President Signs AI Executive Order

whitehouse.gov

Recently (December 4, 2020) the then President of the United States (POTUS) signed an executive order relating to the use of artificial intelligence (AI) software in government agencies. Three areas stand out to me as being significant.

Significant Aspects

1. The insistence that AI applications be understandable
2. The development of a comprehensive inventory of AI applications in government
3. The list of AI situations where these rules do not apply (the loopholes).

My commentary is for your information only. You can read the complete executive order here and come to your own conclusions: https://www.whitehouse.gov/presidential-actions/executive-order-promoting-use-trustworthy-artificial-intelligence-federal-government/  

Understandable

There are many predictable statements that the then POTUS makes in this executive order. Things like "Lawful," and "Accurate, reliable, and effective" are all things that we would expect the agencies within the government would insist upon. Of significance is the requirement that AI applications be understandable.

More specifically, the wording of section 3 states "Sec. 3. Principles for Use of AI in Government. When designing, developing, acquiring, and using AI in the Federal Government, agencies shall adhere to the following Principles:" and then it goes on to include item e: "(e) Understandable. Agencies shall ensure that the operations and outcomes of their AI applications are sufficiently understandable by subject matter experts, users, and others, as appropriate."

As I mentioned in an earlier article about laws covering the use of AI in financial matters, the concept of "sufficiently understandable" will naturally have different meanings to different people. As cited in the aforementioned article, an illustration of this is "the example of forensic scientists explaining themselves to jurors in a trial, and the large gap in the understanding by these laypeople."

How a government body, and all the agencies within it, agree upon what is "appropriate" remains to be seen.

Inventory

There are specific timelines for the implementation of the former POTUS' executive order, and naturally, these too can change with change in administration and staff. Nevertheless, section 4 states: "Sec. 4. Implementation of Principles." among other things, that there will be a timeline for the development of an inventory of AI use cases. Part b of this section reads as follows: "(b) Within 180 days of the CIO Council’s completion of the directive in section 5(a) of this order, and annually thereafter, each agency shall prepare an inventory of its non classified and non-sensitive use cases of AI, within the scope defined by section 9 of this order, including current and planned uses, consistent with the agency’s mission."

This is an ambitious timeline, administration changes notwithstanding. It will be an exciting time to be able to read such an inventory, and I will definitely be keen on combing through such a list; if and when it is published.

Loopholes

As is often the case in government activity, the situations where the rules, laws, and executive orders DO NOT apply are often as interesting as the rules themselves. It behooves us, therefore, to take a peek at "Sec. 9. Scope of Application." and more specifically, the part that states "(d) This order does not apply to:"

Within part d, lower case Roman numerals i and ii read as follows:

"(i) AI used in defense or national security systems (as defined in 44 U.S.C. 3552(b)(6) or as determined by the agency), in whole or in part, although agencies shall adhere to other applicable guidelines and principles for defense and national security purposes, such as those adopted by the Department of Defense and the Office of the Director of National Intelligence;"

"(ii) AI embedded within common commercial products, such as word processors or map navigation systems, while noting that Government use of such products must nevertheless comply with applicable law and policy to assure the protection of safety, security, privacy, civil rights, civil liberties, and American values; and" ... [R&D]

One would expect matters of national security that must be kept classified, secret, confidential, or otherwise restricted from public view; would be one of the loopholes exempted from the order (Roman numeral i), but what about commercial software?

The executive order does mention these commercial products in Roman numeral ii, and extends parts of Section 3 to these use cases, but it does not extend "Understandability" and it does not extend them to bein included in the "Inventory." How far will this definition of commercial products extend, such that they can be exempted? This remains to be seen - but unfortunately, it may never be seen. Just as the military and spy agencies do not need to reveal their use of AI, neither do the producers and government consumers of the "common commercial products." We may never know what these products are doing, nor understand decisions that the AI algorithms make, that may impact our daily lives.

In Conclusion

I found those three areas most interesting, but I am eager to hear your thoughts. Feel free to reach out with a public comment, or send me a private message, as I am eager to learn more about the "Understandable" aspects of AI.