Artificial intelligence (AI) is increasingly being investigated and adopted for medical applications, including the analysis of diagnostic images, electrocardiograms, and other clinical data. These technologies offer opportunities to support healthcare professionals in detecting abnormalities and making informed decisions. However, their effectiveness depends not only on technical performance but also on how reliably they work across different patient populations. One important challenge is algorithmic bias: systematic differences in an AI system's performance that may disadvantage particular groups of patients. How Does Bias Enter Medical AI? AI models learn patterns from the data used to develop them. If these data do not adequately represent the population in which the system will be used, the resulting model may perform less accurately for certain groups. For example, a dermatological AI system developed primarily using images of lighter skin tones may not recognize some skin conditions as reliably in darker skin tones. Similar concerns can arise when datasets underrepresent particular age groups, sexes, ethnic backgrounds, or clinical settings. Bias can also originate from factors beyond the training dataset. Differences in image quality, measurement devices, clinical practices, diagnostic labelling, and the way patients access healthcare can influence the information available to an algorithm. Consequently, high overall accuracy does not necessarily mean that a system performs equally well for every patient group. The Engineer's Role in Building Fairer Systems Addressing algorithmic bias requires a systematic approach throughout the development and deployment of medical AI. Engineers and researchers should seek appropriately representative datasets, evaluate model performance across relevant patient subgroups, and investigate differences in sensitivity, specificity, and other clinically meaningful measures. Validation using data from healthcare settings beyond those used for model development is also important. In addition, explainability methods may help clinicians understand which features influence a model's predictions, although an explanation generated by an AI system does not, by itself, establish that the prediction is correct or free from bias. Monitoring must continue after deployment. Patient populations, clinical practices, and data sources can change over time, potentially affecting model performance. Clear documentation, appropriate human oversight, data governance, and mechanisms for reporting problems are therefore important elements of responsible implementation. From Technical Performance to Ethical Responsibility Biomedical engineering combines engineering principles with an understanding of human physiology, medical technology, and clinical needs. As AI becomes more integrated into healthcare, this interdisciplinary perspective places biomedical engineers in an important position to evaluate not only whether a system works, but also for whom it works and under what conditions. Developing trustworthy medical AI requires more than achieving a high accuracy score. It requires evidence that the technology is sufficiently reliable for its intended clinical use and that its limitations and potential risks are understood. Ultimately, fairness in medical AI is both a technical challenge and an ethical responsibility. By prioritizing rigorous validation, transparency, and patient safety, biomedical engineers can contribute to diagnostic technologies that better serve diverse populations.