Inside Milestone Systems and NVIDIA's Project Hafnia: Pioneering Ethical AI Training
Milestone Systems’ new collaboration with NVIDIA, announced through Project Hafnia, represents a significant step toward responsible and effective AI model training, ensuring data privacy and compliance while advancing technological innovation.
The companies’ aim is to democratize AI model training using high-quality, compliant video data. In an exclusive interview with SecurityInfoWatch, Milestone Systems CEO Thomas Jensen elaborated on the project’s goals and the challenges it addresses.
“One of the biggest challenges with visual language models is to get compliant data,” Jensen explained. “To get data that’s ethically sourced with traceability requirements being fulfilled, knowing where your data came from, are you allowed to use it, how has it been sourced, how has it been trained and so forth. That has proven a general challenge in AI overall.”
Ensuring Ethical AI Model Training
Project Hafnia, named after the Latin name for Copenhagen, leverages NVIDIA’s technology stack, including its NeMo Curator and AI fine-tuning microservices, to ensure data curation and model training are conducted responsibly. “We will only accept data that we are licensed to use in our models. That means we can do conformity testing and secure data lineage,” Jensen stated. "We will not train on models that do not protect your rights as a private citizen.”
The collaboration aims to address the friction between the industry’s push for AI development and the public’s concerns about data privacy. “We see a lot of outcry from academics and private citizens that they’re deeply concerned about this development,” Jensen noted. “We are super focused on ensuring that we will not deviate from our responsible technology principles.”
Project Hafnia's initial focus is on traffic and transportation data, with plans to expand to other sectors. “From a responsible perspective, we are starting with traffic data and transportation data, typically looking at city development,” Jensen said. “Once we have done a full model on city and traffic and transportation, then we will move on to the next use case and expand it.”
Expanding Applications Across Various Sectors
Jensen highlighted the potential applications of visual language model services offered by Project Hafnia. “First and foremost, it will be transportation, road management, traffic management in cities,” he explained. “Secondarily, I think the entire smart city concept will be next. One thing is going to intelligent traffic management and rerouting of emergency vehicles and so forth.”
Other potential applications include hospitality, healthcare and retail. “Hospitality is a good area where you can see leverages from this part,” Jensen said. “You could look at healthcare, which is of course a very prosperous use case, but highly regulated. Retail is always the first to want to adopt new technology but typically the last to really be willing to pay what it costs.”
Jensen emphasized the importance of focusing on specific domains to ensure the development of high-quality AI models. “We've now decided let’s own the domain of traffic data and then we go to the next domain and the next domain,” he said. “We would rather have a highly qualified model that serves the need of a particular segment than to have 10 models that are average.”
Project Hafnia is kicking off with a pilot phase, inviting developers to register for early access here. Initially, the platform will concentrate on traffic video data, with plans to expand its capabilities upon full launch. The platform is owned and managed by Milestone Systems.
