Subjective questions just like humans.
If you think about it, this whole machine-learning thing isn't going to be that great. Humans can already "machine-learn" and computers will be susceptible to all of the same problems and more.
For example, for a computer to learn, it will have to have physical capability to execute or be provided information by humans. Some of the things that it needs to play with won't be practical or possible for it to do. For example, we can''t put a computer in an airplane and let it continuously crash so that it learns how to not crash. So what will we do? We will train it or provide it documents which were written by humans and have errors. All inputs to a computer will be provided by humans or nature and thus it will never get "better" than humans. It may have quicker reaction time and/or be able to process things faster using math, but otherwise, it'll struggle hard. I find it almost laughable that we're talking about having AI do all of these complicated things when Apple and Microsoft can't even get an OS locked down so that thousands of people can't compromise peoples' data.
For example, say I as the computer "Why don't I like that girl?" Or "Why doesn't that girl like me?"
The computer will try to figure it out but it will never know an exact answer. It could go ask the girl.. But some girls themselves don't even know why or what they say may be different than the real answer. What if I lie to the machine? Now it needs algorithms to try and figure out if I'm lying but it can't know for sure, etc...