Aim of this talk is to detect shared features in biological information processing networks, keeping the focus on neural and immune nets. We will proceed by steps: at first we will discuss simplest statistical mechanical models able to store one bit of information only (i.e. the paradigmatic Curie-Weiss and Mattis models) but we will address their thermodynamics using only variational principles stemmed from Lagrangian mechanics (rather than the classical statistical mechanical route). Then we will extend models and techniques toward the Hopfield scenario and we will use it to briefly revise neural networks and Hebbian learning. Next step will be to adapt this scaffold to lymphocyte networks and this will naturally give rise to the theory of autonomous parallel processing, that mathematically mirror the key ability of the immune system to defend its host from several pathogens at once. Finally, we will discuss why so different systems may share such an underlying conceptual description by showing that the transfer functions of ferromagnets (statistical mechanics), neurons (neurobiology), operational amplifiers (artificial intelligence) and lymphocytes (immunology) are identical.