AI Executive Order Sets Foundation for Regulation, Research, and STEM Visa Reforms
President Joe Biden issued an executive order on Oct. 30 that instructs federal agencies to develop security and safety standards for artificial intelligence applications and places new reporting requirements on certain AI developers.
The order focuses on “dual-use foundation models,” which it defines as powerful general-purpose models that present significant security risks, such as by making it easier for non-experts to acquire weapons of mass destruction. Invoking the Defense Production Act, the order requires developers of such models to keep the government apprised of their work.
Another focus of the order is to make it easier for persons from abroad to come to the U.S. to work and study in fields related to AI and other “critical and emerging technologies,” as defined by the White House. It also takes steps toward establishing a shared computing infrastructure for AI researchers, known as the National AI Research Resource.
The order’s provisions rely on expanded use of existing agency authorities and do not promise additional funding, with more far-reaching changes requiring congressional action.
NIST to play pivotal role
Implementation of the order will rely heavily on the National Institute of Standards and Technology, which has taken on increasing responsibilities in AI policy in recent years.
The order instructs NIST to produce best practices for AI development and use within 270 days, including by expanding its AI Risk Management Framework to include generative AI and establishing guidelines for “red-teaming” tests that can measure the security and trustworthiness of AI. Developers of dual-use foundation models are required to follow these red-teaming guidelines and to share the test results with the government.
In total, the order includes dozens of discrete taskings for NIST’s parent organization, the Commerce Department, some of which are explicitly delegated to NIST. Among these other responsibilities, the department is to require entities that acquire or possess “large-scale computing clusters” to disclose their location and computing power.
Immigration provisions hinge on future agency action
The order broadly calls for streamlining the visa process for “noncitizens who seek to travel to the United States to work on, study, or conduct research in AI or other critical and emerging technologies.” Specific provisions include:
- The Department of Homeland Security and the State Department are directed to streamline processing of visa petitions and applications, including by facilitating sufficient visa appointments
- The State Department is directed to consider expanding the categories of nonimmigrants who are allowed to renew their visas in the U.S. without returning to their home countries. The provision explicitly applies to J-1 research scholars and F-1 students in STEM fields generally, not just those in fields linked to critical and emerging technologies.
- The Labor Department is to put out a request-for-information within 45 days on the possibility of adding AI and other STEM-related occupations to the list of Schedule A occupations, which would make it easier for U.S. companies to hire foreign workers into those positions.
- DHS is directed to “clarify and modernize” the visa processes for O-1A, EB-1, and EB-2 visas for advanced-degree holders and noncitizens of extraordinary ability.
- DHS is to consider changing the H-1B skilled visa program to make it easier for experts in critical and emerging technologies to become lawful permanent residents.
Aside from the immigration provisions, the order also instructs the White House Office of Management and Budget to look into using special hiring authorities to onboard and retain AI experts in federal jobs.
AI research initiatives seek funds from Congress
The order calls for the National Science Foundation to launch a pilot program within 90 days to establish a National AI Research Resource (NAIRR). The resource would provide high-power computing and datasets to AI researchers across the U.S., especially those at less resourced institutions. NSF held the first in a series of planning workshops for the pilot on Nov. 7.
Congress initiated formal planning for NAIRR in 2020 through legislation that created a task force to study potential architectures. The task force ultimately recommended Congress allocate $2.6 billion to implement NAIRR over an initial six-year period.
Bipartisan bills to formally authorize the NAIRR are currently pending in Congress but they do not include any funding. The bills are sponsored by the leaders of the AI caucuses in the House and Senate: Sens. Martin Heinrich (D-NM) and Mike Rounds (R-ND) and Reps. Anna Eschoo (D-CA) and Mike McCaul (R-TX).
The order also includes research-focused taskings for the Department of Energy, which has been positioning itself to take on a lead role in federal AI R&D. The order echoes focus areas that DOE labs proposed in a major report this summer on how the department could help develop ways to apply AI in the domains of scientific research, energy technology development, and national security.
Specifically, the order directs DOE within 180 days to “expand partnerships with industry, academia, other agencies, and international allies and partners to utilize the Department of Energy’s computing capabilities and AI testbeds to build foundation models that support new applications in science and energy, and for national security.” DOE is also required within 270 days to implement a plan for building out its AI testbeds and model evaluation tools, with a focus on guarding against risks posed by AI-enabled development of nuclear, chemical, and biological weapons or threats to energy infrastructure.
In addition, DOE is instructed to establish an office that will coordinate AI development across all the national labs and to set up a pilot program to train 500 AI scientists by 2025. The order does not allocate funds for these efforts, instead expecting DOE to use existing resources.
DOE has not offered a budget estimate for its ambitions in AI, though a lead author of the national lab report, Rick Stevens of Argonne National Lab, has envisioned the initiative could exceed DOE’s multi-billion dollar Exascale Computing Project. Stevens has suggested such an initiative could be funded through special legislation focused on AI.
Senate Majority Leader Chuck Schumer (D-NY) has expressed interest in funding a major federal AI initiative and has been holding a series of closed-door forums to inform potential legislation. Describing one of those forums last month, Schumer said participants advocated for $32 billion as the minimum amount that the federal government should spend to spur AI development.