My self-supervised model has a very simple integration area where processed inputs from different sensory modalities meet. I have also created a more realistic model of correlation based association learning to address the issue of concept learning. This very simple Hopfield model was able to answer a crucial challenge in cognitive and conceptual learning; it accounts for qualitatively different behavior to distinct classes of objects (e.g. natural vs man-made objects) with one consistent, biologically plausible learning rule. The network was able to learn the concepts and duplicated human performance differences (data collected by Ken McRae) in reaction time, semantic priming, and property verification time between natural objects and man-made objects. The differences between natural objects and man-made objects arose simply because natural objects consisted of features that were strongly correlated with other features (e.g. HAS WINGS, HAS FEATHERS) whereas man-made ones did not.