Essay by Stefaan Verhulst: “Data openness remains desirable, yet in the age of AI it can also be weaponized. Below, I propose a framework that moves beyond the open-versus-closed binary, complements the FAIR principles, and advances AI-ready data governance through five dimensions of access: availability, accessibility, accommodation, affordability, and acceptability, while clarifying who is responsible for aligning them.
…In 1981, Roy Penchansky and J. William Thomas argued that access to health care is not a property of a clinic or a patient but of the fit between them. They broke that fit into five dimensions: availability, accessibility, accommodation, affordability, and acceptability. A service could exist, be nearby, and be free and still be inaccessible if it was organized around the provider rather than the patient or if the community did not trust it.
That insight has since been carried over to public health data, where the five A’s are used to show what is lost when a common dataset is discontinued. In what follows, I argue that the five A’s deserve a wider application. In particular, they can help us navigate a debate about data and AI that still treats access largely as a door: open or closed, licensed or not, scraped or blocked. Penchansky and Thomas remind us that access is a relationship, and relationships have to be negotiated and maintained.
AI changes that relationship in two ways. First, it adds a new kind of user: models and agents that consume data at a scale and speed no human analyst ever did. Second, it changes the stakes. As I have argued elsewhere, openness has been weaponised, and the defensive response is a data winter in which data have become more enclosed and more scarce. Seen through the five A’s, the data winter is not one failure but five, and they feed on each other. Each also points to a specific task that, today, too often belongs to no one. That task is the work of the data steward.
Reading the five A’s through the fourth wave of open data
In the below, I examine the five A’s through the lens of data for AI, or what we have called the fourth wave of open data. The first wave was driven by freedom of information laws, the second by open-by-default portals, and the third by purpose-driven reuse and data collaboratives. The fourth asks how data can be prepared for generative AI and how generative AI can in turn make data more usable.
That second direction (AI for data) adds nuance to the framework. AI is not only a new consumer of data that puts pressure on each of the five A’s; it is also a tool for improving them. It can help users discover datasets, generate and enrich metadata and documentation, and make access conversational so that people can ask questions of data in plain language instead of navigating portals and schemas…(More)”.
Table: The five A’s of data access in a time of AI
