In the foundations of artificial intelligence we find many scientific disciplines, but it appears that statistical ideas from physics have played an important role. We give a brief summary of some of these ideas and concepts that have led the way to the modern developments. In particular we introduce and describe briefly the Ising model, a fundamental model in statistical mechanics that incorporates interactions among individual spins. We then use the Ising modelIsing model to introduce spin glass models that include additionally disorder in the inter-spin interactions and lead to novel ideas on statistical phase transitions. These ideas take us to the Hopfield networksHopfield network of neurons that model associative memories through the application of the Hebbian ruleHebb rule for learning. The Boltzmann machines are a further refinement of this type of complex nonlinear lattice models that have learning properties. We conclude with the protein folding problem that constitutes a dynamical search in a statistically immense space of possible states that can be actually implemented through artificial intelligence algorithms.