What is my child trying to tell me?

Why this is hard, and why we built this

Behaviour analysts have a name for what you are trying to do. It is called a functional behaviour assessment, and it asks one question: what is this behaviour for? A full one takes many hours of trained observation, spread over weeks. Many families never get one.

What happens instead is usually a questionnaire, filled in from memory, days after the incident. Studies that compare these questionnaires with the experimental method find agreement that varies widely and is often low.

The weak link is the antecedent, what happened in the thirty seconds before. A functional hypothesis rests on it, and it is the part least often recorded. The adult in the room was busy living through it. By the time anyone writes it down, it is a memory of a stressful moment.

A video does not have that problem. It captures the lead up as it happens.

The four things a behaviour usually means

Decades of behavioural research keep landing on the same four functions. In a large study of self injurious behaviour, covering 152 people with developmental disabilities who were each assessed experimentally, a clear function was found in about 95% of cases.

Escape. Getting out of a demand, a task, a place, a transition. The largest single category, about 38% of cases.

Attention or connection. Bringing an adult back into the room, into contact, into looking. Together with requests for a specific object, about 26% of cases.

Asking for something. A thing, an action, a routine resumed, where there are no words to ask with.

Sensory regulation. Seeking input the body needs, or shutting out input it cannot tolerate. About 26% of cases, and the one most often misread as meaningless.

The same behaviour serves different functions in different children, and the same child’s screaming can mean two different things an hour apart. The answer is never in the behaviour alone. It is in what surrounded it.

We also watch for a fifth possibility that the four categories can hide: pain. In a US national parent survey, chronic or repeated pain was reported for 15.6% of autistic children against 8.2% of other children, roughly twice as often. Pain is routinely misread as behaviour, which clinicians call diagnostic overshadowing. If something in your video suggests discomfort or illness, we say so plainly and tell you to see a doctor.

Sources: Iwata et al., Journal of Applied Behavior Analysis, 1994, 27(2), 215 to 240. Whitney and Shapiro, JAMA Pediatrics, 2019.

What your report looks like

Every report follows the same eight part structure. Children differ, and the questions we ask of each video do not.

1. What you told us. Your context, restated, so you can correct us.
2. Before (0 to 30 seconds). Who was in the room, what was asked of your child, what changed.
3. The behaviour. Described in observable terms. Duration, intensity, body, voice. No interpretation yet.
4. After. What followed, who responded, how it ended.
5. What it may be communicating. The four functions, ranked, each with the specific moment in your video that points to it.
6. Where the video is not enough. Named explicitly. Often two functions remain equally possible, and we say which, and what would separate them.
7. What to watch for next time. Two or three specific, observable things.
8. Standing note. This describes one video. It is not a diagnosis.

Part 6 is the one we will not drop. A confident wrong answer about your child is worse than a plain statement that we cannot tell from this.

What we cannot tell you

We cannot tell you what your child is thinking. Nobody can, and any system that claims to is overclaiming.

We cannot give you a diagnosis, a clinical assessment, or a treatment plan. Those belong with your clinicians, and we will not step into that role.

We cannot tell you from one video what is true in general. A behaviour has a function in a context, and one clip is one context.

What we can do is describe, carefully, what is visible, and tell you which explanations it supports and which it rules out.

Why we read these ourselves

Ethan AI builds computer vision systems that detect distress and safety behaviours in real time, running today in a therapy centre and a home. We do not yet have a system that can tell you why a behaviour happened, and we are not aware of one that does. Published research in this area recognises what a behaviour looks like, not what it is for.

So our team reads these videos and writes these reports by hand. It is slower, and it is why intake is limited. It is also the right way to do it for now, and every report we write shows us what the system will eventually have to produce.

If you tick the optional box, your video helps build that system. If you do not, you still get the same report.