Ζητήματα προστασίας της ιδιωτικοτήτας στην τεχνητή νοημοσύνη & στα μεγάλα γλωσσικά μοντέλα

Ασημακοπούλου, Παναγιώτα

Privacy protection issues in artificial intelligence & large language models

Master Thesis

Author

Ασημακοπούλου, Παναγιώτα

Date

2024-09

Advisor

Μήτρου, Λίλιαν

Keywords

Τεχνητή νοημοσύνη ; Μεγάλα γλωσσικά μοντέλα ; Αλγόριθμοι ; Προσωπικά δεδομένα ; Ιδιωτικότητα ; Ρυθμιστικό πλαίσιο ; Γενικός Κανονισμός Προστασίας Δεδομένων (GDPR) ; Πράξη για την Τεχνητή Νοημοσύνη (AI-ACT)

Abstract

The Artificial Intelligence explosion, which includes Large Language Models (LLMs) and related Chatbots, seems to "open a can of worms", posing new challenges and raising serious concerns regarding the protection of personal data and privacy. As these models are trained on vast amounts of data, many questions arise, such as what kind of data is used and collected, how long that data is stored, and whether and to what extent it is being used without our knowledge and approval, ultimately becoming part of a model’s training. In addition, the ability of chatbots to analyze and correlate our various data from different online sources undoubtedly heightens our fear of the existence of various risks, such as that it may be disclosed to unwanted recipients. A typical example of Artificial Intelligence technology is the much-discussed LLM "ChatGPT", which could be said to have been the trigger for the further and substantial protection of personal data and privacy in general, as well as the additional investigation of models, indicating the imperative need for their immediate delimitation. Thus, all possible means are employed, such as the development and implementation of stricter regulations, methods and techniques to ensure data protection in the digital world. In this master's thesis, privacy protection issues arising from the application of Artificial Intelligence are examined, emphasizing Large Language Models. Case studies of these, such as ChatGPT, Claude and Meltemi, are presented, while methods of protecting personal data and privacy are also analyzed through the combination of the regulatory framework and modern techniques, with the aim of strengthening the security and protection of personal data in this field.