Description of the use case

This post is the first of a series of several posts related to our validation cases or pilots. Starting from the pilot led by Telefónica there will be four different use cases entitled to demonstrate different elements of trust of AI. In this case privacy-preserving AI on the edge and beyond.

What is the pilot for?

Our pilot is about an edge platform that ensures protection of personal data while running AI models that use the personal data. Imagine an autonomous driving AI service that tracks the locations and learns the mobility patterns. The pilot will demonstrate how such models can run while keeping the sensitive personal data in the hands of the users. We are conducting research on the architecture, privacy technologies, and interfaces for hosting the AI models. The platform will free AI model developers from building solutions for transferring and protecting personal data and at the same time will relieve privacy concerns of users.

We plan to take advantage of the telecommunication infrastructure which is deployed at a large-scale, and to test the pilot with real traffic. As various AI services are built upon the infrastructure, it will provide a perfect environment to evaluate the developments.

Why is it useful for TELEFÓNICA?

In principle, our pilot provides the opportunity for the partners who build related components to evaluate them in a realistic setting. Although individual partners would of course evaluate the designs and technologies during the development process, it is still challenging to combine all the related pieces from different partners and to test them together.

Our pilot serves multiple purposes. First, it will give a detailed scenario that helps partners think about how their outputs will be used. Second, for the real demonstration, an infrastructure that allows all the related partners to put their developments together will be provided. Third, the infrastructure will run all the pieces together with real user traffic or endpoint data. This not only provides an opportunity of a comprehensive evaluation but also a way of communicating the potential impact of the developments.

What tangible results are expected from it?

We expect to have various useful results. First of all, we will measure the utility and model performance of the AI services that run upon the platform, and privacy guarantees offered to the users. To reuse the example of autonomous driving service used above, we will validate if any external entity was able to identify the location history of a user, and also if the privacy preserving mechanism had any impact on the quality of the service.

Secondly, such measurements made upon a realistic setting of a telecommunication network, will help related players of the market (e.g., telcos, application developers, etc.) to think of further use cases. We will explore various venues to present the demos and results.