Each output is NOT unique. Since it can take in an almost unlimited amount of data and output a set length string we know for a fact collisions will eventually

Each output is NOT unique. Since it can take in an almost unlimited amount of data and output a set length string we know for a fact collisions will eventually occur. The question is - can we predict these collisions or guestimate when they might occur without having to actually try every combination? If the answer is "yes" then the hashing algorithm is broken (see MD5).