An epistemic artificial intelligence

Epistemic artificial intelligence

Although artificial intelligence (AI) has achieved enormous success over the last decade, its inability to deal with severe uncertainty limits its future evolution. While recognising this issue under different names, traditional AI seems unable to address it in radical ways. As a result, even state-of-the-art AI algorithms find it difficult to operate in new situations and environments.

Epistemic AI, which focuses on how knowledge is acquired and used within AI, is the focus of a major venture being coordinated by Professor Fabio Cuzzolin. The Horizon 2020 Epistemic AI project, is a €3M Future Emerging Technologies (FET-Open) project involving top-tier European universities such as TU Delft (Netherlands) and KU Leuven (Belgium).

The project aims to re-imagine the foundations of artificial intelligence by modelling the ‘epistemic’ uncertainty stemming from a machine’s partial knowledge of the world, and to develop a mathematical framework to help them manage this uncertainty as effectively as possible.

It sets out to create new ‘epistemic’ learning settings spanning all the major areas of machine learning: unsupervised learning, supervised learning and reinforcement learning.

Epistemic AI is the flagship project of Cuzzolin’s Visual Artificial Intelligence Laboratory, a fast-growing research unit comprising 35+ researchers and external collaborators, conducting research on AI, computer vision, robotics, autonomous driving and mathematical statistics.

The concept

While traditional ML learns from the (limited) available evidence a model able to describe it, with limited power of generalisation (see the figure below, left), epistemic AI (right) starts by assuming that the task at hand is (almost) completely unknown, because of the sheer imbalance between what we know and what we do not know. Anything is actually possible. Our ignorance is only?tempered?in the light of the (limited) available evidence to avoid ‘catastrophically forgetting’, to use a trendy expression, how much we ignore about the problem or, in fact, that we even ignore how much we ignore. Mathematically, Epistemic AI’s principle translates into seeking to learn sets of hypotheses compatible with the (scarce) data available, rather than individual models. A set of models can provide, given new data, a robust set of predictions among which the most cautious one can be adopted, thus avoiding catastrophic results.

Illustration of the concept of epistemic artificial intelligence. Epistemic AI’s notion of learning (right), as opposed to that of traditional machine learning/artificial intelligence (left).

Epistemic AI’s consortium uniquely combines 3 academic partners from 3 different countries, ensuring a large network for dissemination and significant geographical outreach. Whereas each Partner excels in one or more areas (e.g. KUL in robust optimisation), the consortium assembles all the crucial ingredients (technical expertise,?business and market experience, expertise in end-user domains) to make the introduction of an Epistemic AI finally possible. Several sub-areas are suitably covered by multiple units, such as uncertainty theory (KUL, OBU), AI and reinforcement learning (OBU, TUD), autonomous driving (TUD, OBU). The partners will also prove science-to-technology transfer capability in the intended technology cases, and a living lab for innovation in the field of AI for autonomous driving, thanks to their proven expertise and past track record on this. The project leader (OBU) has prior working relations with all the partners, and demonstrable past experience to qualify to coordinate the project.

Please watch below a brief presentation of the project, given to the University of Luxembourg earlier this year.

For more information, please consult our project website: