ML Co-Design Toolkit

< Back

ML Co-Design Toolkit (coming soon)

Hello,

Thank you for your interest in our toolkit!

The application of machine learning (ML) for patient well-being continues to be a mysterious black-box, with patients often being left unaware that ML is being applied to their medical data and care. Demystifying ML, for patients and end users in general, requires that the field forges the necessary tools to discuss ML and its essential properties with lay people who may not have a strong statistical or ML background. With this in mind, our research team has developed a scalable toolkit for engaging the general public in the co-design of ML that can be applied across various contexts beyond healthcare and offer it as a resource for other research to apply in their work.

The toolkit centers on five critical layers of ML: INPUT, OUTPUT, FIT IN DAILY LIFE, EXPLANATIONS, and UNCERTAINTY. The overarching objective of each layer is to facilitate accessible ML communication through tailored scenarios. Sequentially, these scenarios act as a platform to elicit questions and gather feedback from participants.

GOAL OF EACH LAYER:

  • The goal for the input layer is to capture concerns or comfort levels with input devices (e.g., sensor and survey technology) used to collect data to train the ML model, along with stressors that participants experience.
  • Participants are presented with the output layer to understand their existing notions about engaging with AI and uncover what they deem as dealbreakers.
  • Citing the importance of understanding how the tool fits into the end user’s lifestyle, we include a layer that assesses how the intervention would fit in daily life.
  • In the explanations layer, we probe participants for their preference of global and local explanations. We capture preferences for content and style of communication the user would like from such explanations of why AI might make a particular prediction, as well as their interest in counterfactuals.
  • Finally, we discuss uncertainty with the participants and focus on understanding how the participants perceive uncertainty and risk in ML tools. In this layer, we capture their reaction to false positives and negatives.

Please check back in mid January 2024 when the complete toolkit will be made available to other research teams! We’re hoping our toolkit can be applied across a variety of application areas by other teams aiming to engage end users in the design of ML tools.

 

Sincerely,
The team working on Co-Designing Patient-Facing Machine Learning for Prenatal Stress Reduction