It seems that today, a data-driven computer model might be able to produce a much better artificial language than was previously possible. If you have grammar, vocabulary and text corpus for each language, parameters for number of speakers who would presumably learn the language, then you could perhaps optimize for minimal learning curve for the most people.