Python is a strongly typed language, and to call it untyped is simply incorrect and a bad assumption. Python is dynamically typed, so the interpreter determines types at runtime instead of compile type. But, to say it is untyped, well, is wrong.
Python became popular for three reasons: built-in collection types (lists, dictionaries, sets), support for functional, imperative, and object-oriented paradigms, and finally, it is easy to integrate with existing libraries. Collection types are a big deal because, prior to Python, most languages only offered simple types and fixed dimension arrays. To implement useful data structures that could scale to handle large amounts of data, like linked lists, hash tables, trees, and sets, one had to write lots of plumbing code - comparisons, swappers, whereas with Python, you could declare a list and start list.append()ing away. Python also made it possible to write loops without counters and breaks, which was very different than most languages at the time (for thing in things instead of for thing = 0, thing ++, thing200).
A lot of the popularity in AI comes from managing data with collection types and calling faster code (that usually runs on GPUs). For a developer, you get to an acceptable solution faster. For a user, it's generally fast enough. When it's not, you can always write it in C and call it from Python.