11.2.2 Equity and Fairness in Learning Solutions

Equity and fairness in learning solutions require that LE systems deliver equivalent or better outcomes for historically underserved learners — not merely that they avoid overt discrimination. This is an active design requirement, not a passive absence of bias.

Sources of Algorithmic Bias in LE

Bias in learning engineering systems can arise at multiple points in the pipeline. Training data bias occurs when historical learning data reflects existing inequities — models trained on data from well-resourced schools may generalize poorly to under-resourced contexts. Feature selection bias occurs when proxy variables (e.g., prior GPA) encode demographic disadvantage into model predictions. Deployment context bias occurs when systems designed for one population are deployed with another without validation. [LE-LS-JO-004]

The U.S. Department of Education's AI-in-education report documented that e-proctoring systems with facial recognition technology failed disproportionately for darker-skinned, transgender, and neurodivergent students — with real consequences including false academic integrity accusations. [LE-LS-GL-004] This is the paradigm case for why equity review must precede deployment, not follow it.

Equity as a Design Requirement

The "High-Leverage Opportunities for Learning Engineering" roadmap identified algorithmic equity as one of ten strategic priorities for the field, framing it as a technical and institutional design problem requiring dedicated investment in fairness-aware model development and evaluation. [LE-LS-GL-003] IEEE ICICLE's standards require equity impact assessment as part of the LE quality framework. [LE-LS-CO-001]

Universal design for learning (UDL) principles — multiple means of representation, action, and engagement — provide a design framework for building systems that serve diverse learners from the outset rather than retrofitting accessibility. The Learning Engineering Toolkit frames UDL integration as a HCD competency, not an accommodation. [LE-LS-GL-007]

Fairness Metrics

Practitioners need to choose fairness metrics appropriate to their context: demographic parity, equalized odds, predictive parity, and individual fairness have different implications and trade-offs that must be made explicit in system documentation. [LE-LS-JO-004]

From the Learning Engineering Toolkit

The Toolkit treats fairness as a live concern throughout the learning engineering process. During the iterative creation stage, an early performance check can reveal whether particular features or design choices disadvantage some learner populations — in terms of participation, learning, and engagement — after which the team has to decide whether and how to respond, and be ready to defend those choices. [LET-07] In the investigation stage, interpreting data calls for care to keep bias out. [LET-07]

The book grounds equity in the APA principle of justice, holding that learning engineers should build products, processes, and services that everyone can reach equally, that account for differences in culture and social norms, and that make room for a range of approaches and perspectives. [LET-07] Professional codes likewise stress guarding against bias and prejudice while encouraging diversity. [LET-07]

For design, the Toolkit draws on design justice and anti-discrimination design, both of which zero in on discriminatory assumptions that can get baked into a design. [LET-07] Design justice takes the view that design can entrench existing bias — the answer choices on a multiple-choice item, for example, may carry cultural assumptions — and recasts the designer as a facilitator while the affected community leads the collaborative work. [LET-07] Anti-discrimination design supplies a typology that includes disaffordances, which overlook relevant user identity groups, and dysaffordances, which go further by making unaccommodated users misrepresent themselves — for example, compelling a learner to pick a specific disability, gender, or racial category when no fitting option exists. [LET-07]

Chapter 17's SEEM-ED tool puts justice into practice with prompts such as whether benefits and harms fall evenly across users and nonusers, whether the product or process is accessible, and what its affordances, disaffordances, and dysaffordances turn out to be. [LET-17]