Artificial intelligence is rapidly transforming educational measurement, but its potential hinges on building systems that are not only efficient but also credible, fair, and truly useful for students and teachers. A recent panel of experts highlighted the need for “seatbelts” – rigorous scientific infrastructures – to ensure that AI-driven assessments improve learning rather than simply accelerating existing problems.
The Pillars of Responsible AI in Education
Kadriye Ercikan of ETS argues that three principles must be non-negotiable: efficacy (does it achieve its goals?), validity (is the evidence sound?), and fairness (are results consistent across all student groups?). This means designing fairness into the system from the start, rather than trying to fix biases later. The goal is to shift from assessments that merely describe student status (like a thermometer) to those that drive improvement (like a thermostat).
Reducing the Testing Burden and Increasing Utility
Angela Bahng of the Gates Foundation points out that students already spend up to 100 hours annually on testing, disproportionately burdening students of color and those who are behind grade level. Her work focuses on a “product quality framework” that helps schools choose tools based on their actual usefulness: are they user-friendly, reliable, and directly helpful for instruction? Emerging AI applications – such as speech recognition for real-time feedback and AI reading coaches – show promise, with rigorous evidence expected within the next two to three years.
Beyond Measurement: Respecting Educator Expertise
Michelle Odemwingie, CEO of Achievement Network, argues that validity depends on whether assessment insights actually inform teacher action. The current overload of EdTech tools (over 2,700 in use) creates “information obesity,” hindering educators’ ability to make sense of fragmented data. Odemwingie warns against AI systems that confidently deliver inaccurate information (“reasonable nonsense”), emphasizing that assessment systems must respect teachers’ judgment and expertise to yield lasting value. The core problem isn’t technical; it’s relational.
Prioritizing Human Flourishing Over Optimization
Gabriela López challenges the field to move beyond speed and prediction, designing AI systems that prioritize student growth, agency, and opportunity. She insists that “human variability is signal, not noise” – optimizing for narrow definitions of typical reduces accuracy and trust. True transparency isn’t about exposing code, but about helping people understand what results mean, how to use them, and what they don’t mean.
Ultimately, AI in education must earn trust by demonstrating openness, scientific rigor, and a fundamental respect for the individuals behind the data. The future of assessment lies not in technical sophistication alone, but in building systems that support human flourishing and empower learners and educators alike.
