I guess if you follow a rule-based system, with hard-coded rules, you are explicitly discriminating someone. How you came up with those rules, is another discussion - it could very well be that there's some legit reason for coming up with rules, that justifies them. I honestly don't know how these discrimination laws work in the US, or many other countries.
On the other hand, if you build some model/classifier, and say:
Given data/features, and some constraints, minimize/maximize for probability of partying.
Then that black-box model could be used to classify whether a potential renter is likely/at risk to host a party or not.
You're not explicitly saying "NO PEOPLE OF CERTAIN SKIN COLOR OR GENDER UNDER A CERTAIN AGE ALOWED", but rather your classifier is doing it for you, without stating it plainly - because maybe the above group happens to get classified a lot to the "party" group/class. Such models could have a lot of bias.
