The Decisions Schools Should Never Hand to AI

For the past few years, most school conversations about AI have focused on what the technology can produce. Teachers use it to draft materials, organize information, create starting points for lessons, summarize documents, or think through ideas. Leaders use it to prepare communication, review information, or reduce some of the administrative work that takes time away from other responsibilities. In most of these situations, AI produces something, and a person still decides what happens next.

That distinction is beginning to matter more.

Newer AI systems are being designed to do more than respond to a prompt. They can move through a sequence of tasks, gather information from different sources, identify what needs attention, recommend what should happen next, send communication, update systems, and complete parts of a workflow without requiring someone to direct every step. Higher education is already beginning to experiment with these kinds of systems in advising, enrollment, student support, and administrative work, and it is reasonable to expect similar capabilities to continue moving into K–12 systems over time.

Schools do not need to wait until those systems are fully embedded to decide how much authority they should have.

As AI becomes capable of doing more of the work surrounding a decision, schools need to define which decisions must still be made by a person.

Consider what this could eventually look like in a student-support process. A student has missed several days of school, grades have declined, two teachers have documented concerns, and the student has recently stopped attending an intervention period. An AI system may be able to gather all of that information, identify the pattern, and alert the appropriate staff member much faster than someone could manually review several different systems.

That could be useful.

But the next questions are more consequential. Should the system determine what intervention the student needs? Should it place the student into that intervention automatically? Should it initiate a referral? Should it contact the family? Should it determine which students receive attention first when staff capacity is limited?

Those decisions may appear to be part of the same workflow, but they do not carry the same level of responsibility.

The same issue can surface in discipline. A system may be able to review previous behavior incidents, attendance, teacher documentation, intervention history, and other information and identify students it believes require additional attention. That information might help an administrator notice something that otherwise would have been missed.

But a student is not only the information recorded in a system.

A behavior may be connected to a disability. A student may have experienced something at home that morning. A multilingual learner may have been misunderstood during an interaction. A counselor may know information that cannot appropriately be placed in a general data system. A teacher may understand that the behavior described in a referral looks very different when the events immediately preceding it are considered.

The system may identify a pattern correctly and still be missing the information required to decide what should happen to the student.

Special education makes this distinction even more important. AI may help a team organize observations, compare information across documents, summarize data, or notice areas that deserve closer attention. Those uses may make a complicated process easier to manage.

But should AI determine that a student needs a special education evaluation? Should it determine whether a student is eligible for services? Should it recommend a placement? Should a recommendation generated from available data begin carrying enough weight that team members feel they need a reason to disagree with it?

Those are different questions from whether AI can be helpful during the process.

Schools will face similar questions in course placement. AI may be able to identify students whose performance suggests they are ready for advanced coursework. That could help schools notice students who have historically been overlooked.

But what happens when the same system recommends that another student should not be placed in that course?

Does the recommendation become one source of information for a counselor and teacher to consider, or does it become the decision unless someone intervenes?

The distinction matters because systems make judgments based on the information available to them, and the information schools have about students is often incomplete.

A student may have strong ability that has not appeared consistently in grades. A student may have been absent for reasons unrelated to academic readiness. A multilingual learner may understand far more than one measure suggests. A student with a disability may demonstrate knowledge differently from what the system is designed to recognize.

A person can ask what is missing.

That ability becomes increasingly important as AI moves beyond helping schools organize information and closer to influencing what schools actually do.

Why It Matters

Schools make hundreds of decisions that affect students and staff, but those decisions do not all carry the same consequences. Some determine access to services. Some influence a student's educational path. Some affect discipline, safety, placement, or the way a student will be understood by other adults. Others affect whether an educator is hired, how an employee is evaluated, or whether a concern about someone's performance is escalated.

As AI becomes more capable, schools are going to have to distinguish between work the technology can perform and the authority the organization is willing to give it.

Those are not the same thing.

A system may be capable of recommending which students receive intervention. That does not mean it should determine who receives it. It may be capable of identifying behavior patterns. That does not mean it should determine a disciplinary consequence. It may be capable of analyzing information from a classroom observation. That does not mean it should determine a teacher's evaluation. It may be capable of screening employment applications. That does not mean it should decide who deserves an interview.

The concern is not that AI will always make the wrong recommendation. In some situations, it may identify patterns accurately, surface information quickly, or help adults consider possibilities they would not have noticed on their own.

The larger issue is what happens when a recommendation begins functioning like a decision.

That can happen without anyone formally handing authority to the technology. A system consistently produces a recommendation. Staff become accustomed to following it because it is usually useful. Over time, disagreeing with the recommendation begins to feel like the choice that requires explanation.

The process may still technically include a human, but the human's role has changed.

Instead of independently considering the situation and using the system as one source of information, the person may begin reviewing the AI recommendation and deciding whether there is enough reason to override it.

That is a very different kind of human involvement, and schools should decide whether that is what they want before routine practice establishes the answer for them.

One place to begin this year is by identifying decisions that carry significant consequences for students or employees. District leaders can look across special education, intervention, discipline, student safety, course placement, counseling, hiring, and employee evaluation and ask where AI already contributes information or may soon be able to do so.

Then the conversation needs to become more specific than whether AI is allowed.

Leaders can ask: What role is the technology permitted to have in this decision? Can it organize information for the person making the decision? Can it identify patterns that deserve human attention? Can it recommend possible actions? Can it initiate the next step in a process? Can it make the final decision?

Those are different levels of authority, and schools may draw the line differently depending on what is at stake.

For example, a district may be comfortable allowing a system to flag attendance patterns so a school team knows which students need closer attention. The same district may decide that no student is automatically placed into an intervention because of that flag. A qualified educator reviews the information, considers what else is known about the student, and decides what support is appropriate.

A school may allow AI to help an administrator organize evidence from a classroom observation but establish that the system cannot assign an evaluation rating or determine whether an educator has met a professional standard.

A special education team may use AI to help organize large amounts of information while making it clear that eligibility, services, goals, and placement remain decisions made through the required human process.

Human resources staff may use technology to help organize applicant information while deciding that no candidate can be removed from consideration solely because an automated system ranked that person lower.

The important part is that these boundaries should not be left for individual educators to determine in the moment.

A teacher should not have to decide alone how much weight an AI-generated recommendation should carry in a student referral. A principal should not have to determine independently whether an automated analysis belongs in an employee evaluation. A counselor should not have to guess whether a system-generated risk indicator is information to consider or an expectation to act.

The organization should establish those expectations before people are placed in those situations.

Schools should also begin asking different questions when they purchase or renew technology. For years, technology reviews have appropriately focused on issues such as privacy, security, accessibility, instructional usefulness, and cost. Those questions still matter.

But schools will increasingly need to ask about authority.

What decisions can this system influence? What actions can it initiate? What happens automatically after it identifies a concern? Where does the system stop and wait for a person? Can a person disagree with the recommendation? What happens when they do? Is that disagreement documented? Does the system continue learning from the decisions people make?

Those questions help leaders understand not simply what a product can do, but what role it may eventually occupy inside a school process.

There is also an important distinction between having a human somewhere in the process and making sure that person still has meaningful authority.

A principal who clicks approve on a recommendation generated by a system is technically involved. A counselor who receives a list of students ranked by risk and works from the top down is still participating. A special education team that receives an AI-generated recommendation and discusses whether to accept it still includes people.

But schools need to ask whether those people are actually making the decision or whether they are confirming one that has already been shaped for them.

That is why simply saying there will always be a human involved is not enough.

Schools need to know what the human is responsible for deciding.

There will be many places where AI can appropriately support school work. It can help organize information that is scattered across systems. It can help people notice patterns. It can reduce repetitive administrative work. It can prepare options that a knowledgeable professional then reviews.

Those uses may help educators spend more time on the parts of the work that require relationships, context, and professional judgment.

But there should also be decisions where the expectation remains clear regardless of how capable the technology becomes.

A person decides whether a student needs a particular service. A person decides what disciplinary response is appropriate. A person decides whether a student should be referred for an evaluation. A person decides whether an educator has met a professional standard. A person decides whether an applicant should move forward when the decision affects someone's opportunity to work.

Not because human beings make perfect decisions. They do not.

The reason is that some decisions require someone who can ask another question, consider information that does not fit neatly into a system, recognize when the available data does not tell the whole story, explain the reasoning to the person affected, and accept responsibility for the outcome.

That responsibility should not disappear simply because AI becomes capable of completing more steps in the process.

Schools do not need to predict every new capability that will arrive this year or next year. They can begin with something more basic.

They can decide which responsibilities remain human and make sure the systems they adopt are designed around that decision, rather than allowing the technology to determine the boundary for them.

This is one of many hidden shifts AI is introducing into education. I write these newsletters to help educators and school leaders see them early and respond intentionally. If that matters to you, stay with the series.