Chair of Communications Engineering Prof. Dr.-Ing. habil. Matthias Wolff

C. E. Shannon's  "Electrically Controlled Mouse" Theseus is an early and prominent example of what we call a cognitive or intelligent agent today. Shannon's construction was based on an array of telephone relays which provided the mouse with a simple memory, much like Ariadne's thread, to find its way through a configurable maze. Contemporary cognitive agents use state space models in connection with search algorithms and decision process models to solve similar problems. They have deficits in handling not-yet knowing in terms of formal logic (e.g., stochastic decision processes) and with explainability (e.g., deep neural networks). Both capabilities are, however, essential for cognitive agents. In this project we investigate hyper-dimensional state space models [Miz89, Smo90]—aka. vector-symbolic architectures [Gay06, Kle18]—to devise a so-called inner stage for final cognitive agents [Bis09], on which the agent can safely perform trial actions and plan its strategy. To that end, we develop a representation theory of semantic states, expressed by relational databases or feature-value relations, in Hilbert or Fock space [Foc32]. In this hyper-dimensional representation we can employ quantum logic [BN36] or other quantum-inspired methods to design the agent’s behavior controller and to model higher cognitive functions like, e.g., coping. Fig. 1 outlines the general architecture of a quantum-inspired cognitive agent (see [WHW⁺18] for details).