It has been shown that the neural network properties of systems based on stimuli-sensitive hydrophilic polymers make it possible to raise the question of creating programmable structures based on such materials. It has been shown that the phenomenon of multidimensional hysteresis can occur in systems based on stimulus-sensitive polymers, the behaviour of which significantly depends on more than one thermodynamic variable. In this case, the dependence of the parameters characterising the system's state on the control thermodynamic variables is described by a complex folded surface on which areas corresponding to abrupt phase transitions can be identified. It is shown that the system's state in which multidimensional hysteresis is realised depends not only on the current values of thermodynamic variables but also on the path along which the representing point moves in the space of control parameters. On this basis, it has been proven that analogues of neural networks can be implemented in which information is recorded by submitting the same code sequences to all inputs of the first layer of the network. Physically, this corresponds to the fact that the programming of structures of the type under consideration is carried out using macroscopic influences expressed in the form of changes in thermodynamic variables according to a given law. Simple and illustrative model examples built based on standard stimuli-sensitive polymer systems are considered, proving the possibility of implementing neural networks of this type.