Smells, Spanners, and Switches (2016)

Is this not something modern ML systems are very good at? Solving practically unsolvable problems by approaching the problem with "machine intuition"? I am not versed in ML at all, that is on my list to start playing with this year, but if a computer can play GO succesfully and design new proteins, wouldnt it also be capable of designing molecules that excite these receptors in the correct way? Maybe not now, but theoretically? What is different in this problemspace?