Keywords:-
Article Content:-
Abstract
Generative artificial intelligence models, such as GPT-4, DALL·E and Stable Diffusion are now essential tools for automated text and image production. However, these models are influenced by algorithmic biases resulting from training data , learning mechanisms and user interactions . These biases, often unconscious, can have significant consequences by reinforcing social stereotypes, excluding certain populations and altering the diversity of generated content.
This article provides an in-depth technical analysis of the biases present in generative AI. We first explore their origins , highlighting the biases of corpora training , biases introduced by learning algorithms and those induced by users . Then, we present methods for detecting and evaluating biases , using natural language processing (NLP), computer vision and data modeling tools . Experiments in Python illustrate how these biases manifest themselves in text and image models.
Finally, we propose bias mitigation strategies based on several technical approaches: data rebalancing , embedding debiasing , adjustment of cost functions , regulation of outputs And Model auditability . Integrating these techniques helps make AI models fairer and more transparent. The goal is to provide a pragmatic and rigorous approach to designing responsible generative AI models that respect the principles of fairness and diversity.
References:-
References
Bender, E.M., Gebru , T., McMillan-Major, A., & Shmitchell , S. (2021). On the dangers of stochastic parrots: Can language models be too big? Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, 610-623.
Bolukbasi , T., Chang, K. W., Zou, J. Y., Saligrama , V., & Kalai , A. T. (2016). Man is to computer programmer as woman is to homemaker? Debiasing word embeddings . Advances in Neural Information Processing Systems, 29, 4349-4357.
Gehman , S., Gururangan , S., Sap, M., Choi, Y., & Smith, N.A. (2020). RealToxicityPrompts : Evaluating neural toxic degeneration in language models. Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), 3356-3369.
Joshi, P., Santy, S., Budhiraja , A., Bali, K., & Choudhury, M. (2020). The state and fate of linguistic diversity in the NLP world. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 318-329.
Mitchell, M., Wu, S., Zaldivar , A., Barnes, P., Vasserman , L., Hutchinson, B., & Gebru , T. (2019). Model cards for model reporting. Proceedings of the Conference on Fairness, Accountability, and Transparency, 220-229.
Radford, A., Wu, J., Child, R., Luan, D., Amodei , D., & Sutskever , I. (2019). Language models are unsupervised multitasking learners . OpenAI Blog.
Vaswani , A., Shazeer , N., Parmar , N., Uszkoreit , J., Jones, L., Gomez, AN, Kaiser, L., & Polosukhin , I. (2017). Attention is all you need. Advances in Neural Information Processing Systems, 30, 5998-6008.
Zhao, J., Wang, T., Yatskar , M., Ordonez, V., & Chang, K.W. (2017). Men Also Like Shopping: Reducing gender bias amplification using corpus-level constraints. Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP), 2979-2989.