Collecting Autism Care Insights: MLUX Italy x Accessibility
A collaborative process among AI & Design teams
In the Artificial Intelligence (AI) world, much focus is put on the creation, testing, and deployment of models. Therefore, when AI is put into a product, the design process starts from AI’s constraints, such as the inputs of the model, the type and amount of data it needs to be trained on, and what kind of responses it can return.
Instead of creating these models and then testing them on people, what if we reversed the process and created it from people to satisfy their needs? To try out this idea, Corinne Schillizzi from Tangity shared a project with us to design AI to support stakeholders in the education of autistic children called Artificial Intelligence Design for Autism (AIDA).
User Research
Rather than obtaining a dataset or creating an AI model, this project started with user research. The intention, as outlined in the diagram above, is to integrate this design step, along with the other design steps, with the AI development steps to ensure that they continue to inform each other.
The user research for this project started with identifying all relevant stakeholders due to both their interest in the resulting product as well as their potential to be interviewed during the user research.
These stakeholders include not only the child, but also their parents, their teachers, their classmates, other teachers, and other education and autism experts. Given this stakeholders model, the team decided to interview educators, parents, and therapists. Then they performed collaborative research and design activities with the AI team, and observed stakeholders in practice with the prototypes.
They came to several conclusions about both the needs of autistic children and the needs of the caregivers. In terms of the needs of autistic children, they found that providing structure to the child is mandatory, though how much structure depends on how rigid they are. They also found that the toys and other activities designed for the child need to be personalized, too. Finally, they found that eduction of autistic children needs to be especially focused on making them more independent because so they can avoid stress from unexpected changes in activities.
In terms of the needs of the stakeholders, they found that the caregivers need to share their knowledge about the child. This is because they have different perspectives and knowledge, and sharing it together is essential to caring for the child. They also found it is important to track the possible causes of behavioral crises, and knowing about their progress in becoming more independent or autonomous.
Dataset Definition
Given what they learned in their user research, they determined that the AI would be helpful in creating a personalized education and treatment path for the children. This is because every person with autism has unique characteristics and needs for treatment, and AI can help identify these needs and create personalized paths.
Therefore, the next step was for the design team was to collaborate closely with the AI team to design a recommender system that was adaptable to each child’s characteristics and needs, according to their functioning and flexibility levels (pictured below).
They started this collaborative design process by sharing their research results with the AI team, and then designing a system evolution map. The evolution map (pictured below) maps the system’s knowledge (e.g., contained in the AI model) at the start, the interactions with the system, and the resulting system knowledge from user feedback during the interactions. The initial knowledge in this case was about a child’s functioning level, rigidity, communication style, and other characteristics. The interactions entailed detecting crisis events, such as unexpected events that trouble the child; recommendations from the system to address the event; and user feedback on which recommendations worked well and which did not. Finally, the system knowledge is updated with this information — namely, in this case, that helping the child create an agenda was more effective than isolating them or speaking to them with a quiet voice.
Next Steps
The next steps in the AIDA project are to use the evolution map to collect quantitative data, design the model, and probe the design. More specifically, as pictured above, this will include developing the model, testing it, and then refining all of the integrated concepts, all while making sure to avoid human biases in the data and collaborating closely with the AI team.
Big thank you to our speaker for sharing their expertise with us!
Corinne Schillizzi: UX & Service Designer at Tangity, part of the the NTT DATA Design Network
About the Machine Learning and User Experience (“MLUX”) Meetup
We’re excited about creating a future of human-centered smart products, and we believe the first step to doing this is to connect UX and Data Science/Machine Learning folks to get together and learn from each other at regular meetups, tech talks, panels, and events (held remotely).
Interested to learn more? Join our meetup, be the first in the know about our events by joining our mailing list, watch past events on our youtube channel, and follow us on twitter (@mluxmeetup) and Linkedin.
