Co-PI: Concetto Spampinato (Università degli Studi di Catania) e Aldo Gangemi (CNR)
Human intelligence is capable of learning transparently, continuously and efficiently with limited supervision, adapting to the environment while interacting with other agents. In order for AI to tackle such an ambitious goal, it is necessary to design and develop perception-action detection systems that go beyond the paradigm of mere “imitation by inspiration”. These systems must integrate learning, planning and discovery by rigorously mimicking their biological counterparts across multiple scales, from single neurons to more complex brain networks, to cognitive and social mechanisms. These are the key characteristics that sustainable bio inspired AI systems will need to learn, exactly as humans do, producing efficient (from a data/power consumption perspective) and robust results. The dimensions of efficiency and robustness stand in stark contrast to the recent generation of AI systems whose success has come at the expense of consuming an unsustainable amount of resources and often being fragile.
