Arabizi, a hybrid form of Arabic using Latin characters and numbers, is a dominant mode of informal communication among young Arab speakers on social media. However, its lack of formal structure and cultural nuances presents significant challenges for machine translation, especially across low-resourced Arabic dialects. My thesis explores ChatGPT's ability to translate Arabizi into Modern Standard Arabic and English across Algerian, Egyptian, and Lebanese dialects. Using human evaluation and metrics like COMET, BERTScore, BLEU, and TER, the study found that Egyptian Arabizi translations performed best, while Algerian Arabizi posed the greatest challenges. Translations into English consistently outperformed those into Arabic, highlighting resource disparities. The research emphasizes the need for new metrics tailored to Arabizi and underscores the ethical implications of romanizing Arabic. While ChatGPT showcases impressive potential, it struggles with linguistic complexity, paving the way for future improvements in AI-driven translation for underrepresented languages.