Surrogate models of manufacturing processes are highly useful in the context of digital twins, as they can be considered as a prerequisite for linking between physics and real machines’ cases. However, in many cases of surrogate models, the so-called (hyper)parameters are not easy to be estimated. In this work, the role of AI is investigated in terms of its efficiency in doing that in an (semi) automated way. The case of ARX models is considered for this, where the parameters are clearly related to the physics of the process. In particular, the technique of systemic identification is adopted. The aforementioned investigation is performed in this particular case regarding intuitiveness with respect to the physics, through adopting AI techniques that could be considered to be explicable in some sense. The results indicate the limitations of the AI techniques and their link to the process dynamics as well as their relationships with traditional techniques.