AIDA: the challenge of user research for AI in the autism field
Applying a human-centered and collaborative approach to a complex subject
Written by Lucia Ferretti, Anna Focaroli and Corinne Schillizzi
AIDA is a non-profit research project born within Tangity to help people with Autism Spectrum Disorder (ASD) and their caregivers to communicate effectively.
Why did we focus on communication? First, because it is one of the most challenging skills for people on the Spectrum and, second, because selecting the right communication method is crucial to avoid potential crises. Nevertheless, it is not at all easy to choose the right communication manner to deal with children with ASD because their reactions significantly change depending on the person and on the specific situation. Hence, it is essential for caregivers to know the individual they are interacting with in order to establish a correct relation with him/her. The identification of this necessity is what led us to a fundamental question.
Can an Artificial Intelligence system provide caregivers with information about the child’s condition and suggest the most appropriate communication methods for the individual and the circumstance?
The choice of leveraging on Artificial Intelligence lies in an ongoing evaluation regarding the autism field. On the one side, technology usually has a positive role in facilitating caregivers communicating with or calming down children on the spectrum. On the other side, several studies [Valencia, K., Rusu, C., Quiñones, D., & Jamet, E. (2019). The Impact of Technology on People with Autism Spectrum Disorder: A Systematic Literature Review] have highlighted the need to customize tools and methodologies according to the abilities of the subject with ASD and to the specific circumstance of interaction. In this sense, AI could perfectly meet this need thanks to its capability to learn and adapt.
When there is a teacher replacement, the child finds it hard to accept it, and the teacher has some troubles because she doesn’t know the child.
— Paola Golzi, teacher at the Istituto Comprensivo Statale “Fabio Filzi”
Considering the multiple fields covered by the project, we explored a cross research methodology by applying a human-centered approach to the design of an AI system. Hence, it was necessary to establish a collaborative design process among three teams working together: AI experts, UX/Service Designers, and field stakeholders.
But how to define and apply a collaborative methodology enabling different actors to share their knowledge in order to discover, define, and design a solution for people on the Spectrum and their caregivers?
The application of a new collaborative methodology to a complex field like autism required the team to define a tailored approach to research. In order to face this challenge, we decided to follow some of the principles of “action research”, a discovery strategy including participatory and iterative aspects.
It centeres on doing “with” rather than doing “for” stakeholders and credits local stakeholders with the richness of experience and reflective possibilities that long experience living in complex situations brings with it.
— Davydd J. Greenwood and Morten Levin, Introduction to Active Research, 2nd edition, 2007
First, we applied the principle of active participation of field stakeholders to ensure valuable insights. Hence, the design team involved ASD experts in gathering and analyzing research findings, and, additionally, shared results with AI experts to iterate upon research activities with the aim to achieve a more reliable design of the AI model.
[…] a defining characteristic of action research: its commitment to a process of research in which the application of findings and an evaluation of their impact on practice become part of a cycle of research.
— Martyin Denscombe, The Good Research Guide for small-scale social research projects (2010)
Second, action research entails a cycle of discovery-evaluation process. In other words, following an iterative approach, each research outcome has been evaluated by a cross-functional team to improve further research steps and select the next investigative actions.
As an outcome, both internal and external stakeholders had an active role in the whole research process, from information gathering to insight validation and iteration.
Intending to achieve a systemic understanding of the autism field, the group structured the research phase of the project through different activities:
Desk Research — focused on collecting information about epidemiology, diagnosis and treatment, and new perspectives on the topic. This first activity allowed the team to reach a general understanding of the project’s context that was then deepened through the field research insights.
Field Research — it was critical to deepen our comprehension of the project’s domain and to explore our stakeholders’ needs by conducting three different kinds of activities:
- Interviews — aimed at uncovering details about children’s daily life and caregivers’ experiences by taking into account their different perspectives about ASD. To do so, we structured a different interview protocol for each kind of caregiver to fit his peculiar viewpoint about educating and caregiving people on the Spectrum. Interviews were mostly executed through open-ended questions to outline real-life stories — then translated into personas — and into research outputs to avoid reproducing researchers’ biases.
- Workshops — collaborative sessions with caregivers aimed at giving them the chance to share their experiences with children on the Spectrum and to analyze them together. Workshops represent the principal participatory approach suggested for an action research strategy since it allows testing and evaluating insights in real-time with field experts.
- Participant Observation — it was organized a day in a summer center during which some of us participated in activities together with educators (the same involved in workshops) and their “pupils” with ASD. The main purpose of this experience was to be personally involved in activities and to study children’s daily routines. Additionally, observation aimed at validating previous insights while reinforcing team members’ empathy with stakeholders. Furthermore, this activity enabled the design to map the environment’s dynamics as well as children’s behaviors and interactions within an open social context.
Leveraging on insights collected during the activities mentioned above, the team designed a set of customer journeys mapping critical events involving children with ASD and their caregivers. Most relevant outcomes regard the communication strategies applied by stakeholders and the challenges they faced during interaction. Such aspects constituted a reference for the following phase of solution design by considering different communication methodologies and their effects on the child.
One of the most challenging aspects of conducting research in the autism field was about gathering information regarding young people with ASD. In fact, we had to consider the potential negative impact of the occurrence of new events — such as our interaction — for subjects on the Spectrum. Therefore, to collect data regarding children’s lifestyles and struggles, it was necessary to establish relationships with their closest people, including educators, therapists, and families.
A second issue to be considered is that autism is a broad and multifaceted topic, requiring long-time experience and studies to be fully understood. For this reason, including people surrounding children was indispensable not only to gather information but also to receive specialists’ support in framing the problem.
This research project’s final relevant trial regards the collaborative approach among AI and Design teams that have been adopted since the first research stages. In fact, due to time limits, this new collaborative approach has been defined, tested, and improved while teams were already exploring the autism field.
Work with experts, not for them
Especially when dealing with a complex or sensitive topic, it is important to reach field experts since the very beginning of the project since they may help in defining the best approach and focus for explorative activities.
This strict collaboration supports contextualizing desk research findings and enriches them through first-hand insights. Likewise, these contacts facilitate a collaborative validation of insights while working with field stakeholders on improving their daily life.
Communication is the key (not only for people with ASD)
In order to really unlock the value of including multiple perspectives in every design phase, one needs to adopt multiple communication manners and tools reflecting preferences of the different stakeholders involved.
Hence, it is essential to create and to apply versatile tools and materials designed considering diverse backgrounds in order to accomplish collaboration among various experts. This was vital for us since every change and refinement throughout the research process required a final validation by internal and external stakeholders to ensure the team to achieve a comprehensive picture of the project domain.
Fail, iterate and improve
Ensuring mutual support and collaboration is a challenging task without pre-defined rules. For this reason, in order to define the perfect collaborative methodology, it is necessary a process of test and improvement requiring a consistent effort in terms of time and engagement. It’s crucial that this aspect must concern every team member, not applying a defined hierarchy but rather sharing openly professional resources and suggestions.
Inclusion and co-design lead to an unbiased AI
AIDA has the purpose of creating an AI solution that does not intake the biases of different actors interacting with children with ASD. In other words, the solution should be based on a rich, blended database that weighs the various perspectives of parents, educators, and therapists. In this sense, the extensive stakeholders’ involvement was pivotal to enable the team to analyze the autism topic from different perspectives, collecting a considerable amount of insights about the experience of multiple actors.
The qualitative insights collected during the above-described process have now to be deepened through a quantitative research aimed at creating an AI dataset. This step will require the definition of a new design research methodology that supports the creation of such an outcome.
Additionally, in the next phase, our collaboration with the AI team will be even more central to collect and analyze a significant amount of data.
Want to discover more on the project? Check out the case study.
User Research and Analysis