Solving one of the hardest part of creating an AGI: Sample Efficient and Rapid Learning

This idea is designed to solve one of the hardest part of creating an AGI system, which is Sample Efficient and Rapid Learning, Deep Learning has been really suffering from Sample Inefficiency, while there are alternatives like Active Inference (Karl Friston), Which learns much faster than Deep Learning and requiere fewer samples to generalize the same. However, Active Inference alone is not enough to even get close to human sample efficiency level (which often can do One Shot or Few Shot learning), So this idea proposes combining Active Inference with Core Knowledge or Priors (Elizabeth Spelke) that i (personally) believe was the strongest commitment for human sample efficiency, the way it works is the same like human, which those Core Knowledge explains almost all of the current context they are seeing (or for more detailed explanation of how it works please read the Elizabeth Spelke paper). the most beneficial domain with this idea is Robotics. What makes it innovative is that the model will learn from way fewer samples (even less than Active Inference alone) and will be computationally light (because fewer exploration and planning = fewer memory footprints).