This week, President Trump signed an executive order titled the “American AI Initiative.” While this order lays out some useful first steps toward a larger national policy and course of action for artificial intelligence (AI), the administration will need to do more to ensure its goal of maintaining American leadership in AI technologies. Although the order’s broad “policies and principles” section includes calls to preserve civil liberties, privacy, and American values, it is not entirely clear what those values are or whether they might conflict with the other priorities listed in the order, such as economic competitiveness. Let’s talk first about what the order does before turning to some things it does not do.
What it does:
In terms of concrete actions, the order lays out several steps for agencies to take. Overall, agencies are instructed to consider AI as a research and development (R&D) priority. This includes assessing which programs might or already do involve AI and which will be prioritized, as well as reporting how much money will be spent on those projects. The order instructs agencies to “budget an amount for AI R&D that is appropriate for this prioritization,” but as we’ll discuss later, does not set out additional funds for AI R&D.
Most of the order’s concrete actions center around improving access to federal data and models. The order calls for agencies to review their federal data and models, and identify opportunities to increase access and use by the non-federal AI research community. Agencies are also instructed to improve their data and model inventory documentation. Improving public access to federal data would give researchers a broader, more diverse pool of data with which to build and train machine learning models. Opening federal data sets for public use also allows for more diversity in the uses for that data, potentially leading to applications that might not have been developed by federal agencies. Finally, more publicly-accessible data helps to democratize the development of new AI and machine learning technologies, which should expand opportunities to develop applications that primarily serve public interests.
Additional aspects of the order’s efforts to improve access to and use of federal data require the Office of Management and Budget (OMB) to collect public comments on how agencies can make more data and models available, as well as what kinds of data are most needed. OMB is also instructed to investigate barriers to access or quality limitations of federal data and models that impede AI R&D and testing. We encourage researchers and other interested parties to submit comments once this proceeding officially opens (in 90 days or so).
After the OMB public comment period, agencies are also instructed to consider methods of improving the quality and usability of, and appropriate access to, priority data identified by the AI research community. This may not result in much actual change, but the comment period will at least establish a public record which can help hold agencies accountable. Agencies must also identify barriers to, or requirements associated with, increased access to and use of federal data and models. These include privacy and civil liberty protection for individuals, safety and security concerns, and the interoperability of data and models. This is no small task, and while CDT is glad that the order acknowledges the importance of addressing these concerns with respect to federal data, it gives agencies little guidance as to how to weigh these concerns against other constraints and objectives. More generally, CDT would like to see greater emphasis on and more attention to these concerns as a top-level element of a national AI policy.
Finally, the order instructs the Secretary of Commerce, through the Director of the National Institute of Standards and Technology (NIST), to “issue a plan for Federal engagement in the development of technical standards and related tools in support of reliable, robust, and trustworthy systems that use AI technologies.” Again, while this doesn’t provide much detail or require much more than creating a plan with limited scope, it does require NIST to at least think about what “reliable, robust, and trustworthy systems” might look like and formally engages the Institute in the larger discussion about AI standards and policies.
What it doesn’t do:
Although the order lays out some lofty goals and takes a few preliminary steps toward them, it also lacks several elements that would have resulted in a more substantial addition to AI policy. The most obvious gap is dedicated funding. Granted, federal spending is ultimately up to Congress, but the order could have gone further than asking agencies to consider AI R&D a “priority.” Perhaps the President’s budget proposal will do more to show the administration’s dedication to AI policy.
The order does, in places, call for the consideration of privacy interests and the protection of civil liberties. However, there are no calls to ensure that AI applications serve the public interest, nor any instructions to agencies to even consider what regulatory guardrails might be integral to ensuring that the American AI Initiative primarily benefits the public. The order also fails to call for fairness in AI, even for AI applications developed and used by federal agencies. CDT respectfully suggests that national AI policy should focus on ensuring that technologies benefit human societies and that the underlying technology should fairly and accurately reflect the values of the people it affects. This should be central to, not eclipsed by, the goal of maintaining American leadership in AI. To that end, the order could have been more explicit about the values it intends to promote.
Relatedly, the order seems to be premised on the idea the best AI strategy is to just make more of it as rapidly as possible. The order does little to acknowledge that AI presents both opportunities and risks (other than the risk of falling behind, internationally). Overall, the American AI Initiative moves in the right direction, and CDT looks forward to this and future administrations continuing to engage with AI policy in a detailed, thorough, and balanced manner.