Jerry Petersen
General Dynamics Information Technology Vice President of AI and Data Insights Dave Vennergrund and Vice President of Cyber Matthew McFadden believe that government agencies seeking to ensure the safe and secure implementation of artificial intelligence should work to adopt four best practices based on the National Institute of Standards and Technology AI Risk Management Framework.
First, agencies should work to ensure the trustworthiness of their AI systems by promoting a culture of risk management, Vennergrund and McFadden said in an opinion column posted Wednesday on the GDIT website.
Second, agencies should formulate data management and governance policies that take into consideration the data to be used for building models as well as the data that the models will generate. The same policies must account for different AI use cases.
Third, agencies should frequently and continually analyze and track risks, going beyond traditional security testing and working also to monitoring for data drift as well as outlier behavior.
Finally, agencies should subject their AI models and applications to vulnerability and other types of testing, with the aim of uncovering anomalies under normal and abnormal situations. Tests should be prioritized based on the risks the organization is likely to face.
