سدايا تصدر دراسة عن البيانات المصطنعة: تفيد في تدريب الذكاء الاصطناعي وحماية الخصوصية، بس تحتاج ضوابطSDAIA publishes a study on synthetic data: useful for training AI and protecting privacy, but it needs safeguards

The Saudi Data and AI Authority (SDAIA) has published a study titled "Synthetic Data: Opportunities, Challenges and Governance Mechanisms", according to the Saudi Press Agency (SPA). Synthetic data is data generated to mimic the statistical properties of real data, and it can be used to train AI models.

The study says synthetic data helps with data scarcity and the high cost of collecting and preparing data, speeds up innovation and makes it easier for organisations to share data, especially in sectors that handle sensitive information such as healthcare and financial services. It splits synthetic data into three types depending on how much it relies on real data: fully synthetic, partially synthetic and hybrid. It also says synthetic data can improve the representation of under-represented groups in datasets, which helps reduce bias in the outputs of AI systems.

But the study flags an important point: data that mimics real data does not necessarily remove the risk of re-identifying individuals or of statistical bias. That is why it stresses risk assessments and specialised tests to check data quality and fitness for purpose, and balancing privacy protection against the intended benefit.

The study sets out a roadmap for adopting synthetic data responsibly: defining the purpose, governance, accountability and risk-management frameworks, quality-checking standards, transparency and provenance tracking, and making use of regulatory sandboxes. The full study is published on SDAIA's website.