With the recent discovery of the Higgs-like boson at the LHC, particle physics has entered a new era, where it is of utmost importance to provide accurate theoretical predictions for all relevant high energy processes for signal, background and New Physics production. Crucial ingredients of these predictions are the Parton Distribution Functions, which encode the non-perturbative dynamics determining how the proton's energy is split among its constituents, quarks and gluons. To bypass the drawbacks of traditional analyses, a novel approach to PDF determination has recently been proposed, based on artificial neural networks, machine learning techniques and genetic algorithms. In this approach, PDFs are "learnt" from data without imposing a prior theory, and they "adapt" towards the physically meaningful solution, in the spirit of biological evolution. In this talk we introduce pedagogically the concept of parton distributions, motivate their relevant for LHC physics and describe the latest developments of PDF determinations from LHC data.