Games

Arin the chatbot to help children with special educational needs is born

The project, promoted by the Fondazione Mondo Digitale ETS, in collaboration with IRCCS Fondazione Don Carlo Gnocchi, ITLogiX and Università degli Studi Roma Tre, is aimed at neuropsychiatrists, support teachers, parents, teachers

by Luca Tremolada

8 May 2025

4' min read

It is called Pathway Companion and is the first platform based on artificial intelligence for children with special educational needs. The project, funded by Google.org, is promoted by the Fondazione Mondo Digitale ETS, in collaboration with the IRCCS Fondazione Don Carlo Gnocchi, ITLogiX and the Università degli Studi Roma Tre. It is aimed at neuropsychiatrists, support teachers, parents and teachers and is perfectly suited to the daily support of children between 8 and 14 years of age with reading comprehension difficulties, text comprehension and other special educational needs (BES).

How does it work?

The platform, presented during closed-door demos with Il Sole 24 Ore, was designed together with a team of child neuropsychiatrists. The entry point to the platform is ARIN, the intelligent chatbot - based on ChatGPT - that assists users by exploiting an architecture based on three interconnected generative artificial intelligence engines. The teacher or specialist starts by creating a profile of each individual child with information such as age, class attended and specific difficulties. The pupil profile is entered into the platform, but no identifying data is provided to the artificial intelligence. Based on the training of the model, through conversational logic, ARIN proposes the most effective teaching strategies and compensatory tools to support the student. For example, depending on whether difficulties in narrative comprehension or inferential processes are declared, ARIN will propose a strategy to create customisable teaching content. The teaching material that each teacher can upload onto the platform is 'redesigned' and rewritten according to the specific needs of each individual child, taking into account his or her profile and the strategies identified, e.g. improving readability, comprehension and accessibility (e.g. syntactic simplification, use of images, speech synthesis). The system is able to produce mind maps and teaching sheets in support.

The 'game plan' is that of the human on the loop, which in this case is to be declined as teacher on the loop, in the sense that the teaching materials are those proposed by the teacher and the AI elaboration can be modified by the teacher at any stage, before being proposed to the student and then uploaded on the school's e-learning platforms.

'The teacher is not an obstacle to innovation,' comments Alfonso Molina, scientific director of the Fondazione Mondo Digitale ETS, 'but an integral part of the adaptive process. The system provides data, tracking, indicators of understanding: but it is the teacher, together with the family, who interprets these signals in the real context of the student. And then to ask for the help of an expert."

"In our model, the machine does not judge, but amplifies the ability of the teacher, and the caregiver, to understand and intervene. It is a balance between automatism and humanity, between algorithms and relationships. It is also a sign of responsibility, because with children and with BES you don't just experience technology, you build trust".

In the model presented, which is still in the beta phase, the first two engines are already present: namely, the possibility of creating customised profiles of each individual child with BES, the suggestion to professionals of specific compensatory strategies and tools, and, in support of support and non-support teachers, the ability to adapt teaching materials to the specific needs of the pupil.

Still missing is the third engine, which will be available from September, and involves an intelligent tutor customised to the educational needs of each individual child, who, however, will not be able to use it independently, but only under adult supervision.

The model does not foresee the possibility of using AI to assess children's shortcomings, i.e. humans are not offered the possibility of evaluating AI assessments on any tests and lessons that are offered to children.

"The tool," explains Furio Gramatica, Director of Innovation Development at the Carlo Gnocchi Foundation, "is intended to support teachers in the process of verifying the acquisition of knowledge, through the preparation of customised tests and the measurement of their results, always under the supervision of the teacher.

The third intelligent engine.

In the future, the intelligent tutor will constantly adapt strategies, tools and teaching content on the basis of the caregiver's feedback, thus ensuring personalised, constantly evolving, monitored and privacy-friendly learning. But no evaluation phase is planned for now, also because, adds Gramatica, one of the objectives for now is to contribute to the definition of recommendations, on the basis of existing European policies, starting from the results of the Pathway Companion project, in order to consolidate its sustainable large-scale dissemination.

Why these projects are needed.

Let us start from some data. There are almost 359,000 disabled pupils attending Italian schools, 4.5% of the total number of pupils enrolled (+6% compared to the previous school year), with an increase of 75,000 in the last five years (+26%). The proportion of support teachers with specific training has increased from 63% to 73% in four years. However, there are still many non-specialised teachers (27%, with a peak of 38% in the North), and 11% are assigned late. The use of IT tools is also not widespread: only in half of the schools do all the teachers use technology to support inclusive teaching.