Historically, VLBI imaging is performed with CLEAN, a greedy matching pursuit algorithm. However, CLEAN has well known limitations due to the unphysical representation of the image by delta components and a suboptimal optimization strategy. In consequence, CLEAN is regularly outperformed by forward modelling techniques with respect to dynamic range, resolution and accuracy, particularly for very sparse array configurations as typical for global VLBI experiments. Forward modelling techniques construct an optimization problem consisting of multiple data fidelity functionals and regularization functionals and minimize a weighted sum of these terms by the tools of convex optimization. While this approach typically results in a superior numerical performance, the selection of the regularization weights still is a somewhat arbitrary, ad-hoc procedure done by the astronomer manually. In this talk, I will discuss alternative perspectives on the weighted sum approach that will allow us to waive the arbitrariness in the selection of the weights, the need for parameter surveys and contribute towards an unsupervised, unbiased imaging pipeline. Particularly I will discuss approaches that realize VLBI imaging as a multimodal, multiobjective evolutionary problem, and as the result of a cooperative game respectively.