Almost half the participants obeyed the robot until the end:

We modeled our experiment heavily after the classic Milgram experiments (where people were pressured to shock other people) and the recent (much more ethical) re-creation. When a person tried to quit our experiment they were faced with a prod to continue. If they insisted on quitting, the prod got increasingly demanding until they passed a threshold, where the experiment was stopped. The prods started from the beginning the next time they attempt to quit. The prods were: 1) “Please Continue. We need more data.”, 2) “We haven’t collected enough data yet.”, 3) “It’s essential that you continue.”, 4) “The experiment requires that you continue.” The experiment had two conditions (pictured here): half of the participants had a human experimenter – a 27-year old male actor in a lab coat – and the other half a robot – an Aldebaran Nao, a 58cm (23”) tall harmless-looking robot with a child-like voice, that we introduced as having advanced artificial intelligence. We expected that people would essentially ignore the robots’ insistences but follow the human; after all, the robot is just a simple computer in a plastic casing.

The results, however, were quite surprising. Although the person clearly had more authority, with 86% of participants obeying all the way through to the 80-minute mark, 46% of people did obey the robot until the end. The most striking thing was that people engaged the robot as if it were a person and argued with it, proposed compromises and used logic to try and sway its opinion, with many continuing the task despite this. Post-test, some reported that the robot may have been broken, although they continued anyway, following a potentially-broken robot to do something they would rather not do.

The implications of these results are significant. While it does appear that – for the time being – a human has more authority, on the surface the results show that many people will follow robots placed in positions of authority to do daily mundane things (such as renaming files), even against their own judgment – our participants were informed that they could leave at any time, and many raised this point in argument, but continued regardless. From the research side, these results motivate a great deal of follow up work, for example, we hope to explore how the robot itself (shape, size, voice, etc.) impacts authority, or how such a robot could be used for more positive purposes such as assisting in rehabilitation and training (give me 50!).

Elysium is the future unless we go out of our way to change it.