Why do we assume AI's previous knowledge is worse than ours?

From Meritxell Rogg Barnet:

Regarding previous knowledge, we all expect AI to develop plenty of it. Why do we assume AI's previous knowledge is worse than ours?