That is not correct. Even the so-called "expert systems" are nothing more than programs which operate within the limits of pre-defined algorithms, the only way they can make apparently irrational decisions is when the code leads to a point where the programmer determined that there is no one choice out of the possible choices which is obviously best or correct, at that point the programmer may have chosen to randomly pick where the flow goes. The key points being, it was the programmer and not the program itself which chose the conditions under which the random choice is made and the available choices are limited by the results of previous computations--the program is unable to choose not to make a choice.

The situation with "artificial neural networks" is much the same. In that case the programmer has determined how or to what extent the pseudo-neuron mimics a real neuron, and can change how the system behaves by changing the rules which govern that mimicry.