The only way to achieve this goal is by combining multilevel insights from computational neuroscience, biophysics, philosophy of cognitive sciences, AI alignment, machine learning, dynamical systems theory and neuroAI.

Interdisciplinary collaboration network. Bold names correspond to Work Package leaders.
Solid lines correspond to established collaborations and dotted line to intended collaborations.

The MANTIS project aims to address how interaction with the environment is guided by
autonomous, goal-directed behaviour, delving into the agency of artificial and biological systems as a fundamental dimension of cognition. At the intersection of AI, statistical physics, nonlinear dynamics, biophysics, and philosophy, we seek to unravel the cognitive depth of artificial systems. By exploring neural networks and transformer models from the point of view of nonequilibrium physics, the project aims to shed light on the emergent, autonomous cognitive capabilities of these systems.

Are they mere calculative entities, or do they exhibit emerging patterns that mirror aspects of biological cognition? Furthermore, the inclusion of philosophical and ethical considerations in the project underscores the broader implications of AI cognition. What does it mean for a machine to have autonomy? Can we engineer machines with autonomous capabilities but constrained to behave in alignment to human (ethical) values?

The investigation of cognitive autonomy and agency in artificial and biological systems, as well as the assessment of the control and ethical implications of autonomous AI systems, demands an integrated approach that transcends traditional disciplinary boundaries. The complexity and multifaceted nature of this research topic require a highly cooperative collaboration of experts from diverse fields, including artificial intelligence, computational and cognitive neuroscience, dynamical systems, biophysics, statistical mechanics, philosophy of mind, and AI ethics.

The multidisciplinary nature of this project requires a cross-pollination of ideas and methodologies. Our goal is to build a multidisciplinary network through regular interdisciplinary meetings and discussions, in which perspectives from one field inform and inspire new questions and approaches in others, creating a dynamic environment that drives the research forward.