Data Protection and Bias in AI

TheAIScholars

Daniella DUNGU (Université de Bordeaux), Cloé BOEZIO (Sciences Po Paris, Reims)

Daily life / student life / campus

Being university students ourselves, we are well aware of the growing usage of AI in multiple facets of academic life. Because of this, we aim to address potential issues that could arise with excessive AI use, especially in the domains of data collection and biases. We have thus identified two main areas in which students are likely to use AI for: Purely Academic reasons and personal/psychological reasons.

By using AI for these seemingly surface level applications such as academic help, we almost subconsciously let go of personal information (location, cultural context) that could lead to significant issues regarding privacy and data collection Students also tend to use AI for email writing, giving AI a handle on who you communicate with, information about your social circle We have also identified the issue of using AI for psychological reasons, and may also be more prevalent within students especially first years who often find themselves in a new and unfamiliar environment, and may turn to the use of AI in regards to their psyche due to the lack of a solid social group.

We are especially motivated by the broader questions of how students use AI, especially being university students ourselves who have been exposed first hand to the state of AI. We are also interested in knowing other ways students could use AI as compared to the rest of the population as well as the lack of information they may have regarding its dangers. For instance, what percentage of the student population at both of our universities are fully and 100% aware of just how much information we are giving out in seemingly harmless interactions? And how exactly is this information being handled? This is why we are interested in doing more research in this domain and why it matters.

We thought about the Implementation of an "academic" form of AI. Such as Google Scholar but for AI specifically that would require a collaboration with OpenAI/Google, etc. We thought about making an AI tied to Google Scholar, and will only be able to use sources tied to that specific database in order to find credible sources. This academic AI will be available to all high schools and universities and will be preinstalled on all school computers. It will include a limited form of AI functions, be less warm as to mitigate the risk of using the AI as a form of psychologist. We could use this instead of other AI apps (like Notebook LM) because it will actually look for credible sources within an already existing database instead of taking answers from a potentially uncredible source. By tweaking the code, we can ensure that the AI won't just blindly give answers, but reply to your questions with more questions in order to guide you into finding your own answer on your own. Additionally, to reduce the information gap regarding the dangers of sharing your information and data, we could oversee the implementation of a course for learning more about data risks (protection, lack of traceability)