Yesterday our AI Scientist Erlin Gulbenkoglu took the stage at the Europe’s biggest Data Protection congress IAPP Europe. The discussion entitled “A Day in the Life of an AI Project: Privacy Fields Forever?” included three speeches focusing on the status of regulation on AI and privacy, data protection support model for AI projects.
Erlin is a machine learning expert with a particular interest on explainable AI and data privacy. She has a track record of working with various companies in various industries as an AI expert to help find AI use cases and implement AI solutions.
In her talk, Erlin focused on practical examples of AI projects from her work at Silo.AI and how, when and where privacy could be embedded in projects like these. The other speakers were Monika Tomczak-Gorlikowska, Group Data Privacy Senior Legal Counsel at Shell International Limited, UK and Tobias Bräutigam, Senior Counsel, Bird & Bird, Head of Data Protection, Helsinki Finland.
The discussion was formed around these six phases of an AI project: Scoping, Identifying data sources, Data preprocessing, Modeling, Deployment and finally Requests of Data subjects.
Especially Erlin’s suggestions on data minimization, fulfilling the requirement for the right to get an explanation and right to be forgotten, and having a human-in-the-loop were crucial when thinking about building AI solutions.
Erlin, why are these rights important in particular?
– There has been a lot of discussions on ethical and legal issues related to the use of AI. GDPR also brings up some of these concepts like “right to get an explanation”, “privacy by design”. The phases we demonstrate try to address the requirements/recommendations by the lawmakers and demand from the society.
How can data scientist know about these requests and take them into account?
– Some of these concepts are not easy at all to adopt in AI systems. But in our presentation, we present two projects run by Silo.AI which addressed different aspects of privacy by design concept in AI development to provide examples from real AI projects. I believe most of the data scientists have encountered questions about AI ethics or a related topic. Being aware of these discussions and bringing such discussions to their workplace is very important. Then, to bring those discussions into reality by utilizing methods developed to help with these requirements in the projects they work.
What is the current trend around privacy in AI projects? Is it getting better or worse, what’s your take on the future?
– We can observe that in the top AI conferences, there has been much more interest in related topics like fairness, privacy, model explanation. Researchers propose methods to address the requirements/recommendations by the lawmakers and discuss the limitations of fulfilling the requirements.
Erlin has previously evaluated the concept of privacy by design from the perspective of AI projects. Read more about her thoughts here.