Monsters of Man AI critics
In the movie “Monster of Man”, four combat robots (BR1, BR2, BR3, and BR4) are being tested on the field. Those four were sent to wipe out a village in South Asia by someone working in the American government and an ex-CIA agent to prove a point about the use of AI in war. Three Programmers were controlling the AI by creating and loading new objectives. The problem for the Programmers and the ones who hired them was that the module for control can be removed from the robots. One of the robots (BR4) had a problem with the parachute and fell into rocks then had its module fall out and picked up by children playing around. The robots were programmed in labs and now exposed to the world outside to explore. The environment was partially observable, because they were in the woods and going back and forth to a village they didn’t know. Multiagent and not deterministic, since the innocent people in the village, the small group of armed villagers and the six American doctors exploring the area affected the performance measure of the robots in the effort of saving their lives. The environment was also sequential, dynamic, continuous and unknown to the robots. In this movie, killing innocent people was considered as part of the training process of the robots, which is both unethical and illegal. BR4 on the other hand had to explore the environment on its own without communication with the programmers and the other robots. The agent gathered much more information and learned new skills for combat and communication (voice added). It used a tablet to connect to the internet and learn about humans and conspiracy theories then had an interaction with an ex US soldier living in the village:
Human: “you are made by men”
BR4: “you are made by men”
Human: “The difference is that I am a human….”
BR4 couldn’t understand why the human risked dying to save a child. In the end, BR4 was able to save the remaining humans, hack the programmers, download their files, destroy all other robots and transfer its code and knowledge to all robots in that were being built.
The subjects discussed in Chapter 27 about machines given the ability to select and take-out human life is to consider in this. Agents trained this way (movie) would indeed not consider collateral damage when executing missions. I think that we are close to the creation of such agents. Maybe the combat skills will take a bit time, but we have seen robots dancing, so throwing a punch wouldn’t be much hassle. I think the control part for the agents should be of a great emphasis when making them. For instance, if the control module wasn’t removable, BR4 wasn’t going to disobey, but, then it came out stronger and faster than the others and saved lives.
I like the plot of the movie, how the depict AI learning fast and applying knowledge to the point that it’s scary. The fact that only American lives are considered as lives is a bit concerning though.
