American
International
University-
Bangladesh

Students & Supervisors

Abstract

Artificial Intelligence and Data Science are revolutionizing industries, governments, and everyday life on a scale previously unimaginable. Biases will typically derive from past biases in data sets or technical errors in algorithm development. Unchecked, they can erode trust in AI and widen social inequalities. But as with any innovation, there are nuances to every one. Algorithmic and data bias are most significant because unjust, discriminatory, or unscrupulous decisions from biased AI can have long-lasting implications. Biases happen primarily due to historical bias in the data or technical errors in crafting an algorithm. If not addressed, they can annihilate trust in AI and reinforce social inequalities. The objective of this study is to establish why bias happens in Artificial Intelligence (AI) and how this influences fairness, and to suggest workable solutions on how to advance ethical use of Data Science. With this, AI systems will be able to operate more in line with values of fairness, transparency, and accountability. We were inclusive and interdisciplinary. We began with a systematic review of scholarly literature, industry reports, and policy reports on AI fairness and bias to see what the existing knowledge base was and where the gaps in knowledge were. We employed algorithmic audits of case studies in health diagnosis, recruitment algorithms, and police predictive software to identify and quantify bias. Quantitatively measuring fairness, we used measures like demographic parity that ensures model predictions are well distributed across groups and equalized odds which is a measure of difference in error rates such that models are not predicting against both groups in an unfair manner. Interpretability techniques like SHAP and LIME were used to ensure whether AI reasoning behind decision was correct and investigate hidden biases. Beyond that, qualitative interviews with ethicists, AI practitioners, and policymakers inform us regarding rulemaking regimes, ethics, and governance concerns. Experimental simulations compare biased models with their unbiased counterparts having fairness-conscious algorithms in place so that researchers are able to experiment with fairness gains and accuracy loss trade-offs. Mixed-methodology research with such outcomes renders technically sound as well as socially meaningful results. Our results indicate that most bias in AI is caused by biased training data, ill-specified features, and black-box decision making. For instance, face recognition models trained on lighter faces fail on darker ones, causing racial bias. Fairness-sensitive models such as sample reweighting, adversarial debiasing, and fairness constraints reduced bias with little performance loss. Organizational transparency like bias reports and audit logs increased user trust. Preventing AI bias requires technical and sociotechnical action. Artificial Intelligence bias must be addressed to ensure fairness, equity, and moral responsibility. Discrimination can be avoided through algorithmic fairness, transparency, and control systems. Ethical AI builds trust among users. Fairness-aware AI depends on organizational cultural change and accountability. Multi-stakeholder input, feedback processes, and open reporting help AI align with societal values. Through responsible innovation and ethics, AI can promote human dignity, social justice, and benefit society at large.

Keywords

Publication Details

• Type of Publication:
• Conference Name: 7th International Conference on Integrated Sciences (ICIS 2025)
• Date of Conference: 25/10/2025 - 25/10/2025
• Venue: Eastern University Campus, Ashulia, Dhaka, Bangladesh
• Organizer: Eastern University, Bangladesh