SOLARIS – a project attempting to understand deepfakes

The project aims to create workable policy propositions to push back against machine-generated deep fakes, that threaten the stability of EU democracies

On 7 February, the SOLARIS project kicked off in Amsterdam. The project was funded by Horizon Europe to the tune of 2.6 million euros and it combines the efforts of 12 partners from 8 countries, dedicated to researching the political risks of deep fakes and fake news, generated by machine learning technology.

The research is headed by Federica Russo, a philosophy professor at the University of Amsterdam, but it also involves the Exeter Univeristy, Universidad Carlos III de Madrid, Lumsa University in Italy and the University of Maribor in Slovenia.

The project focuses on Generative Adversarial Networks (GANs) - a class of neural networks capable of creating multimedia content that simulates real content with extreme precision. If a GAN is sufficiently trained on a face, it can make it move and speak in a way hardly distinguishable from real life.

With the rise of access to AI tools, the rise of fake content and algorithm-generated news is becoming an increasing threat to democracy and the well-being of European nations, according to a statement by Amsterdam University.

Infodemic

According to Professor Russo, citizens are exposed to an unimaginable amount of information on the internet at all times. She went as far as to call the situation an ‘infodemic’ - too much information including false or misleading information in digital and physical environments. It also leads to mistrust in health authorities and undermines the public health response, according to a definition by the World Health Organisation, although that was primarily used to describe the information environment during COVID-19.

This is more of a constant state now and it also leads to the spread of conspiracy theories, and social groups become more polarized due to echo chambers in social networks. Moreover, she pointed out that it is often very hard to distinguish between true facts and false or biased ones, without a very specific skillset, making the reliability of any piece of information uncertain.

However, it might turn out that the technology of GANs is useful for generating convincing and value-based content, which could be mobilised to strengthen our democracies. SOLARIS is on a mission to find out.

Studying the effects of deepfakes

The project has three broad objectives during its run. The first is understanding the political risk of GANs as well as understanding their socio-political impact. Understanding is the first set to be able to provide useful policy mechanisms that counties can actually implement.

Consequently, co-creating policy options and regulatory innovations to mitigate the risk of GANs is the second objective of the project. This step is also supposed to produce concrete evidence-based suggestions for lawmakers.

The third objective focuses on channelling the potential of GANs to do good, broadly speaking. Project participants, as well as citizens, will be able to get involved and create value-based GANs to increase awareness of key global issues like climate change, gender and migration.