Actually, the option 2 seems quite flawed to me, like science-fiction.
More likely could be:
2) As soon the AI's mechanics (like Choosing Rival, no name one of the most signigicant in my experience) start to have a human on the receiving end, the AI focuses on survival at any cost, becoming less effective in aggressions but more cohesive while defending and preserving its own borders.
It might be interesting to test an Ironman game with the 3 different Manchurian tribes, like I did by chance.
You choose Haixi, and Jurchin gets allied with Ming and Korea and rivals you.
You choose Jurchin instead, thinking to have outsmarted the AI to get a good starting position, and you find Haixi allied with Ming, and Korea rivaling you.
This is not confirmation bias, anyone can verify that by starting an Ironman game and playing for 2 months.
It's surely WAD, but it does not mean it treats the human like just another AI.
As long as AI is scripted to rival humans, it will have less opportunities to wage war against AIs since rivals are targeted first.
Simple as that, no need to use science-fiction.
Or how about the AI programmer (me) just tells you that you're wrong and experiencing confirmation bias.
- 19