Privacy-Preserving Mechanisms with Explainability in Assistive AI Technologies

With the developing technology and increasing amount of data, Artificial Intelligence (AI) shows the effect in almost every field of our lives. Thanks to AI systems that are revolutionizing the technology world, things only humans could do before can be automated such as visual perception, decision-making, speech recognition, and translating between languages. Even though AI brings to mind the perception of a futuristic future, it also plays a big role in facilitating human life. This role is vital especially for people with disabilities trying to adapt to daily life and be fulfilled with the Assistive Technologies domain. In Model ICT Accessibility Policy Report (Available: https://www.itu.int/en/ITU-D/Digital-Inclusion/Persons-with-Disabilities/Documents/ICT%20Accessibility%20Policy%20Report.pdf. Accessed: 31 Jan 202 [1]), the Assistive Technologies are described as any information and communications technology used to protect, increase, or advance the functional abilities of individuals with particular needs or disabilities. But this brings the potential for simultaneous privacy issues such as spying, exploitation, and data breaches due to context-awareness interfaces and penetrable information everywhere. According to WHO (World Health Organization)’s report on disability, approximately 15% of the world’s population suffers from some form of disability (Summary World Report on Disability, Available: https://apps.who.int/iris/bitstream/handle/10665/70670/WHO_NMH_VIP_11.01_eng.pdf;jsessionid=50C7F4199A25E26711B5A903759B35C6?sequence=1. Accessed: 31 Jan 2021 [2]). Considering that people with disabilities as well as many people in need to use these technologies, we can say that the number of potential users is much higher. Since they are designed to be used by vulnerable individuals with physical and cognitive disabilities, ensuring data privacy is of greater importance. While sensitive data needs to be protected, on the other hand, the use of this data is very critical for the functionality of technology. Besides all these, when decisions of AI systems affect humans’ lives, the need for figuring out how such decisions are taken comes out (Goodman B, Flaxman S, European union regulations on algorithmic decision-making and a right to explanation. AI Magazine 38 (3) 50–57 [3]). Especially with the development of deep neural networks which are considered as black-box models in recent years (Castelvecchi in Nature 538:20–23 [4]), the importance of explainability in Artificial Intelligence systems has come to the fore. In the literature, model explainability and privacy/security by design come together under the notion of Responsible AI (Arrietaa et al. in Information Fusion 58:82–115 [5]). In the scope of this chapter, privacy-preserving solutions for assistive technologies and the relationship between explainability and privacy will be tackled.