Inclusive Practices for Child-Centered AI Design and Testing

Drawing from our experiences as assistive technology designers and researchers for neurodivergent children (Hayes et al., 2010; Cibrian et al., 2020; Dotch et al., 2023; Escobedo et al., 2012), we have compiled some ideas that we are considering when designing and testing new AI technologies. Though not exhaustive, these ideas can catalyze discussion on designing for neurodivergent child-centered AI technologies.

As mentioned earlier, sensory processing difficulties are often a shared experience among many neurodivergent individuals, which can influence their interpretation and processing of sensory input. One way we think about leveraging AI to accommodate this is to incorporate systems that model and predict stimulation levels to help balance under or over-stimulation. An AI system designed to anticipate stimulation levels could use predictive behavior modeling to understand individual patterns of stimulation to predict and adapt to a child’s sensory processing needs. Features of this could look like adaptive user interfaces that can change colors, contrast, or layout based on the user’s preferences, modifiable sensory outputs like varying levels of visual complexity, or predictive models that understand baseline sensory thresholds and use the data over time to anticipate the child’s sensory needs.

Additionally, to design AI technologies inclusively for neurodivergent children, we believe it’s essential to consider their unique attention spans and engagement strategies. Thus, we think about incorporating interactive AI elements that offer novelty and real-time adaptability; we can create experiences that hold these children’s interest more effectively. Such AI systems could utilize data on neurodivergent children’s engagement patterns to proactively adapt novel stimuli when a child’s attention appears to diminish. For instance, these systems might employ dynamic content presentation that alters the way content is delivered based on the child’s current engagement level, or it could employ adaptive challenges that offer different levels of difficulty and types of problems that adapt to the child’s performance, keeping them challenged but not overwhelmed.

Lastly, neurodivergent children often benefit from environments that provide structure yet flexibility and support that minimizes cognitive load. Integrating adaptive algorithms that tailor the AI’s responses to individual children’s reactions to scaffold these areas can lead to more supportive, effective, and neurodivergent-friendly technologies. AI designed to scaffold cognitive load could include several nuanced features like gentle prompts that remind children of the task at hand, guide them through complex activities, suggest taking breaks to prevent cognitive overload, or structured guidance that offers a clear, consistent routine while allowing for adjustments based on real-time data about the child’s performance and state of mind, or memory aids that could help children track tasks, instructions, and goals, supporting their working memory.

As noted above, these ideas are only an entry point for inclusive AI design for neurodivergent children, urging researchers and designers to discuss further what factors should be taken into account when designing with neurodivergent children in mind. In practice, researchers and designers should often include the perspectives of neurodivergent children in their projects to flesh out these design ideas for specific issues and projects further, as discussed in the following section Accommodations and an overall approach towards inclusivity can broaden the scope of who participates in the design and testing of AI technologies. Highly stimulating environments that might ordinarily be considered to spark creativity or be favorable environments for child designers may contribute to access barriers for neurodivergent children, especially those with sensory sensitivities. Thus, inclusive AI design and testing environments must balance sensory stimulation across environmental/physical factors, social considerations, and design tool choices. Factors such as noise levels, group size, and physical space significantly impact how neurodivergent children with sensory sensitivities may engage in design and testing environments. Strategies such as selecting appropriate workshop locations, minimizing stimulants, and providing quiet rooms or self-regulating tools can enhance inclusivity.

During our projects, we took steps to ensure a sensory-friendly design environment for workshops, yet we still had some shortcomings in this regard. As an example, in one project, we worked with groups of three to four children and chose to host workshops in a conference space without windows to minimize excess light. Additionally, we removed the paintings from the room’s walls to minimize stimulants that may become visually overstimulating. While we were prepared to create a sensory-friendly workspace, one point we did not consider was the sensory triggers that our participants might encounter while getting from their vehicles to the workshop room. Because we hosted our design sessions on a university campus, some participants had to walk through campus and/or along a busy road to reach the workshop destination. Following the workshops, some participants shared that it was overstimulating. More specifically, the crowds of students leaving classes, the sound of cars along the streets, and students zipping by on scooters or bikes were overstimulating. With this in mind, in future gatherings, we may consider hosting design sessions away from campus in settings such as a library or other community-centric locations. From this experience, we learned precautions must be taken to ensure participants reach the necessary destinations with minimal sensory triggers.

Finally, we recommend carefully considering materials and their sensory effects, with options like online platforms (e.g., Miro or Google’s Jamboard) or low-stimulating materials offering viable options. For example, in design sessions we have conducted, we’ve noticed how tools such as markers can make uncomfortable sounds or have strong scents that can become problematic for some neurodivergent children due to the noise and scent of the markers. Therefore, having alternative options, such as crayons or pencils, allowed our participants to engage in meaningful design without the added sensory stressors. With the aforementioned change in materials, we also suggest that designers think of how the various materials we use may interact with one another. Additionally, having designated spaces for individuals to decompress and providing self-regulating tools can further support integrating children with sensory sensitivities in AI design and testing. Providing self-stimulating tools can help participants self-regulate when needed. Due to the subjective nature of sensory sensitivities, not all contexts or triggers can be predicted and accounted for; thus, having a separate space to wind down may be necessary. For our workshops, we utilized a zen booth down the hall from the workshop space and offered fidget pea pods to our participants when they joined the workshop.