House Defense Department and Homeland Security officials on their use of Artificial Intelligence technology. They addressed ai Machine Learning, employee training, and potential Cyber Security risks before a House Oversight and accountability subcommittee. This is just over one hour. [captions Copyright National cable satellite corp. 2023] [captioning performed by the national captioning institute, which is responsible for its caption content and accuracy. Visit ncicap. Org] i would like to recognize myself of the purpose of making an opening statement. Good afternoon and welcome to the hearing on the subcommittee on Cyber SecurityInformation Technology and government innovation. At the very first hearing of the subcommittee held earlier this year, expert witnesses told us Artificial Intelligence or ai is likely to bring Disruptive Innovation in many fields and it could and skip eight economic growth, Higher Standards of living an improved medical outcomes in virtually every industry and institution will field the impact of ai. Today we will discuss the impact on the largest most powerful institution in the nation, the federal government. As we know the government today performs an everexpanding swath of activities from securing the homeland to predicting the weather to cutting back benefits checks and many of these functions could be greatly impacted by ai. That is clear from the cases posted by federal agencies pursuant to an executive order issued under the last administration. Federal agencies are attempting to use ai systems to enhance border security, to make air travel safer and to speed up eligibility determinations for Social Security disability benefits, just to name a few cases. Ai will also shake up the federal workforce itself. We hear a lot about how ai could disrupt the private sector workforce transforming our aura limiting jobs or creating others. While the federal government is the nations largest employer and many employees work in whitecollar occupations, ai is reshaping because it can perform many routine tasks more efficiently than humans. That will allow federal employees to focus on higher order work and maximizes productivity. In fact, a study says the use of ai to automate tasks for federal employees could eventually yield as much as 41 billion in annual savings by reducing required labor hours. A separate study by the partnership for Public Service and the ibm center for the business of government identified hundred 30,000 federal employee positions whose work would likely be impacted by ai, including 20,000 irs tax examiners and agents. That of course begs the question whether we need to hire tens of thousands of new irs employees when ai can transform even or replace the work of much of its current staff. I think every american could agree with that. Ai can make work better but it is still just a tool, be it an incredibly powerful one. It can easily be abused when used for the wrong purposes or without the proper guardrails. Ai systems are often fueled by massive troves of Training Data with complex algorithms. These algorithms can yield results in their own designers are unable to predict and struggle to explain sometimes. And we are learning this in real time. So its important we have safeguards to prevent federal government and exercising inappropriate bias. We also need to ensure the federal governments use of ai does not intrude on the privacy rights of its own citizens. The bottom line is we need the government to harness ai to improve its operations while safeguarding against potential hazards. Thats why congress enacted the ai and Government Act in december of 2020 soon after the current nmc patient took office. That law requires the office of management and budget issue guidance to agencies of the acquisition and use of ai systems. Also tasks the office of personal management with assessing federal ai workforce needs. The administration is way overdue in complying with the law. Its no more than two years behind schedule and issuing guidance to agencies and opm is more than overdue in determining how many employees have ai skills and how many need to be hired or trained up. I will also say in the cybersecurity plan before it was made public, i asked the question pointedly to the administration of ai was included at the time and it was not. It was mentioned three times fleetingly, very casually in that document today. The administrations failure to comply with these statutory mandates was called out in a lengthy white paper issued by Stanford University ai institute. The papers authors also found that many agencies had not posted the record ai use case inventories. Others had admitted key use cases including dhs and an important facial recognition program. The paper summed up the noncompliance with various mandates by concluding americas ai ecosystem is certain by imitation of these legal requirements. Most of the ai policy debate is focused on how the federal government should police the use of ai by the private sector, but the executive branch cant lose focus from getting its own house in order. It needs to appropriately manage its own use of ai systems consistent with the law. The subcommittee will keep insisting the Administration Carry out laws designed to safeguard government use of ai. Im developing further legislation to ensure federal agencies appoint ai systems effectively, safely and transparently. We have a huge opportunity before us and i would love to see us harness the technology that is rapidly evolving. I expect this hearing will expect inform many of these efforts and with that, we are going to go to our witnesses and was Ranking Member conley comes in, we will give him time for his opening statement. Im pleased to introduce our witnesses for todays hearing. Our first witness is the director of the white house science and Technology Earning her the designation as a president s science adviser. She is also the first science adviser to be nominated to the president s cabinet. This is her first appearance as a witness before Congress Since her Senate Confirmation last year. We are pleased to have you here today. Im grateful that you showed up. I will tell you, not everybody does and they sometimes send the secretary, undersecretary or the assistant to the assistant and its refreshing to have someone actually sure that weve asked for and i want to thank you for your time today. Our second witness is dr. Craig martel, chief digital nai officer with the department of defense. Our third witness is mr. Eric eisen, chief Information Officer with the department of Homeland Security. We welcome everyone and are pleased to have all of you here this afternoon. So pursuant to Committee Rule 9g, the witnesses will please, if you will stand and raise your right hand. Do solemnly swear or affirm the testimony youre about to give is the truth, the whole truth and nothing but the truth, so help you god . Let the record show the witnesses all answered in the affirmative. We appreciate all of you being here today and look forward to hearing your testimony. I would like to remind the witnesses that we have your written statements and they will appear in full. Please keep your oral seems to five minutes. Press the button on your microphone so its on the members can hear you. When you begin to speak, the light infantry will turn green. After four minutes, the light will turn yellow and when the red light comes on, your five minutes is expired and we ask you to please wrap it up for us. All right. With that im going to yield to our Ranking Member of the subcommittee, mr. Connolly. Thank you madame chairwoman. I have three subcommittee hearings today and two caucus meetings, so im a little out of breath, but thank you for accommodating me. Earlier this week, majority leader Chuck Schumer held his inaugural Ai Insight Forum and senator hickenlooper held one for the need for transparency in Artificial Intelligence. Today our subcommittee returns for a second hearing on the ai to discuss is uses in our own federal government. I think its very clear all members of congress are interested. Im not sure is clear how much members of Congress Know about it. The subcommittee is proud to continue its historical leadership in the ai space. As many of you know, former subcommittee chair will heard held a threepart hearing series on Artificial Intelligence and late former chairman Elijah Cummings focused on facial recognition. These initiatives show that if done right, the federal government can leverage ai to better serve the public. For example, several federal agencies are already using Ai Technologies to cut costs, improve constituent services and strengthen existing systems. The United StatesCyber Command and department of Homeland Security for example employee Ai Technology to protect our networks from counter cyber attacks. United States Postal Service is currently piloting an Autonomous Vehicle project that employs Ai Technology. The department of housing and urban development and Immigration Services are using ai chat bots to facilitate communication with the public, looking for help from the agency. However, like all new tools, if used improperly, ai could result in unintended consequences. For example, Automated Systems can inadvertently perpetuate societal biases such as faulty facial Recognition Technology or opaque sentencing algorithms used by our criminal justice system. Ai can also threaten jobs, proliferate misinformation and raise serious privacy concerns. Thats why i applaud the Biden Administration for proactively taking significant steps to ensure transparency in the governments use of ai. Last october the white house released a blueprint for an ai bill of rights to ensure the protection of civil rights in the algorithmic age. Prior to that, the national Artificial IntelligenceInitiative Act codified the establishment of the American Ai Initiative and the National Advisory committee. The subcommittee looks forward to hearing an update from the panelists before us on the joint work with the secretary of commerce to advise the white house on that ai policy. Everybody can agree the government has a colossal responsibility in developing the necessary guardrails to curb the risk of this incredible technology, while allowing it to flourish. This committee must hold federal agencies accountable and make sure they are making appropriate choices about whether and when ai is right for their mission. The federal government must also intentionally train, recruit and maintain a workforce thats comfortable and confident with this technology. Thats why the chairwoman and i worked to pass the ai training expansion act of 2023, hr 4503. Out of our committee and would expand ai training within the executive branch. Really important. And i commend my colleague for that bipartisan collaboration. Ai is already changing the world around us in so many ways. And we need to step up to the challenge and mitigate the risk. Federal government needs to ensure this technology is created, deployed and used in a safe, ethical, productive and equitable manner. With that, i yield back. Thank you madame chairwoman. Thank you. I ask unanimous consent for representative clay higgins from louisiana to be waived onto the subcommittee for todays hearing for the purpose of asking questions. So without objection, so ordered. I would now like to recognize the witness to please begin your opening statement. Thank you so much, to you Ranking Member connolly and members of the subcommittee. Ive really appreciated the work that you all are doing on Artificial Intelligence and its great to be here with my colleagues and spend this time to focus on these important issues. I have three messages today and the first one is that ai is a top priority for President Biden. Hes very clear that this is one of the most powerful technologies of our times. When we look around the world, we can see that every nation is racing to use ai to build a future that is imbued with their own values and i think we can all agree that we do not want to live in a future that is defined by technology shaped by authoritarian regimes. Thats why the president is very clear that American Leadership in the world today requires American Leadership in Artificial Intelligence. Second, for america to lead an ai, government has some core responsibilities and one of those, one set of those responsibilities is to manage the risks of ai, and your Opening Statements have noted, ais risks are broad because its applications are so broad and the source range from risks related to fraud and information integrity. They include risks related to safety and security. Risks associated with privacy, civil rights, civil liberties, and risks to jobs and the economy. Now, some of these risks can be addressed under existing laws and regulations. Some of these risks can be managed by making sure that government uses ai responsibly, and in some cases, we do expect legislation will be required. Thats about mitigating risks. The reason we are doing all of this work to manage risks is so that our country can seize this technology to build the future. And if you look at what companies are doing, they are racing to build better products and services to transform Industries Using ai. This is a technology that holds equally great promise for the work that government does for the american people. That then becomes the second core responsibility of government. I think both of you have spoken to that as well in your opening remarks youre going to hear from my colleagues about National Security and Homeland Security and theres a lot to be said there. Also briefly touch on the many other Important Services and the Public Purposes that our governments responsibility. When you look across federal government today, youll see that agencies are starting to use the insights that they can glean from these fast troves of data that they generate in the doing of their business. Ai technology is also changing the way Government Agencies interact with their citizens. It can speeded up, can just make those processes work much better. The examples are very wideranging. They include ai for weather prediction. They include ai to help us keep air travel safer. Ai is being used to speed up the processing of disability determinations. Its being used to improve how we process patent applications. Those are just some examples today. If you take a peek inside of labs around the country and look at what is happening with federally funded r d in the world of research and development, youll get a glimpse of where the future is going and ai is playing a huge role there as well because ai can enable the design of the materials that we need for advanced batteries, for hydrogen storage, those things are critical to our clean energy future. Ai can change the way that we predict disasters, the way that we implement plans for resilience as the climate changes. Ai can transform drug design. It can allow us to tailor Clinical Care to each individual patients needs. It can enable major advances in population health. Used responsibly, ai can help us deliver Better Outcomes and to create new possibilities for the american people. My third message for you and ill end with this is that the Bidenharris Administration is taking action to meet this moment. We have moved with urgency on a series of steps that started with the ai bill of rights that we published almost a year ago and i want to emphasize that especially in a time when technology is moving as fast as it is, it is so important to be clear about our values. About the importance of rights. About safety and security. About privacy. That was the Important Role of the ai bill of rights. Recently, because of the president s leadership, 15 companies have now made voluntary commitments to focus on safety, security and trustworthiness in their ai systems that they are developing and driving. Thats a companys responsabilities. Today the white house is working we are going to be voting soon if you could wrap up. We are working today on the white house on an executive order. Working on guidance for departments and agencies. The executive branch. We continue to work with our International Allies because ai does not stop at the borders. Finally, we remain committed to working closely with congress on a bipartisan basis as you consider legislation. Ill finish by saying this work is urgent, its important and i very much look forward to working on it. Dr. Martel, please begin your opening statement. Chairwoman ans distinguished members of the subcommittee, thank you very much for inviting us here today. I like to start just by asking the question, what is ai . So we should have a sort of common definition as we are going through this. When i say the phrase Artificial Intelligence, i mean statistics at scale. We gather massive amounts of data from the past. We use it to build a model and to predict the future. Its really important to think about it that way because its statistics a scale which means its never hundred percent correct, which means for every model that we build, it will always sometimes get it wrong appears a large part of what we have to think about is how we understand when it gets it wrong and what should we do when it does get it wrong. Its important to remember that as im going to my comments. My other panelists here may have different definitions, but thats the operative one for me. I look forward to sharing the ongoing efforts of the chief digital and ai Office Around the use of Data Analytics and ai enable technologies to accomplish our National Defense mission. Data analytics and they are integral to a call pushing the priorities set for the National Defense strategy. Improving data quality, the system used to measure the past. Developing robust performance metrics, that tells us how well we are doing in the future. Providing enterprise ready ai scaffolding, building the Data Integration layer and developing a robust Talent Management plan for the department of defense as a whole. Firstly, quality data and cdaos foundation. We are feet focused on improving the quality of the data that enables most dod use cases. For example, cda o is providing data and Digital Talent teams to the principles Staff Assistance and commanders to the data and Ai Initiative, also called eta. Additionally the cdao is creating verification processes the check data for errors, inconsistency and with respect to bias, class imbalances before ai models are even produced. We are also working closely with the u. S. Cyber command on their fiveyear ai road map for rapidly acquiring ai systems. Secondly, and business performance, cdao in partnership with the dod Performance Improvement officer is defining and data enabling the metrics of of the dod will use to manage its performance. In support of the secretary of defenses priorities, the National Defense strategy and the Strategic Management plan. Cdao is ensuring these metrics are outcomebased and not just how many meetings to go to, but the effectiveness of those meetings. And thirdly, enterprise ai scaffolding consists of the robust environments and tools that enable cuttingedge development to Machine Learning capabilities that we provide the technical and nontechnical Enterprise Services necessary to accelerate secure, reliable and responsible ai development. Fourthly, for c2, we focused on building a Data Integration layer that will enable data centric command and control across the department and with our partners and allies. Cdao is assessing the necessary capabilities for the state integration layer. A series of sermons called guide, Global Information dominance experiments, and the se are in their seventh generation and underway now. Its scheduled for december and a happy to brief the committee on the successes that we have been having. Finally, in order to enable datadriven capabilities across the entire department of defense, we are building a unified Digital Workforce program with the chief Talent Management officer and other undersecretaries or the goal of this program is to develop a Digital Workforce that is globally identifiable and readily accessible for dod use. Ladies and gentlemen, within all of these initiatives, i want to clarify that ai is not a singular, monolithic technology, nor one size fits all solution. Thats extremely important. When we say ai, its not something that we have it, the n we win and if we have it, they lose. We need different algorithms, different success criteria and different data to train the different models underpinning each of our different use cases. Think about the different use cases in your daily lives are talking to your phone, getting shopping suggestions, just the shirt with those pants . And using a Search Engine to find information you need. Each of these require very different Ai Technologies. The same is true for the dod. We need Computer Vision to understand our environments, natural language processing to navigate the departments policies and idiomatic language which is really hard for humans to understand. And reinforcement learning for predictive maintenance. As well as many other Machine Learning algorithms. Its important to remember that ai is neither a panacea nor a pandoras box. If we think about it that way, we are not thinking about it correctly and we will be able to tackle the problem. Its not a one size thing. We need to evaluate its effectiveness and dangers on use case by use case basis. We have to wrap it up. Thats my last word. Id like to recognize to begin your opening statement. Thank you for the opportunity today. I like to know that in addition to serving as a departments chief Information Officer, secretary mayorkas also named me today is the first chief Artificial Intelligence officer. Id like to talk to about three concrete use cases where dhs is already using ai to deliver clear benefits for the american people. And the comprehensive measures we are taking to ensure that our use of ai is safe, responsible and rights respecting. First, dhs is using ai to keep dangerous drugs out of our country. Recently, a car drove north from mexico and approached the santa cedro port of entry in san diego, california. In the past, customs and Border Protection officer that inspected the car would likely have had no reason to give it extra scrutiny, but this time, one of our Machine Learning models noticed a potentially suspicious pattern in the vehicle crossing history. The officer decided to refer the car to secondary inspection where we discovered and seized nearly 60 kilograms of fennel and 16 kilograms of meth concealed in the vehicles rear quarter panels and gas tank. If not for this use of ai, those drugs could be on our streets. Second, dhs is using ai to aid our Law Enforcement officers in investigating heinous crimes. Last month, Homeland Security investigations announced the completion of one of the most successful operations ever against child sexual abuse online. Operation renewed hope resulted in identifying 311 previously unknown victims of Sexual Exploitation and led to the rescue of several victims from active abuse and the arrests of suspected perpetrators. This operation relied on the expertise and dedication of our agents and our partners domestically and abroad, but our agents had an extra tool at their disposal. Machine learning algorithms were used to enhance older images and give investigators new leads. Through this use of ai, we were able to turn formerly cold cases into rescues and arrests. Finally, dhs is using ai to make air travel easier and safer. Tsa has started rolling out touchless precheck and select airports. A new, optional way of going to the airport curb to gate without ever taking out your wallet. Once you opt in, you can check your bag, go through the security checkpoint and board your flight all with just a quick photo. This process and tsas acceptance of mobile drivers licenses in seven states and counting, use thoroughly tested ai powered algorithms to save time, reduce physical touch points and increase security by verifying identity more accurately. While ive highlighted these three examples today, dhs will use ai to transform all parts of our operation, from detecting and mitigating Cyber Security vulnerabilities, to enhancing maritime search and rescue operations and far beyond. Ai will provide smarter and timelier information to our agents and officers to aid them in making decisions, and free them up from routine tasks to focus on higher value work. As we move forward, we will ensure that our use of ai is responsible and trustworthy. That it is rigorously tested to be effective. That it safeguards privacy on , civil rights and civil liberties, while avoiding inappropriate biases. And to the extent possible, that is transparent and expendable to the people we serve. Last month, secretary mayorkas issued are key principles for responsible ai use. We are applying these principles to the dhs Ai Task Force which i lead alongside our undersecretary for science and technology, by issuing comprehensive policies on specific types of ai. As we did just this week with new restrictions on our use of facial recognition. We will work alongside our internal and external oversight partners to include congress and the subcommittee, as we work to implement ai Risk Management framework and remain fully compliant with evolving laws, practices and policies. Thank you again for the opportunity to testify today. I look forward to your questions. Before i gavel out, my colleagues leave to go vote and i will ask my questions before we go and gavel out. Dr. Prabhakar, my first question is to you, dr. Prabhakar. As the president s science adviser and president s top Artificial Intelligence adviser, i expect you regularly to brief him on ai, the ap quoted you saying in a recent interview, you had several conversations with him about ai. How many times have you been able to brief him . He met with his advisers on science and technology i believe in early april. We had a discussion about ai before that and that he spent an extended period with a room full of amazing science and technology experts. People using ai, people generating ai, and had a very extensive conversation there. What did President Biden say to you about ai . President biden has spoken probably many times about ai. What has he said to you in these conversations about ai . Obviously im not in a position to say what he said in the oval. Ill tell you what he has said publicly which is very consistent, which is that he recognizes how fast its moving. How it is part of this pivot point in history and the choices that we make are essential. He is very excited about the potential. How would you characterize his level of understanding of ai . You think he understands . I think its excellent. He grills me and everyone else. He grills you. What is the grill you on . I can talk about the things hes said publicly and its on many topics. Hes talked about the ways that ai can be used. Hes expressed concerns about the way that it can create problems. He has talked about the fact that hes married to a schoolteacher and so he knows not how it shows up in education. Does he understand its uses within the federal government . Thats sort of outside. Education. The federal agencies and how can be utilized, does he understand that . Is it relevant to the progress of ai . The president is very clear about the breadth of applications of Artificial Intelligence and his vast understanding of course, as many years as a legislator and now is president of all the punches of government and the role it plays in National Security. But also in all the other functions of government. He understands, obviously, that its clearly going to be a powerful force. The office of science and Technology Policy is a white house policy shop. Will you and your office plate with respect to federal ai policy . How does that work . We have several roles as ostp. A core role is to be the place where the federal r d enterprise comes together. We Work Together and make sure that people know each other are doing in areas across research and technology, but including Information Technology and Artificial Intelligence. When a massive new shift like this great acceleration in ai happens, one of our Important Roles is to be clear with our colleagues in the white house, with the president , with our departments and agencies about how the technology is progressing. What issues they will need to contend with. What the big opportunities are. And that means that in the case of ai, our national Ai Initiative office which is established at ostp a couple of years ago, the cardroom of people in my organization have been extraordinarily busy mapping out the risks, the opportunities and informing policy. What are some of the operations of your office . It maintains governmentwide ai use case inventory, is that correct . The office of management and budget. Cemented inventory has been criticized in the press as being inconsistent and incomplete for the inventories like uniformity and some significant ai use cases have been omitted. So is your office doing anything to improve the inventories, to improve transparency with the public . What does that look like . The initiative to start cataloguing those use cases was an important one and its very much work in progress. We are getting good insights from what is already in the use case inventory and working with departments and agencies. One last question. 25 seconds. Omb is more than two your site years late to give agencies guidance on acquisition and use of ai for the law requires omb to courtney with your office in drafting the guidance. So why quickly, why is a process stalled and when can we expect to see some guidance . The office of management and budget is working in a very focused manner on what they clearly understand. Two years late. Thank you so much. Our time is up and i yield back. Pursuant to the previous order. I apologize because we are up for votes. The chair declares a committee and recess subject to the call of the chair but we will stand in recess for votes. Thank you