Supervised learning is when the input observations are assumed to cause the output observations, general cause and effect. For instance, touching would be considered input whereas the sensation of being touched is output. It is assumed that being touched is what causes the sensory data.

Unsupervised learning is where latent variables are considered to be the cause of the outputs. These are variables that cannot be measured directly, instead, latent variables are inferred from other "measurable" variables. Compared to supervised learning, unsupervised learning is somewhat like an idealist's philosophy. Instead of assuming all output variables are based on true "real" variables. They stem from individual perceptions of that reality.

It is pretty much what it says on the tin. Supervised is learning based on someone or something defining the answers (x happens because of y) whereas unsupervised learning is based on the individuals own thoughts and then measured by their output. A persons capacity of love could, in principle, be measured by how many people they have loved and/or helped.