1. Besides medical algorithmic audits, are there some additional ways to improve the safety of medical AI systems?
= Besides what classmates mentioned, collecting large amounts of high-quality, real-world data can sometimes be challenging for training AI systems. Therefore, another way to improve the safety of AI systems would be to generate synthetic data that can be used to provide more varied and plentiful datasets. This method enhances the safety of medical AI systems by expanding data availability, decreasing human biases, and safeguarding patient privacy (because it does not include actual medical information). There are, however, limitations as well. As synthetic data is not derived from real-world sources, it may not encompass all potential outcomes, particularly uncommon or unexpected scenarios. It also lacks transparency, making it challenging to comprehend the process when the AI decides the results. Therefore, researchers have to balance synthetic and real-world data, and synthetic data should supplement real-world data rather than fully replace it to maintain the quality and reliability of the results.

2. Please name some key/ideal characteristics that medical AI systems can build trust and confidence in the medical community (for example, in terms of safety, quality, efficacy…)
= Model Transparency: The model should explain decisions clearly and allow us to check and understand the decision-making process.
Ethical Considerations: It’s essential to get informed consent from patients and protect their data privacy.