By Lisa Harmon-Martínez, Director of Learning-by-Doing at Future Focused Education

For years, the measure of student success in our state has been dictated by a narrow set of metrics: standardized test scores and graduation rates. While these numbers offer a convenient snapshot, they fail to capture a student’s true readiness for life beyond high school. They can’t measure creativity, cultural competence, or the ability to contribute to a community. Recognizing this, our state is making a bold move by requiring schools and districts to develop a Graduate Profile—a hyper-local, community-defined understanding of what a student should know and be able to do by the time they graduate. This is a powerful step away from a one-size-fits-all approach and a move toward what truly matters to our communities.

This is where we can leverage large language models (LLMs) to transform our definition of student success. For years, our team has been imagining a future where students could be assessed in a way that truly reflects the knowledge, skills, and attributes that are essential to sustain our communities, our language, and culture. For instance, some schools, like those in Zuni, are already using their Graduate Profiles to assess student capstone projects—deep, multifaceted bodies of work that showcase a student’s journey. This process is deeply insightful, but it requires an incredible amount of people power to find evidence of these qualities in their portfolios and capstone presentations.

AI offers a potential solution. These models have the capacity to analyze a large body of student work—a portfolio spanning from middle school to graduation, or a capstone from start to finish—and identify patterns and themes that align with the indicators of a Graduate Profile. This can dramatically reduce the time and effort required, making this holistic assessment far more manageable and scalable across our state.

The promise is clear, but the peril is equally so. The conversation around AI often focuses on data privacy, and while it's essential that many models we already use in schools, like Khamigo or Magic School AI, protect a user's personally identifiable information, the core issue goes much deeper. The responses and analysis we get back from these models are "normed" according to a vast, global dataset. When we give our students' work and our community's data to these models, we risk losing the crucial, hyper-local nuances that make us who we are.

Imagine if an AI tool, trained on data from all over the globe, were to assess a student’s work. The model would identify what it considers to be "good" or "proficient" based on global trends, potentially overlooking or devaluing unique cultural references, linguistic patterns, or community-specific knowledge. The very purpose of the Graduate Profile is to foster what it means to be a graduate in our local communities—a person who can sustain our language, culture, and economic well-being. Unlike what happens in Zuni, if we as educators and educational leaders give our power over to AI, to assess based on a global standard, we could strip away the very essence of what makes our communities unique.

Furthermore, we're seeing a growing concern in the AI community about model collapse, a phenomenon where models trained on synthetic or AI-generated data begin to lose their accuracy and connection to authentic human knowledge. Our local partner at Oforma AI, Alexander Jacobson, has been sounding the alarm, warning us that our data is our greatest commodity, and we must be careful about how it is given away and for what purpose. We have to ensure that our students’ authentic work—rich with the language, culture, and context of their lives—is not simply used to train new models that will, in turn, strip away the very authenticity they were meant to preserve.

We don’t have to reject AI technology entirely. Instead, we can control how it’s used. The solution lies in building our own infrastructure—specifically, local, closed servers in schools, districts, or at the state level.

This kind of server means that the AI models would operate exclusively within the boundaries of our community or state. Data from student portfolios and capstone projects would never be sent to a global cloud for analysis. It would remain protected, and the AI’s learning would be confined to our specific context. 

This is not about relying on a machine for a final grade. This is about using technology to power up our human judgment. Teachers and staff who know their students best will still be the final arbiters, but the AI can do the heavy lifting of sifting through massive bodies of work to highlight evidence of the qualities we care about.

Ultimately, we are at a profound moment in education, a chance to move beyond standardized tests and truly understand what our students know and can do. The path forward is not a simple one, but it is a necessary one. We can harness the power of AI to create a more meaningful and equitable system of assessment, but we must do so on our own terms—by protecting our data, preserving our culture, and ensuring that our students' success is measured not by a global standard, but by the unique richness of our own communities.

To learn more and problematize the use of AI in our classrooms, consider attending a free professional development session this fall. 

This three-session professional development series on AI use and safety in the classroom will empower New Mexican educators to confidently integrate AI tools ethically and effectively. Participants will explore critical topics such as intellectual property and data privacy when using AI, learn to apply New Mexico PED's AI Guidance to promote academic integrity and responsible integration in their classrooms, and discover how to leverage AI to foster authentic, community-connected student work. The series aims to equip educators with the knowledge and strategies to prepare students as responsible and innovative digital citizens who can harness technology for meaningful learning and positive community impact. You will receive a calendar invitation after completing the form below. 

Thursdays: September 11, October 9, November 6 

***The author admits to using AI to fact check this blog.