subcommittee, as well as their staff. This is a really important session, and I'm honored to be giving this testimony today. I'm Greg Brockman, co-founder and chief technology officer of OpenAI. OpenAI is a nonprofit.
[0:28:32] AI research company with a billion dollars in funding. Our mission is to build safe, advanced AI technology and to ensure that its benefits are distributed to everyone. We're chaired by technology executives Sam Altman and Elon Musk. The US has led essentially all technological breakthroughs of the past 100 years. And they've consistently created new companies, new jobs, and increased American competitiveness in the world. AI has the potential to be our biggest advance yet.
[0:29:06] Today, we have a lead, but we don't have monopoly when it comes to AI. This year, Chinese teams won the top categories in a Stanford annual image recognition contest. South Korea declared a billion dollar AI fund. Canada actually produced a lot of the technologies that have kicked off the current boom. And they recently announced their own renewed investment into AI. So right now, I would like to share three key points for how the US can lead in AI.
[0:29:35] The first of these is that we need to compete on applications. But when it comes to basic research, that should be open and collaborative. Today, AI applications are broadening. They're helping farmers decide which fields to seed. They're helping doctors identify cancers. But the surprising thing is that industry is not just capitalizing on the advances that have been made to date. Companies like Facebook, Google, Microsoft, they're all performing the basic scientific research, the kind of work that you would expect to see just in academia.
[0:30:05] And they're trying to create the new AI building blocks that can then be assembled into products. And even more surprisingly, these industrial labs, they're publishing everything that they discover. They are not holding back any secrets. And the reason they do this is because publication allows them to pool their resources to make faster breakthroughs and to attract world-class scientists. Now, these companies, they stay competitive by
[0:30:31] They publish the basic research, but they don't talk about how they put this stuff together to actually make products, to actually make the things that are going to make dollars for the company. For example, IBM Watson, Microsoft Cortana, there aren't papers on how those are built. And the thing that's happened is that this openness has concentrated the world's AI research and corresponding commercial value all around the United States. This includes attracting many of the Canadian scientists who really kicked off this AI boom. They're here now. In fact, one of them is one of my co-founders at OpenAI.
[0:31:02] And importantly, this has allowed us to define the cultures, the values, and the standards of the global AI community. Now, this field is moving so quickly that basic research advances tend to find their way into products in months, not years. And so the government can directly invest in American innovation and economic value by funding basic AI research. The second thing that we need to do is that we need public measurement and contests.
[0:31:32] There's really a long history of contests causing major advances in the field. For example, the DARPA Grand Challenge really led directly to the self-driving technology that's being commercialized today. But really importantly as well, measures and contests help distinguish hype from substance, and they allow for better forecasting. And so good policy responses and a healthy public debate are really going to depend on people having clear data about how the technology is progressing.
[0:31:59] What can we do? What still remains science fiction? How fast are things moving? So we really support OSTP's recommendation that the government keep a close watch on AI advancement and that it work with industry to measure it. The third thing that we need is that we need industry, government, and academia to start coordinating on safety, security, and ethics. The internet was really built with security as an afterthought, and we're still paying the cost for that today.
[0:32:27] With AI, we should consider safety, security, and ethics as early as possible, and that means today, and start baking these into the technologies, into the fundamental building blocks that are being created today. Academic and industrial participants are already starting to coordinate on responsible development of AI. For example, we recently published a paper together with Stanford, Berkeley, and Google laying out a roadmap for AI safety research. Now, what would help is feedback from the government about what issues are most concerning to it.
[0:32:58] So that we can start addressing those from as early a date as possible. As the chairman said in his opening statement, Accenture recently reported that AI has the potential to double economic growth rates by 2035, which would really make it into the engine for our future economy. The best way to create a good future is to invent it. And we have that opportunity with AI by investing in open basic research, by creating competitions and measurement, and by coordinating on safety, security, and ethics.
[0:33:27] Thank you for your time and I look forward to the Q&A. Thank you Mr. Brockman. I was encouraged by your testimony about the Canadian scientists coming to this country and I will say as someone born in Calgary that I think there are colleagues of mine on both sides of the aisle who have concerns about Canadians coming to this country. Dr. Chen.
[0:33:51] Chairman Cruz, Ranking Member Peters, and members of the committee, thank you for this great opportunity to speak to you on this topic of artificial intelligence and specifically its relationship to space exploration. For the record, I'm here as an employee of NASA's Jet Propulsion Laboratory, which is a federally funded research and development center managed by the California Institute of Technology for NASA. As a senior research scientist in autonomous systems at JPL, I work on the development and application of artificial intelligence to NASA's missions.
[0:34:20] I've had the privilege to lead the deployment of AI software to NASA's Earth Observing One mission, NASA's Mars Exploration Rovers mission, and also the European Space Agency's Rosetta mission. We focus on using AI to improve the effectiveness of conducting science and observation activities in NASA's missions. I know of no better introduction to this topic than to point out that as we speak right now,
[0:34:47] There's a spacecraft called Earth Observing One that's flying about 700 kilometers overhead, weighs about 500 kilograms, and is flying at 7.5 kilometers per second that is fully under the control of AI software. This spacecraft has been under the control of this AI software for over a dozen years and has successfully acquired over 60,000 images under the control of this software and issued over 2.6 million commands.
[0:35:13] The AI software that's used in the operation of this mission includes constraint-based scheduling software to enable the spacecraft to be operated by end users, scientists and people who monitor natural hazards such as volcanoes and flooding. Onboard software, including machine learning classifiers, enables the spacecraft to more effectively monitor these science events. Again, flooding, volcanism, as well as cryosphere, the freeze and thaw of the Earth's environment.
[0:35:41] Furthermore, in a range of collaborations all around the world, this spacecraft has been integrated into a network with other space systems as well as ground sensor networks. And the extent of this multi-agent AI system goes as far as Thailand, Iceland, Sicily, Namibia, and even Antarctica. What this system enables us to do...
[0:36:07] is enable data from one part of the system, such as a seismographic sensor at a volcano in Antarctica, to trigger the observation of the system via space assets. Going even further afield, on Mars, autonomous navigation software is at the heart of all of the Mars rover exploration missions. And this is, at its core, AI-based search software. AI and computer vision software form the core of the Aegis system, which is now operational on both the Mars exploration rover missions
[0:36:37] and the Mars science laboratory rovers. Aegis enables the rovers to automatically target science measurements based on general science criteria such as texture, size, shape, and color without the ground in the loop, dramatically enhancing the science that the rovers can conduct. Machine learning has also had significant impact in dealing with the enormous data sets that space missions produce. Just two examples. In the very long baseline array,
[0:37:06] Radio science is being enhanced by machine learning. Machine learning is used to identify millisecond duration radio transients and reject radio frequency interference events. Here, machine learning allows the automatic triage from thousands of candidates down to tens of candidates for manual review by highly expert scientists. In visual astronomy, in the intermediate palomar transient facility, machine learning is applied to identifying transients.
[0:37:35] Point source transients are basically supernova, and streaking transients are near-Earth objects. Here, machine learning has been used to perform fast daily triage of millions of candidate events down to tens of events, again, allowing the human experts to focus on the most likely candidates and enhance the science. While these examples may give you the impression that AI is commonplace in space exploration, I assure you this is not the case.
[0:38:05] The above examples are a sampling of AI success stories on a small fraction of the overall space missions. Because of the high-stakes nature of space exploration, the adoption of disruptive technologies like AI requires an extensive track record of success as well as continuous contact with the key stakeholders of science, operations, and engineering. However, AI has made tremendous progress in recent years. Instruments on the Mars 2020 rover
[0:38:34] will have unprecedented ability to recognize features and retarget themselves to enhance science. The Mars 2020 rover mission is also investigating the use of onboard scheduling technologies to best use available rover resources. And the Europa multi-flyby mission is also investigating the use of onboard autonomy capabilities to achieve science despite the Jupiter radiation environment, which causes processor resets.
[0:39:01] In the future, AI will also have applications in the MAN program in order to best use scarce astronaut time resources. Past efforts have placed AI in a critical position for future space exploration to increasingly hostile and distant destinations. What we need is sustained resources and a commitment, support, and vision for AI to fulfill its vast potential to revolutionize space exploration.
[0:39:29] Thank you very much for your time. Thank you, Dr. Chan, and thank you to each of you for your testimony. Let me start with just a broad question to the panel, which is what are the greatest challenges and opportunities you see for the continued development of AI? I do think it's very important that we grow our AI workforce quickly, and it's interesting that in a world where we're actually all concerned about making sure there are
[0:40:05] More jobs available that there's such a gap here where we're so short of experts here. Frankly, I look at some of the other major players around the world in this area and China, India, other countries are really pumping out the computer scientists who can form this cohort. So for me, I would feel much more comfortable if we were graduating
[0:40:33] Hundreds of thousands of AI experts every year from our universities instead of just thousands. So let me also complement that work by talking about some technical directions. I mentioned human computer or human AI collaboration. We don't think enough about the human factor.
[0:40:51] It's not all about automation. Of course, there'll be some interesting automation. We can't have people on Mars, for example, looking at those stones and doing the digging. But in general, there's incredible opportunities ahead with co-designing systems so they work really well. They're human aware. They understand human attention. They understand how they can complement human intellect and what people do uniquely and do well. Understanding how...
[0:41:18] to negotiate, to do a give and take, a fluid dialogue in contributions between humans and machines. Lots to be done there, and that includes this piece with explanation, transparency. Many of these answers we get today out of AI systems, the best systems we can build are black box systems. They're opaque to human beings who need to understand to learn how to justify those decisions and how the thinking is done and to understand the reasoning process itself.
[0:41:46] Lots of work to do there. There's another critical direction with thinking through opportunities to take some of the software we've done in the intellectual cognitive space and enter into the real world of physical innovation to teach systems to work in physical spaces. Our robotics today are very, very limited. Even our best practices on Mars don't do the kinds of things that people can do easily.
[0:42:13] There's a coming renaissance in taking some of our advances in AI and bringing them into the physical space. So I believe that the biggest opportunity we have is to continue to move very quickly on the fundamental building blocks, on the fundamental technology. And it really feels like today we're kind of in the vacuum tube era and that the transistor is out there and that we're building very impressive technologies, but that in the future we're going to be, this is really just the tip of the iceberg.
[0:42:39] And I think that the biggest thing to watch out for, I think one of the biggest risks, is that we lose the openness that we have. Today we can have these conversations. We can measure how the technology is progressing. And we can plan for the future. And I think that we can continue to attract the world's best talent by keeping it open. So in the early 2000s, I chaired a series of hearings at the Federal Trade Commission on barriers to e-commerce.
[0:43:03] And I'd be interested in the panel's opinion. Are there particular legal or regulatory barriers or other barriers to entry that are slowing down or impeding the development of AI? One comment I'll make mention is on data. With the growth of AI and the importance of data in AI, there's been a growth of a genuine need for innovation with privacy to secure the privacy for individuals. At the same time, there are...
[0:43:32] Massive data assets that aren't easily available. We had to negotiate at Microsoft Research to gain access to FAA data to help us build new weather maps for the country based on thousands of planes in flight as we speak. We were helped by the OSTP in gaining access to that, but it was not necessarily a simple task. But there are many data sets like this, and we'd love to see public sector data sets, especially with privacy protected, made more available for innovation.
[0:43:59] At the same time, what the NIH requires on NIH-funded projects for data to be released as part of the contracts that are made with researchers, it's very difficult to have medical data shared as a part of the fulcrum of innovation. And so we need to think through HIPAA, altruistic data approaches where patients donate data, new kinds of programs that let us
[0:44:26] Really, maintain patient privacy while gaining access to large amounts of biomedical data. One of the things which I think we are really looking for at the moment is there are some areas such as intelligent braking in cars where there are some legislative questions which might slow us down. For example, it would be tragic if some life-saving technology
[0:45:02] Which would make cars safer couldn't be released because the legal questions about who is responsible got in the way. What I'm talking about here is if I as a software developer invent a new collision avoidance system which unambiguously reduces fatalities by a factor of three, but occasionally, unfortunately, one in a thousand times, maybe there's a disaster.
[0:45:29] There is a difficult question as to how legislatively we make sure we're ready for this. So I can imagine a potential danger of an impasse between insurance companies, policymakers, drivers, and car manufacturers where no one is willing to put life-saving technology into these systems because it's still being left ambiguous who has the responsibility for what. So one final question. General AI.
[0:45:58] has generated some significant fears and concerns from scientists and innovators such as Stephen Hawking, Bill Gates, and Elon Musk. Stephen Hawking has stated, once humans develop artificial intelligence, it would take off on its own and redesign itself at an ever-increasing rate. Humans, who are limited by slow biological evolution, couldn't compete and would be superseded. And Elon Musk has referred to it as, quote, summoning the demon.
[0:46:28] How concerned should we be about the prospects of general AI? Or to ask the question differently, in a non-determinator, does anyone know when Skynet goes online? So my answer to that is that I think that with artificial intelligence generally, that there are a lot of things that we should be careful about and concerned about and think about security, safety, and ethics today. And so I think that...
[0:46:58] The kind of general intelligence that people talk about, I think that my 90% confidence interval on when that kind of technology could arrive is between 10 to 100 years. It's not something that we're at all capable of building today. And today we know that there are concrete safety problems that we can be working on. And so I think that we should be investing in those kinds of questions. And I think that that will help us figure out the right answers for the short, medium, and long term.
[0:47:28] There's been a lot of hyperbole, as you know, stimulated in no small part by Hollywood. These are great themes, and they keep us enamored with interesting possibilities. At the same time, we don't stop at those kinds of long-term outcomes and want to seriously reflect and review possibilities, push to the limit some of these proposals about what's possible, and in advance, proactively work to thwart them.
[0:47:55] to stay on a healthy, secure path. My own sense is these are very, very long-term issues, but that the things we're doing today are actually relevant and interesting in terms of thinking about how AI systems can grapple with unknown unknowns, how it could secure systems from, for example, modifying themselves, their own objective functions, which is one of the concerns that comes up at times.
[0:48:22] In some ways, I am happy to see the growth of interest in the long-term future questions because it raises my confidence that we will track closely and do the best we can when it comes to harnessing AI for the greatest benefits. If I could just add, at the moment...
[0:48:41] Everything that's going on in the current AI revolution is using AIs which are like idiot savants. They're able to search a space that we've prescribed really efficiently. And it is a matter of future researchers, not something immediate to imagine these kinds of self-aware, self-reasoning systems. Those would be really, really important to get right. At the moment, the AIs we're building are all to do with immediately...
[0:49:10] Answering questions about agriculture, safety, people's health, and the students who are being drawn into it are being drawn into it for these idealistic reasons. One thing you will notice, and this is actually under the influence of some of the institutions my colleagues have put into place, is many places such as Carnegie Mellon are actively making ethics and responsibility.
[0:49:34] A central part of the curriculum for these AI experts. Because these kids, they're kids today. They're building the 21st century. We need them to actually understand the human condition while they're doing it. Just to get a sense for the kinds of things that are going on, this coming spring there's going to be an event where we're inviting, or a group is inviting out people who are imagining the most fearful, feared long-term AI scenarios. Call them the red team.
[0:50:03] And then we're inviting out the blue team to disrupt them in advance, and they're going to come together and battle it out. So I would like to take this chance to circle back to one of your earlier questions and tie that in. You asked what are the areas that we need to work in. I would say that one of the key areas that we need to work in is better characterization and understanding the performance of AI systems. And this is something that we have a lot of interest in at NASA because in our space missions, we need to...
[0:50:31] If not prove that they are going to actually perform within certain bounds, we need to have very strong confidence that they will perform in those bounds because these are very high stakes missions. A lot of the applications that people have talked about, health care, self-driving cars, these also are high stakes missions. Before AI can control our critical infrastructure, we need to be confident that it will perform as we want it to perform.
[0:50:58] I think this has been identified before in the OSTP study as a key area of research. Thank you very much. Senator Peters. Thank you, Mr. Chairman. Again, thank you to our witnesses. You're right, we have to make sure this performs. I've been very involved in autonomous vehicles and the research that's going on there. My concern is that we have to make sure the technology gets it right with as few errors as possible because there's already limited consumer acceptance
[0:51:29] Thank you very much.
[0:51:52] Dr. Moore, some of the regulatory aspects of this before you put vehicles on the road to make sure there's proper safety in place or we've thought through how we ensure that is incredibly important. My concern with all of this has always been that there's a disconnect between the speed we're on with public policy versus technology. Right now we are at an exponential rate when it comes to technology and even though we have estimates of AI reaching a
[0:52:21] I think maybe that was Rockwind, a singularity of some sort from 10 to 100 years. We don't know when that is, although things seem to operate a lot quicker than we anticipate. And I believe that we didn't think we could beat the expert player in Go for at least a decade, and I think that just occurred a few months ago. So we just can't fully anticipate what's happening. So I don't know the speed it'll go at, but it'll probably be quicker than we anticipate.
[0:52:46] The one constant in all this is that when it comes to public policy, that operates at a constant speed. It's called snail speed. So it is very slow, cumbersome, and if we are not doing it now, we have no chance of trying to catch up to what's happening with the policies going forward. So I certainly appreciate the comments from several of you that we have to be thinking about this stuff now in a very thoughtful, comprehensive way, because if we wait, it's going to be too late to do that.
[0:53:14] But I want to switch gears to the positive aspects that we want to continue to move AI forward. You've mentioned some of the challenges that we have, the gaps that we have to fill. And I'd like your perspective on where the federal government's role is in terms of research. Mr. Brockman, you mentioned in your testimony, you talk about some, I think you referred to some subfields that need work in it and some other areas. But I'd like to ask each of you,
[0:53:42] Obviously, private industry is already invested. In my opening comments, I mentioned $8.5 billion in 2015. That number is going to continue to go up. So private industry is doing an awful lot of this work, including basic research, which traditionally has been an area where the federal government has supported academic research through grants. But some of that basic research is being done by private industry as well. So that's occurring, not necessarily in other areas. But are there gaps where
[0:54:08] You believe the federal government, there isn't going to be a private industry group out there investing in some of these gaps that we need to figure out. And the federal involvement will be critical to investing in those kinds of research programs for us to make sure that AI moves forward in a societal beneficial way, but also understanding the time constraints associated with the competition that we face from the Chinese and Koreans and other folks.
[0:54:37] So one comment I'll make is that beyond industry, looking at private sector and public sector, academia, there are groups coming together. So I'll just make a few comments about the new partnership on AI. The full name is Partnership on AI.
[0:54:52] AI to benefit people in society. And this is a nonprofit organization that was formed by Facebook and Amazon, Google, IBM, and Microsoft coming together, working with nonprofit teams with balanced board and so on, focused around sets of these long-term challenges and shorter-term challenges with safety, critical systems, ethics, and society, notions of...
[0:55:17] How people and machines work together and even working to stimulate new kinds of challenges and catalyzing new efforts in AI that might not be done naturally by industry. That's one direction. I'm happy to answer questions about that effort which is ongoing. Another couple comments is that there are places and opportunities where we don't necessarily see industry making deep investments.
[0:55:44] I would call these application areas that are rich and ripe for AI innovation. How can we solve homelessness or address homelessness, addiction, related problems in the social science sphere and social challenges sphere? There are some teams in academia right now working hard at
[0:56:03] Very good. I could not agree more with what you're describing. Another example of this phenomenon is
[0:56:28] I have two brilliant faculty in the Human Computer Interaction Institute at Carnegie Mellon who are looking at AI to help people make prosthetic hands easily for folks who've lost their limbs. And they're struggling to find $50,000 or $100,000 here or there to build these things. At the same time...
[0:56:49] Frankly, my friends from industry will be offering the same faculty two or three million dollar startup packages to move into industry anyway. So I do want to make sure that the folks building these things are successful. Another example is in the defense world, tools for helping our war fighters or other...
[0:57:12] Folks domestically who are putting themselves into danger to save other people. There is so much opportunity to use sensing, robotics, and artificial intelligence planning to save lives there. That's an area which it will take a very long time to grow naturally in the private sector. And we have faculty and especially students champing at the bit to work on these kinds of problems.
[0:57:40] There's another area which may sound too theoretical, but I've got to tell you about it because it's so exciting. The big Internet companies are working, their big search engines are powered by things called knowledge graphs, the underlying set of facts about the world which you can chain together to make inferences. A large group of us from academia and industry and from some government agencies want to work to create a
[0:58:09] Public, Open, Large Knowledge Graph, which will permit small AI entrepreneurs to tap into the same kind of knowledge of the world that the big Internet companies have at the moment. So we see in a manner equivalent to how lots of individual networking systems came together to form the TCPIP protocol for the Internet. There's something we can do there. Finally, this one is going to sound really abstract.
[0:58:39] At the moment in machine learning and deep learning came out of mathematics and statistics. Without the fundamental work going on by the mathematicians and statisticians around the world, we wouldn't be where we are. So statisticians, who are often the heroes in AI, need help to progress their field forward as well. So I have three suggestions. The first of these is basic research. And you mentioned that
[0:59:09] Basic research is happening in industry, but I think that we just cannot do too much of it in that we really are at the tip of the iceberg here, and I think that we're just going to find so much value, and that's why the big companies are investing, because they realize that as many dollars are being made today, that there's 100x or maybe more increase in the future. And I think that it's really important that...
[0:59:30] The technology is not owned by just one or a few big companies. I think it's really important that the benefits and that the technology are owned by us all as Americans and as the world. And so I think that the government can really help to ensure that this technology is democratized and that we move faster. The second is measurement and contests. I think that for the reasons I mentioned earlier, that it's really important that we track how it's progressing so we can have a good
[0:59:53] I think that the government has actually been extremely successful in the past with investing in contests. And so I think either creating new measurements or supporting people in industry and academia who are doing the same. And then the third is safety, security, ethics. I think that's going to take everyone. And I think that we all need to work together. I think that that's going to require making sure that there is funding available for people who want to be thinking about these issues. And I think that's going to feed back into all of the questions that everyone has been raising here today.
[1:00:23] Mr. Rockburn, I think I saw that you thought philosophers should be part of that, in addition to technologists. So I appreciated that. Someone with a master's in philosophy, that's good. So I appreciate that. It's going to take everyone. Dr. Chen. Yes, so I would echo some of the statements that the other panelists made. They've identified a lot of great topics that really require a government role. One that I would emphasize is very basic science questions that relate to NASA's mission.
[1:00:52] How did the universe form? How did the solar system form? How did life come into existence on this planet and other planets? These are actually fundamental questions of science and exploration that we really need to leverage AI to go and explore all these nooks and crannies in the solar system. And if you really want to think far out,
[1:01:14] In order to embark on an interstellar mission to see if there's extant life at other solar systems. These are different questions that there's no clear financial motive. So there's a clear role for the government to be able to answer these kinds of basic science questions. And if I could just add one last thing. So I believe that the statistic for the amount of government unclassified dollars that went into R&D and AI was $1 billion, I believe, over the past year.
[1:01:43] As has been mentioned several times, that industry investment is $8 billion. And if this is a technology that's really going to be affecting every American in such a fundamental way, I think that disparity is something that we should act to correct. Thank you for your answers. Appreciate it. Thank you. Senator Schatz. Thank you. Dr. Moore, you talked about mostly the unambiguously positive potential applications of AI.
[1:02:13] We've sort of briefly touched upon the terrifying science fiction possibilities, which I think we're, you know, joking aside, keeping an eyeball on, but that is from 10 to 100 years from now. What I'm interested in is what are the, as Senator Peters mentioned, what are the tough, thorny, short-term public policy and ethical
[1:02:40] Challenges that we're facing right now, not the possibility that machines will overtake us, not even the sort of question of long term unemployment. But I think about doctrine of war. I think about black box algorithms that help with policing or social work or health care. And I'm wondering if you could maybe just going down the line, starting with Dr. Horvitz, give me an example of a short term.
[1:03:06] Ethical, moral, public policy quandary that is upon us now. Well, for one, I think that we'll be seeing interesting legal tests and precedents set up that define new kinds of frameworks for dealing with things like liability. Who or what is responsible? Manufacturers, the drivers of cars, the people who sign various documents when cars were purchased. I think that we haven't, things are unsettled in that space and we'll be seeing lots of interesting work there and there are some very interesting
[1:03:36] Focused workshops and conferences where people ask these questions. When it comes to using various AI technologies, going from machine learning for building classifiers that do predictions and that are used to reason about interesting problems like criminal justice challenges. Should this person charged with a crime have to stay in jail in advance of their court date?
[1:04:05] Well, can they get out early if they can't pay their bail? There are systems out there that have been used and critiqued. And it's pretty clear that there is opportunity for looking very carefully at systems that are used in high stakes situations like this to ensure that there are not implicit biases in those systems, to ensure that there is accountability and fairness. And so long as it's not a government contract where you're working with a
[1:04:31] A subcontractor which says, you know, our algorithm is proprietary. We just spit out our recommendation. That's what you pay us for. Well, that's exactly where I'm going. That's exactly where I'm going. So the question would be, what are best practices, for example, and do we need them when it comes to these kinds of applications? For example, potentially with protecting...
[1:04:52] Privacy, should data sets used in these applications be disclosed and disclosable for study and investigation and interrogation by people who want to make sure that they're fair and that there can be trust in these systems? The basic idea here is that many of our data sets have been collected in advance with assumptions we may not deeply understand. And we don't want our machine learned applications used in high stakes applications to be amplifying cultural biases or any kind of biases that was.
[1:05:21] That was part of the collection process. Why don't we go very quickly because I have one final question, but I'd be interested to hear each one of you quickly answer this question. Very briefly, this AI technology is available to the bad guys too. It is possible to cheaply set up...
[1:05:43] Homemade drones in a bad way. A repressive regime can now use face recognition in a way that they couldn't use last year. We need to actually stay ahead. We can't just sit where we are. So I'd like to actually build on the bias answer and just say that one thing that concerns me is the lack of diversity in the field, especially as we try to think about how can we ensure that these systems are going to do the right things for all of us. And if you look at this panel, that we're actually, I think, pretty representative of what the current field of AI looks like. And I think that we...
[1:06:11] The government and industry and academia need to work together in order to correct that. I would echo Eric's comments on we need to
[1:06:20] Further understand how to characterize the performance of systems.
[1:06:50] Are not biased in performing as you wish. So let me just wrap up with this, and I'll ask a question for the record. My question is sort of mechanical. Dr. Horvitz and many of the other testifiers have made at least a brief reference to the ethical quandaries that we are facing. Blue team, red team. I know, Mr. Brockman, you made reference to safety, security, and ethics.
[1:07:17] And it occurs to me that as this accelerates so fast, that as you do your conferences, as you have your conversations, you may not have fully articulated what kind of system among the AI community you really want to wrestle with these questions, whether it's a public-private partnership, whether it's led by the federal government or convened by the federal government, but primarily driven by private sector actors.
[1:07:44] I don't know, but it occurs to me lots of good thinking is occurring. It also occurs to me that maybe it hasn't been fleshed out from a process standpoint, and we can't take it for granted that it's all going to happen organically. But I will take that question for the record in the interest of time. Thank you. Thank you. Chairman Thune. Thank you, Mr. Chairman.
[1:08:06] for convening today's subcommittee hearing on artificial intelligence. This topic complements our last full committee hearing which explored another nascent technological field, augmented reality. I'm excited by this topic because AI has the potential to catapult the United States economy and competitiveness in both the near and the long-term future. AI presents promising applications in the areas of healthcare,
[1:08:32] transportation and agriculture among others and I want to thank our witnesses today for sharing and highlighting some of those applications today. The recent report and strategy on AI released by the White House Office of Science and Technology Policy provide Congress with important considerations to weigh as we think about what the appropriate role of government is in this promising field so that we ensure that the United States remains the preeminent place for AI in the global economy.
[1:09:01] And so I appreciate, again, the witnesses sharing your insights about what the state of the art is today and where the nation's leading experts see the applications moving forward. I wanted to direct a question, Dr. Horvitz, to you. You mentioned in your testimony that new kinds of automation present new attack surfaces for cyber attacks. And I wonder if maybe you could elaborate on what some of those new cybersecurity vulnerabilities might be. Yes, thanks for the interesting question.
[1:09:31] I'm framing. The systems we build that are doing sophisticated tasks in the world often are assembled out of multiple modules or components that have to talk to one another, ending often in cyber-physical or actual physical activity or effectors like car steering wheels and braking and so on. Every single one of those interfaces presents an opportunity to an attacker to intervene and
[1:10:02] They influence the behavior of a system. There are also whole new categories of attack. I would have been surprised to learn 15 years ago that the community was talking now about machine learning attacks. What's a machine learning attack? The careful injection into a learning system in a sleuthing manner potentially of data that will tend to build a classifier that will do the wrong thing in certain cases. So that just gives you a sense or a taste for the very different kinds of
[1:10:30] Opportunities that are being presented by the systems we're building now. We often think about security in classical ways with verification models and encryption and so on. And these techniques often will apply, but we have to be very careful as we build these systems that we're covering all ground and we're thinking through possibilities. And on the flip side of that, how can the use of machine learning enhance security analysts?
[1:10:58] Ability to catch malicious hackers. Yeah, it's a great follow-on question because it's a yes, yes. I mean, look, I mean, we have to be cautious because human beings and humans plus machines can be very creative in how they attack. So there's a long tail of possibilities we have to sort of account for.
[1:11:18] But there are some very, very promising angles with the use of artificial intelligence and machine learning to detect anomalous patterns of various kinds with low false positive rates. That's one of the goals is to do this well where you don't call everything strange because people are always doing different things that are safe but that seem to be different over time.
[1:11:37] It might seem like a fraudulent event, for example. So I think there's lots of promise. I know that I'm very excited about some recent projects that I reviewed at Microsoft Research in this space. So I think it's an exciting direction indeed. Speaking of someone who, before I was at Carnegie Mellon, I was at an internet search engine.
[1:11:55] This is an area where I would claim that internet search companies are well ahead of what you're seeing happening in the public sector. There actually are some very good technologies out there for doing machine learning versus machine learning warfare, if you like. So it's an exciting area which I would like to see grow in this area. On the bright side, in a recent DARPA challenge, that challenge was about us using artificial intelligence to
[1:12:24] Discover vulnerabilities autonomously and using machine learning in other systems, which sounds like a kind of frightening thing, but A, it is actually important for our national defense that we have these capabilities.
[1:12:37] B, that is one of the ways in which we keep ourselves safe is by having our own AIs trying to break into our own systems. So this is another area which just wasn't there two years ago. Carnegie Mellon, University of Michigan, and plenty other major computer science universities are heavily involved now in using AIs to both try to break and warn us about breakages in our own computer security systems.
[1:13:04] Mr. Moore, very quickly here because I'm out of time, but could you build a little bit on your written testimony about how the United States can win the AI talent war? In other words, what are the best ways to sustain enough AI talent at universities to conduct basic research and drive innovation while also filling what is a growing demand for AI jobs in the private sector? So I think this begins in middle school, and even the U.S. government can really help here if we just help.
[1:13:34] Kids in middle school understand that one of the most important and interesting things they can be doing with their lives right now is to learn mathematics so that they can be building these kinds of robots and systems in the future. This is something which needs training. It's not that you need to be a genius. You need to be trained in math from about the age of 13 or 14 onwards. Understand that that is even cooler as a career move than going to work in Hollywood.
[1:14:02] Once we've got the kids' minds in the right place, we can bring them through the university system, scale that up, and then we'll be in good shape. What I don't want to do is keep us in our current situation where the talent...
[1:14:17] Crunch is being dealt with by this massive bidding war for this small amount of talent. That's not sustainable when the other continents are doing such a good job of producing AI experts. And I think one thing that's really important is that we can continue to attract the best AI researchers in the world by having an open, basic research community that just draws everyone in. It's been working, and I think that we can grow that and strengthen that community. Thank you. Thank you, Mr. Chairman. Thank you, Chairman Thune. Senator Daines.
[1:14:47] Thank you, Mr. Chairman. Thank you for holding this hearing today. Very timely. Before I came to the Senate, I used to have a legitimate day job in the technology sector for about a dozen years. And our company had several patents for AI design. We won a national award for AI innovation. That was back in the early 2000s. So we're not talking about something that's new. It's been around for quite some time. In the Senate, we often talk about...
[1:15:15] What we need to do to ensure the U.S. maintains leadership, looking at global competitiveness and innovation, technology, whether it's broadband, smart cities, medical research. So I'd like to start my question with Mr. Brockman. Could you expand your testimony about what other countries are doing in terms of encouraging AI?
[1:15:37] And follow on there would be, what do you see some of the competitive disadvantages that we face right now in our country as it relates to ensuring that we maintain global leadership in this important area?
[1:15:46] So in other countries, I think there's a mix. And so that, for example, you see companies like China's Baidu who want to scale up their investment in this field. And the way that they've been doing it is that they've actually opened a US-based research institution and have joined our research community and our publishing and kind of following our lead. With South Korea, I think that...
[1:16:10] You know, sort of around the same time as the AlphaGo match that they announced that they were going to make this billion dollar investment into AI. And Canada recently has been talking about that they're starting to increase their national funding. And so that's the flavor that you're seeing both companies and the governments stepping up their investments and trying to find, because I think everyone sees there's so much activity happening. And I'm sorry, I actually missed the second part of the question.
[1:16:37] Well, just look at what do you see as some of the headwinds relates to create competitive disadvantage for our country. So, I mean, I think that the thing we should be aware of, and so there's a stat mentioned about the number of papers that are published, and I think that that's actually, it's a true fact, but it's not necessarily the most important fact. The most important fact is where do the fundamental breakthroughs come from? Because the thing that happens is if you are the one who creates...
[1:17:03] The sort of underlying technology, it's like discovering electricity. You're the one who understands everything that went into that, that the papers that get published, that those are in your language, and that you really get to set the culture for how people build on top of it, that you control. You're probably the one who published the underlying code that people are using. And so I think that the thing that we need to watch is the question of for the actual fundamental advances, the capabilities that we just did not have before but that we have now, where do those come from?
[1:17:30] And as long as that's us, then I think we're in really good shape. And so I think that we need to continue to make sure that that's the case. I want to go back to the point that was brought up earlier on the cyber threat. In 28 years in the private sector, I never received a letter from the Human Resource Department that said that my information had been hacked until I became a federal employee, elected to the US Congress. And I, like millions of other federal employees, got a letter.
[1:17:57] From OPM, talking about the fact that my information has been hacked. I spend a lot of time working and chatting with some of these very innovative, smaller tech companies. They're doing some amazing things in relation to advancing the speed of business, relates to protecting our assets. I am concerned, and you mentioned the fact that the federal government can lag, is not always leading in that area.
[1:18:23] And I know it's frustrating because we have solutions here. We can't sometimes penetrate our own firewall, figuratively speaking, as it relates to trying to get these solutions in front here to get our government to move at the speed of business. Because I know when I was in the cloud computing business, you wanted to make sure you were never on the front page of the Wall Street Journal because of a hack and what that does to valuation of companies. It's been very obvious in the last few years. So what do we need to do to ensure that the best technology is moving so fast right now?
[1:18:51] is in the hands of our best people in the federal government. This is not a critique on the people working in the federal government. This is oftentimes the barriers we look up here to ensure that we're protecting our national assets. Who'd like to answer that one?
[1:19:04] So if I may, so I've actually been extremely impressed with the work that the USDS and OSTP have been doing to solve problems like this, to start injecting some of, and I think it really starts with getting the best technologist in the door, and then secondly, giving them the power and empowering them to make changes within the federal government. And so I think that it really starts with the people, making sure that we're attracting them and making sure that the structures within government exist. And I think that as long as there's an attitude that's receptive.
[1:19:32] Within the agencies or wherever you want to upgrade, I think that that's the best way to get this to happen.
[1:19:39] I'd like to jump in there also. I think one of the key things is for the government to be at the forefront or at least participating in the forefront of technology, there has to be an active interchange in what I would call a vibrant ecosystem that includes multiple kinds of institutions. And I'm very happy to say in the AI in space arena, there's a large amount of interplay between the commercial sector, between the government sector, between small companies.
[1:20:06] It seems every week there's another company being started up to do real-time space imaging of the Earth for business intelligence. I think that all of this is indicative that there's a good structure with this interchange of information, and I think that's the key to making sure that the government stays in the right location and able to understand and be smart in how it uses this technology. All right, thank you. I'm out of time. Thank you, Mr. Chairman.
[1:20:32] Thank you, Senator Daines. I'd like to thank each of our witnesses for coming to this hearing, which I think was informative and productive and will be just the beginning of what I expect to be an ongoing conversation about how to deal with both the challenges and opportunities that artificial intelligence presents. The hearing record will remain open for two weeks. During this time, senators are asked to submit any questions they might have for the record and upon receipt.
[1:21:01] The witnesses are requested to submit their written answers to the committee as soon as possible. Thank you again to the witnesses, and this hearing is adjourned.