6. AI Governance

Theme six: AI Governance

Stories of AI implementations having all the restraint of the wild west aside, there are substantive efforts at governing AI (or more accurately, governing the use of AI) at multiple scopes, from the single institutional level to the international. In this theme, we will survey some attempts at AI governance as well as some metastudies which attempt to explore one aspect of AI governance over a broad area.

With deference to Stan Lee, "with great power there must also come - great responsibility". The City of Albuquerque, New Mexico's AI Working Group developed an AI policy document that seeks to "improve lives, enhance public services, preserve public trust, and is rooted in our shared values of access, advancement, and accountability" while positioning Albuquerque to be a "national leader in the responsible and sustainable use of AI".

A quandary that each publicly traded corporation must wrestle with is "whose benefit supersedes whose?" In a two thousand four interview, cofounder and then-chairman of Southwest Airlines Herb Kelleher maintained that placing employees first led to satisfied customers which ultimately drove shareholder value. Likewise, a two thousand twenty-three paper by Tay et al. echoed and expanded this claim by pushing back against performance-based valuation in favor of what they called "optimal functioning" in which worker, organization, customer, and societal/environmental well-being are all equally regarded. We start our examination of corporate AI governance with a paper by Cordeiro et al. which similarly advances the claim that organizations that integrate AI into their operations will find themselves poorly served by the so-called "shareholder primacy" which has been at the center of corporate operations from the nineteen seventies through the early two thousands. The use of AI, they say, will move organizations away from "shareholder primacy" to a "stakeholder-centric" model.

A common response to concerns about corporate AI governance is for tech companies to publish a variety of documents inventorying the ways in which they are responsible, transparent, effective, reliable, and inclusive: and how customers who use their AI products can also manifest those traits. A critical reader might ask themselves: "Are these claims proactive or reactive?" Microsoft's Responsible AI Principles and Approach site is a good example of such content. Should we conclude anything from the fact that the transparency report has been updated in both two thousand twenty-four and two thousand twenty-five, while the Responsible AI Standard hasn't been updated since two thousand twenty-two? This site is a gateway to a variety of materials ranging from those intended to look like industry standards, to quasi-marketing case studies, and training. What does the choice of topics, and the choice of topics omitted say about Microsoft's goals for this site, and their view of their AI products' strengths and weaknesses?

Of course, we see AI being implemented by more types of institutions than just companies. Jobin et al. conducted a scoping review using eighty-four instances of AI ethics documents in grey literature from private companies, government agencies, academic and research organizations, NGOs, intergovernmental/supranational organizations, non-profits, professional and scientific societies, and so forth. Common themes emerged from nearly all of the documents: transparency, justice and fairness, non-maleficence, responsibility, and privacy. On the surface, this would seem to indicate commonality, but the operational details of these themes varied wildly between documents. The authors, at least in two thousand nineteen, found their search methods yielded documents in the global north more easily than elsewhere. They also found that documents tended to focus on harm prevention rather than promotion of the benefits of AI, almost as if the goal of the entire ethics exercise is to manage exposure rather than advance more effective ways of working.

As we've seen a few times already, AI could be regarded as a confounding variable. As the Financial Services Sector Coordinating Council observes in "Published Documents: Financial Sector Artificial Intelligence Executive Oversight Group Deliverables", AI tends to be an accelerant. For some actors, it provides an enhanced ability to compromise financial systems and engage in a variety of cyberattacks. For others, it provides a set of tools for enhanced transparency and development of innovative products. At nine intervals of various lengths since November two thousand twenty-three, the FSSCC has issued a set of reports regarding identified gaps in the financial sector's use of and preparedness for AI. The cited reference is this year's master document which describes and links to each of the subordinate gap documents. This year, they address a lexicon; a risk management framework; identity and authentication; explainability; data quality and alignment with state, federal, and international governance; and enhanced fraud.

The next two resources examine artificial intelligence from an international perspective.

Like the FSCC resource above, the UN's Artificial Intelligence Subject Briefing is a launchpad for several in-depth documents relating to AI. Those twelve resources are "Resources" heading, about seventy-three percent of the way down the main page. The author(s) of this briefing see AI as helping the UN fulfill up to eighty percent of its sustainable development goals as well as strengthening the work of the entire UN system. These possibilities, however, come at the cost of "coordinated global governance", which includes pushing back against the "shadows" of AI: disinformation, human rights violations, voter and public opinion manipulation, and the undermining of trusted institutions. Aspirationally, the UN calls for all governments and stakeholders to work together to develop a common governance framework that supersedes any existing criteria and resolves existing gaps in AI governance.

The next resource with International scope is The National Judicial College's Massive Open Online Course, "Artificial Intelligence and the Rule of Law". Developed with UNESCO, The Future Society, and the IEEE Standards Association, this six-module course's topics include adoption of AI across justice systems, online courts, algorithmic bias and safeguarding human rights, and AI ethics and governance relevant to judicial operators.