ReaCog, a Minimal Cognitive Controller Based on Recruitment of Reactive Systems

Area of Science:

• Neuroscience
• Artificial Intelligence
• Cognitive Science

Background:

• The prevailing view suggests large neuronal systems are necessary for cognitive abilities.
• Planning ahead is a fundamental cognitive function.
• The relationship between system size and cognitive capacity remains an active research area.

Purpose of the Study:

• To challenge the notion that only large neuronal systems can exhibit cognitive planning.
• To propose and validate a small-scale artificial neural network capable of planning.
• To demonstrate how planning emerges from the exploitation of existing reactive mechanisms.

Main Methods:

• Development of an artificial neural network (reaCog) for six-legged walking behavior.
• Minor expansion of the reaCog system to incorporate planning capabilities.
• Analysis of how novel solutions arise from combining memory elements.
• Investigation of internal simulation grounded in embodied experiences.

Main Results:

• The reaCog network successfully performs a specific behavior (six-legged walking).
• A modified reaCog system demonstrates the ability to plan ahead.
• The system generates novel solutions by recombining existing behavioral elements.
• Planning emerges from the exploitation of reactive control structures, not a separate module.

Conclusions:

• Small neuronal systems can possess cognitive planning abilities.
• Planning can emerge from the internal simulation of embodied experiences within a reactive system.
• The reaCog model provides a proof of concept for embodied planning in small-scale artificial systems.