Gains:
- Ability to summarize data such as absence, academic and behavioral observation with artificial intelligence support and prepare a follow-up table and observation notes
- Ability to write prompts that leave the decision to the expert while producing summaries that make it easier to spot patterns for early warning
- Understanding that artificial intelligence reduces the student to a risk and the danger of bias, and being able to see that early warning should always be combined with expert observation and interview.
One of the important jobs of the PDR professional is to monitor students over time: attendance patterns, academic progress, behavioral observations, social relationships, interview history. This monitoring aims to recognize early when a student may need support; This is called early warning. Early warning means activating support based on observable signs (sudden increase in absenteeism, falling grades, withdrawal) before a problem escalates. But this field can be both very useful and very dangerous: if used well, it allows reaching a student on time; Used poorly, it traps a student in a “risk label.”
In this unit, we will learn to use AI as an aid that summarizes scattered tracking data and makes the pattern visible. The most critical principle: AI can point out the pattern, but it cannot evaluate the student or declare him or her "at risk." Early warning is always combined with the specialist's observation and discussion with the student; The final evaluation and decision lies with the expert. Let us also recall the crisis boundary: if a follow-up data indicates a serious risk (for example, a student's situation indicates immediate danger), this is a crisis situation and the protocol in unit 9 comes into play; Not AI.
The fine line between seeing patterns and labeling them
Tracking data alone is not meaningful; context gives meaning. Three-week increase in absenteeism; It may be related to an illness, a family situation, or a difficulty at school. AI can make this increase visible in a table or summary (“absenteeism has increased significantly in the last three weeks”); this is a useful cue that draws the professional's attention to that student. The danger is that the AI (or the expert, based on the AI) will jump from this signal to a judgment: “this student is at risk/problem/neglected.” This jump can be both wrong and stigmatizes the student.
Rule: AI signals, expert constructs meaning. A summary is a justification for starting a conversation, not a conclusion. No sign becomes an evaluation without talking to the student.
Another subtlety is to be careful when combining data sources. Attendance, grade trend, and behavioral monitoring are individually meaningful, but having the AI combine them as a single “risk score” is misleading. A student's absenteeism may have increased because he or she has a chronic health condition; This is not a sign of academic failure. Cramming different types of data into a single number obscures the reality of the learner and creates a false prioritization. The correct approach is to see each type of data as a separate "sign" and leave the whole to the interpretation of the expert. AI can collate data; but it should not do the work of welding them into a single judgment. Likewise, "how long" a sign lasts is also important: a one-week fluctuation and a trend lasting three months have very different meanings, and it is again up to the expert to read this time dimension.
Caution: A printout that lists a student as “at-risk” may permanently impact that child's perception at school. To prevent such labels from entering circulation, early warning printouts should be kept confidential and used only for support purposes and under expert supervision.
Bias: who counts as a "risk" matters
Bias in early warning is a serious justice issue. If the system (or the AI brief) disproportionately flags students from a certain socioeconomic group, a certain neighborhood, or certain behavioral patterns as “at risk,” this institutionalizes a bias. The expert must distinguish which signs truly indicate support and which merely point to a pattern; Each student should be approached individually and fairly. The "risk" language produced by AI should not be adopted without questioning.
Step by step: from data to support
- Collect data anonymously. Organize data such as attendance, observation, and grade progress with "Student A/B/C" instead of ID codes.
- In summary with AI. Summarize the pattern and notable changes at the phenomenon level.
- Choose signs. The expert decides which abstracts deserve an interview.
- Opinion. Talk to the signaled student; learn the context.
- Evaluate and plan. Expert establishes meaning and support.
- Keep it dark. Sharing outputs outside of support purposes; Do not circulate the label.
Copiable templates
1) Tracking data summary:
Your role: PDR data editing assistant. ADD JUDGMENT, "risk" label or recommendation.Task: Summarize the following anonymous absence/observation data at the FACT level by student; State any notable CHANGES (increase/decrease). The evaluation and decision are mine; you just make the pattern visible.Data: [paste anonymous data]
2) List of reasons for meeting:
From the anonymous summaries below, list facts that might constitute grounds for initiating a SUPPORT CONVERSATION. Write a one-sentence NEUTRAL reason for each (such as "absenteeism has increased in the last three weeks"). Don't use labels like "risky/problematic." Summaries: [anonymous summaries]
3) Tracking table editing:
Arrange the following anonymous follow-up notes into a table with columns for date, observed phenomenon, planned step, and outcome. Don't add comments, keep fact.Notes: [anonymous notes]
4) Preparation before the interview:
Your role: PDR assistant. I am preparing for a SUPPORTIVE meeting with a 14-year-old student whose attendance has increased. Suggest 8 preparation questions that are non-judgmental, open-ended, and do not make the student defensive.
Weak prompt / Strong prompt
Weak:
Look at this class's absence list, mark at-risk students.
It makes the AI the judge, asks it to produce a “risky” label, no context and no conversation.
Strong:
Your role: PDR data editing assistant, adding "risk" label. In the anonymous absence data below, summarize (at the phenomenon level) students who showed SIGNIFICANT CHANGE in the last 3 weeks. For each, just write what has changed; I will make the comment and decision.Data: [anonymous data]
It signals, it does not label; The decision is in the expert's hands.
three mini cases
Case 1 — Helpful sign. A specialist was having difficulty tracking individual absences in a class of 200 students. He had the anonymous data summarized by the AI; A student who would normally go unnoticed, who showed a quiet but steady increase in absenteeism, stood out. The expert spoke; He learned that the student had a transportation problem in the morning and produced a practical solution. AI gave the sign, the expert and the student built the solution.
Case 2 — Return from label. One AI summary described one student as “high risk for behavior”; whereas in the expert's data there was only the phenomenon of "request to change class twice". The expert rejected this label, spoke with the student, and found that the request stemmed from a desire to distance himself from a group of friends. The label could unfairly stigmatize the child; expert filter prevented this.
Case 3 — Bias check. An expert noticed that the "risk" list the AI produced disproportionately included students from a particular neighborhood. He did not use the list, thinking that this might reflect bias; instead, it applied fair, fact-based criteria to all students. Thus, support was distributed according to actual need, not stigma.
Common mistakes
- Making the AI mark "risky". Subverting the judiciary to the tool. Solution: AI shows the phenomenon/change, the expert makes the judgment.
- Evaluation without interview. Not just looking at the data and talking to the student. Solution: every sign gains context through a conversation.
- Ignoring bias. Adopting a pattern-based “risk” list. Solution: control of justice and individuality.
- Circulating the label. Sharing the "risky student" list. Solution: privacy and support purpose limit.
- Confusing it with a crisis. Treating a sign of serious risk as routine monitoring. Solution: crisis protocol (unit 9) is activated.
Table: Sign or judgment?
output
Type
what to do
"Absenteeism has increased in the last 3 weeks"
Sign (phenomenon)
Reason for meeting
"This student is risky"
Judgment (label)
Reject, return to reality
"Grades have been falling for two semesters"
Sign (phenomenon)
Reason for meeting
"His family neglects him"
judgment/inference
Reject, return to observation
"The frequency of meetings has decreased"
Sign (phenomenon)
Expert evaluates
In summary
In student tracking, AI is a useful aid that summarizes scattered data and makes patterns visible; But it cannot evaluate the student or declare it "at risk". AI gives signals, expert constructs meaning. Contextualize each sign with a conversation, check for bias, circulate labels, and switch to crisis protocol at signs of serious risk. The purpose of early warning is not to stigmatize, but to provide timely and fair support.
Application task
Create anonymous attendance/observation data for a fictional classroom. Have the AI summarize at the fact level with the “tracking data summary” template. Examine the output: Has the AI added any “risk/issue” labels or judgments? Mark, remove, and create a fact-only, neutral list of reasons to discuss. Then ask the bias question: does the list stack up in a certain pattern?
checklist
- [ ] I anonymized the data before giving it to the tool.
- [ ] I wrote the "add risk label/judgment" limit to the prompt.
- [ ] I kept the AI's output at the fact/sign level, making the judgment myself.
- [ ] I contextualized each sign with an interview.
- [ ] I checked for bias and kept the labels private.