Procurement as a Change Agent for AI Governance

Global attention has been drawn to the multiple ways artificial intelligence (AI), affects society. A growing area of interest is the role of AI procurement in addressing the impacts of this technology within both the public and private sectors.

Analyzing the text of soft-law programs is one way to evaluate developments in AI procurement. Soft law is an initiative that sets out substantive expectations about AI’s design and deployment, but which is not enforced directly by the government. They are not as strict as hard law, which takes the form of legislation or regulation.

These programs are soft law and reflect the views of many entities, including government, private sector and non-profits. They include strategies, recommendations and principles as well as best practices.

We collaborated with Gary Marchant, Arizona State University. This database contains AI soft-law programs from 2001 through 2019. This database includes all AI-related text for 634 programs in 64 different geographic areas. This corpus contains the word “procurement” 84 times across 44 programs.

The 86 statements that were relevant to AI procurement were then identified and compiled into a new database. The figure below shows that the majority of statements regarding AI procurement were issued by governmental entities (48%), or jointly with other organizations (40%). It is interesting that most procurement mentions are from soft law initiatives, which were released in the average year of 2018.

Two dimensions were used to analyze each procurement-related statement. The first dimension determined the directionality and influence. Statements can be used to direct stakeholders in the entity’s sphere of influence (internal), guide individuals or groups (external), and target both internal stakeholders (both)

The largest group was the internal statements, accounting for 72% of the total. These statements are only found in government-led organizations or alliances of governments such as national governments, states or provincial institutions, or multilateral institutions like the EU. Surprisingly local government is represented by guidance created by New York City’s Automated Decision Systems Task Force.

22% of programs contained external statements. These could be governmental pronouncements that target the private sector, associations, or private sector entities providing guidelines for the AI stakeholder community or non-profits offering advice to government agencies. A small number of statements (6%) are directed at both internal and outside stakeholders.

The second dimension is about the substance of the soft-law text. The relevant statements were divided into four categories: protection of rights, acceleration of technology advancement, creation of soft laws, and improvement of processes. Each statement could have up to two purposes.

Sixty-six percent of the statements reflected an improvement of processes as the most common purpose of soft law procurement programs. A statement must suggest ways to improve a procurement process in order to qualify. The guidelines, which were created by the United Kingdom’s government in collaboration with the World Economic Forum (WEF), were a great example for government agencies. The UK and WEF gave concrete guidance to government agencies regarding their AI procurement practices in this document. They direct government agencies to “consider when purchasing a tool that contains AI; testing the application over its lifetime is crucial” and to “ensure interoperability and require open licensing terms in order to avoid vendor lock-in.”

Similar advice can be provided by people outside the government. Microsoft, for example, provides a list with regional procurement standards and guidelines developers should consider when developing AI-based technologies.

The creation of soft laws, which is present at 30% of the statements, was second in popularity after improving processes. Organizations urged the improvement or creation of soft law mechanisms to help guide AI procurement. New York City’s task force on ADS recommended advancing a set of “best practices” in ADS to assist agencies. This included data retention and sharing, ADS procurement and data sharing.

Nonprofit organizations also made statements urging governments to act in the area of procurement. C-Minds was one such organization. It requested Mexico to “develop guidelines for smart AI procurement.” They explained that government must embrace technology in order to provide citizens with high-quality services.

In 27% of the samples, the purpose was to ensure and protect rights. The need for society’s use of procurement to ensure that individuals and organizations have their rights protected was a common theme across all statements.

The Council of Europe, for example, dedicated an entire paper to algorithmic impacts on human rights. It urged countries to ” encourage the creation of algorithmic systems, technologies that enhance equality access to, and enjoyment, human rights through procurement.” UNESCO, however, has recommended that governments “use government procurement and funding to drive gender equality in AI.”

In 17 % of the statements, the purpose to accelerate technological development was identified. These statements included ideas on the role procurement plays in creating and deploying AI technology in the market, mostly from the private sector. For example, the French government stated that it would support European industries with its procurement policies. Korea, on the other hand, issued a statement in which it said that it would “enable businesses to take early advantage of intelligent IT ecosystems.”

Many trade associations worldwide support public procurement for funding AI development. A regional association representing the medical technology industry said that public procurement should be used to both build “innovation” and provide trustworthy AI.

This research shows how much conversation there is about procurement in the world of AI governance soft law. These discussions are dominated by the government as the primary participant and target.

This protagonism is well-deserved. The best marketplace conditions for AI are set by public entities. They can create incentives to encourage responsible AI usage by large companies as they are big customers. They can advocate for soft laws to improve the quality of life in the rest of the country by being arbiters of social norms. They can also improve their operations by deploying AI as stewards of public resources.

This article’s bespoke procurement section reflects the increasing interest in AI and this field. As AI technology improves its ability to serve humanity, the importance of this relationship will increase.