SOLARIS : Strengthening democratic engagement through value based generative adversarial networks

生成对抗网络（GANs）是一类能够创建类似现实的媒体内容（音频和视频）的人工智能模型。虽然gan有不同的应用领域，如视听制作、人机交互、讽刺、艺术创造性表达，但它们目前和可预见的误导用途同样众多，令人担忧。主要的担忧与所谓的“深度造假”有关，即以极高的精度模拟真实事件的虚假图像或视频。如果对一张脸进行训练，甘斯可以让它以一种超现实的方式移动和说话。这项技术带来了紧迫的政治威胁，因为gan可能——而且已经——被用来传播假新闻和虚假信息。这对民主治理和监管提出了紧迫的挑战：提高政府间机构的问责制、透明度和可信度。然而，gan也构成了一个提高民主意识和扩大积极和包容的公民身份的机会。SOLARIS以两种方式应对这些挑战。一方面，我们分析了与这些技术相关的政治风险，以防止对欧盟民主国家产生负面影响。因此，solaris将建立监管创新，以检测和减轻深度造假风险。另一方面，我们评估了ganses为重振公民的民主参与所带来的机会。我们将共同创造涉及公民科学、基于价值的gan内容，以加强民主参与。SOLARIS涉及三个用例：第一个目的是理解感知可信度的心理方面。第二种方法模拟威胁gan内容在社交媒体上的传播，以检测风险并设计缓解策略。第三，共同创建基于价值的gan内容，以提高对关键全球民主主题（如气候变化、性别维度、人类移民）的认识，最终增强积极和包容的数字公民身份。

Generative adversarial networks (GANs) are a class of AI models able to create media contents – audio and video – resembling reality.Although there are different promising areas of application of GANs – e.g. audio-graphic productions, human-computer interactions,satire, artistic creative expression – their current and foreseen misleading uses are just as numerous and worrying. The main concern isrelated to the so-called “deepfakes”, fake images or videos simulating real events with extreme precision. If trained on a face, GANscan make it move and speak in a hyper-realistic way. This technology poses an urgent political threat since GANs could be – and havealready been – used to spread fake news and disinformation. This raises an urgent challenge to democratic governance and regulation:to improve GANs accountability, transparency, and trustworthiness. Nevertheless, GANs also constitute an opportunity to enhancedemocratic awareness and expand active and inclusive citizenship. SOLARIS reacts to these challenges in two ways. On the one hand,we analyse political risks associated with these technologies, to prevent negative implications for EU democracies. As a result, SOLARISwill establish regulatory innovations to detect and mitigate deepfake risks. On the other hand, we assess the opportunities raised by GANsfor reinvigorating the democratic engagement of citizens. We will co-create, involving citizen science, value-based GANs contents toenhance democratic engagement. SOLARIS involves three use cases: the first aims at understanding the psychological aspects of GANsperceived trustworthiness. The second simulates the circulation of threatening GANs contents on social media, to detect risks and designmitigation strategies. The third co-creates value-based GANs contents to boost awareness on key global democratic topics (e.g: climatechange, gender dimension, human migration), to ultimately enhance active and inclusive digital citizenship.