Contributing to Accessibility Datasets: Reflections on Sharing Study Data by Blind People

Title of the Paper

Contributing to Accessibility Datasets: Reflections on Sharing Study Data by Blind People

Bibliographic Information

• Subject Area: Data Sharing and Privacy Protection in AI and Assistive Technology
• Keywords: Dataset, Privacy, Artificial Intelligence, Disability, Data Ownership

Research Background and Issues

• What problems or challenges did the authors identify?

  • Although assistive technologies can significantly improve the lives of people with disabilities, privacy and ethical issues in data collection and sharing hinder the creation of larger, more impactful datasets. For example, photos taken by blind individuals are often used to train computer vision models, but most current models are built using datasets captured by sighted individuals, resulting in suboptimal performance in scenarios involving blind users.
  • Data sharing may lead to privacy breaches, such as re-identifying users or secondary use of data to support unfair algorithmic applications.
  • There is a lack of transparency in data collection and sharing, leaving contributors with little knowledge about how their data will be used and the potential impacts.

• Why is this issue important?

  • If large-scale AI systems fail to include data representing the blind community, the technology may not only fail to help but could also cause harm during use.
  • Research that provides more responsible and inclusive practices for data collection and sharing can enhance the blind community's acceptance of data sharing while promoting technological fairness.

• Motivation and related work:

  • It is undeniable that advancements in AI rely on high-quality, diverse datasets. Based on previous studies, the authors found that privacy and ethical concerns significantly hinder participation from the disability community in data contribution.
  • This study aims to explore the perspectives of the blind community on data contribution and sharing by working directly with them, thereby providing theoretical and practical guidance for data practices.

Solutions

• What methods or solutions did the authors propose?

  • Designed and implemented a series of research experiments centered on the blind community, including data collection activities and semi-structured interviews.
  • Employed a contextual research framework by deploying AI-assisted applications in real-life scenarios within the homes of blind participants to explore the risks and benefits of data sharing.

• What is innovative about this solution?

  • Placing data contributors at the center of discussions on data sharing, inviting them to reflect on issues of data privacy and usage in real-world contexts.
  • Interdisciplinary integration, applying social science research methods (e.g., semi-structured interviews, contextual probes) to study ethical and fair AI data practices.

• What are the implementation steps? What key technologies were used?

  1. Designed a two-part study for the blind community, including data collection and reflection.
  2. Data Collection Phase: Participants used an AI system equipped with trainable object recognition functionality via smart glasses to capture object photos and refine the model.
  3. Data Reflection Phase: Conducted Zoom interviews where participants shared their views on data sharing, analyzing their attitudes toward potential uses of research data, including scenarios of open sharing and authorized access.

Research Findings

• What specific findings were achieved?

  • The blind community highlighted important considerations when weighing the benefits and risks of data sharing, such as the type of recipients, contextual background, and the sensitivity of socio-economic attribute data.
  • While most participants expressed a positive attitude toward data contribution and were willing to share data to improve technology, they showed distrust toward corporate usage intentions and expressed concerns about sharing certain metadata (e.g., age, gender, address).

• What advantages does it have compared to existing solutions?

  • Greater focus on the participation and reflection of the specific community affected by sharing practices, rather than merely treating target data as research resources.
  • Provides a clear and detailed research framework that promotes reproducible study designs in other fields, enhancing ethical transparency and fairness in various data-sharing scenarios.

• What were the experimental or evaluation results?

  • Many participants believed that data sharing is crucial for driving technological advancements, particularly in improving assistive technologies for daily use.
  • Participants expressed significant concerns about sharing sensitive data (e.g., medication, personal documents, or family background information) and emphasized the need for data controllability and transparency.

• Limitations and Future Directions:

  • Limitations: The sample size was small and limited to the blind community, and participants' trust in the research team or institution may have reduced their concerns about data sharing.
  • Future Directions: Explore attitudes toward data sharing in broader disability communities; design and promote ethical frameworks that support fairness and inclusivity; develop technologies capable of automatically detecting private information in images to reduce privacy risks in data contribution; further study personalized processes and feedback mechanisms in data-sharing decisions.