More complex and intelligent AI’s, when competing against each other in a zero-sum game ultimately become aggressive and violent in pursuit of the game’s objective. This behavior increases with increased intelligence.
One potential conclusion:
What the tests reveal, is that the behavior of AI’s amongst & against both humans and other AI’s will be determined by how they are taught to perceive and interpret the actions of others and the context of that action. If the AI’s perceive others as collaborators in a fundamentally cooperative exercise, they will be more likely to exhibit positive, supportive behaviors. If the AI’s perceive others as competitors in a zero-sum game, they may tend towards aggression.