Autism and EU AI Act

By now, all of those who know me, also know that I am deeply interested in how AI can be beneficial in improving behaviours of autistic children. In my past articles I have also listed a few use cases. I have also, in the recent past, written articles on the evils of enacting laws before technologies become settled; the main evil being curbing innovations. Let us look at one example covering all these aspects here

Article 5 of the EU AI Act states that the following artificial intelligence practices shall be prohibited:

(dc) the placing on the market, putting into service for this specific purpose, or use of AI systems to infer emotions of a natural person in the areas of workplace and education institutions except in cases where the use of the AI system is intended to be put in place or into the market for medical or safety reasons.

Annex III of the Act, provides a classification for all High Risk AI Systems. One of those high risk items classified are AI systems intended to be used for emotion recognition. (Article 1-ab of Annexure III)

(Italics/Bold italics mine)

To start with, going by all of these above, we can safely presume that those AI systems used for emotional recognition where it is intended for medical, or safety reasons is only permitted by the Act and even those are classified as ‘High Risk’.

·???????? Medical reason means ‘relating to illness and injuries and to their treatment or prevention’.

·???????? Safety reason means ‘the state of being safe from harm or danger’

Let us look autism now. What is it?

Autism is not an illness. Signs of autism might be noticed when you're very young, or not until you're older. If you're autistic, you're autistic your whole life. Autism is not a medical condition with treatments or a "cure". But some people need support to help them with certain things.

What is this emotion recognition software?

Emotion recognition software that uses artificial intelligence (AI) and machine learning algorithms to analyze and interpret facial expressions and emotions

?How can AI in general and emotional recognition software in particular help autistic children?

A paper at https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8875834/ by the US National Library of Medicine will be relevant here. Without going into details, the paper lists various methods by which emotion recognition tools augmented by AI can be used to improve lives of autistic children.

To sum up, emotion recognition AI is clearly banned by EU-AI Act. Talk of curbing innovation!

Let me resubmit my five use cases that I made for AI and Autism

?·???????? All institutions like ours struggle to draft reports on the children’s progress as they have various limitations-Like resources, time, language issues etc. Reports are sent to many authorities like insurance companies, schools, hospitals and of course the parents. Presently my company, like scores of others have many such reports already in our data bases. Whilst the software applications related to therapy does produce reports, they do not consider various other factors like reports of other children or even earlier reports of the same child.? Can AI take these inputs and provide reports of high quality based on verbal/written inputs from therapists? I think this is answered with a resounding ‘Yes’ by all who know even rudimentary AI

·???????? Can AI also constantly try to improve these reports based on the additional inhouse data available and publications/tools/research papers/studies available online? I think this gets slightly more complicated but can be done

·???????? Can AI do partial work of therapists/supervisors in providing a better ‘way forward’ for their therapy sessions in future? For this the AI should analyse data inputs from therapists, take inputs from their supervisors, consider all available resources available online including research papers, latest trends etc. This is getting towards AI doing part of therapists work and is complicated IMHO

·???????? Can AI determine the session out puts of therapists/supervisors and determine whether this is proving beneficial to the child? This should be based on actual progress the child is making with goals set for him/her and what was achieved. This also relates to step 3. In case there are gaps, AI should address these and change the session formats of therapists/supervisors

·???????? Lastly, can AI grade therapists/supervisors to show how effective they are in their work and point out weaknesses/strengths? IMHO this is the toughest one of them all

?