The Machine Vision Game

Making Computer Vision Development Trade-Offs Tangible

Laura de Groot

This project opens up the discussion about the acceptability of a machine vision system during the development phase, using the scan car development process in Amsterdam as a use case. Acceptability is explored based on various trade-offs made during this development phase (e.g., Accuracy versus Interpretability). By providing a tangible approach to explaining and interacting with the system, the design improves the understanding of non-experts citizens about machine vision systems and nurtures a deliberative debate.

The final prototype, a tangible user interface to explore the machine vision model in its system, was evaluated with pre- and post-game knowledge tests and post-game focus groups. The results indicate improvement in participants' subjective understanding, enabling them to form opinions about what is acceptable and creating a shared language. Ultimately, this design contributes to participatory approaches to the responsible design of artificial intelligence (AI) by providing a practical example of how to involve non-expert citizens in the development phase.

There is a need to align technology with its operating context to increase acceptance. During the machine learning (ML) development phase, various trade-offs influence the behavior of a system created. While this might initially seem purely technical, human judgment is involved. This human judgment should involve other (in)direct stakeholders in addition to the usual developers, programmers, and specialists. This means including citizens in the development of scan cars and their machine vision models.

However, this involves discussing complex and abstract concepts that the layperson does not understand, which hinders meaningful conversations. This project provides a tangible representation of machine vision development and related ideas to help citizens inform and articulate their opinion.

In this way, they can be part of the conversation and thus align the technology with public opinion before it is deployed. Their involvement informs and steers the development process of a machine vision model to increase acceptance and legitimize choices made.