The way we learn as humans is fundamentally different from the way we can teach machines to learn. When we train sophisticated models such as deep neural networks to mimic our visual perception we show them a huge amount of complex data such as images and videos, and let these models automatically learn an internal representation of the world from such input. This does not consider that when we, as humans, learn new concepts, we do not only use our visual perception, but we also consider connections to what we already know, information that we have distilled and stored as memories (our knowledge). And as a consequence, our decision making abilities rely on both neural activities (processing what we see, for example) and cognitive processes (linking what we see to what we know).
I want to replicate this into Artificially Intelligent systems by combining neural and cognitive approaches to AI from a holistic perspective, so that one can complement and support the other in different circumstances.
I believe this will not only enhance their ability to learn and reason, but also make them more robust and trustworthy, providing new opportunities for better understanding their outcome in human terms, as well as facilitating detection and mitigation of errors and bias.
