DeepWorld is a compilation of “artificial countries” generated by neural networks which used data of all existing countries (around 195) to generate new anthems, flags and other descriptors. The project is a hybrid of critical reflection on national identities in combination with practical research in deep learning applications, such as Generative Adversarial Networks (GAN) and Recurrent Neural Networks (RNN). This generated ‘world model’ acts less as utopian alternative to our physical world, than as a view of our world from the ‘eye’ of an alternative intelligence.
[audioplayer file=”http://kim.hfg-karlsruhe.de/deepworld-wiki/audio/deepland_anthem1.mp3″ titles=”National Anthem from Itu (generated by melody rnn)” align=”aligncenter”]
National Anthem from Itu (generated by melody RNN)
With the results of this process, we want to reflect on our relation to national identities and to explore the patterns and themes emerging today from broad processes of algorithmic generalisation. With this project we are also investigating the human and social biases that are invisibly passed and re-enconded in artificial networks. Yet by moving within these limits of bias, and actually “borders” of comprehension, we still want to question the Gestalt nation. Does an artificial neural network really recognise patterns of social organisation? Is it possible to create a new design on the basis of given real world data? Is it possible to project some kind of new utopia or are we in danger of being forced into submission by economic efficiency and statistical models?
The project is presented in the form of an online encyclopedia.