Equity in AI for Public Health: Part 1 – Why Fairness Must Come First
Table of Contents
Plain-language takeaway: AI can assist public health teams in discovering patterns within complex information. Equity makes sure these patterns are developed, understood, and used in ways that are fair, safe, and helpful.
Public health teams often need to understand how different aspects of people’s lives are interconnected. Mental health, family support, school experiences, access to services, discrimination, and other factors rarely occur in isolation. However, these connections can be difficult to identify when relevant information is dispersed across numerous survey questions.
Artificial intelligence, often referred to as AI, can be used to solve those problems. This involves computer-based techniques that can detect patterns, organize information, or make predictions. In the field of public health, such tools enable teams to examine large datasets more efficiently.
Yet the fact that a pattern has been found does not mean that it has been understood. A computer cannot judge whether the pattern shows an unmet need, an unfair social issue, a decision made during the design of the survey, or even a data error. Similarly, it cannot evaluate whether presenting the pattern will be beneficial or harmful to a community.
This is why Equity in AI needs to guide the work from the start.
At Wellington-Dufferin-Guelph Public Health (WDGPH), the WHY Survey Personas project is a real-world example. The project searches for patterns in the Wellbeing and Health of Youth (WHY) Survey to help staff see how youth needs, strengths, risks, and supports overlap.
The survey is a joint effort involving WDGPH, the Upper Grand District School Board, and the Wellington Catholic District School Board; for the 2025–26 survey cycle, WDGPH also collaborated with Conseil scolaire catholique MonAvenir, a French-language Catholic school board. It gathers information from students in Grades 4 to 12, parents or guardians, and school staff to help with school and community planning. The student portion of the survey lasts approximately 30 minutes and is available in separate versions for Junior (Grades 4 to 6) and Intermediate/Senior (Grades 7 to 12) students. The survey includes topics such as mental health, relationships, school climate, physical activity, substance use, and equity.
The project does not try to identify individual students, diagnose youth, predict their futures, rank schools, or make automatic decisions about services. Instead, it helps public health teams learn, plan, and discuss important issues.
The article includes neither any findings nor any results concerning sensitive subgroups.
What is the meaning of equity in the field of public health? #
In the field of public health, equity involves taking steps to reduce health disparities that are unfair and preventable, and it also means acknowledging that people have different opportunities, resources, supports and barriers from the beginning.
These broader factors are generally referred to as the social determinants of health and include income, housing, disability, racism, gender, sexual orientation, language, geographic location, school climate, and access to services. They shape youth health and well-being. Survey response patterns may therefore reflect not only individual choices, but also the broader social, economic, and environmental conditions and systems that influence young people’s opportunities, experiences, and access to resources.
For instance, imagine that a computer method detects a pattern linked to greater health needs. Public health staff might use this insight to ask better questions regarding the available supports. On the other hand, if the pattern is labelled in a way that blames students or is treated as a fixed category, it could result in stigma.
Therefore, when it comes to equity in AI the question goes beyond ‘Did the method find a pattern?’ and also includes:
Whose situation is being shown? Whose might be left out? What conditions might have shaped the pattern? Who might gain? Who might suffer? Which uses should be forbidden?
In the WHY Survey Personas project, these questions guide every step, from setting the purpose and preparing data to analysis, interpretation, and communication. Equity is part of the process from the beginning, not just something checked at the end.
How can unfairness enter an AI project? #
AI systems are taught using information that has been selected and prepared by humans, and they are given the aims and rules that are established by humans. Because of this, unfairness can appear at a number of stages in the process.
A survey might not reach every group in the same way. Some youth might feel uneasy about answering sensitive questions. Students could get different questions based on their grade or earlier answers. Larger averages can hide small groups. Choices made while cleaning data might erase important differences. Methods used to study the data may focus on common answers and fail to represent smaller groups clearly.
Human decisions likewise influence how results are interpreted and put to use. A technical finding might be given a degree of certainty that is unwarranted. A persona label could highlight the challenges. An exploratory summary might be misapplied to decisions even though it was never intended to support them.
These risks do not mean public health should avoid using AI. Instead, they highlight the need to keep reviewing the project’s purpose, limits, evidence, and possible harms.
A simple guide to NIST AI RMF 1.0 for beginners #
A useful resource is the Artificial Intelligence Risk Management Framework issued by the U.S. National Institute of Standards and Technology.
This article discusses NIST AI RMF 1.0. This voluntary guide helps organizations identify and manage the possible benefits and risks of AI. It does not provide a simple pass-or-fail test. Instead, it organizes responsible work into four connected actions.
| Action | Plain-language meaning | Application to the WHY Survey Personas project |
|---|---|---|
| Govern | Set the rules, responsibilities, and limits. | Define who reviews the work, how privacy is protected, and which uses—such as student targeting or school ranking—are prohibited. |
| Map | Understand the purpose, setting, people, and possible effects. | Consider the survey design and how youth, families, schools, communities, and public health staff could be affected. |
| Measure | Check performance and look for problems. | Test whether patterns remain similar when reasonable methods change and examine missing information, representation, privacy, and possible unfairness. |
| Manage | Respond to what was learned. | Revise, restrict, pause, or reject an approach when the evidence is weak or the remaining risk is too high. |
These actions set up a cycle: if a fairness concern is identified during testing the team may have to go back and consider how the information was prepared; if a privacy concern is discovered during communication then a less detailed summary may be needed; and if a new use is proposed then the project’s purpose and limits will have to be reviewed once again.
For public health agencies, this is an important shift in thinking. Responsible AI is not just about choosing the right method. It also means focusing on governance, public purpose, data quality, fairness, privacy, transparency, human review, and proper use.
How Equity in AI and the WHY project strengthen one another #
The WHY Survey project puts the principles of equity in AI into practice.
For instance, the project has to distinguish between a response that a student chose not to provide and a question that the survey never asked. The team also asks three questions: Does the group appear often enough to discuss? Can it be explained clearly? Could its description unfairly define youth by a problem?
The project is also improved when viewed from an equity perspective since this enables the team to identify who is included and to safeguard smaller groups; it also requires the setting of clear boundaries, honest acknowledgment of uncertainty, and the use of respectful language.
This relationship goes both ways:
- The project provides WDGPH with a genuine opportunity to put responsible AI practices into practice.
- Equity adds safeguards that keep the focus on the youth and communities behind the data.
Privacy and confidentiality are part of fairness #
It is particularly important to regard privacy since the WHY Survey is about young people and involves sensitive subjects.
Privacy concerns how information is collected, used, and shared. Confidentiality means protecting information entrusted to the project so it is not accessed or disclosed beyond approved people, systems, and purposes.
Removing names is necessary, but may not be sufficient. While names are not captured in the WHY Survey, a rare combination of grade, location, identity, or experience could still make it possible for a small group or even an individual to be identified if the results are released without careful consideration. For this reason, complete records, written responses, and results about very small groups must stay within approved systems and be accessible only to authorized staff.
There is also a balance to strike. Too much detail can make it easier to identify someone, while too little detail can hide unfair differences. The team may need to combine information, hide a result, or keep it private. They should also say when the information does not support a clear conclusion.
The fact that names are removed does not ensure that a result is fair. Just as posing an important equity question does not mean that a result can safely be shared, privacy, confidentiality, and fairness all have to be taken into account.
What success should mean #
For the WHY Survey Personas project, success is not just about making clusters or naming them.
Success starts with a clear public health purpose. The work must respect the survey’s design, protect youth privacy, and look for unfair effects. It must also explain uncertainty, include human review, and stay within firm limits.
NIST AI RMF 1.0 helps organize these responsibilities through four recurring actions: govern, map, measure, and manage. Equity gives those actions a public health purpose.
Part 1 explained why these safeguards matter. Part 2 will move from principles to practice. It shows how the safeguards change the WHY Survey workflow, from handling “Not Asked” responses to testing possible clusters and writing respectful personas.