Dominika Dlugosz

Dual process theory asserts that human thought processes can be modelled in terms of a dual system, with one part devoted to fast, automated, associative thinking, and the other to deliberate, effortful, logical inference. However, most Machine Learning models and Artificial Intelligence systems up to date focus exclusively on the first mode of reasoning. In this project, we propose that an AI system built upon a robust dual-process approach can substantially advance AI's cognitive capabilities with respect to the state of the art. In addition, we want to explore a paradigm shift in automated reasoning in which the process is performed in continuous latent spaces rather than token space. Our final point of focus is explainability - a fundamental characteristic necessary for the model to be applicable in critical areas of human activity. We hope that the work in this project will set new directions for future research in cognitive abilities of AI and form a dependable foundation for further developments.