To what extent can machine learning algorithms predict the outcome of random number generators?

I choose this topic to write my thesis about. I already did my research but haven't found a lot about it. (I'm now familiar how pRNG, but nothing more) I even saw some blogs saying this isn't possible, but if I must write a whole thesis about it it must be a little predictable right? I'm wondering how I could approach this.

Thanks!

Greetings, Miguel F.