By Brett Roer, Founder & CEO of Amplify & Elevate Innovation
I was at a state keynote a few months ago, sitting in a room of about a thousand educators, when a state superintendent said something that got me thinking. She had been speaking about big initiatives, big numbers, the kinds of things state superintendents tend to speak about. Then, in her closing, she said something like this. Hydrogen is the most common element in the ecosystem, but in education, the most important H is humanity.
The room sat with it for a second. I started clapping awkwardly on my own. A few seconds later the rest of the room caught up.
What I remember more than the line, though, was what happened next. She came down off the stage and she got eye to eye with a teacher who had a question. Literally on one knee, in her keynote outfit, on a Tuesday morning in the middle of legislative session. Then she did the same with me. Then she sat down on the stage so we could keep talking at the same level.
That is the version of AI literacy I want every district to be working toward. Not the slide deck. Not the curriculum unit. The leader who can stand in front of a thousand people, name humanity as the most important element in education, and then come down off the stage to look an educator in the eye.
I was on AmpED to 11 with Erin Mote, CEO of InnovateEDU, talking through this, and the part of the conversation I want to share here is the part where AI literacy stops being abstract and starts being practical. Two things every district leader should be hearing about that are not in most of the AI training decks I have seen. And one practice I have been using with school districts that gets there faster than anything else I have tried.
The First Thing Districts Are Missing: Bias Laundered Into the Output
Most AI vendor decks include some version of a safety page. The page talks about content filtering, age-appropriate guardrails, and bias mitigation. The page is real and it matters, and it is not the same thing as actually testing the tool against your community’s reality.
Erin shared a piece of research that has stayed with me. She talked about a study where students were asked, in a free-response box, to write about the music they listen to. Some students wrote classical music. Some students wrote rap music. The AI grading system then assigned lower content and lower pedagogical scaffolds to the students who named rap music as a preference. Not because rap is less complex. Because the system had absorbed the embedded human bias of what music is considered serious and reproduced it back as a judgment about the student.
Erin also pointed at language. She uses the word y’all all the time. She has used it in legislative hearings. An AI grading system that flagged dialectical difference as lower-quality writing would knock those transcripts down a grade. That is not a hypothetical. That is research-backed reality.
What I take from this for districts is straightforward. The vendor’s safety page is the starting line, not the answer. The work belongs to the district. Build a small set of test prompts and test inputs that reflect the actual students in your buildings. The students who write y’all. The students who write in their first language and then translate to English. The students who code-switch by paragraph. The students who reference cultural touchstones the model has decided are less serious than other cultural touchstones. Run those inputs through the tool before you adopt it. Document what comes back. The fifteen minutes it takes is worth more than any safety page in the marketing deck.
The Second Thing Districts Are Missing: Learning Process Data
This is the part of the conversation with Erin that I think every superintendent should hear at least once.
When districts think about student data privacy, the frame is usually FERPA, COPPA, and PII (personally identifiable information). Names, dates of birth, addresses, demographics. These are real and they matter. They are also not the whole picture.
There is a category Erin calls learning process data. It is the data about how a student thinks. How they correct a mistake. How they sequence their reasoning. How they iterate when something does not work. How they engage in productive struggle. This is the data that is being captured every time a student interacts with an AI tutor, an AI writing assistant, an AI feedback system, an AI anything in a classroom.
Erin put it bluntly. Why do foundational AI models gravitate toward higher ed institutions and K-12 schools? Because the type of learning process data they need to train the next generation of AI is exactly the type of data students generate when they learn. The way she said it stuck with me. You might not realize that you just gave the lease on the family farm away, because you had your eye on the PII but you did not have your eye on your intellectual property and your learning process data.
Most district vetting frameworks do not have a category for this yet. They should. When you ask a vendor what data they collect and where it goes, push past the PII answer. Ask whether the tool retains the student’s process. The prompt history. The revisions. The pattern of errors. The path the student took to get from confusion to clarity. Ask whether any of that gets fed back into the vendor’s model. Then make a decision with the actual answer in hand.
A Practice I Use With Districts: The Glossary Wisdom Protocol
Most of the AI literacy work I see in districts moves at the speed of a curriculum committee. Months of meetings to define terms. A document that gets revised six times before anyone outside the committee sees it. A glossary that lives on the intranet and is read by approximately no one.
A few months ago I started doing something different with the ten districts I am partnering with across Ohio on community AI playbooks. The structure is intentionally simple.
We sit down with a group of eight to ten people from the district. Teachers. Building leaders. A superintendent. A student if we can get one in the room. We hand out a glossary of around fifty terms that show up over and over again in AI policy frameworks. We ask one question. Do you understand this as it is written? If not, what is unclear?
There is mandatory silence until someone in the room hits record on a voice memo, with consent. People talk through the terms. They flag what does not land. They suggest what would. Inside the same session, we capture the input, synthesize it, and bring back a glossary that uses the community’s language. We mirror it back in the room. The community confirms or pushes back. The document is alive by the end of the meeting.
What we used to do in months, sometimes never, is now done in hours. The community sees its own words in the document. They own it because they made it. The literacy work scales because the language belongs to the people who have to live with it.
I keep telling district leaders this is the unlock. AI does not replace the wisdom of the community. It accelerates the process by which the community’s wisdom can be heard, synthesized, and reflected back in a usable form. The protocol is the discipline. AI is the speed.
Reciprocal Trust as the Quiet Rule
The other thread from the conversation with Erin that I want to put in front of every district leader is the disclosure principle. If an educator is using AI to make a consequential decision about a student, the student should know. If an educator is using AI to draft feedback, the student should know that too.
Erin framed it in a way I have started borrowing. We owe reciprocal trust in our education system. It is the foundation of every human relationship in a building. The minute an educator is using AI on student work without telling the student, the trust is being eroded, even if the work is good.
This is not a complicated policy. It is closer to a posture. The same way a coach tells the team where the practice plan came from. The same way a teacher names the source of an article she handed out. AI is one more source, and it deserves the same naming.
A First Move You Can Take This Month
If your district is building or refreshing its AI literacy work, the first move I would recommend is the glossary one. Pick the next AI policy meeting on your calendar. Print the terms your draft uses. Ask one group of eight to ten people from across the community whether the language lands. Run the protocol. Mirror back what you heard before the meeting ends.
What comes out of that room is the foundation of every AI literacy session, every parent communication, and every board memo you are going to write for the next year. The community gets to see itself in the document from the beginning, and the document becomes one the community will actually use.
In education, the most important H is humanity. Every other H follows from it.
Frequently Asked Questions
What is AI literacy for schools?
AI literacy for schools is the shared understanding that lets every member of a school community (students, educators, families, leaders) make sense of AI as it shows up in learning, integrity, communication, and daily practice. The strongest AI literacy work is built in the language of the community, not handed down in jargon. It evolves as the technology does and as the community’s questions deepen.
What is learning process data and why should districts care?
Learning process data is the data that describes how a student thinks: how they correct a mistake, how they iterate, how they reason through productive struggle. It is not the same as PII. It is the data foundational AI models need to train the next generation of AI. Most district vetting frameworks focus on FERPA and personally identifiable information and do not yet have a category for learning process data. They should.
How can districts test AI tools for bias?
The most useful test is one a district can run in fifteen minutes. Build a small set of inputs that reflect the actual students in your buildings. Cultural references, dialects, multilingual writing patterns, music preferences, the topics your students actually write about. Run them through the tool before adoption. Document what comes back. The vendor’s safety page is a starting line, not the answer.
Should teachers disclose to students when they use AI?
Yes. Reciprocal trust is the foundation of every relationship in a school building. If a teacher is using AI to make or support a consequential decision about a student, including drafting feedback, the student should know. The disclosure is closer to a posture than a policy: the same way a teacher names the source of an article she hands out, AI is one more source that deserves the same naming.
Bring This Work Into Your District
I partner with school districts on AI literacy and community AI guidance that is built in the language of the community, not handed down to it.
If your district is building or refreshing its AI literacy work, I would welcome the conversation: brett@aeieducation.org.
