>>12
>What are the sensors and the actuators?
That's not even part of the problem. They could be anything, for instance, depending on whether we were talking about a robot doing the flipping AND making the predictions, or whether we were talking a human doing the flipping and inputting the results into a program. If that detail was important, they would have told you in the video.
>Are we observing the coin flips from distance and updating the model based on the flips?
Again, not part of the problem, but what do you mean by "model"? If you mean the algorithm, then no; the dataset that the algorithm learns from, then yes.
>Or are we actually flipping and the probability depends on physics as well?
Not part of the problem but at least somewhat related to whether this is a fully or partially observable problem. Now, you can philosophize all you want about whether some omniscient being could predict the outcome of a coin flip by calculating all the variables in the universe, but you can't and neither can your computer, so we're clearly dealing with a partially observable problem.
Try not to think too much about the problems other than what is being asked or you're going to lose sight of what you should be learning. They will get the bigger questions later, I assure you.