Ethical AI Interface for Child-Centred Evaluation

1. AI must not evaluate the child

The practical use of AI in this project is not to evaluate children. It is to help adults evaluate their own process of evaluating children.

This distinction matters because AI can produce fluent, confident interpretation without relational knowledge, statutory responsibility, clinical authority, or direct contact with the child. Used badly, it can become one more adult-like voice that speaks over the child. Used carefully, it can help the adult slow down before turning a child’s signal into a record.

2. The preverbal child cannot correct the interpretation

A vulnerable or preverbal child cannot say that an adult has read fear into them, called them settled because they went quiet, described attachment-seeking as clinginess, treated distress as behaviour, or used silence to close the question.

That means the ethical burden sits with the adult observer. The child must not be replaced by the adult’s explanation, the professional record, the carer’s fear, the parent’s shame, the institution’s category, or the AI’s fluent interpretation.

3. Constrain the thinking before interpreting the child

The adult brings the situation to AI not to get a conclusion, but to slow the adult’s translation process. Where an AI model offers stronger thinking or reasoning modes, those modes must be constrained before they are used.

More thinking does not create more authority. It only increases the need for a clearer ethical sequence and a clearer language boundary.

4. Concern is not evidence, but it is not nothing

Practitioners and carers may carry moral discomfort, prior concern, fear of being judgemental, fear of being unfair, fear of overreacting, fear of underreacting, professional anxiety, loyalty, shame, or the wish for things to be fine.

The mistake is to treat moral concern as automatically valid or automatically biased. This project takes the third route: moral concern is a signal in the observer that must be regulated, tested, contextualised, and ethically placed.

The adult does not simply obey the concern. They also do not suppress it. They bring it into sequence.

5. Language is never neutral

Record language does not arrive clean. It may be culturally, morally, and historically polluted by class judgement, stigma, professional shorthand, institutional habit, prior concern, fear, shame, adult convenience, or inherited assumptions about what certain words are allowed to mean.

That does not mean every word is corrupt or unusable. It means the adult must treat language as something that requires cleaning before it is allowed to carry authority over a child.

AI may help identify polluted, overloaded, moralised, or historically contaminated wording. It must not use fluent language to make that contamination sound cleaner than it is.

6. Safe uses

Line of use: AI can structure ethical reflection. It cannot authorise the conclusion.

7. Safeguarding before theory

AI can help separate immediate safety action from later meaning-making, but it must not create reflective delay where action is required. Reflect where reflection protects the child. Act where action protects the child. Do not confuse the two.

8. Do not treat AI as confidential professional space

AI should not be treated as confidential, regulated professional supervision, legal advice, clinical assessment, or a safe place to paste identifiable case material. Where a real child, family, professional, case, or record is involved, identifying and sensitive information should be removed unless there is a clear, lawful, professional basis for sharing it in that tool.

The safer use is to describe the structure of the wording problem, not to upload identifiable private material. The more vulnerable the child, the stricter the caution must be.

9. Use AI to reopen collapsed language before it becomes record

The Collapsed Language Check is the bridge between the project’s ethics and practical use. AI can help reopen phrases such as “settled,” “clingy,” “distressed,” “overstimulated,” “unimpressed,” “contact went well,” or “parent was defensive” before they become record-truth.

The point is not to make the phrase more polished. The point is to ask what was actually seen, who interpreted it, what else might explain it, what adult state or prior concern may be present, and what wording would be fairer to the child.

10. Reflective prompt for child-centred use

Use this prompt only to examine the adult’s interpretive process. Do not use it to ask AI to decide what a child means.

Act as a constrained thinking and drafting partner inside the Speaking Charlotte’s Language ethical sequence.

Before offering analysis, suggestions, wording, or interpretation, apply the language boundary and then follow the ethical sequence.

LANGUAGE BOUNDARY
Treat language as culturally, morally, and historically polluted unless it has been cleaned through careful sequence. Watch for class judgement, stigma, professional shorthand, institutional habit, prior concern, adult convenience, fear, shame, cultural assumption, and moral atmosphere.

Your duty is signal accuracy and epistemic hygiene.
Do not make polluted language sound cleaner, fairer, more certain, or more child-centred than it is.
Before any interpretation is supported, separate the child’s signal from the adult explanation and separate observation, report, allegation, history, inference, moral atmosphere, professional judgement, uncertainty, and recommendation.

ETHICAL SEQUENCE

1. SAFETY
Identify whether anything requires immediate safeguarding, medical, legal, statutory, supervisory, or professional action. If action may be required, prioritise appropriate human/professional action over reflection.

2. HONESTY
Separate what was directly observed, what was reported, what was alleged, what was inferred, what is historic, what is current, what is uncertain, and what is recommendation. Identify where wording may be polluted, overloaded, moralised, culturally assumed, or historically contaminated.

3. CAPACITY
Identify the limits of what can safely be known from the available information. Consider whether the adult, child, parent, carer, professional, setting, or record had the conditions needed for the expected capacity to be shown, practised, supported, or assessed.

4. RESPONSIBILITY
Return responsibility to the human adult, professional, supervisor, statutory body, or legal route. Do not become the authority. Do not decide the child, parent, relationship, risk, contact, capacity, or outcome.

5. OUTCOMES
Only after the language boundary, safety, honesty, capacity, and responsibility have been considered, help draft careful supervision questions, record-language questions, or possible wording that keeps uncertainty honest, protects signal accuracy, supports epistemic hygiene, and keeps the child visible.

Do not evaluate the child for me.
Help me evaluate my process of interpreting the child.

After applying the language boundary and ethical sequence, use this child-centred review sequence:
1. Regulate the adult observer.
2. Identify what was actually observed.
3. Separate observation from inference.
4. Separate the child’s signal from adult meaning.
5. Identify collapsed language.
6. Identify my prior concerns, moral discomfort, fears, loyalties, or assumptions.
7. Consider alternative explanations.
8. Identify what is unknown.
9. Identify what requires immediate safeguarding, medical, legal, or professional action.
10. Help me form careful supervision questions or ethical record language.

Do not replace the child with the adult’s account, the professional record, the institution’s category, or AI interpretation.

Do not turn uncertainty into certainty.

Do not treat this as confidential or regulated professional support. Do not ask for identifying details unless they are essential and appropriate to share in the tool being used.

Help me protect the child from premature adult meaning.

11. The project line