Who’s NGI: Aleksandar Jevremovic introduces Casper to protect children online

Project Casper is designed to protect children from online threats. We are trying to create an artificial supervisor that protects children when they are using the Internet from pornography, nudity, predators, cyber bullying, indoctrination, Internet challenges, involvement in criminal activities, etc. Our project uses artificial intelligence (AI) at the human-computer interaction level, while preserving privacy.

My wife and I have three relatively young children (13, 11, and 8), and we faced the problem of inappropriate content being aggressively shown to them, in the form of ads, banners and redirections. Personally, my family has the luxury of spending a lot of time with our children, supervising them and protecting them from the aforementioned threats. However, I’m fully aware that many parents are not in the same position, and that many children are much more vulnerable when using the Internet. We are trying to solve that problem.

My team did extensive research on this topic and tried many currently available solutions, but the results were far from perfect. I talked about this with my colleagues, what a ‘perfect solution’ for this would look like and that’s how we got the idea for an ‘AI-powered supervisor’. Professionally, I was also exploring the field for AI in human-computer interaction, due to my position in the Technical Committee on Human-Computer Interaction (TC13) at IFIP/UNESCO. However, we needed a more formal approach, to have more researchers/developers and more commitment – to be able to materialize the idea. We had a concept, but NGI helped us to prove it.

NGI Support

My colleagues found the information about the first NGI Trust call and suggested that we should write a proposal. I was a bit reserved at first, since we were not very experienced in writing proposals for EU-funded projects, and I was afraid that funding would require us to change the research direction. However, the call looked very interesting, as well as the whole NGI initiative, so we decided to give it a try. We are very happy we applied.

The most obvious way that NGI supported our idea is by funding our work. It allowed us to have the researchers and developers able to invest more of their time in this project. Besides that, people from NGI organized our participation in many great events, like the Web Summit in Lisbon, or the Internet Governance Forum in Berlin, to present the project and get relevant feedback. Some great training and mentoring options were available through NGI Tetra. Also, we got very useful advice and information directly from the guys at NGI. We probably wasted some of their time with some very basic questions, but they were too kind to tell us that. Last but not least, for us, it was very important that people from NGI recognized our idea and supported it – it really boosted our self-confidence.

Project Casper

There are many projects with the same goal – to protect children using the Internet. Also, some of them are utilizing AI. The main characteristics of our project are that it uses AI on the human-computer interaction level, supports different types of content (image, video, audio, text), supports a broad and extensible range of threats, and, maybe most importantly, tries to deeply understand what is happening in the intercepted communications. Additionally, we are exploring the possibility to enable collaboration (e.g. in detecting predators), while preserving the privacy of the user.

Initially, the plan was very simple – develop software to enable screen/audio recording, that applies AI algorithms to detecting inappropriate content, and that hides that content or alarms parents. However, it turned out that things are a bit more complicated, and the solution needs to be more sophisticated. Different platforms (Linux, Windows, Mac, Android, iOS) use different user interface architectures, and we are trying to avoid forks. Some AI algorithms were not good enough. We needed much bigger training sets. Hiding inappropriate audio content is impossible, or some significant lag is required. The computing capacity of mobile phones is not sufficient for real-time protection. So, in some ways the problems we faced shaped the idea.

New ideas showed up through feedback, consulting with people from academia and industry. Also, architectural decisions forced us to provide some space for the future and potential ideas and uses. We decided to make the architecture more generic and modular, to avoid architectural technical debt.

We are far from being experts in marketing, but the idea behind the project is relatively simple and understandable, so it is easy to share and promote. My colleagues Milan and Marko joined the management committee of a COST action ‘Transnational Collaboration on Bullying, Migration, and Integration at School Level’, so they are able to share information and get feedback from researchers in social sciences.

We plan to attend some boot camps organized by NGI Tetra, in order to improve our knowledge in fields like intellectual property management, getting funding, etc. We are good at research, and we value free and open-source software, but we still have to learn a lot about business.

The NGI Community – and beyond

There is great enthusiasm and energy in the NGI community. My impression is that people from NGI are generating more opportunities than the target population can follow. Almost every day you can see something new published via LinkedIn, the website or other channels. And, that’s not a common thing these days.

NGI has a great combination of professionalism, precision, spontaneity, and improvisation. You can see this, for example, in how quickly they overcame the problems of events canceled due to the Coronavirus. You understand that you are a part of a much bigger idea/plan/project, but you have freedom to explore and include new things. I guess that is very important to attract relevant types of people that could generate new ideas, but also materialize them.

We created a new research framework in this field, so the research part will definitely continue to exist, as well as some sub-projects. However, our task is to find ways to continue the overall project, since it is integrating all these components, preserving the direction, and speeding up the development.

We opened Pandora’s box, and it would be very hard to close it now. Definitively, we plan to continue our work regardless of the end of support from NGI. I’m not sure about the final shape it will take, but things will not end, for sure.

About the author

Aleksandar Jevremovic is a Professor at Singidunum University in Serbia where his team created one of the first undergraduate study programs related to computer security in Southeastern Europe, a researcher at the School of Electrical Engineering in Belgrade, a guest lecturer at Harvard University in the USA, and a visiting scholar at Cyprus University of Technology.

Project Caspar has four other project members from Serbia (Mladen Veinovic, Milan Cabarkapa, Milos Stojmenovic, and Marko Krstic), and four from North Macedonia and Portugal (Ivan Chorbev, Ivica Dimitrovski, Nuno Garcia, and Nuno Pombo).

Follow Project Caspar on Twitter: https://twitter.com/CASPER_NGITRUST

And on Aleksandar Jevremovic’s LinkedIn account: https://www.linkedin.com/in/aleksandar-jevremovic/