Videos and images from the public space are helpful – for example, for steering passengers at train stations. This data is also essential for the training of AI systems. However, recorded persons must be anonymized for data protection reasons. bright AI has developed a software that makes this extremely realistic thanks to AI. Standardization is helpful for the start-up in many places, as CEO Marian Gläser explains in an interview.
Mr. Gläser, in order for companies to be able to legally use recordings with personal data, this data must be anonymised. What makes bright AI's solution different from other anonymization software?
Marian Gläser: Many other providers pixelate faces and license plates on videos and pictures or blur them. In contrast, our "Deep Natural Anonymization" provides people with artificial faces in pictures and videos. The original facial data is extracted using AI and replaced by artificial image data. And a license plate becomes one with different sequences of letters and numbers.
What's the advantage?
Our images remain more realistic, which makes the data much more valuable. For example, if a pedestrian's face is captured and pixelated in a classic way, information such as age and gender is lost. But also the direction of view, facial expressions and attention to a situation. Our solution preserves all this without being able to draw conclusions about the real person. It is precisely this information that can be relevant for the analysis of traffic situations. And to optimise systems in cars, for example, an autonomous vehicle must recognise whether it has been perceived by the pedestrian or not.
They are involved in AI standardization. What role do norms and standards play for bright AI?
Standards and standards give us orientation in terms of data protection, they set the framework where the legal formulation stops. In addition, GDPR compliance must be explainable at the application level. Standards support us in doing so because they translate legal regulations into more understandable recommendations. This uniformity also helps in cooperation with partners. In addition, standards and standards make suppliers comparable worldwide. So being certified according to a globally recognized standard such as the DIN EN ISO/IEC 27001 series can be a competitive advantage if it is not the competition.
Your start-up is a member of the AI Federal Association and the AI Committee at DIN, what is important to you when working?
AI needs interfaces for different systems. In standardisation, these are defined together with other stakeholders. That's where we want to be. And there is another overarching advantage: standardisation offers us the opportunity to create awareness of our issues in politics. That means we can get involved on a larger scale. It's often more difficult as a lone warrior.