The roadmap outlines policy areas the working group believes “merit bipartisan consideration” and summarizes findings from that group’s insight forums held last year with AI leaders from industry, academia and advocacy groups. In addition to boosting AI spending, the report also covers deepfakes, upskilling workers, and fully funding a National AI Research Resource in its priorities.

“We hope this roadmap will stimulate momentum for new and ongoing consideration of bipartisan AI legislation, ensure the United States remains at the forefront of innovation in this technology, and help all Americans benefit from the many opportunities created by AI,” the working group members said in the roadmap.

The Bipartisan Senate AI Working Group is composed of Sens. Mike Rounds, R-S.D., Martin Heinrich, D-N.M., Todd Young, R-Ind., and Senate Majority Leader Chuck Schumer, D-N.Y. The roadmap comes as legislators in both chambers have explored myriad ways to address the risks and potential of the booming technology but haven’t yet passed broad AI legislation. 

Previewing the announcement last week, Schumer said their approach isn’t to develop one comprehensive plan but rather targeted legislation that addresses specific issues. In a press conference Wednesday, Schumer said the working group’s deliberations were never meant to supplant the work of congressional committees.

“We are very, now, hopeful that the bipartisan momentum that we fostered and the recommendations we made will extend into the committees and their process,” Schumer said. “If anything is going to be accomplished, it has to be bipartisan and it’s going to be done by the committees.”

Schumer also said he plans to meet with House Speaker Mike Johnson, R-La., “in the very near future to see how we can make this bipartisan effort bicameral.”

Among the recommendations in the roadmap, the working group encouraged the executive branch and the Senate Appropriations Committee to reach “as soon as possible” the $32 billion in annual spending on non-defense AI innovation that was proposed by the National Security Commission on Artificial Intelligence in its final report. 

That panel, which was made up of people from industry and academia, was tasked with making recommendations to the president and Congress on AI and issued its conclusions in 2021. At the time, their recommended investment would have doubled government research and development spending.

Lawmakers also underscored the need to fund accounts that haven’t reached their full funding potential under the CHIPS and Science Act, “particularly those related to AI.” Among the accounts the lawmakers listed was the National Science Foundation’s Directorate for Technology, Innovation, and Partnerships, which is aimed at boosting U.S. competitiveness in critical and emerging technologies through research.

Additionally, authorizing a full-scale National AI Research Resource was included as a policy priority. The NAIRR, which operates under NSF, is currently in a pilot phase and is providing access to industry and federal tools and data needed for AI research, such as access to supercomputers and generative AI models. Lawmakers and administration officials, however, have stressed the need for legislation to codify and fully fund the resource.

The roadmap was immediately met with praise and criticism Wednesday.

Linda Moore, president and CEO of TechNet, applauded the roadmap’s support for funding, including for the AI Safety Institute and legislation to authorize the NAIRR. TechNet, a network of technology CEOs and senior executives, worked to advocate for the passage of the NAIRR legislation with the bill’s House sponsor, Rep. Anna Eshoo, D-Calif., Moore said in a prepared statement. 

By providing funding for those initiatives and others, “Congress will empower a new generation of AI leaders, expand innovation and opportunity beyond Silicon Valley, and keep America at the forefront of scientific development for generations to come,” Moore said.

Meanwhile, Nicole Gill, co-founder and executive director of Accountable Tech, called the roadmap “another proof point of Big Tech’s profound and pervasive power to shape the policymaking process.” Accountable Tech is an organization focused on reining in Big Tech.

Gill called the insight forums a “dream scenario for the tech industry” and alleged that companies “played an outsized role in developing this roadmap and delaying legislation.” She also said the roadmap “is most concrete in offering a roadmap for industry priorities while merely hand-waving toward many of the most pressing challenges associated with the widespread adoption of AI.”

This story was updated May 15, 2024, with comments from Schumer’s press conference Wednesday.

The post Government AI funding among priorities in Senate working group roadmap appeared first on FedScoop.

The projects range from research into language model safety and synthetic data generation for privacy, to developing a model for aquatic sciences and using AI for identifying agricultural pests, according to a release from the NSF. Of those projects, 27 will be supported on NSF-funded advanced computing systems and eight projects will have access to those supported by DOE, including the Summit supercomputer at Oak Ridge National Laboratory.

“You will see among these 35 projects’ unbelievable span in terms of geography, in terms of ideas, core ideas, as well as application interests,” NSF Director Sethuraman Panchanathan said at a White House event. 

The NAIRR, which launched earlier this year in pilot form as part of President Joe Biden’s executive order on AI, is aimed at providing researchers with the resources needed to carry out their work on AI by providing access to advanced computing, data, software, and AI models.

The pilot is composed of contributions from multiple federal agencies and private sector partners, including Microsoft, Amazon Web Services, NVIDIA, Intel, and IBM. Those contributions include access to supercomputers; datasets from NASA and the National Oceanic and Atmospheric Administration; and access to models from OpenAI, Anthropic, and Meta.

In addition to the project awards, NSF also announced the NAIRR pilot has opened the next opportunity to apply for access to research resources, including cloud computing platforms and access to foundation models, according to the release. That includes resources from nongovernmental partners and NSF-supported platforms.

Panchanathan described the appetite for the resource as “pretty strong,” noting that 50 projects have been reviewed as positive. But he said there aren’t yet resources to scale those 50 projects. “There is so much need, and so we need more resources to be brought to the table,” Panchanathan said.

While the pilot continues, there are also bipartisan efforts in Congress to codify and fully fund a full-scale NAIRR. Panchanathan and Office of Science and Technology Policy Director Arati Prabhakar underscored the need for that legislation Monday.

“Fully establishing NAIRR is going to take significant funding, and we’re happy to see that Congress has initiated action,” Prabhaker said, adding that the White House is hopeful “that full funding will be achieved.”

The post NSF, Energy announce first 35 projects to access National AI Research Resource pilot appeared first on FedScoop.

The President’s Council of Advisors on Science and Technology voted unanimously in favor of the report Tuesday following a nearly hourlong public discussion of its contents and recommendations. The delivery of PCAST’s report will fulfill a requirement in Biden’s executive order on AI, which called for an exploration of the technology’s potential role in “research aimed at tackling major societal and global challenges.”

“Empowerment of human scientists” was the first goal presented by PCAST members, with a particular focus on how AI assistants should play a complementary role to human scientists, rather than replacing them altogether. The ability of AI tools to process “huge streams of data” should free up scientists “to focus on high-level directions,” the report argued, with a network of AI assistants deployed to take on “large, interdisciplinary, and/or decentralized projects.”

AI collaborations on basic and applied research should be supported across federal agencies, national laboratories, industry and academia, the report recommends. Laura H. Greene, a Florida State University physics professor and chief scientist at the National High Magnetic Field Laboratory, cited the National Science Foundation’s Materials Innovation Platforms as an example of AI-centered “data-sharing infrastructures” and “community building” that PCAST members envision. 

“We can see future projects that will include collaborators to develop next-generation quantum computing qubits, wholesale modeling, whole Earth foundation models” and an overall “handle on high-quality broad ranges of scientific databases across many disciplines,” Greene said.

The group also recommended that “innovative approaches” be explored on how AI assistance can be integrated into scientific workflows. Funding agencies should keep AI in mind when designing and organizing scientific projects, the report said.

The second set of recommendations from PCAST centered on the responsible and transparent use of AI, with those principles employed in all stages of the scientific research process. Funding agencies “should require responsible AI use plans from researchers that would assess potential AI-related risks,” the report states, matching the principles called out in the White House’s AI Bill of Rights and the National Institute of Standards and Technology’s AI Risk Management Framework.

Eric Horvitz, chief scientific officer at Microsoft, said PCAST’s emphasis on responsible AI use means putting forward “our best efforts to making sure these tools are used in the best ways possible and keeping an eye on possible downsides, whether the models are open source or not open source models. … We’re very optimistic about the wondrous, good things we can expect, but we have to sort of make sure we keep an eye on the rough edges.”

The potential for identifying those “rough edges” rests at least partially in the group’s third recommendation of having shared and open resources. PCAST makes its case in the report for an expansion of existing efforts to “broadly and equitably share basic AI resources.” There should be more secure access granted to federal datasets to aid critical research needs, the report noted, with the requisite protections and guardrails in place.

PCAST members included a specific callout for an expansion of NSF’s National Secure Data Service Demonstration project and the Census Bureau’s Federal Statistical Research Data Centers. The National Artificial Intelligence Research Resource should also be “fully funded,” given its potential as a “stepping-stone for even more ambitious ‘moonshot’ programs,” the report said.

AI-related work from the scientists who make up PCAST won’t stop after the report is edited and posted online next week. Bill Press, a computer science and integrative biology professor at the University of Texas at Austin, said it’s especially important now in this early developmental stage for scientists to test AI systems and learn to use them responsibly. 

“We’re dealing with tools that, at least right now, are ethically neutral,” Press said. “They’re not necessarily biased in the wrong direction. And so you can ask them to check these things. And unlike human people who write code, these tools don’t have pride of ownership. They’re just as happy to try to reveal biases that might have incurred as they are to create them. And that’s where the scientists are going to have to learn to use them properly.”

The post Scientists must be empowered — not replaced — by AI, report to White House argues appeared first on FedScoop.

The policy, released Thursday, corresponds with President Joe Biden’s October executive order on AI and establishes a framework for federal agency use and management of the technology. Among the requirements, agencies must now vet their AI uses for risks, expand what they share in their annual AI use case inventories, and select a chief AI officer.

While the final version largely tracks with the draft version that OMB published for public comment in November, there were some notable changes. Here are six of the most interesting alterations and additions to the policy: 

1. Added compliance time: The new policy changes the deadline for agencies to be in compliance with risk management practices from Aug. 1 to Dec. 1, giving agencies four more months than the draft version. The requirement states that agencies must implement risk management practices or stop using safety- or rights-impacting AI tools until the agency is in compliance. 

In a document published Thursday responding to comments on the draft policy, OMB said it received feedback that the August deadline was “too aggressive” and that timeline didn’t account for action OMB is expected to take later this year on AI acquisition. 

2. Sharing code, data: The final memo adds an entirely new section requiring agencies to share custom-developed AI code model information on an ongoing basis. Agencies must “release and maintain that code as open source software on a public repository” under the memo, unless sharing it would pose certain risks or it’s restricted by law, regulation, or contract.

Additionally, the memo states that agencies must share and release data used to test AI if it’s considered a “data asset” under the Open, Public, Electronic and Necessary (OPEN) Government Data Act, a federal law that requires such information to be published in a machine-readable format.

Agencies are required to share whatever information possible, even if a portion of the information can’t be released publicly. The policy further states that agencies should, where they’re able, share resources that can’t be released without restrictions through federally operated means that allow controlled access, like the National AI Research Resource (NAIRR).

3. AI Talent Lead: The policy also states agencies should designate an “AI Talent Lead,” which didn’t appear in the draft. That official, “for at least the duration of the AI Talent Task Force, will be accountable for reporting to agency leadership, tracking AI hiring across the agency, and providing data to [the Office of Personnel Management] and OMB on hiring needs and progress,” the memo says. 

The task force, which was established under Biden’s AI executive order, will provide that official with “engagement opportunities to enhance their AI hiring practices and to drive impact through collaboration across agencies.” The memo also stipulates that agencies must follow hiring practices in OPM’s forthcoming AI and Tech Hiring Playbook.

Biden’s order placed an emphasis on AI hiring in the federal government, and so far OPM has authorized direct-hire authority for AI roles and outlined incentives for attracting and retaining AI talent. 

4. Aggregate metrics: Agencies and the Department of Defense will both have to “report and release aggregate metrics” for AI uses that aren’t included in their public inventory of use cases under the new memo. The draft version included only the DOD in that requirement, but the version released Thursday added federal agencies.

Those disclosures, which will be annual, will provide information about how many of the uses are rights- and safety-impacting and their compliance with the standards for those kinds of uses outlined in the memo. 

The use case inventories, which were established by a Trump-era executive order and later enshrined into federal statute, have so far lacked consistency across agencies. The memo and corresponding draft guidance for the 2024 inventories seeks to enhance and expand those reporting requirements.

5. Safety, rights determinations: The memo also added a new requirement that agencies have to validate the determinations and waivers that CAIOs make on safety- and rights-impacting use cases, and publish a summary of those decisions on an annual basis. 

Under the policy, CAIOs can determine that an AI application presumed to be safety- or rights-impacting — which includes a wide array of uses such as election security and conducting biometric identification — doesn’t match the memo’s definitions for what should be considered safety- or rights-impacting. CAIOs may also waive certain requirements for those uses.

While the draft stipulated that agencies should report lists of rights- and safety-impacting uses to OMB, the final memo instead requires the annual validation of those determinations and waivers and public summaries.

In its response to comments, OMB said it made the update to address concerns from some commenters that CAIOs “would hold too much discretion to waive the applicability of risk management requirements to particular AI uses cases.” 

6. Procurement considerations: Three procurement recommendations related to test data, biometric identification, and sustainability were also added to the final memo. 

On testing data, OMB recommends agencies ensure developers and vendors aren’t using test data that an agency might employ to evaluate an AI system to train that system. For biometrics, the memo also encourages agencies to assess risks and request documentation on accuracy when procuring AI systems that use identifiers such as faces and fingerprints. 

And finally on sustainability, the memo includes a recommendation that agencies consider the environmental impact of “computationally intensive” AI systems. “This should include considering the carbon emissions and resource consumption from supporting data centers,” the memo said. That addition was a response to commenters who wanted the memo to expand risk assessment requirements to include environmental considerations, according to OMB.

The post AI talent role, releasing code, deadline extension among additions in OMB memo appeared first on FedScoop.

The roughly $3 billion requested for AI investments largely reflects the priorities in Biden’s October executive order on the budding technology, which outlined a path forward to harness AI’s power while also creating standards for responsible use. The request would direct some of the biggest sums to agencies like the National Science Foundation, the Department of Energy and the Department of Commerce.

In total, the Biden administration requested $75.1 billion for IT spending across civilian agencies in fiscal 2025, a small uptick from the $74.4 billion it asked for in 2024.

The president’s budget comes a week after Congress avoided a shutdown by passing a package of six appropriations bills for the current fiscal year. Notably, those bills included cuts for agencies like NSF and Commerce’s National Institute of Standards and Technology, which were both given key tasks under Biden’s AI order.

Here are five AI-related takeaways from the request:

1: Research at NSF

The budget includes more than $2 billion in funding for NSF’s research and development in AI and other emerging technology areas, including “advanced manufacturing, advanced wireless, biotechnologies, microelectronics and semiconductors, and quantum information science.” It also includes $30 million to fund a second year of the pilot for the National AI Research Resource, which is designed to improve access to resources needed to conduct AI research. The pilot, which began in January, was required under Biden’s order and bipartisan, bicameral legislation pending in Congress seeks to authorize the full-scale NAIRR.

2: AI cybersecurity at DOE

The budget also includes “$455 million to extend the frontiers of AI for science and technology and to increase AI’s safety, security, and resilience” at DOE. The funding would support efforts “to build foundation models for energy security, national security, and climate resilience as well as tools to evaluate AI capabilities to generate outputs that may represent nuclear, nonproliferation, biological, chemical, critical-infrastructure, and energy security threats or hazards,” according to the document. It would also support the training of researchers.

3: AI guardrails at Commerce

The budget seeks $65 million for Commerce “to safeguard, regulate, and promote AI, including protecting the American public against its societal risks.” Specifically, that funding would support the agency’s work under the AI executive order, such as NIST’s efforts to establish an AI Safety Institute. The recently passed fiscal year 2024 appropriations from Congress included up to $10 million to establish that institute.

4: AI talent surge

The request also seeks funding for the U.S. Digital Service, General Services Administration and Office of Personnel Management “to support the National AI Talent Surge across the Federal Government.” The budget estimated that funding to be $32 million, while the analytical perspectives released by the White House put it at $40 million. Those talent surge efforts were outlined in Biden’s executive order and have so far included establishing a task force to accelerate AI hiring, authorizing direct-hire authority for AI positions, and outlining incentives to maintain and attract AI talent in the federal government. 

5: Supporting chief AI officers

Finally, Biden’s request also provides funding for agencies to establish chief AI officers (CAIOs). According to an analytical perspectives document released by the White House, those investments would total $70 million. Agencies are required to designate a CAIO to promote the use of AI and manage its risks under Biden’s executive order. So far, many of those designees have been agency chief data, technology or information officials. Specifically, the budget mentioned support for CAIOs at the Departments of Treasury and Agriculture, in addition to funding a new AI policy office at the Department of Labor that would be led by its CAIO.

The post From research to talent: Five AI takeaways from Biden’s budget appeared first on FedScoop.

“I think the prospects for the legislation are, I would say, very good to excellent,” Rep. Anna Eshoo, D-Calif., lead sponsor of the legislation, told FedScoop on Tuesday outside a hearing exploring federal science agencies’ use of AI for research. “We want to get this done this year.”

Eshoo, isn’t currently a member of the House Committee on Science, Space and Technology, but was able to participate Tuesday in a joint hearing of its energy and research and technology subcommittees. Eshoo said she anticipates a markup of the legislation in “hopefully in March.”

Similarly, Rep. Jay Obernolte, R-Calif., who is a co-sponsor of the House legislation, told FedScoop outside the same hearing that the legislation is a priority. 

“When I look at the landscape of potential AI legislation that should pass this year, I think the CREATE AI Act is right at the top of that list, and so I’m cautiously optimistic that we’ll see some traction,” Obernolte said.

A spokesperson for the full committee didn’t immediately respond to a request for comment on timing for a markup.

While the National Science Foundation recently launched a pilot for the NAIRR to inform the creation of the full-scale resource, the bipartisan and bicameral bill — called the CREATE AI Act — would enshrine it in federal statute. 

Eshoo said she welcomed the pilot launch and was “eager to see what comes out of it,” but also noted that “the full force of it is through the legislation.” 

The idea behind the NAIRR is to provide researchers with the resources needed to carry out their work on AI, including advanced computing, data, software, and AI models. The pilot, which was a requirement in President Joe Biden’s AI executive order, is supported with contributions from 11 federal agencies and 25 private sector partners. 

Evolution, metrics for success

The NAIRR was a central topic of discussion at the Tuesday hearing, which featured witnesses from NSF, Oak Ridge National Laboratory, Georgia Tech, Oakland University, and Anthropic. Lawmakers’ questions indicated interest in the pilot and capabilities of the full-scale resource. 

Rep. Frank Lucas, R-Okla., chairman of the full committee, for example, probed panelists about how the resource could keep up “with the rapidly evolving industry standards for advanced computational power.”

Jack Clark, co-founder and head of policy at Anthropic, which has been supportive of the legislation, said making sure researchers can do ambitious research will be key. 

“They should not be able to run into a situation where they’re unable to do their research due to running into computational limits,” Clark said. “And how you achieve that in a fiscally responsible way is to make sure that the NAIRR is allocating a portion of its resources for a small number of big-ticket projects each year and adopt a consortium approach for picking what those are.”

Meanwhile, Rep. Scott Franklin, R-Fla., asked about what metrics Congress should be watching for to evaluate success of the pilot as they weigh the estimated $2.6 billion the full resource would require.

In response, Tess deBlanc Knowles, NSF’s special assistant to the director for artificial intelligence, pointed to the number of users the pilot will serve, whether it reaches communities that don’t typically have access to the resources, how many students it can train, and the impact of the resources on projects “in terms of access to computational data resources that they are able to access through the pilot.”

DeBlanc Knowles also noted that experimenting with types of resources and modes of accessing them in the pilot will help the agency “design and scope the plan for the full-scale NAIRR.” 

The post House lawmakers optimistic about NAIRR legislation prospects as pilot moves forward appeared first on FedScoop.

The pilot for the resource, referred to as the NAIRR, is composed of contributions from 11 federal agencies and 25 private sector partners, including Microsoft, Amazon Web Services, NVIDIA, Intel, and IBM. Those contributions range from use of the Department of Energy’s Summit supercomputer to datasets from NASA and the National Oceanic and Atmospheric Administration to access for models from OpenAI, Anthropic, and Meta.

“A National AI Research Resource, simply put, has the potential to change the trajectory of our country’s approach to AI,” NSF Director Sethuraman Panchanathan told reporters on a call ahead of the launch. “It will lead the way for a healthy, trustworthy U.S. AI ecosystem.”

The idea for a NAIRR has been under discussion for some time as a way to provide researchers with the resources needed to carry out their work on AI, including advanced computing, data, software, and AI models. Supporters say a NAIRR is needed because the computational resources that AI demands aren’t often attainable for prospective academic researchers.

Katie Antypas, director of NSF’s Office of Advanced Cyberinfrastructure, underscored that need on the call with reporters, saying “the pilot is the first step to bridging this gap and will provide access to the research and education community across our country — all 50 states and territories.”

The launch comes ahead of a requirement in President Joe Biden’s Oct. 30 AI executive order for NSF to establish a pilot project for the resource within 90 days. According to an NSF release and accompanying call with reporters, the two-year pilot will serve as a “proof of concept” for the full-scale resource. 

Creating a pilot that would run parallel to a full buildout was among the options the NAIRR Task Force, which was co-chaired by NSF and the Office of Science and Technology Policy, presented in its implementation framework for the resource roughly a year ago. 

The pilot is divided into four focus areas: “NAIRR Open,” which will provide access to resources for AI research on the pilot’s portal; “NAIRR Secure,” an AI privacy- and security-focused component co-led by DOE and the National Institutes of Health; “NAIRR Software,” which will facilitate and explore the interoperable use of pilot resources; and “NAIRR Classroom,” which focuses on education, training, user support, and outreach.

Antypas said anticipated uses of the pilot might include a researcher seeking access to large models to investigate validation and verification or an educator from a community college, rural, or minority-serving institution who’s able to obtain AI resources for the students in their classroom.

When asked how resources are being vetted for the NAIRR, Antypas said there will be a process for datasets that become part of the resource. “We are going to be standing up an external ethics advisory committee to be providing independent advice on, you know, what are those standards? How do we develop those with a pilot?” Antypas said.

Quality of datasets came into focus recently after a Stanford report flagged the existence of child sexual abuse material on a popular AI research dataset known as LAION-5B. FedScoop previously reported that NSF doesn’t know if or how many researchers had used that dataset — it doesn’t track this aspect of principal investigators’ work — but highlighted the need for a NAIRR to provide researchers with trusted resources.

Among the support from industry, Microsoft is contributing $20 million in compute credits for its cloud computing platform Azure, in addition to access to its models, and NVIDIA is contributing $30 million in support, including $24 million in computing access on its DGX platform.

Some contributions are tied to specific uses. OpenAI, for example, will contribute “up to $1 million in credits for model access for research related to AI safety, evaluations, and societal impacts, and up to $250,000 in model access and/or ChatGPT accounts to support applied research and coursework at Historically Black Colleges and Universities and Minority Serving Institutions,” according to information provided by NSF. Anthropic, meanwhile, is providing 10 researchers working on climate change-related projects with API access to its Claude model.

The list of partners could grow as time goes on. Tess deBlanc-Knowles, special assistant to the director for AI in the Office of the Director at NSF, noted on the call with reporters that the pilot came together on “a really ambitious timeline” and said “it’s important to note that this is just the beginning.”

deBlanc-Knowles said NSF hopes to bring on more partners and add more resources after the launch “so that we can serve more researchers, educators, and more places, and start to really make progress towards that bigger vision of the NAIRR of democratizing AI.”

The post National Science Foundation rolls out NAIRR pilot with industry, agency support appeared first on FedScoop.

AI models are often trained on datasets that represent billions of samples from the internet — like text, images, and links — that are scraped from the open web. These databases are helpful because they’re extremely large and represent a diverse range of topics, but they can end up collecting illicit imagery that goes undetected. This risk raises critical ethical and legal issues — and also poses a challenge as federal agencies ramp up their efforts to support AI research.

The National Science Foundation, a federal agency that provides funding to scientific researchers, pointed to the need for a National Artificial Intelligence Research Resource in the aftermath of a major Stanford report that highlighted the presence of child sexual abuse material on LAION-5B — an open dataset created by the Large-scale Artificial Intelligence Open Network that’s popular among researchers and some generative AI systems. The report highlighted a major vulnerability with AI research, experts told FedScoop.

(LAION doesn’t have a relationship with NSF, but a researcher affiliated with an NSF-funded AI research institute appears to have collaborated on a paper presenting the dataset.) 

According to the Stanford Internet Observatory, a research center that studies the abuse of the internet, the LAION-5B dataset represents nearly 6 billion samples, which include URLs, descriptions, and other data that might be associated with an image scraped from the internet. A report published in December determined that “having possession of a LAION‐5B dataset populated even in late 2023 implies the possession of thousands of illegal images,” and in particular, child sexual abuse material. While LAION told FedScoop that it used filters before releasing the dataset, it’s since been taken down “in an abundance of caution.”

“We collaborate with universities, researchers and NGOs to improve these filters and are currently working with the Internet Watch Foundation (IWF) to identify and remove content suspected of violating laws. We invite Stanford researchers to join LAION to improve our datasets and to develop efficient filters for detecting harmful content,” a spokesperson added.

Still, the issue highlights the challenge that working with openly available data presents, particularly as the Biden administration pushes federal agencies to develop their own models and conduct more AI research, per the White House’s October 2023 AI executive order. In addition to the immense trauma inflicted on victims, the potential use of child sexual abuse materials in AI systems raises legal concerns, since interacting with the material can constitute a crime. 

The NSF doesn’t have plans to release dataset guidelines right now, but a spokesperson said the issues with LAION-5B exemplified why building alternative resources for AI development is important. The agency spokesperson pointed to the importance of the National Artificial Intelligence Research Resource, which would create new tools for academics to research and develop these kinds of technologies. A roadmap for the NAIRR was released by the White House early last year and the first meeting for the pilot program led by NSF and partners took place last November. 

“This incident demonstrates the critical role for the independent research community in building a trustworthy and accountable AI ecosystem,” an NSF spokesperson said in an email to FedScoop. “It is essential for the research community to be funded to investigate, examine, and build trustworthy AI systems.” 

“When established, the National Artificial Intelligence Research Resource (NAIRR) is envisioned to provide the academic community with access to trustworthy resources and tools that are needed to move towards a more equitable and responsible AI ecosystem,” they added. 

Illicit, dangerous, and disturbing content is frequently included in open web databases. But child sexual abuse materials, often referred to as CSAM, presents a specific challenge: searching for or looking at the material within a dataset, even to remove it, raises obvious legal complexities. Along with government investigators, only certain organizations are legally allowed to study the presence of CSAM on the internet. 

Researchers use creative technical tools, such as a form of hashing, to determine what images are in this system without viewing the images themselves, as the Stanford report did. 

“If you’re pulling all of this material off of the web into a dataset, then you’re also scraping a lot of really undesirable content because it is a reflection of human activity online and humans aren’t always doing great things, right?” said Eryk Salvaggio, an AI researcher who has written about this issue for Tech Policy Press.

“You throw out a net into the ocean, you catch fish, but you’re also going to pick up garbage,” Salvaggio added. “In this case, that garbage is highly traumatic and dangerous material that humans have posted online.”

The long-term impacts of a dataset like this could be wide-ranging. A primer on generative AI published by the Government Accountability Office in May of last year pointed out that AI image generators have now been built using the LAION dataset. In addition to concerns that datasets can include child sexual abuse material, they also introduce the risk that sexual abuse material can end up shaping the output of generative AI.

The risk for federal agencies

The LAION-5B incident raises questions for federal agencies looking to support AI research. NSF, which has a series of AI research initiatives and supports a network of National AI Research Institutes, does not track what datasets are used on the specific projects pursued by its principal investigators. The Department of Energy, which oversees the national lab system, declined to provide a comment. 

The Intelligence Advanced Research Projects Activity “is aware that tainted datasets pose serious risks, which is why the data used in IARPA’s programs undergo review by Institutional Review Boards to ensure that it meets quality, legal, and privacy standards,” a spokesperson for the organization, which is housed in the Office of the Director of National Intelligence, said in an email to FedScoop. 

“ODNI’s Civil Liberties, Privacy, and Transparency Team also routinely reviews and monitors research programs to confirm that they meet these benchmarks. If problematic data were to be identified, IARPA would take immediate steps to remediate it,” the spokesperson added.

The LAION-5B database analyzed by Stanford was sponsored by Hugging Face, a French-American AI company, and Stability AI, which made the image-generator Stable Diffusion, according to an appendix released with the paper. LAION told FedScoop that it did not have a relationship with the NSF. 

Still, there is a real risk that federally funded researchers could or will use tools similar to LAION-5B. Many research institutions have cited LAION-5B in their work, according to Ritwik Gupta, an AI researcher based at the University of California at Berkeley who shared a database of such institutions with FedScoop. 

Notably, a researcher affiliated with the NSF-funded Institute for Foundations of Machine Learning (IFML), an AI Research Institute based at the University of Texas, is listed as an author of the paper announcing the creation of the LAION-5B dataset. A blog post from the IFML about researcher collaborations details work related to LAION and LAION-5B, and the LAION 5-B paper is also listed in the NSF’s public access repository. Neither the University of Texas at Austin, the researcher, or IFML responded to requests for comment. 

Inadvertently including CSAM in foundation models — the Biden administration executive order highlighted the building of foundation models as a priority — can be a risk if proper precautions are not taken, according to David Thiel, the chief technologist at the Stanford Internet Observatory and author of the analysis of LAION-5B. 

“Images need to be sourced appropriately and scanned for known instances of CSAM and [Non-Consensual Intimate Images], as well as using models to detect and categorize potentially explicit material and/or imagery of children,” Thiel said. 

Gupta, the Berkeley researcher, expressed a similar sentiment, saying that “the NSF, and all other government agencies which are funding AI research must thoroughly mandate and enforce the implementation of the [National Institute of Standards and Technology’s] Risk Management Framework into their grants. The NIST RMF is vague when it comes to screening for CSAM, but it largely covers how any dataset curated for AI should be properly acquired.” 

NIST did not respond to a request for comment by the time of publication.

The post Child sexual abuse material found on popular dataset shows risks for federal AI research appeared first on FedScoop.

A pilot for the NAIRR, a resource aimed at improving access to the computational power needed for AI research, is expected to be established in the next three months under Biden’s Oct. 30. order. The NSF on Tuesday held the first in a series of meetings “to engage the broader community in the design of the pilot,” a spokesperson told FedScoop.

The meeting of roughly 100 federal agency, private-sector and nonprofit organizations was focused on the sharing of “insights from experts and prospective providers of computational, data, software, and other resources that will be made available through the NAIRR pilot program,” the spokesperson said.

In an interview with FedScoop, NSF Director Sethuraman Panchanathan said the pilot is an investment in partnerships with relevant industry to be able to make the research infrastructure NAIRR will create “available for a set of partners so that they might then look at the pilot program and learn from it as they are building the full scale in the future.”

Panchanathan said the agency is currently working with the Office and Management Budget and Congress on the budget requirements to ensure they have the resources to complete the pilot and, eventually, the full-scale resource.

The efforts will build upon an implementation framework for the NAIRR that a task force created by Congress and co-chaired by NSF and the Office of Science and Technology Policy produced in January 2023. That framework included a pilot option that could run in parallel to a full buildout of the NAIRR, while expediting the availability of resources.

The executive order also requires Panchanathan, within 45 days, to identify heads of agencies who will submit reports that highlight the agency resources that could be developed and integrated into the NAIRR. 

The spokesperson said there are ongoing efforts to determine federal agency contributions to the pilot, noting a “NAIRR interagency working group of 14 agencies has been working together since the spring, discussing the plan for the pilot and potential contributions from agencies.” 

While the NAIRR requirements are among the most pressing deadlines for NSF under the executive order, the agency has many other responsibilities in the action. Those include establishing a pilot program to train scientists, creating at least four new National AI Research Institutes, and instituting a Research Coordination Network with the Secretary of Energy.

Panchanathan also said the NSF plans to have further meetings in the spring and fall of next year, focused on elements included in the executive order including trustworthy AI and science and security.

The post National Science Foundation convenes first meeting on NAIRR pilot following Biden order appeared first on FedScoop.

The new legislation in the House and Senate would create the National Artificial Intelligence Research Resource (NAIRR), a “national research infrastructure” that would give researchers access to data and tools needed to create trustworthy artificial intelligence. Details of the bill were first reported by Bloomberg Government.

“AI offers incredible possibilities for our country, but access to the high-powered computational tools needed to conduct AI research is limited to only a few large technology companies,” Rep. Anna Eshoo, D-Calif., a lead sponsor of the bill, said in a written statement. 

Establishing the NAIRR would change that by providing “researchers from universities, nonprofits, and government with the powerful tools necessary to develop cutting-edge AI systems that are safe, ethical, transparent, and inclusive,” Eshoo said. 

Advocates for NAIRR say the resource will be crucial to the future of AI research because of the vast computational resources the technology requires. The research field is currently dominated by big tech companies capable of providing such resources.

The legislation comes as Congress and the White House pursue strategies for AI use and regulation. It also follows recommendations the NAIRR task force — which included government bodies, academics, and private organizations — submitted to Biden and lawmakers in January. That task force estimated the budget for the NAIRR if established would be $2.6 billion for the initial six-year period.

The bill, named Creating Resources for Every American To Experiment with Artificial Intelligence Act of 2023, or CREATE AI Act, was introduced in the House by the co-chairs and vice chairs of the Congressional AI Caucus: Eshoo and Reps. Michael McCaul, R-Texas; Don Beyer, D-Va; and Jay Obernolte, R-Calif. Its Senate companion was introduced by Sens. Martin Heinrich, D-N.M.; Todd Young, R-Ind.; Cory Booker, D-N.J.; and Mike Rounds, R-S.D.

Under the legislation, the NAIRR would be overseen by the National Science Foundation and the daily operation would be managed by an outside organization that would be selected through a competitive process. 

That outside organization would be “a nongovernmental organization, which may be an independent legal entity or a consortium of 1 or more partners (which may include federally funded research and development centers),” according to the bill text. 

“To maintain United States leadership over our adversaries in technology, including AI, we must unleash the full potential of American innovation,” McCaul said in a written statement. “The legislation will open up resources to allow more great American minds to work together to develop smart, safe, and secure AI.”

The post Lawmakers introduce bipartisan bill to create a federal AI research resource appeared first on FedScoop.