How and to what end may consciousness contribute to action? Attributing properties of consciousness to an embodied, minimally cognitive artificial neural network
Frontiers in Psychology June 18, 2013 H. Cruse, M. Schilling
An artificial neural network called reaCog, built on a decentralized, reactive architecture for hexapod walking and insect-like navigation, can invent new behaviors and plan ahead through internal simulation. The authors argue that, under a monist view that sets aside the phenomenal aspect of mental phenomena, reaCog exhibits properties such as bottom-up and top-down attention, intentions, volition, and some aspects of Access Consciousness—including planning, guiding new actions, and global accessibility of memory elements. These properties emerge from network cooperation, not explicit implementation. reaCog enables both reactive/automatic and access-conscious control, but lacks metacognition or Reflexive Consciousness. The authors suggest that even simple networks can display properties of consciousness when the phenomenal aspect is excluded.