The goal of our experiment was to gain a deeper understanding of human-robot
interaction. If we are to believe the Media Equation humans treat computers
as social actors. In other words we treat computers in a similar way as
we would tread another human being.
We wanted to know if this pattern would also hold true for the more negative
sides of human behavior. The capacity of humans to torture each other
has been demonstrated by various experiments in the past. So the question
we tried to answer is: “Will a human torture a robot differently than
it would another human?”
The Milgram Experiment
In 1965 Stanley Milgram performed a series of experiments called Obedience.
As the name suggest Milgram wanted to investigate the relationship between
authority and obedience.
The cover-story told to the subjects was that it was a memory test. During
the experiments the subjects were told to give electrical shocks to another
person. This second person was in fact an actor and didn’t actually get
shocked. The actor had to perform a test and every time he made a mistake
the subject would have to administer a shock with every mistake the voltage
would be increased. The experiment proceeded the actor would seem to suffer
more from the shocks. He would start complaining and this would go on
to screams of pain. He would plead with the subject to stop the experiment.
The subject would in turn be urged by another actor playing the experimenter,
to go on with the test. The main measure of the test would be at which
voltage, if any, would the subject refuse to go on.
What immediately stands out in the results is the fact that all participants
continued until they reached the maximum voltage. In Milgram’s original
experiment only 16 out of 40 participants applied the maximum shock. Of
course this does not necessarily mean that none of them felt compassion
for the robot. During the original Milgram experiments there were enough
subject who were really troubled by what they were doing and yet also
continued all the way to the maximum voltage. There were many subjects
who expressed pity or compassion towards the robot, one even tried to
cheat so that he would not have to administer the shocks. But the urges
of the experimenter were always enough to make them continue all the way
to the end. What we can get out of these results is that humans can ignore
their feelings of compassion easier when dealing with robots than with
It would be wrong to assume that because of the results that the Media-Equation
does not apply to human-robot interaction. However, what we did notice
is that it only applies to a certain degree, especially in the cases of
negative human behavior. Contrary to interaction with other humans, it
would seem that when dealing with robots, humans will disregard their
own feelings of compassion if they believe that no permanent damage will
come from their actions. This knowledge will help in the design of robots
that will have to interact with humans on a regular basis, e.g. house-robots.
One requirement that we would be able to derive is be that such robots
have to be “torture-proof”.
Download the full article [*.pdf]: Click