Soft law standards can promote responsible procurement of artificial intelligence systems.
The many ways in which artificial intelligence (AI) affects society has attracted a lot of attention worldwide. One area of growing interest is the role AI procurement can play in addressing the effects of this technology within the public and private sectors.
One way to evaluate developments in AI procurement is to analyze the text of soft law programs. By ‘soft law’ I mean any initiative that creates substantive expectations about how AI should be designed and deployed, that is not directly enforced by the government. They are “soft” compared to hard legislation that takes the form of law or regulation.
These soft law programs reveal the views expressed by many entities — the private sector, nonprofits, and government — about AI, and take the form of strategies, recommendations, principles, and best practices, among other things.
In collaboration with Gary Marchant of Arizona State University, we have created a database of AI soft law programs from 2001 to 2019. The database contains all AI-related text associated with 634 programs from 64 unique geographic areas. The word ‘procurement’ in this corpus occurs 84 times in 44 programs in the database.
Subsequently, a total of 86 relevant judgments were identified and compiled in a new database focused on purchasing. As can be seen from the figure below, the vast majority of AI procurement statements came from government agencies (48 percent) or government agencies that issue statements with other organizations (40 percent). Interestingly, most of the tender mentions come from soft law initiatives that were made recently, as the median year of publication is 2018.
Any purchase-related statement in the database was analyzed using two dimensions. The first dimension identified the direction of influence. In other words, statements may be intended to direct stakeholders within the sphere of influence of the entity making them (internal), individuals or groups outside of it (external), or target both internal and external stakeholders (both).
|Table: Two Dimensions in AI Procurement Soft Law|
|External||22||Creation of soft laws||30|
|Both||6||Protection of rights||27|
|Acceleration of Technology Development||17|
Internal statements made up the largest group, accounting for 72 percent of the total. They are only found in organizations run by the government or by governmental alliances, such as national governments, state or provincial institutions, or multilateral institutions such as the European Union. Surprisingly, the local government is represented in the form of guidance developed by the New York City Task Force Automated Decision Systems (ADS).
External explanations were found in 22 percent of the programs. These include government statements targeting the private sector, private sector entities or associations that provide guidance to the AI stakeholder community, or non-profit organizations that provide advice to government agencies. Only a diverse but small mix of statements (6 percent) target both internal and external stakeholders.
The second dimension describes the substantive purpose of the soft law text. For this dimension, relevant judgments were categorized into four groups: improvement of processes, acceleration of technology development, creation of soft law and protection of rights. It is important to note that each statement can be categorized with up to two purposes.
Process improvement was the most popular goal of soft law procurement programs, reflected in 66 percent of statements. Eligibility required a statement with a suggestion for improving a purchasing practice. For government agencies, a good example was the guidelines established by the Government of the United Kingdom (UK) in conjunction with the World Economic Forum (WEF). In this paper, the UK and WEF provided concrete advice to government agencies on their AI procurement practices. Specifically, they instruct government agencies to “consider during the procurement process that acquiring a tool with AI is not a one-time decision; testing the application throughout its lifetime is critical” and “ensure interoperability of AI solutions and requires open license terms to avoid vendor lock-in.”
Similar advice is given by actors outside the government. For example Microsoft offers a list of regional procurement standards and guidelines developers should consider when deploying their AI-based technologies.
After improving processes, the second most popular goal that emerged in the database was to create soft law, which is present in 30 percent of statements. Among these statements, organizations advocated creating or improving soft law mechanisms to guide AI procurement. For example, in New York City, a task force devoted to ADS suggested promoting a set of “ADS best practices, including ADS procurement, data retention, and data sharing, to serve as a resource for agencies.”
Nonprofits also issued statements urging governments to take action on procurement. This was the case with C-Minds, which: early the Mexican government to “develop guidelines for smart AI purchasing” and explained that “in order to better provide high-quality services to citizens, the government must embrace technology.”
Ensuring and protecting rights was an explicit goal found in 27 percent of the sample rulings. All rulings in this class stated the need for society to use procurement as a means of ensuring the rights of individuals and organizations are guaranteed.
For example, the Council of Europe dedicated an entire document on the impact of algorithms on human rights. In it, the body urged countries to “promote the development of algorithmic systems and technologies that improve equal access to and enjoyment of human rights and fundamental freedoms through procurement.” UNESCO has in the same vein recommended that governments are “using public procurement and funding to drive gender equality in AI.”
Finally, 17 percent of the surveyed statements identified the purpose of accelerating technological development. This group contained ideas about the role of procurement in driving the creation and adoption of AI technology in the marketplace, primarily in the private sector. For example, the government of France confirmed that its procurement policy should “support European industries”. Similarly, Korea gave a pronunciation to “enable companies to establish early dominance in intelligent IT ecosystems.”
A number of trade associations around the world advocated the use of public procurement to fund AI development. This is the case with a regional medical technology industry association that: stated: that procurement should be used both to “build innovation and to ensure reliable AI”.
This research shows to what extent the conversation about procurement exists within the world of soft law AI governance. For the most part, the government is the main participant and target of these discussions.
This protagonist is well deserved. Public entities have different powers to determine the optimal market conditions in AI. As major customers, they can design incentives for creating and growing responsible AI use by companies. As arbiters of social norms, they can push for favorable soft laws that spread to the rest of the country. As public resource managers, they can improve their own operations through AI deployment.
Customized purchasing database this piece reflects the growing interest in the relationship between this field and AI. Attention to the relevance of this relationship will undoubtedly grow in importance as AI technology expands its capabilities to serve humanity.
This essay is part of a nine-part series entitled Artificial Intelligence and Procurement†