Value-Sensitive Design in Action: Designing Student-Centered Intelligent Tutoring Systems with Community College Students and Instructors

Synthesis: Ko and Hughes (2026) apply Value Sensitive Design (VSD) to the design of student-centered intelligent tutoring systems (ITS). Working with community college students, instructors, instructional designers, and field experts (developers and data scientists), they ran the VSD cycle of conceptual, empirical, and technical investigations and produced a value-oriented prototype. The work documents how learners and instructors -- a stakeholder group historically left out of learning-platform design -- articulate values, surfacing persistent value tensions (transparency vs. interpretability, privacy vs. instructional insight, student agency vs. system-guided Scaffolding) that designers must manage rather than resolve.

Relevance to AI in Education: This study is a concrete example of human-centered design practice for AIED: it operationalizes how Ethics values can be built into an ITS from the outset (rather than retrofitted), and it directly addresses the privacy, explainability, and autonomy concerns raised by data-intensive tutoring platforms -- in the specific, often-overlooked context of developmental mathematics at a community college where adaptive tutoring drives the entire course.

Key Findings

• 16 value-aligned design features across three families. The final prototype (Step 7) integrated explainable AI features (E1-E5: comprehension-check interpretation, setting expectations on the learning science behind the ITS, practice-question selection details, concept and learning-path connection, communicating the AI's confidence level), human-in-the-loop features (H1-H9: student control over re-assessment, review, personalized goals/pace, bookmarking, confirming confidence over mastery, personalized support after long answer times or accuracy declines, feedback to the ITS, communicating receipt of external help, and AI-assistance involvement-level control), and privacy-control features (P1-P4: learning-data repurposing control, permission to share learning analytics, permission to share affective states, and privacy-level control).
• Value tensions shaped every design decision. Rather than treating transparency, privacy, and agency as independent requirements, the study found them in tension: students preferred collaborative/humanized explanations over raw model transparency (transparency alone has little value unless it supports learning); sharing emotional/affective analysis with instructors was controversial among both students and instructors; and opinions on privacy settings split along student/instructor lines (students wanted control, instructors wanted visibility to support learning).
• The historically overlooked stakeholder. The paper argues students and instructors are rarely included in designing the AI learning platforms they use, and most studies do not clearly explain how values such as fairness, transparency, or autonomy are built in. Here, direct engagement of these stakeholders produced value-rich design inputs (needs, pain points, value tensions) that were mapped through the VSD process into concrete technical features.
• Context-specificity is explicit. Findings are context-bound to a single institution that is unusual in fully utilizing its ITS's adaptive/AI capabilities in developmental math (most U.S. community colleges disable adaptive tutoring and use the ITS mainly as a question bank) -- and to adult, returning students (many with multi-year gaps from schooling). This is a strength for authenticity but limits generalizability.
• VSD is a viable, structured route for human-centered AIED design. The study demonstrates the full VSD cycle (conceptual investigations of stakeholders/values, empirical investigations of activity and value tensions, technical investigations of value-aligned features) can be applied in education, extending VSD beyond its prior mainly-HCI applications and addressing the gap between AI-ethics problem identification and actionable solution creation.

Connected Concepts

• Intelligent Tutoring
• Explainable AI
• Human In The Loop AI
• Privacy
• Agency
• Ethics
• Stakeholders
• Design Based Research
• Human AI Collaboration

Connected Articles

• Hazra Safetutors Pedagogical Safety 2026 — SafeTutors: pedagogical safety guardrails for tutoring
• Making AI Tutoring Productive Mastery Math 2026 — design of productive AI math tutoring
• Zhang Tutormoments 2026 — context-appropriateness of tutor help (scaffolding vs. rigor)

Citation

Ko, E. G., & Hughes, J. E. (2026). Value-sensitive design in action: Designing student-centered intelligent tutoring systems with community college students and instructors. Computers and Education: Artificial Intelligence, 10, 100560.