Advanced Computing in the Age of AI | Wednesday, June 19, 2024

Feds to Smaller AI Vendors: Help Us Help You on AI 

US Capitol Building, Washington, DC

As AI use and innovation continue to increase globally, getting small and medium-size AI companies more involved in the financially lucrative and still-nascent AI revolution is a developing goal of the US government.

One federal agency, the National Security Commission on Artificial Intelligence (NSCAI), is asking smaller AI vendors to send in comments and suggestions by Oct. 23 about how to make it easier for them to work with the US government to bolster commercial AI innovation.

With the comment deadline looming, the NSCAI says it seeks detailed input from small- and medium-sized AI vendors, particularly when it comes to working together to catalyze AI development, expand the national security innovation base, and make it easier for them to do business with the federal government. The agency says it is specifically seeking recommendations involving needed statutes, regulations, policies, budgets, organizations, and cultures, as well as other related issues.

The NSCAI’s appeal for comments and input from smaller AI vendors was published Sept. 23 in The Federal Register. The responses will be used to identify critical areas for improvement and recommended changes in the government's approach to technology procurement and support for commercial innovation, according to the agency.

Some of the sample questions the agency proposes for smaller AI vendors include:

  • What are the challenges or obstacles you face in seeking to do business with the federal government, to include scaling successful solutions? What changes could be made to reduce or remove those challenges or obstacles?
  • How do you weigh the tradeoffs between accepting financing from US firms versus foreign firms? What role could the US government play in connecting US firms with trusted investors in the United States and allied countries?
  • When is the federal government a compelling customer? When is it not? What steps could the federal government take to become a more compelling customer?
  • How could the government better communicate (1) national security challenges to industry and (2) opportunities for industry to demonstrate and iterate potential solutions? How could the government structure engagements with industry to foster innovative and unexpected solutions?
  • If your firm were to initiate or expand its national security or national interest work, what large capital investments over the next 24 months would your firm consider making in the United States? How much financial support and in what form (e.g., non-dilutive capital, loan guarantees, equity stakes, or other financial instruments) would be required from the U.S. government to undertake those investments?
  • What would you hope to gain from temporary talent exchanges between the federal government and industry? What are the challenges or obstacles in conducting such exchanges and how would you recommend they be overcome?
  • How can industry and the federal government better collaborate through all stages of product development to safeguard against bias in AI systems?
  • How can the federal government incentivize responsible AI development through acquisition?

Two comments have been received so far by the commission, and more are expected to come in before the Oct. 23 deadline, Tara M. Rigler, the director of strategy, communications and engagements for the NSCAI, told Enterprise AI. The input "can provide an opportunity for the commission to hear perspectives from a broad array of firms and illuminating any areas the commission could address in its final report," she said.

"The NSCAI views small- and medium-sized enterprises as a key source of America's strength in developing and applying AI," said Rigler. "To inform our final report, we recently released a public request for comments from small- and medium-sized AI firms regarding how the U.S. government can be a better customer. The commission has also already made a number of recommendations to diversity and democratize access to hardware, software, algorithms, and talent so that small- and medium-sized US AI businesses can thrive. Highlights include a National AI Research Resource to expand access to computing and a national microelectronics laboratory and incubator to assist promising, early-stage microelectronics startups."

Analysts Weigh In

James Kobielus, principal analyst of Franconia Research, told EnterpriseAI that one of the chief barriers for smaller AI vendors who want to do business with the federal government is that their products often can’t scale to the large needs of federal agencies.

“For many AI opportunities in the public and private sector, this fact tends to favor the larger vendors who provide it all,” he said. “One way to reduce barriers for AI SMBs to address such opportunities with federal customers is for the feds to require that the bigger cloud AI vendors, such as AWS, Microsoft, Google, IBM and others, maintain open partner ecosystems and marketplaces. To the extent that the cloud vendors and other providers of comprehensive AI ecosystems fail to maintain open partner ecosystems and marketplaces, federal agencies should disqualify such larger players from bidding on government AI business.”

In addition to resolving those issues, the federal government should also consider cracking the whip and passing laws that require government agencies to only procure AI solutions that align with the thematic principles outlined by the Partnership on AI or other industry and user alliances, said Kobielus.

“Alternately, government procurement agents could be required to only purchase AI solutions that are certified by an impartial group such as Underwriters Laboratories," he said. "These principles and mandates could be made a standard requirement to respond to in any AI-related RFI or RFP from a government agency. However, that may be an onerous requirement for small- to midsized AI vendors, favoring larger vendors who have the resources to gain those certifications.”

Still, with many AI SMBs aware of public calls for tighter regulation on AI, the specter of that will deter some of them from getting too deep into the market and will raise questions those companies will have to answer, Kobielus added. “For instance, has the federal government, in order to incentivize AI SMBs to seek federal contracts for AI apps/systems, considered indemnifying or limiting the losses of these firms from downstream legal risks related to privacy, bias, and other controversial impacts of these technologies?”

Karl Freund, senior analyst for AI, machine learning and HPC for analyst firm Moor Insights & Strategy, said that in terms of AI hardware today, the only small- and medium-sized AI businesses are startups, and some of them are already targeting US Department of Energy (DoE) supercomputing centers and their users. "Providing additional funding to the DoE and Department of Defense targeting startup equipment procurement would help a lot, otherwise the playing field is tilted to the advantage of the big players such as Nvidia, AMD and Intel."

Another analyst, Rob Enderle, principal of Enderle Group, told EnterpriseAI that if the NSCAI’s efforts are executed well, they should increase the possibility of more collaborative outcomes between the government and smaller AI vendors.

But while the effort is sound, seeking comments on developing AI collaboration ground rules just before a presidential election will be problematic in terms of focus and progress, said Enderle.

“A different administration could have a completely different approach, and folks know this, which will make it difficult to gain much traction right now,” said Enderle. “This outcome is a shame because I think an effort like this is much needed.”

The NSCAI was established in 2018 to promote AI, machine learning and related automation technologies to meet U.S. economic and national security needs.

The agency will collect the comments and data from AI vendors through Oct. 23 and will issue a final report in March, 2021.

SMB AI vendors can contribute comments via email, traditional mail or by fax. Commenters must submit their remarks with the tag, Docket No. 09-2020-01, in the subject line. Mailed comments may be sent to the National Security Commission on Artificial Intelligence, Attn: RFI COMMENT—Docket No. 09-2020-01, 2530 Crystal Drive, Box 45, Arlington, VA 22202. Fax submissions, which must include the docket number on the fax cover page, may be sent to +1-571-778-5049.

Editor's Note: This story was updated at 1 p.m. PT to add a late response from the NSCAI.