by Carl Schonander

Interim AI report offers mix of ambitious and practical recommendations

Nov 08, 2019
Artificial IntelligenceGovernmentTechnology Industry

If we want US, rather than Chinese, values to infuse the worldu2019s approach to AI, then high powered diplomacy is an essential tool. A new NSC report breaks down the approach.

The National Security Commission on Artificial Intelligence released an interim report on November 4, 2019. The report offers a blend of deservedly bold recommendations coupled with actions that lie within the power of the Executive without all of them involving major funding.

The recommendations involve investments in AI R&D; applying AI to national security missions; training and recruiting AI talent; protecting and building upon US technology advantages; and, marshaling global AI cooperation. The Commission may be somewhat less optimistic than warranted with respect to the prospects for the West to share data as US and allies’ approaches to preserving privacy are converging.

However, the Commission is spot on with respect to the need for high powered diplomacy. The US has used diplomacy to forge Western approaches to AI through the OECD AI Principles. Chief Technology Officer Michael Kratsios’ November 7, 2019 Lisbon Web Summit speech makes clear the Administration emphasis on working with allies. More of this approach is essential if US, rather than Chinese, values will infuse the world’s approach to AI. 

Investing in AI R&D 

It is true, as the Information Technology & Innovation Foundation posits, that: “Dwindling Federal Support for R&D Is a Recipe for Economic and Strategic Decline.”  In 2017, the US government invested 0.62% of GDP on R&D. In 1962, five years after Sputnik, the federal government invested 2.2% of GDP. Yes, it’s quite likely that China is exaggerating its basic R&D spend. But we know what we spend and the trendline should be going upwards.

How to spend is critical in this and any context. The Commission proposes that the National Science Foundation (NSF) fund a series of AI R&D Institutes. It also proposes an entity within the NSF analogous to the National Cancer Institute. The Commission suggests that the government leverage federal installations around the country to attract talent to AI hubs. It suggests funding mid-career faculty awards for AI researchers to incentivize them to remain in academia.

The report correctly emphasizes the 2019 Executive Order on Maintaining American Leadership in AI, although it notes that net new resources are not afforded per the Order. The Commission correctly identifies deepfakes as concerning and appropriately advocates for more research into digital forensics. There are self-imposed handicaps as well such as funding and staffing restrictions that hamper the government’s ability to maximize the return from DoD’s federally funded R&D centers.

Applying AI to national security missions

Accelerating AI adoption for defense purposes will require policymakers to engage in the hard work of changing acquisition practices. This is hard to do, but it is always worthwhile trying to make processes more efficient. Appendix two in the interim report provides examples of what needs to change. Startups in the AI space cannot devote the sheer amount of time it takes to win a DoD contract. (Note: The Commission does not propose this, but it would be useful for the Congress and DoD to consider new reductions in requirements to permit smaller contracts. Analogizing from the “regulatory sandbox” concept for financial innovation, DoD should be empowered and encouraged to take chances and risks with technologies.)

The Commission notes that it is challenging for DoD to transition to scaled production once a pilot has been implemented. That needs to change as well. DoD needs to find a way to build the infrastructure to share scaled successes.  Finally, publicized examples of how DoD can be a rewarding customer are needed to overcome selection bias against DoD as a potential customer.

Training and recruiting AI talent 

The major factor here remains, as the Commisison puts it: “The American AI talent pool depends heavily on international students and workers. Our global competitiveness hinges on our ability to attract and retain top minds from around the globe.”

Everybody knows this, but it is hard to do anything about it because the issue gets caught up in the broader immigration debate. In the meantime, there are more small bore proposals that could make a meaningful difference. They include mapping skills in the reserves; financial incentives for completing AI coursework; including computational thinking in the Armed Services Vocational Aptitude Test; and, treating coding like proficiency in a foreign language to incentivize DoD skills acquisition.

The Commission also suggests including AI in possible future legislation to facilitate public-private exchanges in cyber security. Finally, but certainly not least, the Commission encourages greater diversity talent acquisition. Industry has certainly not done a good enough job in this respect, but the US military has historically been a pacesetter with respect to, for instance, racial integration so this would seem to be an area where the US national security establishment could conceivably excel and perhaps even lead society at large.

Protecting and building upon US technology advantages 

There is a lot to talk about here, but the big issue is what to do about export controls. It is essential  to ensure that new US export controls on “emerging” and “foundational” technologies are targeted in the sense that they restrict exports of US technologies that truly affect US national security. And that such controls do not affect only US technologies.

It makes no sense to prevent China from acquiring something if it can get substantially the same technology from another country. So controls do have to be multilateral and focus on “choke points” in the technology acquisition process. The Commission suggests that restrictions on hardware, specifically semiconductor manufacturing equipment (SME) might be warranted because the chips such equipment make possible are increasingly needed to deploy high-performing algorithms.

This is consistent with recommendations from Georgetown’s Center for Security and Emerging Technology. Today, 90% of SME production is located in the United States, Japan, and the Netherlands. With diplomacy (more on that below), it should be possible to craft an export control policy with respect to SME that preserves US (and Western) competitive economic advantages, while at the same time addresses national security equities.

Marshaling global AI cooperation and building American diplomatic capability

Some of the publicity around the release of the interim report has centered around the recommendation that the United States share data with allies. This should certainly be encouraged, but the Commission says that “divergent views on data privacy present significant hurdles, including with respect to the European Union’s General Data Protection Regulation.” There are hurdles, but they will probably become less significant with time. The United States will likely adopt a federal privacy law in the next couple of years and there is consensus – including from US industry – that the use of AI requires measures to avoid discrimination.

The United States, the EU and other Western allies will not adopt exactly the same kind of law and regulations with respect to using AI for decisions affecting people. But they will share largely the same goals. The Commission rightfully says: “Strong US-led diplomacy will play a critical role.” The Trump Administration has a creditable record with respect to using multilateral diplomacy in achieving consensus with respect to the OECD AI Principles. The United States should build on this success.

One way of elevating the US game in this respect would be to, as the Commission supports, implement “plans for the State Department to establish a bureau focused on cyberspace security and emerging technologies.” In this context, it is critical that this bureau be tasked with integrating the totality of US interests (strategic, economic, human rights etc.). Ideally therefore, this bureau should report to the Under Secretary for Political or Economic Affairs.

Given the Commission’s support for using science diplomacy “in promoting AI collaboration,” on balance it would probably be best to have this bureau report to the Under Secretary for Economic Affairs because this Under Secretary oversees US science diplomacy.     

The United States will not win the AI race without investing more in basic R&D. But there are a lot of other things that can be done as well that can make a meaningful difference.