Synthetic data has emerged in the past decade as a new solution to privacy concerns. While the collection of personal information from records and online activities remains a serious concern, this type of data can serve as a useful tool to reduce the risk of identifying individuals.
Synthetic Data: An Opportunity to Alleviate Concerns
Synthetic data is generated computationally and mimics the statistical patterns of real data to provide useful information without putting individuals' identities at risk. For example, in the UK, researchers can use synthetic cancer data created from NHS registration information without accessing actual patient records. However, the existence of real data is essential for training models that generate synthetic data, which in itself comes with new privacy challenges.
Read more: Marie Ghadrian, a pioneering lawyer and model for women's rights, left a great legacy
Governance Challenges and Responsibilities
Key questions regarding the governance of synthetic data include who designed this data and which populations and behaviors are considered “normal.” These issues become particularly important in the fields of health, welfare, policing, and finance, where synthetic data can influence how disease, vulnerability, and suspicious behavior are defined. For instance, synthetic transaction data is used to develop and test systems that may identify fraud and money laundering.
A report published by the UK financial regulatory body emphasizes that companies must assess and mitigate the risks associated with synthetic data, including risks related to discrimination. Meanwhile, the social risks arising from synthetic data may not simply be a privacy violation but can lead to the creation of an illusion of objectivity.
This type of data can be shaped by human choices that may be discriminatory, as it appears scientific and systematic. If the data-generating model lacks sufficient accuracy, it may overlook the experiences of minorities or exceptional cases.
Ultimately, the governance of synthetic data requires not only transparency but also the ability of institutions to explain why synthetic data is used, what may be distorted, and how it has been validated. This issue becomes particularly significant when data is used for decision-making regarding rights and opportunities or how public money is spent.
The experience of the 2020 exam results in the UK illustrates how a system can create serious errors using non-real data. In this case, the algorithm used to moderate exam scores led to a decrease in scores for high-performing students in low-income areas, and ultimately the government had to revoke this algorithm.
In conclusion, synthetic data should not be viewed merely as a privacy technology but should be considered as part of modern knowledge policies. This data can carry both promise and risk, and it is essential that its governance is conducted properly to prevent negative consequences.
Read more: Trump and Xi must reconsider how to manage AI competition · Robert Malley analysis: Trump has only 3 ways to deal with Iran
