PIA – Privacy Implications of AI-based Personalized Learning Tools

AI-based tools are increasingly used across contexts – including learning environments in which they provide benefits such as personalized learning, but also make use of sensible data related to personal characteristics, competency developments, and grades. This leads to privacy-personalization trade-offs, e.g. also shown for personalization of online content (Garcia-Rivadulla, 2016; Lin and Unger, 2012).

To make the complex privacy implications of AI-based learning tools graspable – as compared to seldom read and complex privacy policies – we provide PIA as a personalized learning tool that provides feedback on privacy implications in an interactive way. Combining the focus areas «Teaching with AI» and «Learning about AI», it enhances understanding and reflection on privacy aspects of AI. In line with ETH’s aim for responsible digital transformation, it also provides students with an option to control the privacy-invasiveness of the tool instead of forcing users into a «agree-as-is or opt-out» decision.

Variable content allows for the integration of PIA in a wide range of ETH courses either focusing on AI or privacy such as in various SiP courses or large courses across all departments aiming to offer a personalized learning tool to enhance the individual experience and exam preparation in a privacy-aware way.

1. The first project aim is to offer students the benefits of personalized learning allowing them to learn flexibly in terms of their current state of knowledge and workload to enhance their learning outcomes.
2. The second aim is to enhance students› understanding and to allow for a critical reflection on privacy aspects of AI by making privacy implications graspable within an AI-based tool that students actually use in a safe learning context.
3. The third aim is to provide students with control over their privacy by offering options to vary the degree of privacy-invasiveness and the related degree of personalization within the tool, i.e. to influence the personalization-privacy trade-off.

Thereby, the project aims to provide a personalized AI-based learning tool that offers the learning-related benefits of personalization while allowing for critical reflection of privacy aspects.

Our AI-based learning tool PIA aims to deliver a personalized learning experience while making the privacy implications of personalization graspable, e.g., through direct feedback in the communication with the user or through highlighting changes in the selection of learning contents based on the chosen level of privacy-invasiveness and the corresponding degree of personalization. The innovation not only lies in delivering feedback on privacy aspects moving beyond difficult-to-understand and seldom-read privacy policies but especially in providing privacy control options. Whereas many tools only provide an option to either accept the conditions or not use the tool, the PIA tool offers varying degrees of personalization and related privacy-invasiveness making the personalization-privacy trade-off transparent and manageable. Thereby, the tool offers students of all degree programs an option to better understand and critically reflect on privacy of AI-based tools within the Science in Perspective program.
In addition, the tool can be used by faculty of all departments within the courses either with the provided privacy-related content or as a privacy-conscious learning tool including course-specific content. This enables faculty to design their classes to include more interactive discussions and complex problem solving activities instead of lengthy lectures, as base knowledge can be more easily acquired by students independently using the PIA tool.