The goal of the EU H2020-FETOPEN project Epistemic AI is to create a new paradigm for next-generation AI providing worst-case guarantees on its predictions thanks to a proper modelling of real-world uncertainties.

Although artificial intelligence (AI) has improved remarkably over the last years, its inability to deal with fundamental uncertainty severely limits its application. The Epistemic AI project re-imagines AI with a proper treatment of the uncertainty stemming from our forcibly partial knowledge of the world. As currently practiced, AI cannot confidently make predictions robust enough to stand the test of data generated by processes different (even by tiny details, as shown by ‘adversarial’ results able to fool deep neural networks) from those studied at training time. While recognising this issue under different names (e.g. ‘over-fitting’), traditional Machine Learning struggles to address it in non-incremental ways. As a result, AI systems suffer from brittle behaviour, and find difficult to operate in new situations, e.g. adapting to driving in heavy rain or to other road users’ different styles of driving, e.g. deriving from cultural traits. The project reimagines AI through a proper treatment of the uncertainty stemming from our forcibly partial knowledge of the world. Epistemic AI’s paradoxical principle is that AI should first and foremost learn from the data it cannot see.

E-pi website