• AANYA DABI

Generative AI models are becoming integral in shaping decisions and societal perceptions. However, these models are trained on extensive datasets that are historically biased, overrepresenting certain communities while underrepresenting others. This paper explores the paradoxical challenge of addressing bias in such models without resorting to data manipulation or deletion, which could compromise authenticity and historical integrity. Through the generation of 840 images across seven occupational prompts using Meta AI, this study highlights disparities in racial and gender representation, revealing systemic inequalities embedded in generative outputs. Professions like nursing disproportionately depict white women, while blue-collar roles exhibit relatively diverse representation, though rooted in racial stereotypes. Addressing these biases requires balancing historical awareness with equitable representation. Potential solutions include incorporating human oversight, leveraging data preprocessing and reprocessing methods, and designing algorithms that maintain data integrity while mitigating bias. This research underscores the importance of navigating the intersection of historical accuracy and equitable representation, offering actionable steps to build inclusive AI systems. The findings emphasize that achieving fairness in AI is not merely a technical challenge but a socio-ethical imperative

AI DATA BIAS IN GENERATIVE MODELS: NAVIGATING THE PARADOX OF DATA MANIPULATION VS. OMISSION

"AI DATA BIAS IN GENERATIVE MODELS: NAVIGATING THE PARADOX OF DATA MANIPULATION VS. OMISSION", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.11, Issue 12, page no.g500-g510, December-2024, Available :http://www.jetir.org/papers/JETIR2412657.pdf

"AI DATA BIAS IN GENERATIVE MODELS: NAVIGATING THE PARADOX OF DATA MANIPULATION VS. OMISSION", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.11, Issue 12, page no. ppg500-g510, December-2024, Available at : http://www.jetir.org/papers/JETIR2412657.pdf