Name of the participant: Stefan Mager
Description of the IT research project: The spread of Intelligent Digital Assistants (IDAs), also known as Smart or Virtual Assistants, has increased significantly in recent years, especially in the private sector. However, very few of the speech-based digital assistants are based on the privacy-by-design principle postulated by the European Data Protection Regulation (DSGVO). This is because the most common IDAs currently in use send user data such as voice recordings into the cloud so that the machine learning model on which the “intelligence” of the IDA is based can be trained (“cloud-based IDAs”). For this reason, the use of IDAs is unattractive for users for whom privacy is important. This practice is also unsuitable for companies that want to protect trade secrets or even have to protect the confidential conversations of their customers and employees from unauthorized access for legal reasons.
However, according to current knowledge, IDAs no longer need to send personal user data into the cloud to reach the performance level of cloud-based IDAs in different contexts. “Edge-based” IDAs process personal user data decentrally on the end device – e.g. smartphone, tablet or loudspeaker – and by default do not forward personal data into the clouds of IDA providers. These privacy-by-design IDAs learn either via the Federated Learning Model or a Decentralized Data Generation Concept and also have the advantage of working offline, which shortens their latency, increases their functional reliability and reduces their power consumption.
The aim of the project is both to develop an edge-based IDA prototype for the context of stationary trade and to provide design guidelines on how stationary traders can use IDAs in a legally compliant manner (i.e. taking into account the DSGVO and BetrVG) without losing control over the data collected by the IDAs. An important aspect for the development of the guidelines is to test with the prototype how edge-based IDAs should be used in stationary trade to ensure the acceptance of the technology by employees and customers.
Software Campus partner: LMU Munich, MediaMarktSaturn
Implementation period: 01.02.2020 – 30.06.2021