The whole point of a robot is to do exactly what you say. Or is it?
Image a creature that exists for its own purpose. A robot that may have artificial emotions, its own motivations and it own personality. Such a robot would make a much more interesting companion than, say, an intelligent kitchen appliance.
We are current working on a model of Artificial Personality. Here, personality is revealed by the choices a robot makes when it is insufficiently constrained to perform a given action.
Humans have a wide variety of personality traits, which can be identified and can be tested for.
We have an experimental platform called Cogbot. This robot cannot be commanded to do anything. Instead, a person needs to make friends with Cogbot first, and then if Cogbot likes the person, he may be more biased to follow a suggestion of a person.
However, the personality can also be tuned . For example Cogbot can be made to be more conscientious. If so, Cogbot is more likely to stay on task. If Cobbot is made more agreeable, it would be more likely to take the suggestions of others. And so on.
The internal architecture of Cogbot is based on a massive number of Baysian Rules (right now more than 1000). Each rule acts on a field of probabilities including emotional probability, motivational probabilities and so on. The second to second behavior of Cogbot cannot be predicted.
However, the overall dynamics of Cogbot gives one the impression of clear human personality traits.