[0:07:08] The joint hearing will come to order. Before I get started, I'd like to say, man, what a difference a week's make. And thank you, Ms. Anita, and lemon juice. This time last week, I didn't know if I'd be vertical this week, but I feel like a new man today. Without objection, the chair is authorized to declare a recess of the subcommittee at any time. I also ask unanimous consent that Reps Beyer and Ishu be allowed to sit in for this hearing.
[0:07:39] Without objection, so order. Welcome to today's hearing entitled Federal Science Agencies and the Promise of AI in Driving Scientific Discoveries. I am going to now recognize myself for five minutes for an opening statement. Good morning. I would like to thank our witnesses for taking the time to join us. Today's joint subcommittee hearing will examine how we can harness artificial intelligence to drive new scientific discoveries.
[0:08:08] We will also discuss the obstacles academic researchers and startups face in gaining access to these AI research resources and the proper role of the federal government in ensuring they are readily accessible. All of us have heard about the rapid pace of technological process in AI. I have seen it firsthand how applying AI to the trucking industry has profoundly improved driver productivity and efficiency.
[0:08:36] America is a world leader in AI development, and according to the Stanford University AI Index report of 2023, private businesses have invested roughly three and a half times the amount invested by China in cutting-edge AI systems. The U.S. is also the global leader in newly funded AI companies.
[0:08:58] The federal government spent nearly $3 billion in 2022 on AI research and development. The federal government is the leading source of support for basic research and areas critical to national security. Congress, led by this committee, passed the National AI Initiative Act in 2020. This initiative laid the groundwork for the research the National Institute of Standards and Technology and the National Science Foundation are conducting in AI.
[0:09:29] I am pleased to see Georgia Tech represented among the panelists as it houses three of the 25 NSF AI institutes in the US. These institutes are applying AI to the real world challenges. Solving these challenges is limited by the high cost of compute infrastructure. It is estimated that the cost of computing has increased annually by a factor of three between 2009 and 2022.
[0:09:57] These costs have widened the gap between big tech academic AI researchers and entrepreneurs. Facilitating public-private partnerships can help narrow this gap and efficiency maximize the development and use of responsible AI systems. This committee looks forward to seeing how the NAIR pilot at the National Science Foundation progresses over the next two years.
[0:10:22] With 11 federal agencies and 25 private sector and nonprofit organizations partnering in the pilot, it looks to be off to a good promising start. At the end of the day, our adversaries are not waiting for us. The Chinese Communist Party is implementing AI industrial policy on a national scale. It is Congress's responsibility to ensure that America maintains its leadership role in AI.
[0:10:50] That will require the academic community, industry partners, and small businesses to all work together to shape the future of this technology. I want to again thank the witnesses for being here today, and I look forward to a productive discussion. I now recognize the ranking member, the gentlewoman from Michigan, for an opening statement. Thank you, Chairman Collins.
[0:11:16] You've had a remarkable first term in Congress and you're such a valued voice in this institution and it's a delight to work with you on this committee and certainly thank you Chairman Williams as well for hosting today's hearing. AI, the conversation that possesses Congress but compels action and we're here to explore the promises of artificial intelligence for scientific discovery and how
[0:11:43] The incredible scientific agencies of the United States government can harness and enable these opportunities. So we've got just a remarkable set of witnesses who are going to provide testimony and it's a real delight to have Dr. Luhe Sharma here from Oakland University, the pride and joy of Michigan and really the
[0:12:13] School of Engineering that is doing so much in Oakland County. So thank you, Dr. Sharma. And look, the Science Committee has just had a great track record in terms of passing legislation to minimize the downsides of artificial intelligence and maximize the good for this revolutionary technology. And while responsible development of AI will continue to be
[0:12:42] Certainly a long-term challenge and something that this committee will work on for a long time to come. AI has also given us new and exciting opportunities.
[0:12:54] Many that just haven't been tapped into yet. So we're going to hear about that today. And in my home state of Michigan, advances in artificial intelligence by automakers are accelerating the development of autonomous vehicle technology that could potentially, you know, that has some real significant opportunity to...
[0:13:17] minimize traffic and certainly increase safety you know the goal of zero auto accident fatalities is something that we should be championing here from the halls of Congress and so AI systems are also going to be analyzing as we know massive amounts of data to accelerate research in fields from cosmology to
[0:13:39] Engineering and biology and neurology and those examples present just a tiny slice of the ways in which artificial intelligence is already benefiting many aspects of our lives and supporting our economic and national security. There's another committee here in the United States Congress, a select committee on U.S. strategic competition with the Chinese Communist Party.
[0:14:06] And out of many core technology areas, artificial intelligence is one that we are honing in on. And we realize that it's not going to just be the private sector alone that is going to be compelled to invest and act. It's our partnership with academia and federal agencies that are not only authorized, but invested in. We passed a chips and science bill. We did a risk management framework for artificial intelligence.
[0:14:35] We're working here on this committee to reauthorize Quantum Initiative Act to harness not only the computing power and how capital intensive that is, but also its applications.
[0:14:48] And while we say AI, AI, AI over and over again, and trust me, I'm in an AI meeting multiple times a week, it feels like when we're in session, it is so important to also be talking about what this committee is working on with quantum. And so as we have prepared testimony here today, certainly those of us on the committee who are dedicated to the technology prowess of the United States of America vis-a-vis our
[0:15:18] Our grand global competitions that we will not just sleep through, that we will harness through our capabilities as a free and open society, as a democracy, and as a country that has the willingness to co-invest in strategic competition. We're really very proud of the researchers, the efforts and the energy, and we look forward to what transpires.
[0:15:46] In today's hearing, thank you, Mr. Chair, and I yield back. Thank you, Ms. Stevens. I now recognize the chairman of the Energy Subcommittee, the gentleman from New York, for an opening statement. Thank you, Mr. Chairman. Glad to be here. Good morning. Today, the Subcommittee on Research and Technology and the Subcommittee on Energy will be examining the federal government's science agency's ability to drive innovation and deliver scientific achievements using artificial intelligence.
[0:16:14] AI will play a crucial role and an important role in our society and in our economy in the coming decades. To be clear, artificial intelligence is like the first powerful microscopes that allowed humans to see details into the physical world hundreds of years ago. Microscopes today continue to unlock secrets in medicine, material, and countless other innovations.
[0:16:39] AI is also like the emergence of computer technologies that, in just the last 75 years, have done amazing things. Today, machines help us solve problems at unprecedented speed and unimaginable complexity in just seconds, like how to safely navigate to the moon and get home safely, or like how to examine DNA information, looking for clues and diseases and potential cures, just some of the things that computer technology has done.
[0:17:09] AI is simply a tool that allows humans to see complexity and insight into huge amounts of data of our modern digital age. For example, artificial intelligence may vastly reduce the times it takes to increase the accuracy when doctors, sorry.
[0:17:28] Vastly reduce the time it takes and also to increase the accuracy when doctors evaluate cancer cells, increasing diagnosis and treatment options at a critical moment. AI is a powerful and exciting tool. And like all new tools and technologies, the risks and opportunities must be understood. And we also must confidently move forward. Given its untapped potential, the federal government and its agencies have an important role to play. Why?
[0:17:57] Simply put, the billions of dollars that the federal government has invested in science over a very long period of time has produced invaluable data sets that hold secrets that AI alone can unlock. These AI-enabled discovery could be transformational in energy, medicine, materials, and many other areas. Similarly, the federal government has tools like
[0:18:25] High performance computing resources that are unique and powerful for creating AI generated algorithms from this remarkable data set that they possess. So let me dive in here for just a minute.
[0:18:36] The National Science Foundation supports university research while fostering next-generation workforce. The National Institute of Standards and Technology, part of the Department of Energy, develops standards for trustworthiness in collaboration with industry. The Department of Defense uses its own research laboratories, as well as the industrial suppliers, to develop AI technologies to keep our nation safe. And of particular relevance to
[0:19:04] My Energy Subcommittee, the Department of Energy maintains and develops the federal government's AI infrastructure through the national laboratories. Of particular importance, the national laboratories are critical in developing next generation workforce that will be critical to our nation's future with artificial intelligence.
[0:19:22] Given its unique data sources and its dedicated access to computing and unique experimental facilities, that's where data comes from, the Department of Energy is strategically positioned to lead and accelerate AI development. The Advanced Scientific Computing Research Program, located within the Office of Science, conducts cutting-edge computational and networking research. It also manages the most advanced computing systems in the world.
[0:19:50] Top 500 list of the fastest supercomputers. The Department of Energy operates four of the top 10 of these supercomputers, including the only two exascale computers in the world, Frontier at Oak Ridge and Aurora at Argonne.
[0:20:11] Because of these extraordinary tools, the DOE is well positioned to lead and to play a larger role in this ecosystem as it relates to artificial intelligence. Today's hearing, as we dive into NAR and other programs that are in place, we're going to try to explore where the federal government should and could lead and use our unique resources to
[0:20:37] Drive AI-fueled innovation here in America for Americans. So, you know, as Congress continues to support this research, I hope that there's new innovation that comes out of even this dialogue. And I look forward to our conversation today. I yield back. Thank you, Mr. Williams. I don't know why I couldn't get that out. I now recognize the ranking member of the Energy Subcommittee, the gentleman from New York, for an opening statement. Thank you so much.
[0:21:07] Good morning, and thank you to all of the distinguished witnesses for joining us here today, as well as to Chairman Collins and Chairman Williams for holding this important hearing. While we have all gained awareness of some of the risks associated with AI, and we will certainly explore these risks further in our discussion this morning, this hearing will also present us with an opportunity to discuss the benefits of using these capabilities to further drive innovation and advance our national scientific expertise.
[0:21:35] DOE and its network of national laboratories employ advanced computational systems and powerful supercomputers that enable them to be at the forefront of AI development. With their computational resources and tools, DOE researchers are well positioned to fully benefit from the capabilities of artificial intelligence.
[0:21:54] These advances may help us solve humanity's most pressing problems by enabling researchers to analyze huge data sets, rapidly generate and test new designs of life-saving medical treatments, and optimize new advanced manufacturing techniques. DOE's supercomputing ecosystem and AI tools serve as a critical resource for academic and industry users from the US and around the world. However,
[0:22:20] Outside of a handful of well-resourced tech companies and universities, there is a general lack of access to computing resources that would allow them to train AI systems of comparable complexity with chat GPT. This is particularly concerning with regard to institutions serving marginalized communities. To ensure that we are sharing the benefits of AI and advanced computing equally, we need to pursue an agenda of scientific computing for the people.
[0:22:48] As I have said before, we need a skilled and diverse workforce to maintain the vitality of DOE's scientific computing ecosystem long into the future. To that end, I am looking forward to discussing with our witnesses how expanding access to these resources can tackle these challenges. I would also note that while DOE's AI and supercomputing capabilities hold tremendous promise for accelerating scientific discovery, we need to use these capabilities responsibly and ethically.
[0:23:17] As the use of artificial intelligence becomes more commonplace in our everyday lives, we must ensure that its fundamental algorithms are designed to protect people's privacy and eradicate bias. I repeat, and eradicate bias. We must also stop these tools from fortifying the structures of systemic racism as we have seen happen with things like predictive policing and facial recognition technology.
[0:23:45] This will only become more important as AI is further trained to be able to process, analyze, and store sensitive information such as biomedical data sets. All that said, I want to thank again our panel of witnesses assembled today, and I look forward to hearing your testimony. And I just want to further emphasize the issues of bias,
[0:24:09] And structures of systemic racism that persist in many of our institutions. We all carry biases as we create these new technologies. It's important for us to be aware of that up front and have safeguards as it relates to that going in. Thank you and I yield back, Mr. Chairman. Thank you, Mr. Bowman. The chair now recognizes the ranking member of the full committee for a statement.
[0:24:34] Thank you, Mr. Chairman. Today's hearing will explore the promise of artificial intelligence in supporting scientific innovation. Now, this is a topic the Science Committee knows well. In 2020, this committee led the development of the National AI Initiative Act to support research, development, and standardization for trustworthy AI.
[0:24:55] The bill created a whole of government approach to fostering AI driven innovation. And since it became law, the initiative has resulted in some truly wonderful outcomes. Near my district, NSF and the Department of Agriculture supported the University of California at Davis to launch an AI research institute dedicated to next generation food systems that are more resilient to climate change, disease, and supply chain disruptions, healthier for our communities,
[0:25:25] And more productive for our farmers. AI is also being applied to minimize resource consumption and waste in food systems. One of the most interesting aspects of these AI institutes, which we authorized in the initiative, is that each of them fosters wide reaching ecosystems around their respective topic. Not only does UC Davis include partnerships with six institutions, but it supports agricultural competitions, hackathons, and internships to teach students
[0:25:54] Students from the surrounding area to learn about AI and robotics for farming. These competitions include participants such as Hartnell College, a wonderful community college in my district well to the south of Davis. Colleges like Hartnell will play an important role in AI ecosystems. They'll train technical workers to use AI technology in fields like food and agriculture.
[0:26:20] However, small universities and community colleges face significant resource constraints and are often reliant on large universities like UC Davis to give their students opportunities for hands-on learning. Even most research universities run into barriers with access to computing and data resources that's going to be required to conduct AI research.
[0:26:43] Now, a major focus of this hearing will be the National AI Research Resource, or NAIR. NSF launched a pilot for this program a few weeks ago. The goal of NAIR is to democratize access to AI tools. To date, most of these resources exist in industry, specifically in a handful of very large companies. To truly achieve the promise of AI for societal benefit and develop effective guardrails against harm,
[0:27:12] Talented and passionate researchers, students, startups from across the nation will need access to the kind of computational and data resources that are currently available only to a few, to the biggest companies. Achieving that kind of scale
[0:27:29] will require significant federal investment. To justify this investment, we need to make sure our resources can go as far as possible, but also make sure they're used efficiently and judiciously. I look forward to the discussion today to learn more about the NAREP program and how Congress can help address the resource needs.
[0:27:50] of the AI community, including startups, large research universities, as well as community colleges like Hartnell. Thank you to our distinguished witnesses, and Mr. Tehrman, I yield back. Thank you, Ms. Lofgren. I will now introduce our witnesses. Our first witness today is Ms. Tess DeBlanc Knowles, who is a special assistant to the Director for Artificial Intelligence at the National Science Foundation. Our next witness is Dr.
[0:28:20] Georgia Tarasi, who is the Associate Laboratory Director for Computing and Computational Sciences at the Oak Ridge National Laboratory. Next, we have Dr. Abdallah, who is the Executive Vice President for Research at the Georgia Institute of Technology. Then we have Dr. Shamra, who is the Dean for the School of Engineering and Computer Science at Oakland University.
[0:28:49] And our final witness is Mr. Clark, who is the co-founder and head of Policy of Anthropic. With that, I now recognize Ms. DeBlanc Knowles for five minutes to present her testimony. Chairman Collins, Chairman Williams, Ranking Member Stevens, Ranking Member Bowman, and Ranking Member Lofgren, and the members of the subcommittees.
[0:29:14] It's a privilege to appear before you today to discuss the important role the US National Science Foundation plays in pursuing leading-edge AI research and harnessing AI to drive discoveries in scientific domains. For more than seven decades, NSF has invested in research, researchers, innovations, and innovators, and world-class scientific research infrastructure that has garnered incredible benefits to the nation.
[0:29:38] Many of the technologies and the industries that are the focus of national conversations around competitiveness today, including artificial intelligence, are rooted in sustained NSF support, in many cases over several decades. As a leading non-defense federal funder of AI research, NSF invests more than $800 million each year on AI research, education, and infrastructure across all 50 states, DC, and Puerto Rico.
[0:30:04] NSF's ability to leverage expertise across disciplines allows the agency to bring diverse groups of scientists and engineers together with private industry, communities, and others to identify problems and use science, engineering, and technology to develop solutions. NSF is driving cutting-edge innovations that expand our understanding of AI concepts and techniques, accelerating trustworthy AI development, applying AI across scientific domains,
[0:30:32] Democratizing access to AI resources and preparing the next generation AI workforce. For example, in the field of biotechnology, NSF funded researchers are using AI to identify strains of crops with favorable characteristics to improve crop yields. In engineering, NSF funded researchers are leveraging AI to address supply chain gaps, test quantum networking hardware, and optimize aviation maintenance and inspection. In the health sector,
[0:31:00] NSF-funded researchers are developing AI tools to monitor patients with lung disease and alert medical staff to patients at risk for sepsis. NSF has also invested in large-scale transformative efforts. In 2020, NSF launched the National AI Research Institutes, bringing together government, industry, academia, and others to advance AI-based technologies across a wide range of important application areas.
[0:31:25] Today, there are 25 AI institutes representing a $500 million investment throughout the country, working to advance weather and climate modeling, education, health care, agriculture, and many other critical national priorities. Importantly, they are also developing the AI leaders of tomorrow. And just last week, we announced the first ever NSF Regional Innovation Engines, investing in 10 teams spanning 18 states.
[0:31:51] These investments intend to harness science and technology and regional level resources to spur innovation, leading to breakthrough technologies and new jobs. Among these 10 inaugural NSF engines, seven involve a significant focus on the application of AI in areas such as sustainable water intensive industry, next generation agriculture, and regenerative medicine. However, our ability to continue to lead in AI and innovate with AI
[0:32:19] requires an investment in the infrastructure required for current and future generations to harness the power of this technology. Access to such infrastructure is currently limited to well-resourced institutions and companies. And Congress recognized this issue in the passage of the National AI Initiative Act of 2020, which spurred the creation of the National AI Research Resource, or NARE, task force. That task force recommended the creation of a NARE to expand access to the resources and tools that make AI innovation and discovery possible.
[0:32:48] Just a few weeks ago, NSF and collaborating agencies across the federal government launched the Nair Pilot. This pilot is a first step towards realizing the vision for a shared research infrastructure that will strengthen and democratize access to critical resources necessary to power responsible AI discovery and innovation. The Nair Pilot will bring together computing,
[0:33:10] Data, software, and training resources contributed by 11 agencies and 25 private sector nonprofit and philanthropic partners to support work that pushes the boundaries of AI technology and that applies AI to other fields of science and engineering. The pilot will run for two years, serving as the proof of concept for a full-scale Nair that will allow for researchers throughout the country to access world-class infrastructure to use AI to accelerate innovation.
[0:33:37] While AI raises significant challenges that must be addressed and risks that need to be mitigated, we cannot lose sight of the extraordinary promise it holds. More thorough and faster cancer screening, more accurate weather modeling and forecasting, more secure cyber infrastructure. These are the benefits we're already seeing today. By providing more access to AI tools and fostering innovation across fields of science and engineering, we can unlock even greater advancements for the nation. Thank you for the opportunity to testify before you today.
[0:34:08] Thank you. I now recognize Dr. Taurasi for five minutes to present her testimony. Chairman Collins, Chairman Williams, Ranking Member Stephens, Ranking Member Baumann, Ranking Member Lofgren, and distinguished members of the committee, thank you for the opportunity to discuss the important role the Department of Energy and its national laboratories play in advancing a national artificial intelligence capability.
[0:34:42] I'm a biomedical engineer and computational scientist by training, with three decades of experience as a translational AI scientist. I was also privileged to direct the Oak Ridge Leadership Computing Facility and help deploy Frontier in 2022, the first exascale system in the US and the world's most powerful supercomputer dedicated to open science. Frontier marked an impressive trajectory for our facility, for DOE, and for the nation.
[0:35:10] Demonstrating that in the past decade we improved computational power by a factor of 500 while limiting the growth of energy consumption which increased only by a factor of 4. Frontier currently supports researchers from academia, industry and federal agencies in training very large AI models to advance both basic and applied science and offer new insights into quantum materials, chemistry, physics, biology, fusion and more.
[0:35:40] Leveraging its unique scientific user facilities and research capabilities, DOE has supported multi-agency collaborative projects to advance shared national priorities throughout the years. Now, DOE is supporting NSF's NEAR pilot by providing access to its leadership computing resources, AI testbeds, and data assets.
[0:36:02] This includes the Oak Ridge Leadership Computing Facility's Summit supercomputer, currently ranked number 7 fastest supercomputer in the world. Summit offers an accredited environment well-suited for the NIR Secure focus area. Co-led by NIH and DOE, NIR Secure enables academic scientists to pursue AI research using data requiring privacy protections and extra security.
[0:36:29] In addition to working with NSF to provide greater access to AI resources, DOE is also taking steps to fill the gaps in performance, efficiency and trustworthiness needed to fully deliver on the promise of AI. The immense computing power required to train increasingly complex AI models demands even more energy-efficient, high-performance computing solutions.
[0:36:53] DOE has been making vital R&D investments to address these challenges and extend the transformative efficiency and performance gains achieved through its exascale computing project. But more progress is needed. That's why the DOE National Labs collaborated to develop a programmatic roadmap known as the Frontiers in AI for Science, Security and Technology, or FAST. Recognizing that AI will accelerate innovation,
[0:37:22] FAST targets complex mission-critical applications in energy, science, and security. It does so by leveraging DOE's unique compute resources, vast scientific data assets, and multi-domain scientific expertise. Although interest in AI is dominated by applications, more research is needed to improve AI performance, efficiency, safety, reliability, and trustworthiness.
[0:37:47] FAST aims to address these gaps as well, leveraging DOE's long history in scientific machine learning and deep experience in balancing risk and opportunity when translating solutions from open science to national security. In addition, FAST aims to drive progress and amplify innovation in AI through open partnerships with academia and industry, as DOE has successfully done many times before, including with the Exascale computing project.
[0:38:15] Nair and FAST take different but complementary approaches to secure the nation's global competitiveness in AI. While Nair focuses on building AI capacity for the nation, FAST focuses on building AI capabilities by targeting complex, mission-driven challenges. In summary, as a science and technology powerhouse, the DOE system of national laboratories possesses unparalleled compute and data resources and multidisciplinary talent pools to advance
[0:38:44] Cutting edge AI from research to fielded applications in challenging mission domains. A coordinated government-wide strategy leveraging DOE's capabilities will be key to the United States maintaining its edge in the AI race unfolding worldwide. Thank you again for the opportunity to testify. I welcome your questions on this very important topic. Thank you. I now recognize Dr. Abdallah for five minutes to present his testimony.
[0:39:13] Chairman Collins and Williams, ranking members Bowman and Stevens, and members of the committee, thank you for the opportunity to testify. Today, I would like to address three points. The imperative of long-term planning and funding to fully unlock AI's potential, the necessity of complementing U.S. leadership in basic science and engineering with the capability to develop AI hardware, software, and systems, and the importance of expanding our talent pool.
[0:39:41] I note that the business successes of AI technologies today are the results of decades of research at universities and industrial laboratories funded by NSF and other federal agencies. While industry directs R&D funding into well-defined businesses, universities focus on discovering and innovating in areas not yet created. Let me describe one example of the symbiotic relationship between federal agencies and labs, academia and industry. Georgia Tech, as you heard,
[0:40:10] was awarded three NSF AI institutes and an EDA grant for AI in manufacturing. One such institute focuses on adult learning and online education, using generative AI to personalize learning and transform education for distant learners. Another institute is combining AI and optimization tools for complex systems design, such as power grids, large supply chains, as well as to the design of the next generation of chips.
[0:40:36] The third institute is developing collaborative AI systems to assist aging adults through personalized interaction that will lead to more fulfilled lives and substantial healthcare savings. Georgia AIM is an EDA-funded network of universities, technical colleges, industries, and community organizations that is assisting 38 small manufacturing businesses around Georgia with cybersecurity compliance.
[0:41:01] Georgia AIM also created AI robotics technician training curricula for the Veterans Education Career Transition Resource Center to help them transition into civilian jobs. Challenges faced by academia and AI research include the need for large numbers of graphical processing units, GPUs, and large local storage and memory. As one example, an AI institute allocated $1,000 per month
[0:41:25] for using ChatGPT 3.5, but one researcher exhausted all of it in one day by instead using ChatGPT 4. As another example, a recent global ranking of AI programs placed 11 universities, seven of which are in China, ahead of the top American university, pointing to a growing advantage of international competitors.
[0:41:48] Pressures on building and sustaining large academic computing capacity are exasperated by how universities piece together the funding for their research computing infrastructure from smaller grants. The AI research resource, NER, will broaden access for computing infrastructure in a coordinated, equitable, and hopefully sustainable way. Transformative advances in AI also require talented contributors.
[0:42:14] Nair promises to bring together industry, government, and academic partners to help drive equitable access to education, training, and advances in AI. While the establishment of Nair and similar activities across the federal and industrial funding landscapes are steps in the right direction, I believe the following are important additional steps. First, a commitment to the long-term funding and certainty in supporting Nair.
[0:42:38] Realizing the importance and promise of Nair to our national and economic security, its startup funding should be supplemented with stable operating support. Congress enacted chips in science, funded chips, and put down a payment on the science aspects, but now is the time to realize that science and AI are just as critical to American security as are semiconductors and manufacturing. Second, reclaim our ability to translate our fundamental research into usable products.
[0:43:06] Along with a renewed focus on the reshoring and friendshoring, federal agencies are encouraging universities to translate their fundamental research into products, therapeutics, and companies. We must do so, however, while accounting for the social implications of AI development. Values such as fairness, transparency, explainability, privacy, and accountability must be designed for at the onset. Third, increase our efforts to attract and retain more talent into AI.
[0:43:35] The competitive advantage of US high-tech companies is often limited by the availability of talent. We must regain our role as the strongest magnet for talent and creativity from around the world. There is also an economic and moral imperative to attract more domestic students into higher education and more specifically to attract underrepresented groups into STEM in general and AI in particular.
[0:44:00] Lastly, the potential of AI for reskilling and upskilling displaced American workers cannot be overemphasized. I end with a strong endorsement of the stated goals for NERD, namely to spur innovation, to increase diversity of talent, to improve capacity, and to advance trustworthy AI. And thank you. Thank you. I now recognize Dr. Shamroff for five minutes to present his testimony. Chairman, ranking members, and committee members.
[0:44:29] Thank you for the invitation to testify before this committee. The opportunity to discuss the potential of artificial intelligence in driving scientific discoveries and enhancing federal science agencies' capabilities is both an honor and a privilege. AI stands at the forefront of a technological revolution. Its potential to transform research across a spectrum of scientific fields is unparalleled. However,
[0:44:58] Realizing this potential requires a concerted effort in harnessing AI capabilities while addressing the challenges it represents. Given these are related dynamics, it's crucial that we recognize the role of our two research institutions similar to Oakland University, especially for emerging technologies such as AI. These institutions are crucial for advancing AI research and workforce development.
[0:45:27] The collaboration between Oakland University and the automotive industry to train employees on cybersecurity, data science, and AI exemplifies the synergy between academia and industry. This partnership not only enhances workforce skills, but also aligns with the industry evolving needs. Federal agencies play a crucial role in the scalability and success of such programs by providing funding and support
[0:45:55] that can help expand these educational initiatives, making them more accessible to a broader audience. This not only enhances the workforce competencies, but also contributes to national competitiveness in a global technological arena. To further this initiative, Oakland University has introduced stackable certificate programs targeted at both high school and adult learners.
[0:46:24] These programs are designed to offer flexible, progressive learning pathways that enable individuals to accumulate and build upon their skills over time, thereby fostering a versatile and adaptable workforce equipped to meet the evolving demands of the tech industry. Federal agencies and departments such as National Science Foundation and the Department of Energy
[0:46:49] can be instrumental in bolstering AI initiatives at R2 institutions through targeted funding, partnership, and resource sharing. By providing dedicated grants and programs, these agencies can enable R2 institutions to enhance their research capabilities in AI. Furthermore, fostering collaboration between R2 institutions and industry partners can bridge a gap between academic research and real-world applications.
[0:47:18] Thus enriching the AI workforce. Investing in AI education and training programs at our two institutions is equally important. Tailored curricula, internship and mentorship opportunities can prepare a diverse and skilled workforce equipped to meet the evolving demand of the AI industry and other opportunities.
[0:47:42] We call on the House Committee and the National Science Foundation and other stakeholders to foster support to AI initiatives, particularly at R2 institutions. This includes increasing funding for AI research, facilitating the development of AI infrastructure, and promoting educational programs to prepare the next generation of AI experts. Enhanced federal support is crucial for maintaining America's leadership in AI and ensuring its positive impact on society.
[0:48:12] In conclusion, the advancement of AI offers a promising avenue for scientific discovery innovation. By recognizing the critical role of our two research institutions, encouraging industry-academia collaboration, and addressing the challenges inherent in AI, we can harness the full potential of AI. We urge the House Committee on Science, Space, and Technology to champion this initiative, ensuring the United States remains at the forefront of this transformative field.
[0:48:43] Thank you for this opportunity to highlight these critical initiatives and I look forward to discussing how we can further harness these opportunities for our nation's scientific and technological advancement. Thank you. Thank you. I now recognize Mr. Clark for five minutes. Chairman, ranking members, and distinguished members of the committee, thank you for the opportunity to testify for you today about the importance of American leadership in the field of AI.
[0:49:12] My message is simple. It has become more expensive to develop frontier AI systems, and the federal government should keep pace with this rise in costs and invest in initiatives like Vinare. For this testimony, I'm drawing on my background as a co-founder of Anthropic, an AI research company that builds and deploys safe and useful AI systems.
[0:49:36] We've spent many millions of dollars developing systems that sit at the frontier of large-scale AI work today. And before Anthropic, I worked at OpenAI, where similarly I built and deployed large-scale systems. Through this experience, I've seen up close the immense value and scientific possibility of AI, as well as its rising costs. To put it in perspective, the costs to develop cutting-edge generative AI systems
[0:50:06] has risen staggeringly in recent years. In 2020, it would cost you between $1 and $5 million to develop a large scale system. In 2021 and 2022, that rose to between $10 and $20 million. In 2023, last year, it cost between $50 and $100 million to stay on the frontier and the costs keep rising.
[0:50:30] To illustrate the scale, Meta CEO Mark Zuckerberg said last month that his company plans to acquire 350,000 AI chips by the end of 2024. That is a retail price of about $10 billion. It's not inconceivable that AI systems costing hundreds of millions apiece become commonplace soon.
[0:50:55] Vinay can serve as a counterweight to industry's outsized role in AI development by providing crucial infrastructure for public interest AI research. Specifically, Vinay should fund research into beneficial applications neglected by industry, such as education, healthcare, fundamental science, and so on. It should enable computational resource access for studying the safety properties
[0:51:22] of large AI systems and support research into ways to strengthen oversight of these systems and reduce their potential harms. And it should facilitate the study of open source and openly accessible AI systems to include benefits measurement, societal impact and bias analysis, evaluation science, misuse assessment, model robustness, and more.
[0:51:46] In essence, Venere can diversify the AI future so progress benefits for common good and also be a sustainable source of scientific advancement for positive applications of AI. Venere can also expand government capacity to develop its own cutting-edge systems. By doing this, Venere can provide essential infrastructure to unlock targeted research at positive applications of AI and a better talent base for the federal government by creating valuable infrastructure
[0:52:15] Where researchers can develop and hone their skills. And it can reduce dependencies between the government and the private sector when it comes to understanding the state of art of AI. Information asymmetries are dangerous and Vinay represents a way to reduce them. I'm honored to be here and based on my experience working on the frontier of AI in industry, I asked the committee to consider the following policies.
[0:52:41] Number one, make targeted investments in the National Science Foundation, the Department of Energy, and the National Institute of Standards and Technology. Number two, find ways to better utilize the government's supercomputing assets for large-scale AI training, as Oak Ridge has recently done with its Frontier supercomputer. And finally, be bold. We are at the start of a technological revolution, and the more ambitious the government is here,
[0:53:08] Thank you for your time. Thank you. I want to thank all the witnesses for their testimony. The chair now recognizes himself for five minutes.
[0:53:24] You know, as I was sitting there listening, I was writing down some things just from my personal experience of being in the trucking industry for over 30 years. And a lot of the advancements that we've seen, I mean, it's been an incredible industry to be a part of. But to see the advancements, you know, we've got equipment now that AI has embedded inside the transmissions that will actually put the transmission in neutral. And then when it knows it's coming to a hill, it knows which gear to put it in.
[0:53:53] I was just told about a meeting that we had last week where there's new technology, AI technology out there that'll be able to read and help make better dispatch decisions. So as the truck approaches the place where he's unloading or she's unloading, it'll be able to better make a decision on the next load to make them more efficient. And I was reading an article here recently where soon in the near future, the trucking industry will have the ability.
[0:54:22] I don't want to put any onerous on you four-wheelers out there, but we'll be able to read what you're actually doing at any given time so that we can make better adjustments as we're going down the road so that we can avoid obstacles or incidents. Also, I have a small university in my district, Dr. Abdallah, I'm sure you're aware of him, University of Georgia.
[0:54:46] And the agriculture department there has made so many significant gains. And it's crazy about the AI when you go up there and you look at how they're spot fertilizing each individual weed or whether they are advancements in tractor technology and how they're able to better make improvements in soil and yields. And I just had a meeting yesterday with the technical schools.
[0:55:14] They're in Georgia. I'm a big technical school fan. And their main concern was artificial intelligence. And the reason that was their concern is they don't have educators that know how to teach what the technical schools are going to need to provide in the future. So it's a very worthwhile meeting. This is a huge hearing, in my opinion, and a personal one. Ms. DeBlanc-Knowles.
[0:55:40] How is the NAIR pilot program supporting small businesses in Georgia? And how will NAIR balance the interest of basic research and entrepreneurs who need access to these technologies?
[0:55:55] So we talked a lot about this resource divide between researchers in well-resourced institutions and private sector companies and academia. And that resource divide applies as well to researchers and startups and small businesses who don't have access to the adequate computational resources as they try to prototype and scale and really commercialize the products that they're developing. And so per the recommendations of the Nair Task Force that we're using as the framework for the Nair pilot,
[0:56:23] Small businesses and startups who have existing federal grants, so SBIR, STTR, and similar grants, will be eligible to apply for access to resources through the NAREPilot.
[0:56:36] I do want to emphasize that the first R in Nair is research. So even as we are equipping small businesses to do this kind of access through the Nair pilot, it is focused on the earlier stage prototyping and initial scaling of the product versus support for kind of production level development.
[0:56:57] Thank you. Dr. Abdallah, the Baptist came out. I mean, I took up a lot of time. Can you elaborate on how Georgia Tech's leading the way in AI basic research to new technologies and industries? Yeah, thank you for that, Mr. Chair. So in our space, we've been doing AI when it wasn't cool. You know, there were multiple AI winters, as you...
[0:57:21] Probably no. But some of that work, probably one of the most famous one is something we call Jill Watson. Jill Watson was actually an AI agent that helped in scaling up our computing program. Right now, we have a computer, a master's in computing, online master's in computing that has almost 15,000 students in it. And we're using AI tools in that, for example. Jill Watson.
[0:57:51] When it was implemented initially, it started out as a TA in a class, and the students initially did not like how this TA was behaving. By the end of the semester, they felt it was one of the most empathetic TAs. It was such a great thing that they're used to. Some of them even, they didn't know it was an AI agent. They asked her out.
[0:58:15] On a date, for example. Because at that point, it was delivering such great response. So that's one of the areas. Of course, I mentioned in the supply chains and microelectronics design, all the things that our colleagues at other universities are working on, too. Thank you. I appreciate that. I now recognize Ms. Stevens for five minutes for questioning.
[0:58:41] I am here for the Truckers for AI Caucus. This is so exciting to hear about that. And as a Metro Detroiter, we salute your interest in autonomous or AI innovations for the mobility sector. Lots to dig on here. And Dr. Sharma, I'd love to start with you because it's just, again, just such a treat to have you here representing
[0:59:11] Michigan and Oakland University. And, you know, the challenges faced by the non-Research 1 institutions in states with major R1 institutions, you know, we've got a lot of diverse research institutions in Michigan, each bringing, you know, kind of importance and a voice to the federal research ecosystem.
[0:59:37] It's critical that all of these institutions have access to federal resources and funding. So what do you think of the NSF, DOE, and other federal agencies? What should they do to ensure that non-R1 institutions like OU and its amazing researchers have to get access to federal funding resources? And how do you see Nair helping with these resources or these efforts?
[1:00:07] Thank you. And I really think the number one challenge we have right now is access to high computing computers. Because when we talk about AI and research, a lot of these institutions, especially similar to Auckland University, we have a challenge about access to these computing powers. So this is extremely important for us to work.
[1:00:30] And develop a workforce actually of the future. And the other thing is in terms of funding to develop curriculum and curricula to expand our training and our opportunity to diverse the workforce that we need for AI and moving forward. So federal agencies can be important in terms of funding these initiatives. Would that be a consortium type of model?
[1:00:57] Sharing resources is very important. So consortium could be where even our one schools can collaborate with our two schools and give them access to high computing powers. So I really think all it's about collaborating and all putting our resources together in order to move forward and to always stay competitive at the national level. So the federal agency could be a catalyst to bring all these constituents together as a consortium working together and
[1:01:27] Improving access to these technologies. And also scaling the AI certificate program, which you mentioned in your testimony. Is there a role for NSF or DOE to play? Obviously, we always think of Department of Labor with certificates. But what about scaling what you've been able to do with these AI certificates? So basically what we do right now with the industry, when we do these trainings, we have about 20 seats.
[1:01:56] Federal agency can scale all of this to hundreds or thousands. And I really think this is important, especially in an area like Detroit where the need for technology jobs are increasing day after day. And even they're demanding that we have these trained employees in the automotive industry. Train the trainer. Right. And we also want to get...
[1:02:22] Ms. DeBloch-Nolson, because you're obviously overseeing some of this effort at NSF, and so how does NSF engage institutions like OU to address the barriers to access? So that's one of the fundamental driving motivations behind the NAR pilot and then the full-scale NAR is really overcoming some of these barriers. So less well-resourced institutions have access to the necessary computational
[1:02:51] And also really important in this space is the training to equip those researchers to use those resources effectively, whether they're pushing the frontiers of AI or applying AI to their fields of study. Yeah. Well, you've certainly given us a lot to think about here. And just let it be known for the record yet again that Oakland University in Rochester Hills, Michigan, has an AI certificate.
[1:03:21] Thank you, and I yield back. Thank you, Ms. Stevens. I now recognize Mr. Williams for five minutes. Thank you, Mr. Chairman. Dr. Abdallah, I was concerned about your statement that Georgia Tech students were duped by AI into thinking it might be available for a date. Do you think that reflects more the dating culture at Georgia Tech than it does the advancement of your AI? And before you answer, I'll warn you, my son is a senior at Georgia Tech in aerospace engineering.
[1:03:52] You don't really have to answer that. Thank you. Thank you. So I've worked with GTRI in the past and seen the excellent work that you do, both commercially as well as inside the institution. How much federal funding flows through Georgia Tech each year related to research? So Georgia Tech does about $1.5 billion in research expenditures. About 80-plus percent of that is federal.
[1:04:22] And thank you. And how much of that is currently going towards AI or AI-related projects? A few hundred millions. In fact, I think the latest numbers that I sent with my statement is about 200 million. And just so you can flex, what kind of work is Georgia Tech doing in AI with those federal dollars? So as I mentioned, we have the three AI institutes that do work in learning. We do in manufacturing.
[1:04:51] We do work in the supply chain and design for materials and others. GTRI, as you mentioned, is a UARC. It's a university-affiliated research center. They do a lot of work for supporting national security and AI also. And we do also have teams or researchers focusing on trustworthiness and ethical AI. Thank you. Dr. Taurasi, I'm interested in...
[1:05:20] The role that memory plays in the architecture of an AI system relative to compute, we all talk about the GPUs and the crunching capability of creating these models. Can you explain the importance of particularly chip-based resident memory, RAM, that kind of thing?
[1:05:44] So there is certainly for the deployment of the next generation of high performance computing resources in the Department of Energy, memory bandwidth is what is at the forefront of our attention because we've seen the number of applications that either are directly artificial intelligence or they integrate artificial intelligence with modeling and simulation which has been primarily driven by more flops.
[1:06:10] So this is something that the Department of Energy is making strategic investments because this is a new frontier. And that's what exactly differentiates the Department of Energy in terms of how they leverage their investments. We have to be always at the bleeding edge of technology, understanding where the gaps are, and keep pushing forward. Right now, memory bandwidth is where we will pay our attention.
[1:06:40] In addition, of course, to energy efficiency. Many of you brought it up. This is something that we cannot ignore. And the Department of Energy has always been driven by investments and technology advances that speak to energy efficiency. I gave the example for Frontier. Thank you for that. That's very helpful. And it's also important to note that the global leader in memory technology is Micron, US-based.
[1:07:08] I couldn't tell you, Congressman. Do you want to guess how many automobiles the federal government
[1:07:40] I bet it's a multiple of that number, would be my guess. And I think that's true actually throughout its entire history. So when you talk about the escalating costs and the incredible innovation that's happening around AI systems, why is it that you suggest that the federal government needs to own
[1:08:06] A significant amount of the leading AI systems in order for America to compete. Excellent question. As you build up these systems, you discover a hundred different ways that they break, and you discover a hundred different ways to fix them. We've seen this with Frontier's own work scaling itself up to the frontier of possibility. If you don't build this stuff and develop the skills and the cuts and bruises from learning how to build it, you're not going to... Sorry, my time's expired, but only the government can do that?
[1:08:36] Private industry is not doing that. Google, they don't do that at Microsoft. Industry does a fantastic job of this, and it leads to those skills mostly sitting within industry, and you have less of a talent base that can be shared across government and academia. I think your colleagues have shared really excellent examples of federal dollars being used for the most impact for America to continue to lead in AI, and maybe owning the automobile factory is not the best way. Thank you, Ms. Williams.
[1:09:07] Now recognize Mr. Bowman for five minutes. Thank you, Mr. Chairman. This first question is for Mr. Blanc-Knowles and Dr. Taurasi. Much of the educational background that informs AI research and development can be taught well before the college and university level. Concepts such as what data is and how to logically go from a general idea to a specific conclusion can be taught as early as elementary school.
[1:09:32] What course materials, tools, and resources can NSF and DOE provide through the AI programs to equip our K-12 teachers with the ability to instill this insight? So NSF is focused on building the skills and technical capacity at each level that we see as needed to really move towards this AI-ready workforce and the incorporation of AI as we see coming across different sectors of society and the economy.
[1:10:03] We just recently launched the Educate AI initiative that focuses on making opportunities for the development of more inclusive AI educational resources at the K-12 level, community colleges, undergraduate levels, as well as for adults who are interested in reskilling.
[1:10:22] And this builds on work that we've already done through our AI for K-12 initiative. And that developed guidelines to help teachers who are looking to incorporate AI into their classrooms and created a repository of open resources that as teachers look to build these baseline skills that are critical, not just for those who want to pursue AI careers, but for students as they learn how to understand outputs of AI systems or integrate AI just into everyday use. And we're going to continue to build on this work going forward.
[1:10:55] So the Department of Energy offers many training opportunities and internships across different levels of development, starting with high school students to undergraduate, graduate, and faculty. These resources, we're making them available now by expanding our educational programs and internship programs to institutions that are minority institutions because we know that there are huge gaps there.
[1:11:20] And our resources are really at the bleeding edge of technology. So in order to elevate everyone, we have established programs such as pathways to high performance computing that help bring individuals along to make the most of our resources.
[1:11:38] For the educators specifically, we do not have, to the best of my knowledge, a program that is related to AI, something for us to discuss, but our user community is targeted scientists and we try to create the culture of science as early as possible.
[1:11:58] With high school students and engaging also with our local communities, pretty much every laboratory has a strong network of collaborations with local schools, community colleges, and universities. Thank you. Will the NAIR pilot program, excuse me, will the NAIR pilot NS programs provide education resources for K-12 educators?
[1:12:21] Educational resources are a key theme that we're launching the NeuroPilot with in terms of supporting access of educators to the resources needed to bring into their classrooms.
[1:12:33] Educate the next generation of leaders. For the Nair pilot, because of its limited scope, we are focused on undergraduate institutions, really because of that connection to the computing resources. We've heard from professors that they have to scale down their instruction to kind of fit within their limited computing and data frameworks. And students don't have enough computing to finish their AI term papers. So we really want to pilot equipping these university professors with the resources that they need so that
[1:13:02] American students are being able to train, experiment with, evaluate models at the scale that's at the cutting edge right now. Anyone else care to comment? Thank you. Thank you. The chair now recognizes the very stealthy chairman of the full committee, Mr. Lucas, for five minutes. Just watch and learn, Mr. Chairman. Just watch and learn. Mr. Clark.
[1:13:32] Cutting-edge AI research used in universities, but today this research is dominated by the industry, particularly due to the high cost of computing needed to train advanced models. To address this problem, some have advocated, of course, for the creation of Nair. Given the fast pace of progress in AI, how can we ensure that the development of Nair keeps up with the rapidly evolving industry standards for advanced computational power? Thank you.
[1:14:03] The key thing will be to make sure that researchers in academia are able to do research that they consider appropriately ambitious and large scale. They should not be able to run into a situation where they're unable to do their research due to running into computational limits. And how you achieve that in a fiscally responsible way is to make sure that Venere is allocating a portion of its resources for a small number of big ticket projects each year and adopt a consortium approach.
[1:14:33] for picking what those are. Would anyone else like to touch on that question? I would agree with the notion of the consortium because we've seen its success with the HPC COVID consortium a few years ago. We all leaned in with the resources that we had available, but there was a good coordination of the projects as they were coming in to match the scale of the project and the complexity of the project with the appropriate resource.
[1:15:02] Continuing with you, doctor, the Nair pilot program is a two-year program. However, the summit supercomputer is due to be decommissioned in less than nine months, October of 24. What are NSF and DOE's plans to address this computation gap? That's a very good question. I want to emphasize that the Department of Energy and the Oak Ridge Leadership Computing Facility, LinkedIn,
[1:15:27] To donate effectively the supercomputer and keep its operations without additional funding. That is a question now to see how the NSF NIR will evolve. And if there is interest and support, we will lean in once again. OK. Following up on that, if.
[1:15:52] If the pilot is being shifted to another supercomputer, how will this move affect the current research being run on that computer?
[1:16:05] Transition from one supercomputer to another, this is something very common for us because every four to five years we deploy new supercomputers and we work with the scientific community to ensure that their codes can transition from one system to the next. This was actually part of the Exascale computing project that worked with different scientific communities, different scientific codes to make sure that when Frontier and Aurora are deployed, the systems are usable on day one.
[1:16:35] So this is DOE's philosophy of how to deploy and operate resources and make sure that they are impactful from the go.
[1:16:46] And we've seen that with Frontier with so many applications and great successes for the nation, just with the Gordon Bell Awards a couple of months ago in supercomputing. That was possible because Frontier became available on April, beginning of April of 2023, and these teams were able to run largest scale computations since day one. It takes a village, and the Department of Energy and the National Labs provide not only the hardware,
[1:17:14] But actually the computational scientists who work with the users. That may be the most heartwarming answer of the day right there. Thank you. Miss DeBlank Knowles, the American scientific enterprise is open and collaborative with basic research published without restrictions. How can we maintain a balance between ensuring the openness of basic AI research while mitigating potential threats to national security that could arise from sharing such information?
[1:17:45] So in alignment with NSF policies that generally apply to NSF funding opportunities, there are pilot resources that require access requests. So these are the allocatable resources that are being made available through the pilot, the computing resources, a proprietary data set, software licenses.
[1:18:03] Those will be subject to our standard eligibility requirements. So researchers who are applying for funding must be based at US institutions. As we move forward in this process, the Nair Program Pilot Management Office, based out of NSF that involves interagency partners, will work very closely with NSF's office for the Chief of Research Strategy and Security Policy to make sure that we're integrating best practices in research security as we move forward to scale the pilot and optimize our resource allocation processes.
[1:18:31] Thank you very much. And I yield back, Mr. Chairman. Thank you, Mr. Chairman. The chair now recognizes the ranking member of the full committee, Ms. Lofgren, for five minutes. Thank you, Mr. Chairman. This has been a very important hearing, I think. There are so many issues here, but I want to pick one, Dr. Taurasi.
[1:18:52] In your testimony, you highlighted the application of AI to potentially improving the performance of fusion reactors. Specifically, you note that scientists from several national labs are using the leadership computing facility at Oak Ridge to train AI models that can predict and potentially
[1:19:14] Prevent plasma disruptions in the reactor. Can you discuss this in more detail? And are you seeing interest from fusion's emerging private sector in taking advantage of these models yet? So this is certainly one of the most active areas of research across the DOE complex. And certainly we see that in our supercomputing center with Frontier.
[1:19:39] You know that the modeling and simulation of nuclear reactors, it's an activity that has been going on for decades. And with every generation of supercomputer, we're pushing boundaries one step further. Now with Frontier, we are able to simulate the full lifetime of a nuclear reactor. And AI plays a very important role in terms of being able to optimize the design of the nuclear reactors, predict its lifetime, and
[1:20:08] Different events that need to be prevented. So what that means effectively is that the certification of nuclear reactors, that we design them, but then the process of certification for deployment is extremely lengthy. And this is now an environment where we can do the digital design process.
[1:20:34] of building, exploring different optimization designs to identify the ones that need to move forward for production with the appropriate certification. This is a turning point for the industry and in terms of accelerating production. If I can just follow up with that.
[1:20:54] A few weeks ago, I went up for a day trip from San Jose to Seattle to visit a number of the startup fusion companies. Hellion, they've been a witness here, and Zap, and some others. How do we connect those startup...
[1:21:13] Companies exploring fusion with the resources that you have. I didn't get the sense in talking to any of them that they were aware of the resources that you have, nor had they used them as they are developing really very interesting efforts in fusion, which is one of the toughest nuts to crack.
[1:21:35] That's a very good question, because as great as these resources are across the DOE complex, certainly there are many blind spots and communities that do not know us very well. We work with industry, big industry and startup companies. It is part of engaging
[1:21:55] Broadly and speaking to different communities, the demand is huge. Our systems are oversubscribed, but we are ready to support particularly startup companies and particularly new users through what we call the director's allocation program, discretionary allocation program, because we can bring them on board. So certainly there's more work for us to do in terms of socializing these great resources across so many communities.
[1:22:25] Let me just, you know, the San Jose Sharks play in my district, so I get to ask this question. Wayne Gretzky said you need to skate to where the puck is going to be. And in this committee, we're looking at a number of issues, but some of the most important are fusion, quantum, and AI. And it seems to me that quantum and AI are going to converge. That's where the puck is going to be.
[1:22:54] Expand access. And I'll just say some of the private sector individuals, for example, OpenAI, are skeptical that we can actually provide the computational power for non-big companies to do the research and compete. And I don't think you have to be opposed to the private sector to...
[1:23:14] I believe, as I do, that this should not be owned by the biggest companies in the world. This should be open to startups. This should be open to researchers, not just the big companies. Can you address that issue, or is that too far-fetched? For me, in terms of, if I understand correctly, your question is about these emerging technologies, AI and quantum, and access to these resources.
[1:23:42] Right? Correct. We're talking about access to AI. With Quantum, this is a technology that is still in its infancy, but it holds a lot of promise. I can tell you what we're doing at the Oak Ridge National Lab because we see that promise and we want to build the user community that can leverage these resources when they become.
[1:24:02] production quantum computers. We have the quantum computing user program that gives access to users via a cloud model to different quantum computing platforms. And again, with our computational scientists who understand both high performance computing and quantum computing, we'll try to lift these new users into a paradigm that will take time,
[1:24:26] To mature the same way as high performance computing, but certainly holds the promise. Mr. Chairman, my time is expired. Thank you, and I yield back. Thank you. I now recognize the gentleman from Florida, Mr. Franklin, for five minutes. Thank you, Mr. Chairman. Dr. Abdullah and Dr. Chamra, you both represent institutions with a significant number of students, and I have a polytechnic university in my district that has 2,000 students. How would you...
[1:24:53] How would you recommend that smaller research institutions position themselves to better compete for the federal university partnership funds that are out there? Thank you for that question. So as I mentioned, many of our AI institutes and our work is in collaborations with technical college systems and with smaller institutions. I think the best way for smaller institutions to access both the expertise and to scale up
[1:25:22] And their knowledge and applications is to really work closely with the larger institutions.
[1:25:51] Anyone in the world can get to, with the exception or with the limitation of this large resources which NER is addressing. But the knowledge to try to, the basic knowledge, the foundational knowledge in order to be able to use some of these tools is available a lot of different places. What is lacking is the access of resources, and I think the best way to do that is to team up, be part of the...
[1:26:18] Nair, for example, or other such initiatives. Dr. Charmer. I really believe that Nair and the federal agency should dedicate certain funding to these smaller schools because it's important to recognize that these smaller schools, whether 2,000 or up to 10,000 students, they have different type of students.
[1:26:47] Attract different students than these students. And actually, these smaller schools attract students from all geographical areas, from all walks of life. And I really think a federal agency should give them more resources to develop curriculum as well as research. And I agree with Dr. Abdullah that we should collaborate, but at the same time, there should be funding specific to these smaller schools where actually they train.
[1:27:10] A diverse group of workforce development and they train a lot more diverse student body. And I really think smaller school have an important role to play in this arena. And this is how we ensure that workforce development in AI and other technology are accessible to everybody. Thank you.
[1:27:31] All right, thank you. Ms. DeBlanc-Knowles, the NAR Task Force report recommended a $2.6 billion investment over the first six years. What criteria over the course of this two-year pilot project should we be looking at to evaluate success and decide is that the right number? Should we fund more? I mean, obviously, the answer to the question is always going to be, you know, how much more can we do? But how can we, what metrics should we be establishing along the way to gauge success? So the NAR pilot is being...
[1:28:01] Funded out of existing agency resources and then bolstered by in-kind contributions from that network of 25 partners. One of the aspects of the pilot that we're really excited about is as we bring together that diversity of resources from federal partners as well as private sector partners, we get to experiment with the types of resources that work well for the research community as well as different modes of them accessing that.
[1:28:28] And we think that that will help us better design and scope the plan for the full-scale NAR. I think some of the metrics that we can use to evaluate the impact of the NAR pilot is in terms of the number of users served, if we can reach into these communities that don't typically have access to these resources, and if we can
[1:28:49] On that educational thrust, you know, how many students we can train. And then really looking at the projects themselves that we are funding and looking at the impacts to those projects in terms of access to the computational data resources that they are unable to access through the pilot. Okay, thank you. Dr. Taurasi.
[1:29:10] I've got a security question for you. What procedures or guardrails has DOE put into place to ensure Summit doesn't become compromised as we're sharing the resources with outside models from folks like OpenAI, Meta, and others who run things on DOE assets and infrastructure? So certainly the Department of Energy with the leadership computing facilities has always been very mindful about cybersecurity because we're prime targets. There is no doubt about that.
[1:29:40] And there are lots of policies and controls that are in place that we continuously revisit because the world is a changing world. With the Summit supercomputer, that was the first supercomputer that we established authority to operate so that our users can bring projects that have sensitive data. And that is a capability that is the synergistic development of technical innovation.
[1:30:09] Policies and administrative controls, because it takes all of these three. And we're continuing that with the frontier. Now, with the new user communities, the policies that we are trying to put in place, they are related in terms of the data they bring in and the third-party data.
[1:30:29] Because for us, our users are scientific users. We bring scientific data. But in the AI era, and particularly under the near pilot, that scope is broader. So we need to be mindful and we're looking into the policies and what the expectations are from the users who will be granted allocation to the systems. There is a thorough process where we vet the users, the project, and the data assets that they want to bring in.
[1:30:57] All right, thank you. And I'm over my time. Mr. Clark, I had one for you, but I'll have to pass to later. Now you're back. All right, the chair now recognizes the gentlewoman from North Carolina, Ms. Ross, for five minutes.
[1:31:08] Thank you, Chairman Collins, Chairman Williams, Ranking Member Stephens, and Ranking Member Bowman for holding this very important hearing. And thank you so much to all the panelists for joining us today. I represent a district in North Carolina that's home to businesses, universities, an NSF AI Institute, and other organizations that are helping drive innovation and scientific discovery in AI.
[1:31:36] In November, I hosted a roundtable in my district with my fabulous colleague, Congresswoman Valerie Foushee, where we met with local and national leaders in artificial intelligence. And I'm looking forward to the next one that she will be hosting at an HBCU.
[1:31:55] Today's hearing highlights the importance of continued collaboration between Congress and the AI sector as we stand at the precipice of unprecedented technological advancements, and you have really shed a lot of light today. I want to pick up on a workforce issue that we haven't discussed.
[1:32:20] As we know American leadership in AI begins with a strong workforce that both nurtures domestic talent, which we've discussed a lot today, but also attracts the best global.
[1:32:33] Many international students come to the United States to conduct innovative research in emerging technologies. In fact, a 2019 report by the Center for Security and Emerging Technologies revealed that two thirds of graduate students in AI related programs are international students. I myself taught international students at Duke Law School and understand the expertise that they bring.
[1:33:02] And the number of domestic students in these programs unfortunately has not increased. And we've talked a lot today about how to increase that. And I'm tremendously interested in doing that. But many of these international students wish to stay here.
[1:33:20] And they face several obstacles to permanent residency and even citizenship. And we know that our immigration system with its visa backlogs is just broken. I've advocated for documented dreamers, kids who come here with their parents on H-1B visas and then have to self-deport to other countries with their entire education that they got here to compete.
[1:33:48] I'll start first with Dr. Abdullah. Can you discuss how the United States can better attract and ultimately retain international talent in the American AI enterprise? Thank you. I myself am an immigrant. I think if I look around this table, I think four of us out of the five are immigrants.
[1:34:15] How to attract it is by continuing to be as welcoming as we have been for the last, I don't know, hundreds of years. How to retain them is a different question because others right now, including competitors, allies, and adversaries, are becoming a lot more savvy about attracting even graduates from the United States.
[1:34:39] When you apply as a foreign student or as an applicant to a US university, you have to show to the embassy that you are going to go back. You have to prove that you are going to go back in order to get the visa, which defeats the whole purpose. So I think what Congress and others have been discussing around trying to make sure that
[1:35:01] Either the H1B or other visas are available in some targeted areas. I think that's one approach. But really, most appropriately, I think, is to try to keep the welcome mat and to try to figure out
[1:35:17] Not just from the companies or not just from a particular group, but to make sure that universities and graduates of these programs, especially in the areas that we're really interested in, have a path, have a legal path to try to contribute in here. Canada is doing that by taking some of our graduates and other countries are doing so. So I think just going back on that international talent.
[1:35:45] Initiative is really important. But I do want to add to also, you know, there is an opportunity with U.S. students. The reason I say that is because there's not enough of them going into these areas. So focusing on that is actually also another dial that we can use. Yes, as we say, both and. Yes. Thank you so much, and I yield back.
[1:36:14] The chair recognizes Mr. Obernolte for five minutes. Well, thank you, Mr. Chairman. Ms. DeBlanc-Knowles, I would like to highlight something that you touched on in your testimony. We have a rich tradition in this country where cutting-edge research is done in our academic institutions, and we have a tradition of publication where that research is transparent, it's shared with other people, and that leads to verification.
[1:36:44] By peer review and people in other institutions will literally stand on the shoulders of giants in reaching new frontiers in the advancement of human knowledge. But I fear we are in danger of changing the way that advanced research in artificial intelligence is done because as Mr. Clark testified, the cost of training frontier AI models.
[1:37:10] is on this exponential curve and it's already beyond the ability of most academic institutions to pay for. So that's why the National AI Research Resource is so important in establishing a pool of compute and data to enable this research to continue to be done in academia rather than by private companies.
[1:37:35] The pilot program for Nair is wonderful. Congratulations on getting that kicked off. Obviously, we are trying to codify that. We have the Create AI Act, which Congresswoman Eshoo, Congressman Byers, and I are co-sponsoring to try and make that permanent and attach it to a permanent source of funding. Will that be enough by itself?
[1:37:59] To guarantee that that research continues to be done in academic settings as well as in commercial settings. So making the transition from the Nair pilot to a full-scale Nair will have incredible impacts on our academic AI research environment, as well as the ability for scientists in other domains to apply AI to achieve and accelerate breakthroughs. And I think the impact can't be understated. And you mentioned the
[1:38:28] The role of the academic research community in terms of verifying and validating results. And so we see a lot of potential for the NARA, not just enabling a small group of users to train frontier models, but enabling many more researchers to work with the cutting edge models of today. We hear from researchers in academia that
[1:38:48] Even when working with an existing open large language model, they don't have access to enough compute to perform inference on that model and to really kind of kick the tires on how is it working, how can we improve the safety and reliability. And so the NAR is going to have an impact in terms of enabling researchers with just, they don't need tremendous amounts of compute, but they need adequate compute in order to work with the cutting edge models. Right. Thank you. Mr. Clark, very nice to see you again.
[1:39:15] I've enjoyed our conversations in the past. I just want to highlight how remarkable I found your testimony, because one of the problems that Nair is designed to solve is this problem that we might have a concentration of market power into just the few of the biggest companies who could afford to monopolize access to advanced GPUs and the energy needed to feed them and the data needed to train frontier models, which would limit competition.
[1:39:45] And also limit entrepreneurialism when it comes to AI. But I mean, you're here as co-founder of Anthropic saying this would be a terrible idea. You know, we support more competition. And I find that incredibly refreshing. Could you take a minute and explain why you think that Nair and in a larger sense, more competition is good, not only for the industry, but for the country? Thank you. I care about this deeply.
[1:40:12] I think ecosystems die when they become dependent on a very small number of entities within them. And this is true of private sector dominance for a really important part of science. We are fortunate to be in opposition. We publish research and we seek to share it. But we are one type of species in an ecosystem. There need to be others within us. And I think that's crucial for us to have really good ideas as a collective. And to your comments about...
[1:40:37] Open publication, the incentives of industry do not point in the direction of full publication of scientific information. We do our best, but we have incentives. And you need to both change those incentives and create people with different ones. That's why it's so important that we fund this and take it seriously. Well, thank you. I wish I had time for another question. I could talk about this all day. But thanks to all of our witnesses for your testimony on this really critically important topic. I'll go back.
[1:41:09] The chair recognizes Ms. Salinas for five minutes. Thank you, Mr. Chair, and thank you to our ranking member also for holding today's hearing and to the witnesses for sharing your value.
[1:41:21] Valuable insights. Ms. DeBlanc-Knowles and Dr. Taurasi, as you know, Oregon is a global leader in the semiconductor industry. The CHIPS and Science Act created a regional tech hubs program, and last year we were super excited to see the Department of Commerce announce support for what we call the Corvallis Microfluidics, or CoreMic, tech hub, led by the Oregon State University. CoreMic focuses on microfluidics technology, which helps cool down semiconductor temperatures and improve energy efficiency.
[1:41:50] As I understand it, this high-powered background computing that enables AI uses an increasing amount of energy and heating in computing. This poses a significant barrier to expanding our AI capabilities, particularly while considering our climate goals. So my question, how do your agencies approach leveraging other federal programs like the tech hubs or tackle some of the challenges facing AI research such as the enormous energy requirements?
[1:42:20] So from the NSF perspective, I mentioned in my testimony our recent launch of the inaugural regional innovation engines. And that's being run out of our new technology innovation and partnerships directorate, which was authorized in the chips and science legislation.
[1:42:34] Called the TIP Directorate. And the TIP Directorate has been working very closely with EDA as we stand up the regional innovation engines and the regional tech hubs so that we can create synergies between these investments. Regional innovation engines focused on the earlier stage research and kind of seeding that innovation environment, beginning to establish the jobs and train the workforce with the EDA tech hubs coming in and really building that infrastructure at scale.
[1:43:03] And I'll also mention on the environmental sustainability side of things.
[1:43:12] Energy efficient computing from the hardware to the algorithms is an important area of research and one that NSF is supporting. And we've supported investigators that have developed algorithms, methods for training machine learning systems that are more efficient and also as they explore how to use machine learning to make computing itself more efficient. Thank you.
[1:43:38] The Department of Energy has a long history of investing in microelectronics, in the design of new hardware, actually working very closely with industry, different vendors.
[1:43:48] Setting always higher the North Star. What is the next big challenge that we need to work together? So this is part of DOE's culture and we're bringing that with every new supercomputer that we are asked to design and deploy. There is a lot more to be done because as Tess said and you described, we're certainly at a point where
[1:44:16] There is so much that we can get out of the existing hardware. This is a tipping point for science, and we all need to lean in. At the same time, I agree that we need to be looking at the whole ecosystem of how energy consumption is driven not only by the hardware and the data center, but also by the programming languages, the software, and
[1:44:45] The behavior of the users. This is something that the Oak Ridge National Laboratory is looking very closely. Actually, we are the first ones to design a digital twin of our data center that includes both the building, the machine, and tries to understand what behaviors and how we can improve energy efficiency not only based on hardware that...
[1:45:09] It will take some time to get us to the next big advancement, but also in terms of the software, hardware applications, and user behavior. So these are some areas of research that are very prime for AI. Clearly, artificial intelligence can play a place already, an important role in accelerating that. Bottom line, DOE has an ecosystem that is driven by energy efficiency for decades.
[1:45:38] Thank you both. I yield back. The chair recognizes Mr. Baird for five minutes. Thank you, Mr. Chairman, and I appreciate the witnesses being here. I always learn something in this committee from having you testify. So my first question goes to Mr. Clark. You know, you identified money and cost as a key challenge for the AI systems and mentioned that pay the cost of every search.
[1:46:10] That you run through your system. So given your organization's experience scaling at large AI platform and the need to save money, what recommendations do you have for the NAR program to ensure operating costs are kept at a reasonable level and make sure that there's an efficient allocation of research resources? Mr. Clark. Thank you for your question. I was temporarily distracted by the sound.
[1:46:42] The important thing is to have a small number of kind of large-scale projects which use a large number of VNIR resources, and you closely instrument and analyze these projects, and you make your incentive be make them as efficient as possible. And then you use the tools and science that you advance there to make all projects on VNIR more efficient. This is how we do it at the company, and it can be scaled to VNIR. Anyone else have any comments? If not, I have another question.
[1:47:15] Anyone else have comments on that question about the cost and how we pay for it? Okay. So, Dr. Terasti, my question. This past November, we reauthorized the National Quantum Initiative with a particular focus on the development of quantum computing and communications. Moreover, we amended the initiative to study the feasibility of a national quantum corridor.
[1:47:44] That would connect federal laboratories, institutions of higher education, and private sector stakeholders in order to rapidly transmit large quantities of information. Originally, the Department of Energy recommended that the Oak Ridge Summit supercomputer be slated for decommissioning at the end of 2023. But now the supercomputer will run through October of 24 and assist with NARES pilot program.
[1:48:13] So could the Oak Ridge Summit supercomputer utilize quantum technology to drive AI machines learning more efficiently? Dr. Torres, that's a loaded question. It's a loaded question, yes. Often when we talk about these technologies, high-performance computing, quantum and AI, we use them interchangeably. High-performance computing and quantum computing are two distinct forms of...
[1:48:43] One is very mature and the other in its infancy.
[1:48:48] To your question, the existing supercomputers, Summit or Frontier, they are already used to design new quantum materials that are extremely important for the deployment of quantum computers. They are used to simulate and design quantum processors that are extremely important for the evolution of quantum computing. So certainly there is a synergy between these technologies. And I would say high performance computing
[1:49:17] is a building block to advance both AI and quantum computing. Whether it is the summit supercomputer or the frontier supercomputer, this is secondary. Leadership computing is necessary for these advances. And we need these advances as we are looking into the 10-year horizon and the 20-year horizon about new forms of computing that could deliver on the promise of energy efficiency and help us with certain
[1:49:46] Classes of problems that are suited to quantum computing. As an intermediate step, the integration of high-performance computing with quantum computing, that's the next step we need to take. Anyone else have any comments in that regard? Yes, sir. I just have an addendum to the previous answer. You can also measure the success of Veneer by the savings it creates. So through using Veneer, scientists may not have to spend money on resources they would have had to buy in.
[1:50:19] So thank you. And with that, I yield back, Mr. Chairman. The chair recognizes Ms. Foushee for five minutes. Thank you, Mr. Chairman, and to the chairs and ranking members for holding this hearing today, and to the witnesses for your valuable testimonies.
[1:50:44] Mr. Clark, in your testimony, you noted the importance of, and I quote, federal initiatives and funding that promote equitable access to the critical inputs to generative AI models. How does Anthropic's corporate governance structure reflect equitable access or diversity, equity, and inclusion? We are incorporated as a public benefit corporation, and what that means is that we...
[1:51:12] We operate as a private sector company, and we make money, and we also have the ability to have a duty to societal benefit alongside a fiduciary benefit. So as we grow more successful, we hope to use that to do more things like our participation in Binair Pilot and donation of in-kind resources and third-party research access programs. Are any of your initial trustees African American? We are participating in...
[1:51:42] 10 initial NAR pilots and I would have to get back to you or your staff about the composition of people in those pilots. Okay, thank you. And how is Anthropix currently, how are you currently working to promote equitable access to AI systems? Are you currently involved in any partnerships or initiatives with HBCUs or MSIs or community and tribal colleges?
[1:52:09] We are actually in the early stages of conversations about access partnerships with a couple of HBCUs. I'll be able to get back to you about the specific names. And we have a job we're hiring for, for a program manager to scale up those initiatives and administer them.
[1:52:26] And finally, you said that we should research techniques to strengthen oversight and to mitigate harm. Should federal agencies like NSF and NIST help research and develop technical capabilities that safeguard or detect copyrighted works from being used or trained on without consent in generative AI models?
[1:52:53] In short, yes. And just a brief expansion is we believe that testing and measurement is how you make policy go from something we talk about to something that you can enforce. And so we should be developing tests and measures for issues like those you mentioned, as well as discrimination, misuse, and benefits as well.
[1:53:13] Thank you. Ms. DeBlanc knows, does NSF have any projects or efforts to develop tools or technical capabilities that safeguard or detect copyrighted works from being used or trained on without consent in foundation models?
[1:53:31] So NSF has a critical role to play in this space in terms of supporting the cutting edge research to develop these type of techniques. And right now, NSF funded researchers are working on ways to annotate documents, to provide documentation of proof of copyright, ownership.
[1:53:46] And content authentication. At the same time, NSF-funded researchers are developing mechanisms for detecting whether text comes from a large language model itself. For decades, our secure and trustworthy cyberspace program has supported research in many methods for keeping data safe and secure while detecting modifications. And Dr. Taurasi, how about at Oak Ridge National Labs?
[1:54:18] So first of all, the Department of Energy, part of its mission is national security. So clearly technologies that can protect us from these adversaries and these attacks are part of the research portfolio of the laboratories. The same is true with the Department of Energy, with the Oak Ridge National Lab, and there are ongoing activities in these domains. I come from the open science.
[1:54:47] I can speak to how we protect scientific data and how we protect the leadership computing centers that are dedicated to open science. And certainly, this is something that we manage, as I mentioned earlier, with a combination of technical solutions that involve R&D, but also policy and administrative controls that are absolutely necessary. Thank you, Mr. Chairman. I yield back. The chair recognizes Mr. Keene for five minutes. Thank you.
[1:55:21] Mr. Chairman, and thank you to all of our witnesses for joining us today. I am hopeful that today's hearing will guide legislation and AI towards a brighter future with responsible innovation, protection of children, and paving the way for the next generation. I'd like to start out by saying I am all for advancing AI because of the vast and great potential it has to benefit American society.
[1:55:51] However, we must have a balanced approach that encourages innovation without excessive regulation while prioritizing necessary safeguards to protect the public at large, and most importantly, our children. Just last year, in my district, a high school was involved in a disturbing scandal where students were caught using artificial intelligence to create derogatory images of their fellow classmates. As a father,
[1:56:21] I am deeply disturbed by this disgusting act. The ease with which these students could alter images of their classmates at the push of a button is unimaginable. Schools are meant to be a place of learning. This is not normal, and it cannot become the norm for any student at any school. The absence of proper AI labeling
[1:56:53] Without objection. Thank you, Mr. Chairman.
[1:57:24] At a time when generative AI systems are increasingly capable of fabricating high quality media, the visible and transparent labeling of AI generated content offers one potential safeguard against deception and confusion. Additionally, they highlight the emphasis on labeling AI generated content should be considered in light of academic work demonstrating the utility of warning labels in addressing misleading and deceptive content.
[1:57:54] This has led me to introduce HR 6466, the AI Labeling Act in the House of Representatives. It's a common sense approach for full transparency when interacting with AI. My bill would require developers generated of AI incorporate a prominently displayed disclosure to clearly identify content generated by AI and establish a working group of government, AI developers,
[1:58:24] This creates opportunities for both good and bad actors alike. Advocating for proper guardrails and transparency to foster a sense of responsibility among AI companies.
[1:58:53] And individuals utilizing this technology. Mr. Blank, Nolus, how is the National Science Foundation actively supporting or funding research initiatives that aim to develop effective and standardized AI labeling or watermarking systems to ensure transparency for the public when the content is generated by AI algorithms? In this scenario, as with many
[1:59:22] Questions as we move forward towards a more responsible AI future. There are still open issues that the research community is poised to help address in terms of developing that.
[1:59:32] Thank you. Thank you, Mr. Clark.
[2:00:00] As AI technologies continue to advance, what voluntary industry-led initiatives and research do you find are most effective in promoting ethical AI practices, privacy protection, and transparency without imposing overly restrictive regulations that could stifle innovation?
[2:00:19] We're very supportive of coming to agreement on norms for labeling AI-produced content and also developing systems for identifying AI content that is in the wild and can then be attributed back to the provider. We focus on text, but agree this issue covers all modalities. Thank you, Mr. Chairman. I yield back. The chair recognized Ms. Sykes for five minutes.
[2:00:42] Thank you, Mr. Chairman and ranking member for convening this joint subcommittee hearing today. And to our witnesses, thank you for your testimony and being with us. In 2022, the National Science Foundation awarded a $2 million grant to fund graduate-level traineeship program for University of Akron students studying mechanical engineering, computer science, and polymer sciences. These sort of cross-sector partnerships are crucial for ensuring access to critical resources necessary to power both responsible and innovative AI discovery.
[2:01:10] My question is for Dr. Abdallah and Dr. Shamra. You're both at universities, and you both know the power and the need of graduate students and how their studies and their postdoc scholarship produces much of our innovative STEM research. Can you please talk about the education and research tools needed for federal agency programs to best equip our trainees to learn how to incorporate AI into their projects? I completely agree with you that
[2:01:42] Graduate students are essential for the innovation and technology. And actually NSF and DOE, they have many programs that can fund actually and give full scholarship for a student to go for a PhD program. But schools like Oakland University and others, we always find it challenging access to high computing powers. So as a student, actually get that scholarship from NSF.
[2:02:10] Still, we really need the high computing power in order to generate the research and provide an opportunity for these students to produce quality program and learn and actually get the outcome they need. R1 schools and other universities, they have the resources, and of course, Dr. Abdullah talked about Georgia Tech with over a billion dollars in research. That provides an enormous resource to spend on these. Small schools don't have that.
[2:02:40] So I really think a federal agency democratizing access to high computing power could lead to a lot of advancement for graduate students and others. Thank you. Thank you. So I would agree with that, that providing access to compute power and other resources across the board, even at a university as well resource as Georgia Tech.
[2:03:09] Or many of the R1s. We do run into these absolute limitations. Going back to the question about what programs or what exactly we can do to help produce or educate more students in this space. I think in talking to some industry colleagues, and maybe Mr. Clark could add to that, is that a lot of times what they're interested in is not the specific
[2:03:37] Training in a particular area, but the fundamental knowledge in math, statistics, computing, and so on. So AI is a tool. AI is now the flavor of the year or the decade, and it's been extremely successful. But a lot of the work that goes into building these models, not into the training necessarily or not into the technology, or specifically the applications, have to do with a solid foundation of an undergraduate education.
[2:04:07] And math, statistics, and sciences. Thank you for that. And we talked about specifically graduate and PhD postdoc education. But what about for folks in undergraduate, two-year degrees, maybe even high school? What opportunities do those students have? I can tell you an example. At Oakland University, what we do, actually we offer certificate program and summer program for where we bring actually young kids.
[2:04:35] Thank you. I'm going to pivot a little bit and I appreciate you mentioning that because we talked
[2:05:04] Quite a bit about the missing million in this committee and getting more people of diverse backgrounds, geographical backgrounds, racial backgrounds, because we need it not only to be competitive but for our national security. But I'm from Northeast Ohio. We recently received a Tech Hub designation for polymers and would be interested to know from anyone on the panel, how does AI incorporate itself into the polymer science industry? AI, as I mentioned, is a tool that is being
[2:05:34] Applied across everything, polymers, drugs, materials, hypersonics, almost anything that you can think of that today requires measuring something and then making a decision or a design of something else. The availability of data and availability of large data and being able to discover not just
[2:05:59] The connections between this data, but something that you can project forward, including in the design of, again, polymers or other things, is extremely important. We have a saying in science and computing that says we overestimate the short-term effects of a technology, but we underestimate its long-term effect. This applies very much to AI right now. We're in the hype cycle where it could apply to everything.
[2:06:28] All what we say applies to things. But in the future, it's going to be just like the Internet that started out as a means of communication. It's now our way of doing business and conducting our business. Thank you. Thank you very much. I think we could use, we probably use that saying in Congress as well. Mr. Chairman, I hear back. The chair recognizes Ms. Lee for five minutes. Thank you, Mr. Chair.
[2:06:59] And thank you to all of our witnesses for sharing your testimony today and for your important work addressing both the potential and risk of AI as it relates to scientific discovery. While harnessing AI to drive scientific discoveries and pursue leading edge research is promising, I'd like to direct attention to foundational and often unrecognized labor that has underlaid this technology for so long.
[2:07:24] The exploitation of both labor and mineral resources in Africa that fuels the AI industry and the tech industry more broadly must be a focal point of the AI regulatory conversation. As the National Science Foundation and the Department of Energy look to stand up major data and computational resources to support US AI research and education, we must ensure federal resources don't fuel labor exploitation.
[2:07:51] Recent news have highlighted troubling concerns around the exploitation of labor and the development of AI programs. Large language models like ChatGPT require human reviewers by workers before they're suitable for use by the general public. These workers often are based in the global south and working for very little pay, are responsible for filtering through disturbing materials to eliminate it from the answers supplied by ChatGPT. Others work for little to no pay.
[2:08:19] To hand label thousands of images that go into the data sets training AI systems. These are serious moral and ethical dilemmas. It's documented that employees performing these tasks suffer from anxiety, depression, and post-traumatic stress disorder due to constant exposure to horrific content. While we race towards innovation in the digital economy, let's not forget every person at every step along the way to building this future.
[2:08:46] That said, Ms. DeBlanc-Knowles and Dr. Taurasi, do your respective agencies have a policy or a plan to ensure federal resources that support data annotation for artificial intelligence research and education, including those provided through the National AI Research Resource Palette, don't support labor exploitation?
[2:09:10] So NSF vets all research proposals to the agency for responsible and ethical conduct of that research as part of the decision-making process for funding. We also require that all institutions that receive NSF funding certify that they have a plan to provide appropriate training and oversight in the conduct of that research. In terms of the NAR pilot, together with our partners, including NIST,
[2:09:36] We are developing guidelines and expectations for any data sets that will be made available to the research community through that pilot. Right now we are starting with federal agency data sets that are already open to the public.
[2:09:47] One of the benefits we see from the Nair pilot is really bringing together the community to develop this consensus around what are those right policies for data sets, for models, as we move towards this responsible AI future. A lot of these questions are new, and so we think that there's a benefit here to really bring the community together and build consensus. Thank you. Dr. Tarassi, please. In addition to what you heard, I will pivot differently. Data quality.
[2:10:17] When it comes to data curation, which includes labeling, determines the quality and trustworthiness of the AI models. This is one of the guiding principles for the Department of Energy and the laboratories that we believe data curation needs to be done internally. We're talking about scientific data. We touch scientific data, and they need to be curated and occasionally labeled by domain experts, scientific domain experts. The example you gave
[2:10:46] It's more appropriate for the NIR pilot, right? Because we don't know exactly what are the third-party data sets that NIR users may bring in. But as you heard, there is a vetting process. Now, whether there is an explicit policy, there is no such policy right now. But something for us to think about, how we can ensure that data curation as a whole is done in a legal and ethical way.
[2:11:15] Yes, thank you. I'm sorry, just with the remainder of my time. Reports on AI industry exploiting African labor include the traumatizing working conditions of data workers. African countries like Ghana and Nigeria being used as dumping grounds for toxic e-waste and forced evictions and human rights abuses from cobalt and coal tan mining in the Democratic Republic of Congo. Responsible AI begins with ethical business practices, which unfortunately are not happening. The exploitation of black labor is not lost on me, especially, of course, during Black History Month.
[2:11:44] Mr. Jack Clark, while your company may not directly produce goods or extract physical resources, the technologies used by your firm can be considered beneficiaries of sex exploitation. How does Anthropic, a company that claims to develop AI research and products that put safety at the frontier, ensure its AI research and products don't benefit from the exploitation of African labor and natural resources as part of AI safety mission?
[2:12:05] Thank you for your question. We take this really seriously. So there are two main things that we do. On the technology level, we work to make sure that our stuff doesn't generate or minimizes the amount of harmful or troubling content it generates, which humans do need to review. And on the economic side of things, the majority of our longstanding data suppliers pay California minimum wage or above, regardless of where they are in the world. And we're constantly seeking to improve that. On your resource question,
[2:12:34] We are following industry best practices here, and we've also, with the Nair Pilot, put our AI system access towards environmental analysis work. Thank you. Thank you, Mr. Clark, and thank you for allowing me that time. I go back. Mr. Issa is recognized for five minutes. Thank you. The Constitution says Congress shall make no law abridging freedom of speech. When our founders wrote that,
[2:13:08] They didn't say, Congress shall make no law abridging the freedom of speech, and the NSF shall award no grant that abridges speech. But is there any real difference between paying to abridge speech and directly abridging speech? Hearing no response to that, I'll take it as a yes. And for the record, the Supreme Court's already held that
[2:13:40] Congress, in this case, is equal to the executive and the judicial branch, that all three branches are bound by the Constitution as are the states. So the question is, hasn't the NSF already violated its own transparency rule in doing this and doing it, well, hidden from sight that they were in fact abridging free speech? Thank you for being brave.
[2:14:13] I confess that I'm not familiar with the scenario that you're talking about, but I'm happy to take it back to my colleagues and come back to you with... Well, when you openly say you're censoring, what is that? To censor is to take down. To take down is to abridge. Why would that not be abridging free speech?
[2:14:37] So NSF does not censor. We do support research. Well, it's certainly in the words of people addressing your people in a room, and they weren't corrected. So if you sit in a room and, you know, people receiving money say they're censoring, that that's what they're doing. And let's be honest, anytime something does not appear as a result of a system that limits it,
[2:15:05] Or even when you pay for students to learn how to do that. What is that if it's not, in fact, potentially abridging free speech? And for the record, Mr. Chairman, I just left judiciary. I've been going back and forth because we're having the same hearing in judiciary on the same subject. And there the witnesses were more forthcoming that, in fact, you can take down speech.
[2:15:35] Based on very narrow guidelines, those guidelines are nowhere in the work you're doing. As Chairman Jordan said, in no uncertain terms, if you questioned the origin of the COVID
[2:15:51] If you questioned the six feet of separation, if you questioned the effectiveness of the vaccine, if you did any of that, you got taken down during COVID. So there's no question at all. We would not have formed a special select committee on the weaponization by government, if not for organizations like the National Science Foundation, in fact, participating in.
[2:16:15] What we call weaponization, but let's just call it the abridging of free speech is what I'm choosing to do here today. So you have a scientific integrity policy. If words like the study asserted that the broad swaths of the public cannot effectively sort truth from fiction online, that means that, and I'm just a direct question for you,
[2:16:43] You're making a decision on what the truth is and deciding what people get to see as the truth. You're making their decisions for them. Is that correct? So NSF's investments in this space really aim to enhance authentication and trust and information, and we're supporting the study of how information flows. Again, in my position, I have been focused on standing up the National AI Research Resource. Okay, but let's go through the AI. Isn't AI simply automating what human beings do perhaps better?
[2:17:14] Perhaps faster, but ultimately there's nothing in AI that is going to be different than humans other than you're attempting to have artificial intelligence, meaning a machine, do what humans are already doing or would like to do. Is that correct? In certain cases, AI is even more advanced than humans in terms of we see a lot of complementarity between the capabilities that AI brings. We've talked a lot today. Okay, but let me go through. I know you talked a lot, and I apologize. I have only a few seconds left.
[2:17:45] is open to bias. If you produce or encourage or fund the creation of bias, then in fact what you've done is create a generation of people, artificial people, who in fact will
[2:17:59] Determine what's right or wrong, what people learn, particularly rural people, veterans, the elderly, because that's what your studies show you think are vulnerable because they can't figure out, they're too dumb to figure out the truth, so you're figuring out the truth for them. That's what it says, when it says military families, veterans, et cetera, rural communities, you're specifically making up.
[2:18:24] And in the case of AI, that bias often comes from the data sets that the models are trained on, which gets back to this conversation about the importance of high-quality data that has been vetted for use in that model in the training. Vetted meaning censored.
[2:18:50] Vetted in terms of ethical standards. The gentleman's time is expired. I think the gentleman will yield back.
[2:18:56] Ms. Bonamici is recognized for five minutes. Thank you to the chairs and the ranking members, and thank you to the witnesses. We all know that automated systems such as predictive, generative, artificial intelligence present tremendous opportunities to bolster U.S. innovation, reimagine the role of technology in our everyday life. And as the technology and applications of AI systems develop so rapidly, we need a durable and responsible federal roadmap to guide AI research. The problem is I have...
[2:19:24] Five hours of questions in five minutes. So I'm going to start with Ms. DeBlanc-Knowles. The NEAR Task Force published in its final report in January of last year with a recommendation that a single federal agency should serve as the administrative home for NEAR operations. Should these operations be housed in an existing agency, and if so, which one? Or is a new federal agency best suited to oversee NEAR's operations?
[2:19:47] So the Nair Task Force report recommended that the National Science Foundation serve as that administrative home, but that the Nair initiative really be an interagency effort. I mean, you see us taking that model forward in the Nair pilot, where NSF is serving as that operational home, bringing together the interagency to provide access to resources to the research community. That makes sense, and I need to move on. And certainly we've had conversations, and many of my colleagues represent,
[2:20:17] Representative Iza mentioned judiciary. We were just talking about AI in education. So it makes sense that there's a broad view of where this is happening. I'm going to go to our representatives from academia, Drs. Abdullah and Tarma. AI is already affecting all sectors of life. The responsible deployment of AI systems will require equipping researchers, educators, and students with core competencies to navigate the systems. In your testimony, you both discussed the importance of developing ethical guidelines for AI.
[2:20:46] So what are colleges and universities doing to incorporate AI ethics in curriculum, and is there a universal understanding of what ethics means in this context? So I'll start that we have people who are both researching these issues as well as implementing them in classrooms. So we do have guidelines, for example, about
[2:21:09] How you can use some of these tools in coursework or in research, but also we train our students also in terms of data quality and so on. I don't want to interrupt. I just want clarity. Is that the definition of ethics according to your institution then? Well, I mean, usually the ethics or the ethical dimensions have to do with
[2:21:35] The data quality, meaning does it apply across? Has it been tested on, is the data coming from a large pool, or is it being trained on a man's cancer and applied across, for example? So those are some of the questions. Thank you. I want to hear from Dr. Charma. So basically what we do, we offer every semester seminars to all our students about how to use
[2:22:05] ChatGPT in the classroom in an ethical way. But at the same time, we collaborated with the Department of Philosophy that every student in the College of Engineering and Computer Science cannot graduate without taking an ethics course. So this is a requirement for graduation because I really completely agree with you. Not only in AI, the ethical application of engineering, computer science, data science, and anything we do
[2:22:35] Our engineering students must have an ethics course before they graduate. Thank you. I appreciate that very much because there was a mention earlier in the testimony about how it's important to have the math, statistics, science background, but I also submit that philosophy, history, other fields are going to...
[2:22:50] That would be a big part of defining what ethical use of AI is. I want to follow up on Representative Salinas' question about energy consumption. We know that energy consumption is a concern. In fact, the International Energy Agency projected that AI electricity demand is expected to increase by about 10 times in the next two years. So some in the industry, like Ampere Computing in Oregon, in the district I'm honored to represent, are developing processors aimed at reducing energy consumption associated with AI.
[2:23:19] What steps are you taking to make AI hardware more energy efficient and what resources from a fully reauthorized NIR would help your work in this space?
[2:23:37] So NSF is supporting work to make computing more energy efficient from the algorithm side of things in terms of machine training of the machine learning systems, as well as the application of AI to computing systems to make them more energy efficient. There's still work to be done in this space, and we're supporting kind of the research foundations that can get us there. Thank you, Dr.
[2:24:00] Terasi could just briefly, because I'm . So DOE has demonstrated that it is at the bleeding edge of pushing energy efficient technologies. I gave the example about the frontier supercomputer.
[2:24:10] You are absolutely right. There needs to be renewed emphasis on energy efficient computing as a whole and a program that needs to focus on this regardless of what technology we're talking about because this is a fast-moving field. Absolutely. Thank you. And Mr. Chairman, as I yield back, I request unanimous consent to enter into the record the recommendation on the ethics of artificial intelligence.
[2:24:32] Without objection. Thank you. I yield back. The chair recognizes Ms. Eshoo for five minutes.
[2:24:51] Well, thank you, Chairman Williams and Collins and ranking members Bauman and Stevens for holding what I think is really one of the most important hearings that I've attended in some time. I want to add to that that
[2:25:12] Chairman Lucas and Ranking Member Lofgren have my unending gratitude for all the courtesies that they've extended to me, including allowing me to wave onto the committee today. I have very fond
[2:25:28] Memories of this committee because this is where I served in my very first term in Congress. And as I look around the walls here, there are many that are in these portraits that I had the privilege of serving with in the House. And to the witnesses, you've done a superb job today. Really outstanding job. Now, we all know that AI can
[2:25:55] They provide tremendous benefits. They've been spoken of, so I'm not going to repeat all of that. I think in one sentence, AI holds great promise as well as peril. And it's up to us to shape policies that are going to advance what is good.
[2:26:18] And make sure that all Americans are the beneficiaries of that and to address what is in the lane of peril because we take an oath to protect and defend. And that oath must always be upheld. It is worth each witness repeating what it takes.
[2:26:45] In terms of the main ingredients of AI, good data, computing power, and a lot of it, and people. People that know what they're doing. And a whole set of human beings that we are going to need to be trained and educated in this. And that the resources that are needed for this are really huge. And they become larger and larger with...
[2:27:13] with each year. I'm the mother of NAIR, of the legislation, of the CREATE Act, as well as the previous legislation that Congress passed, the National AI Initiative Act of 2020, which really set up, set the table for the bill that is being, having its legislative hearing today.
[2:27:41] Why are we doing this? First, I want to call out Stanford's human-centered institute of AI, because I've worked closely with them on the previous legislation and this. Because it's essential that all sectors in the country have access to this. I thought the chairman asked a very good question a while back in this hearing about how many automobile
[2:28:09] How many companies does the federal government run? Well, the answer is zero and that's appropriate. But it's also appropriate to recognize that we have many sectors in this country and they too need to be a part of this. Not only the public, the private, the non-profit, the academic.
[2:28:35] The medical, I could go on and on. All of these sectors. The philanthropic sector as well. And we need to be in a position where all researchers and innovators across the United States have the tools. And that's really, in essence, what the...
[2:28:57] What the NAIR is. And I'm so pleased to carry the legislation. It is bipartisan. It is bicameral. As well, it should be. So, my question to Ms. Knowles. What specific, I'm going to trim down the length of the question. What specific resources would an applicant receive with the NAIR?
[2:29:27] And what additional resources could be available to researchers through a fully authorized NIR in your view? First, let me thank you for your leadership on this concept of the NIR.
[2:29:40] So in terms of the Nair pilot, researchers will be able to request access to computational resources, and those are both cloud-based computing resources as well as resources on premise like those in our DOE supercomputers. They also will have the ability to request access to advanced data sets, models, both open models and proprietary models, and training resources. And I think one of the important
[2:30:07] Resources that our partners are bringing to the table for the pilot is the time of their own researchers to be able to work with the research community that are provided access to the resources to really maximize the benefit that the research community can get from access to the resources. As we look towards the full scale in there, those resources that will be available will be
[2:30:28] Larger, more diverse, there'll be an integrated data discovery portal, higher levels of computational resources, and then also a really robust user support mechanism that would be built in to the Nair, as well as a community of researchers that can use that platform to collaborate and learn from each other and access training resources.
[2:30:50] Well, my time has certainly expired, Mr. Chairman. Thank you for the courtesy that's been extended to me today. And I ask that the SIIA letter to the committee be placed in the record along with my statement. And again, all of my thanks to you. Without objection. The chairman recognizes Mr. Beyer for five minutes. Mr. Chairman, thank you very much. And thank you, Mr. Chairman and ranking member for allowing me to.
[2:31:19] To wave on. Ms. DeBlank-Knowles, we're very excited about NIR. If she's the mother, I'm the uncle. Now that it's up, what has this done for the demand for data sets under the pilot program? Are you getting more and more people wanting more and more data sets as you build out the NIR? So the NIR launched on January 24th.
[2:31:43] And with that, then our pilot launched on January 24th. With that launch, we did open the first opportunity to apply for resources that are being made available through the pilot. And those are NSF and DOE supercomputing resources. So in a couple of weeks, we'll have a good sense of the level of interest from the community to access those computational resources. And we've scoped that call around the theme of safe, secure, and trustworthy AI as a priority area that we want to support research on by connecting it to those necessary computational resources.
[2:32:12] I can't believe you can't give me a more comprehensive answer after 13 days. Let me move on. I'm going to come back to you in one second. But I want to thank Mr. Clark. When we had our opening kickoff this past year of the AI caucus, you were our guest speaker. Terrific. I think it was the biggest one we've ever had, 150 people. And I want to really thank you for your testimony today for being brave enough to follow up on what my pal Jay Obernolte said.
[2:32:41] To represent the private sector with all that money involved and encouraging us to make sure that there is the democratization, the public sector investment. And thanks for the endorsement of the Create AI Act, too, because with no disrespect to the president's executive order, we also want Congress to act on most of this stuff and build it in.
[2:33:04] But a couple of years ago, the National Science Foundation had a really interesting report on the fact that the federal government had all these amazing databases at Census and IRS and the Bureau of Labor Standards, BLS rather, et cetera, BEA, that didn't talk to each other. Completely different file formats, different coding things, legacies. And so you suggested that we...
[2:33:28] Put together a pilot program to make them talk to each other so the researchers could do that. And we passed the National Security Data Service Act, which set up that pilot program. Can you give us an update on where we are on the National Security Data Act pilot? I don't have deep expertise in the National Security Data Service pilot that's being stood up, but I'm happy to come back to you in writing after consulting with my colleagues. Okay, great. You have short answers today. Let me move on then to one more question.
[2:33:59] And this is for Mr. Clark. You brushed over very quickly the existential problems. And, of course, we tend to look at use cases. We don't want to regulate the math or the computer science. So we look at privacy. We look at bias. We look at job elimination. When do we begin to worry about the existential parts of it? Should we worry? You, more than probably any person in this room, are...
[2:34:30] Deep into this, the science itself on it. We'd love your perspective on that. So at Anthropic, we've implemented something called a responsible scaling policy. What that means is that before our next really large scale, tens of millions model AI system arrives, we've built a load of tests for potential misuses or security issues that model could generate. And the purpose of this is to know if that kind of
[2:34:58] Potential for threat is coming and be able to intervene on it as soon as possible. And I think there's a lesson here which is mostly we're going to get a huge amount of benefits from AI but if there are risks we want to have the tests ready so that we can run them on frontier systems as they come along. And that's going to tell us if there is anything to be concerned about. I don't believe that there is concern today. I do believe we urgently need to invest in the testing and evaluation infrastructure.
[2:35:26] Mr. Clark, you didn't sign the pause letter that Dr. Hinton and many others did. How was that pause supposed to work? We didn't sign the letter partly because we were not quite sure how a pause was meant to work, but also...
[2:35:53] When you're turning into a bend on a road, you don't slam on the brakes. That's very bad. You actually need to accelerate into the turn to get a handle on it. I think what we can choose to do in America is invest aggressively in the future of AI research so that we can study it and understand any potential issues it creates. Pausing or not doing this just buys us surprise, which seems risky. Thank you very much. Mr. Chair, I yield back. The chair recognizes Mr. Mullen for five minutes.
[2:36:22] Thank you, Mr. Chair. I want to start by thanking my good friend and neighboring colleague, Congresswoman Eshoo, for her leadership on many things, but particularly the Create AI Act, of which I am a proud co-sponsor. As we've seen here today, we should continue to take steps to democratize access to the computational power needed to advance AI research, and I strongly support her legislation and her partnership over the years. Thank you all.
[2:36:50] Thank you for joining us today at this hearing. I live in South San Francisco, which is the birthplace of biotechnology. My district overall is home to hundreds of biotech and life sciences companies doing vitally important work, developing therapeutics for thousands of diseases, improving people's lives.
[2:37:09] Given that, I am particularly interested in learning more about the potential for the use of AI in the development of medicines and other types of treatments. I'd like to better understand how NSF is already supporting this work and how we can further advance these.
[2:37:25] Promising areas of study. So Ms. de Blanc-Knowles, you noted in your testimony earlier that NSF's ongoing support for R&D heavy sectors such as biotechnology have been instrumental in their development in the U.S. Could you tell us just a bit more about how NSF is funding biotech R&D that leverages AI and how exactly does the use of AI lead to the development of new medicines?
[2:37:52] So it used to be the case that we funded AI through our computer science directorate. And now in recent years, we've seen AI funding flow out through all of the eight NSF directorates that cover the full frontier of science and engineering domains. In the biotechnology space, our directorate in that space is funding work that's really applying AI to transform that discovery process and accelerate it from the
[2:38:21] Determining of which...
[2:38:24] New drug therapies to test to developing kind of novel proteins design and drug design that can drive new outcomes. And the opportunity of space is really almost endless in terms of biotechnology being such a data rich area that AI can really transform the whole discovery process, accelerate it, and lead to new discoveries.
[2:38:52] Thank you for that. And I'm also proud to represent three community colleges in my district. One thing I hear consistently from many innovative tech and life sciences companies in my district and across the country for that matter is that there aren't enough people in the pipeline to do the jobs of the future. As we've been discussing here, job and career pathways in AI are underdeveloped as compared to other IT jobs such as cybersecurity.
[2:39:21] I completely agree with you that the landscape has to come from high school,
[2:39:50] Two years community college and four year colleges and that's why actually with Auckland University we always do articulation agreement with community colleges that it's seamless to get education at community college and come to Auckland University to finish with four years. But this is why we do stackable certificate. I really think it's important that for the workforce development that we break our curriculum and as four years colleges we need to evolve our curriculum.
[2:40:19] These four years that we have to have for students to graduate and join the workforce, it's outdated. We really need to start taking certificate, and as they advance, they come back, they take another certificate. If they accumulate four, they can graduate with an undergraduate degree, but we really need to collaborate high school, community colleges, and four-year colleges to ensure that the future of workforce development right now.
[2:40:45] Instead of talking about existentialist question, how many jobs are going to be eliminated? We really need to be prepared right now for the future. I appreciate that very much. Speaking about the collaboration and linkages from high school up the ladder. So thank you all for your information today. Mr. Chair, I yield back. Thank you. As we wrap up, I just want to offer the ranking member, Ms. Stevens, make a short closing statement.
[2:41:14] Well, thank you, Mr. Chair, and thank you so much to our incredible committee members as well as just this fantastic panel of witnesses. I really want to commend all of you for taking time out of your busy schedules, your responsibilities to share with us all. And I think of note, we had two members wave on to this committee from, you know, from...
[2:41:40] A-committees is what we call them, Ways and Means and Energy and Commerce. Ms. Eshoo, as Mr. Mullen mentioned, has had just a remarkable 30-year career in the Congress and incredible leadership in the AI space.
[2:41:57] We are on to something here on the science committee with this hearing and what it has commanded from the Congress and those who don't formally sit on this committee. And so with that, we took a lot of notes, looked up a lot of great stuff as you were all talking. We've got our marching orders on what we need to do here and look forward to the continued dialogue with all of you as we move forward. And thank you, Mr. Chair. I also want to just take a minute.
[2:42:28] I come out of the tech industry and spent time looking at infrastructure projects for HPC across Internet 2 networks that are shared. The technology background is orchestration and orchestration from a full stack infrastructure management, lifecycle management in compute.
[2:42:54] What's happening in artificial intelligence is unlike anything else. It's different, yet again, from the dot-com boom in the amount of investment, the speed of development in these tools. It's absolutely extraordinary. My comments about an automobile manufacturing plant are simply that this is going to overwhelm every resource of government.
[2:43:21] The investments that we make in Nair are absolutely essential and exciting. We build all kinds of great tools, like the National Ignition Facility. We send probes to outer space and to the other planets for discovery. James Webb Space Telescope, an extraordinary tool. Nair provides that tool for the research because our science community has to understand
[2:43:51] Artificial intelligence, where it's going, what the pitfalls are. But the Nair, it seems, provides research and access, you know, democratization as it's been described. You know, on the other side, I think Dr. Tarassi and Dr. Abdallah, you mentioned, you know, the tools that are enabling AI development if it's a wide band.
[2:44:19] I just want to try to explain the context for both my enthusiasm, my support, but also my caution that we can't compete even at a government scale with what's going to happen in private industry. We're going to have to partner with them.
[2:44:45] And fortunately, if it's bare metal access to get resources that the government can do its own research, it's going to be complex. And unlike anything else, the government simply cannot afford to compete of the amount of investment that's going on. But these are extraordinary times. I want to thank you for your valuable testimony.
[2:45:09] For the excellent questions and dialogues from my colleagues, the record will remain open for 10 days for additional comments and written questions from members. And finally, this hearing is adjourned. Thank you so much.