According to the experiment, all 'permissions' were gained before the video was uploaded. The little girl's parents permitted it for her.
There is no data about how many people they actually experimented with, and there is no data on how many people refused to give their permission. If someone didn't show a great deal of emotion, refused to translate it for him without any sympathy or just read it aggressively to him, they obviously wouldn't give their permission to have that uploaded on the internet. So the only people who would have given permission are the ones who were nice to him. Secondly, even from all those who gave permission, they can only put a very small number of people in the video to avoid making it several hours long. Obviously they selected only those who showed the most empathy.
One thing that shows this is that there's only one little girl. Why not more? It is pretty obvious that if they were going to invite little girls/children, they would invite more than one. Yet only one is shown on camera.
The experiment specifically shows those people who are being the nicest, without any information on how the rest behaved. Also, it is never told to the participants that the african american is an applicant. Their assumption would be that he might be an applicant but he also may be an employee. For all they know, from their perspective he could be their future boss. They'd be inclined to be nice to him.
tl;dr Video is biased and shows only those who were the nicest and showed the most empathy. No data on how many people behaved how, only that out of an unknown number of people, a few were nice. Which is entirely to be expected.