This sounds like science fiction: can AI detect ADHD through movement? 

I have not been this intrigued by a research paper in a long time! It sounds almost like science fiction: could artificial intelligence recognise ADHD simply by watching the way we move? And yet, who knows where this kind of research may lead.

A new study published in Scientific Reports explored whether artificial intelligence could identify patterns associated with ADHD using ordinary video recordings made at home. The researchers analysed recordings of 78 young adults who spoke freely to a camera for approximately five to ten minutes. The AI was not listening to what they were saying. Instead, it converted their movements into a digital representation of the upper body and examined patterns in the movement of the head, eyes, shoulders, arms and hands.

From these recordings, the researchers extracted almost 600 movement-related features, including movement speed, acceleration, range, coordination and variability over time. The best-performing model identified participants with and without ADHD with an accuracy of 85.7%. That is impressive.

It is also important to understand what this result does and does not mean.

The study does not show that ADHD can now be diagnosed by watching someone during a video call. The total sample was small, and only 21 participants were included in the final test set. The available data also did not include detailed information about medication, treatment, ADHD presentation or co-occurring conditions.

Anxiety, autism, fatigue, pain, excitement, cultural communication styles and many other factors can influence the way a person moves. An algorithm can detect a pattern, but it does not necessarily understand why that pattern is present.

What I find particularly interesting is the move away from assessing ADHD only in highly controlled clinical or laboratory settings. ADHD does not exist only during an appointment, a questionnaire or a computerised attention test. It is experienced in everyday movement, communication, motivation, emotional regulation and interaction with the environment. Traditional assessments provide important information, but they may not always capture the complexity and variability of real-life functioning.

Technology like this could eventually provide clinicians with an additional source of information. It might help us observe changes over time, understand individual patterns or evaluate how symptoms respond to treatment. However, the same technology could also be misused.

There is a significant difference between using AI to support a careful clinical assessment and using it to monitor employees, students or children and make assumptions about their brains based on how much they move in front of a camera. A person is not a collection of skeletal coordinates.

The most promising future for AI in ADHD is not replacing clinicians, questionnaires or lived experience. It is combining different sources of information to develop a more complete and personalised understanding of the individual.

The technology is fascinating. The science is preliminary. But sometimes science fiction is simply science that has not reached clinical practice yet.