This guide reflects the stages of the development and implementation process:
Within each section, this guide breaks down the process into the specific activities that are necessary during that phase.
Pre-Development includes the initial planning activities required to develop equitable AI solutions (i.e., problem definition, goal setting, use-case definition, stakeholder mapping, infrastructure assessment, and risk assessment).
Model Development includes data availability, model selection and design, pre-processing and in-processing considerations, specifically the potential bias injection points, and the implications of and mitigation strategies for bias.
Post-Development includes testing, model deployment and monitoring best practices, model retraining, output adjustment, and maintenance. This section will address the practical ways to mitigate bias in the development of and use of algorithmic outputs.
Post-Implementation addresses how to interpret outputs, educate end users, and collect and incorporate feedback for ongoing strategic monitoring of the tool.
Are you a non-technologist?
(EdTech leaders, institutional leaders, project managers)
The summary guide on this website is focused on non-technical stakeholders involved in the creation of AI solutions. You’ll find guiding questions, examples, templates, worksheets, and guides to support you as you lead the development of responsive and equitable AI solutions.
Are you a technologist?
(algorithm creators and developers)
Our Github page is a one-stop shop for developers working to incorporate equity at every stage of the machine learning pipeline. You’ll find technical resources, including, among others, bias detection libraries like Aequitas a Bias and Fairness Audit Toolkit, illustrative examples, and sample code.