There are two antagonistic point of views available. The first one is that a robot is a machine similar to a dot-matrix printer which doesn't know anything but has to execute a program. The other point of view is that a robot is some kind of artificial life which has its own will and goals. Then the robot is able modify its program and takes decisions not preprogrammed before.
The general problem with highly developed artificial intelligence is that it has tendency to act like artificial life. That means a simple printer will always be a machine, but recent models which have a complicated firmware and large amount of sensors have more in common with an elaborated character than with a normal automation technology. Suppose robots were used in the future for highly complicated tasks, then they will develop their own personalities, comparable to a virtual human which can't be predicted in advance. This results into a high-risk technology which is out of control. Hans Moravec wrote an interesting book about this topic which was criticized heavily in the public perception.
Evidence On the first look, it's hard to tell what robots can do and what not. Who is defining if a system gets too complicated? To get a more realistic impression of human-robot-interaction, computergames are providing experimental evidence. They are available for all computers and simulate a wide variety of domains. The proposed distinction between low complexity automation and high-complexity artificial life is available in computer games too. A simple non-player-character behaves static. For example, the enemies in Super Mario have a simple forward-backward pattern. They are acting similar to a traffic light and are not able to improve themself. They can be treated as automation technology which is easy to predict.