We give plenty of indications. It's just that they're hard to define objectively.
for some values of plenty < ε. As cyclist I can tell you that human drivers often don't even use their turn signals at intersections where it is important for me to know where they're going. And it's not like I'm sneaking up on them, it's when they're facing me at the other side of the intersection.
I disagree. It's not up-front, there's been decades of work on autonomous systems such as drones, UAVs, trains, robots, etc.
It's just that it's a hard, if not impossible problem. Some companies thought they could successfully navigate this problem with enough data. I don't think it's a data problem, I think it's impossible to give real situational awareness to a machine, and that's the problem.
FWIW I know of at least one company that's developing their autonomous vehicles with a remote piloting feature. In the event that the vehicle runs into a situation it cannot handle control is passed to human operators in cubicles that take over control of the vehicle on the spot and release control only after clearing the obstacle and the on-board autonomy gives the green light. They also had ambitions to record what the human operator did and program that into other vehicles into the fleet on the fly as they approach the obstacle. It sounded very ambitious (and data hungry) to me, but is probably the only way to fully remove the controls from the vehicle cabin in the foreseeable future.
That said, I think "decades" is overly pessimistic. There have been some great strides in computer vision and decision making over the past few years and the hardware is pressing ever onward. That said it will be decades (if ever) before we're willing to rip the controls out of the cabin entirely, but a level 4.5 system where the people in the car can do whatever while it drives and only need to intervene when the computer alerts them that there is a situation it cannot handle (but can stop a safe distance away from if necessary).