Published in The Journal of Robotics, Artificial Intelligence & Law (May-June 2020)

To begin 2020, the Office of Management and Budget ("OMB") released a memorandum (the "AI Memo") with the subject "Guidance for Regulation of Artificial Intellgience Applications."  In its introduction, it explains that it "provides guidance to all Federal agencies to inform the development of regulatory and non-regulatory approaches regarding technologies and industrial ssectors that are empowered or enabled by artificial intelligence (AI)1," pursuant to Executive Order 13859, "Maintaining American Leadership in Artificial Intelligence" (the "AI Executive Order")2.  The AI Memo establishes 10 broad principles for federal agencies to use when formulating "regulatory and non-regulatory" approaches to AI (a dichotomy that I'll return to later):

  1. Promote public trust in AI;
  2. Provide ample opportunities for the public to participate in and provide feedback on rulemaking governing AI;
  3. Leveraging scientific and technical information and processes;
  4. Assess risks in subject AI;
  5. Consider the costs and benefits of any AI;
  6. Maintain a flexible approach to adapt to changes and updates to AI applicatioins;
  7. Consider impacts AI may have on fairness and discrimination;
  8. Incorporate disclosure and transparency in the rulemaking process to increase public trust and confidence in AI applications;
  9. Promote AI systems that are safe, secure, and operate as intended; and
  10. Coordinate with other federal agencies on AI strategies.3


The AI Bar has been waiting for the AI Memo since last year's AI Executive Order, which called on the OMB director to issue a memorandum "to the heads of all federal agencies that will inform how they develop regulatory approaches to AI technologies."4 When the White House issued the AI Executive Order, I welcomed the development, noting that "the lack of any concrete federal efforts to implement laws and a regulatory framework governing AI has been a serious problem for several years now."5 I worried that if we want AI that benefits the largest number of people, "we have to introduce regulatory rules into the sector to guide it in that direc¬tion," to set the playing field.6

Within the AI Memo's principles are some values that should be the basis for the AI regulatory framework, and that is very good. But the OMB did not make any effort to institute a federal vision for AI, which means that the AI Memo does not provide the guid¬ance that AI regulation needs to provide.

Relying on the Right Values

The AI Memo represents a good introductory framework for federal agencies to use when beginning to regulate AI. Although I have advocated for more regulatory action that is stronger than an introductory framework,7 it is possible to see how federal agen¬cies would use the values in the AI Memo to produce decisive AI regulations. The AI Memo's call for promoting trustworthy AI applications, relying on open and public consideration (to the extent national security, intellectual property, and other consider¬ations permit this), and the mitigation of bias in any AI to promote fairness and nondiscrimination point agencies toward the sorts of values that promote the rule of law and AI's role in strengthen¬ing it.8 Matt Sanchez, Chief Technology Officer of CognitivScale, commented that the 10 principles "are a good first step to drive more transparency and trust in the AI industry. They address the elements of trust that are critical for safeguarding public data—privacy, explainability, bias, fairness, and compliance."9

Other experts in the field have also noted that the AI Memo is a promising start to regulating AI, although it does not represent genuine government action. Terah Lyons, the executive director of the non-profit group, Partnership on AI, thought that the AI developer community may see approaching AI regulation with values like trustworthiness and fairness as a "positive step in the right direction," but added "It's a little bit hard to see what the actual impact will be."10 Ms. Lyons is not wrong. Unless there is rulemaking or regulatory action the follows the AI Memo, the values it emphasizes to federal agencies will be exemplary, but will not have much tangible impact.

I do not want to discourage federal agencies and policy makers from thinking carefully and fully about AI regulation, but I worry that the current chain of baby steps—AI Executive Order, the AI Memo, the final memorandum that OMB will release after the public comment period, etc.—will give key decision makers the impression that a lot of work has been done to govern AI, when in reality the work of regulating AI has only begun. The private and public sectors are investing heavily in AI research and develop¬ment, and if the federal government continues AI rulemaking at this pace, any eventual AI regulation will be unable to keep pace with AI applications and innovation.

Having said that, the values in the AI Memo hopefully suggest key considerations that AI developers should incorporate into their AI even without formal regulation.

Failure to Mark the AI Sector's Playing Field

The slow march of AI governance is a systemic concern that is not unique to the AI Memo. However, the AI Memo adds to that problem by elevating "non-regulatory oversight" to appear equal to regulatory oversight. In almost every section, the phrase "non-regulatory oversight" is paired with "regulatory oversight." This is due, in large part, to the AI Memo's emphasis on innovation and growth.

I think that emphasis makes sense, to a degree. Innovation and growth are important, but the AI Memo appears to place them above almost all other concerns:

Federal agencies must avoid regulatory or non-regulatory actions needlessly hamper AI innovation and growth. Where permitted by law, when deciding whether and how to regulate in an area that may affect AI applications, agencies should assess the effect of the potential regulation on AI innovation and growth. Agencies must avoid a precautionary approach that holds AI systems to such an impossibly high standard that society cannot enjoy their benefits.11

The AI Memo touches on this later, explaining:

Promoting innovation and growth of AI is a high priority of the United States government. Fostering innovation and growth through forbearing new regulations may be appropri¬ate. Agencies should consider new regulation only after they reached the decision . . . that Federal regulation is necessary.12

I agree with the former quote from the AI Memo that whatever the federal government does with regard to AI should be done with the intention of letting as many people as possible in society share AI's benefits.13 I also agree with the other sections of the AI Memo that advise federal agencies to consider the potential risks, benefits, and costs when looking at regulation and to have a clear understanding of the existing federal, state, and local regulations.

But a problem emerges by the end of the latter quote. Per that text, agencies should only consider new regulation after deciding that new regulation is necessary to foster innovation and growth. That approach significantly limits the United States' ability to make qualitative decisions about outcomes and benefits concerning AI. By that I mean that per the guidelines in the AI Memo, if an agency determines that it would be best for the public for AI applications to incorporate the right to an explanation or each individual's right for an audit of all automated decision making performed by a company's AI using his or her data or about him or her, the agency would have to prove that considerations concerning costs, risks, and benefits outweigh the potential hindrance those requirements could pose to AI growth and innovation.

That might sound like a small detail, but consider the differ¬ent burdens of action. Without the AI Memo's guidance, a federal agency could determine pursuant to the Federal Trade Commission Act that any company relying on AI that does not provide an audit of automated decisions is engaged in unfair or deceptive acts or practices in or affecting trade. With the AI Memo, before making that determination, the same agency arguably has to determine that that specific requirement outweighs its potential impact on AI inno¬vation and growth. It not only serves as a separate required step, slowing down regulatory action further, but it also has a chilling effect on regulatory action. Despite the great values contained in the AI Memo, the emphasis on growth and innovation may push agencies at a conceptual level to avoid thinking of AI regulation as a way to set the rules for the AI sector's playing field. Non-regulatory oversight becomes a much more attractive option.


The AI Memo's chilling effect could be devastating for the regulation of AI and how AI affects most people. At times in this space, I have compared the development of AI to the Industrial Revolution, noting that AI could introduce as much or more change in America as the factory system did in the 19th century. That type of manufacturing radically changed the nature of work in this country, as concentrating that much labor in one production facility was almost entirely unheard before then. It took a century for legal innovations like minimum wage requirements, work place safety and environmental regulations, and prohibitions against child labor and excessive work hours to help spread the benefits of industrialization. One of the reasons why those legal changes took so long was that governments were reluctant to make qualitative decisions that might affect innovation and growth, refusing to set rules for the manufacturing sector's playing field. I would hate to see something similar happen as AI changes the nature of our lives in the near future.


1. Draft Office of Management and Budget Memorandum, Guidance for Regulation of Artificial Intelligence Applications (January 13, 2020), at 1 available at (the "AI Memo").

2. Exec. Order No.13,859, 84 Fed. Reg. 3967 (February 14, 2019), Sec. 1; available at

3. AI Memo, supra note 1, at 3-6.

4. John Frank Weaver, "Everything Is Not Terminator: What Does the Executive Order Calling for Artificial Intelligence Standards Mean for AI Regulation?" The Journal of Robotics, Artificial Intelligence & Law (Vol. 2, No. 5; September-October 2019), 377.

5. Id.

6. Id.

7. John Frank Weaver, Robots Are People Too (Praeger Publishing, 2013), 13-15; 175-186; John Frank Weaver, "Regulation of Artificial Intelligence in the United States," in W. Barfield & U. Pagallo, eds. Research Handbook of Artificial Intelligence and Law (Edward Elgar, 2018), 162-178; John Frank Weaver, Everything Is Not Terminator: The Importance of Regulating AI As Soon As Possible," The Journal of Robotics, Artificial Intelligence & Law (Vol. 1, No. 2; March-April 2018), 131-136.

8. See AI Memo, supra note 1, at 3-4, 6; John Frank Weaver, "Everything Is Not Terminator: Is China's Social Credit System the Future?" The Journal of Robotics, Artificial Intelligence & Law (Vol. 2, No. 6; November-December 2019), 445-51; John Frank Weaver, "Evidencing a Lack of Bias: Using Artificial Intelligence to Promote the Rule of Law," Shanghai Law Research (Proceed¬ings of 2019 World Artificial Intelligence Rule of Law Forum) (Vol. 9, 2019) (forthcoming).

9. White House Releases 10 AI Principles for Agencies to Follows, AI Trends (January 9, 2020),

10. Matt O'Brien, "White House proposes guidelines for regulation the use of AI," Associated Press (January 7, 2020),

11. AI Memo, supra note 1, at 2.

12. Id. at 3.

13. This is not exactly what the AI Memo says, as it only notes a concern that "society cannot enjoy [AI's] benefits," but I'll give the OMB the benefit of the doubt that it is concerned with the people in society and not society as a single entity.

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.