Towards the Neuroevolution of Low-level artificial general intelligence

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Evolutionary Computation

Background:

• Intelligent behavior in nature arises from organism-environment interaction and adaptation.
• Learning is hypothesized to occur via sensory feedback interpretation during agent action.
• A embodied agent and a reactive environment are essential for this learning process.

Purpose of the Study:

• To propose and evaluate a framework for low-level artificial general intelligence (AGI).
• To investigate the evolution of biologically-inspired neural networks that learn from environmental interactions.
• To benchmark the adaptivity and generality of evolved controllers in mutable environments.

Main Methods:

• Neuroevolution of Artificial General Intelligence (NAGI): evolving spiking neural networks with adaptive synapses.
• Agents with evolved controllers are instantiated in mutable environments.
• Tasks include food foraging, logic gate emulation, and cart-pole balancing.

Main Results:

• Successful task completion using small, evolved neural network topologies.
• Demonstrated adaptivity and generality of the neuroevolutionary approach.
• Validation of the hypothesis that learning occurs through interpreting sensory feedback.

Conclusions:

• Evolving low-level AGI from basic principles is feasible.
• The NAGI framework provides a viable method for developing adaptive and general AI controllers.
• This approach opens avenues for tackling more complex AI tasks and curriculum learning scenarios.