Fair, but If we are demanding a level of AI that is beyond the state of the art for AAA games, then we would not be making a reasonable criticism.
AI quality (or intelligence for that matter) does not really mean much out of context. It is really constrained by technology and resources. When we say AI is good or bad, we usually think it is good/bad compared to a human. If we use this criterion we can basically tear down to shreds every AI which has ever existed. So it helps to also compare an AI with what we can expect from a good AI.
Today, there are 2 extreme kinds of AI, depending on the type of game. And a lot of in betweens.
On one hand we have chess like games, were the game can be solved by computational power. Which basically means that you first make an AI to win the game, and then your AI design consists in basically coding mistakes on the AI so the player has a chance. And Immersive-like AI, where AI is not programmed to win, but to behave in a way that creates an illusion of challenge and understandable behavior.
Most games are on the second part of the continuum, and AI consist usually on building a decision scheme with several decision branches, intending to code decisions that are reasonable from a human perspective, with different degrees of complexity.
On one hand we have great AI usually relying on very complex hand crafted models, usually behavioral trees combined with priority optimization techniques and limited POV information. These models work great, but by definition the more complex they are the weirder behaviors you will see. Because they are based on hierarchies of hierarchies of hierarchies of behavioral rules, and is simply impossible to plan for every possible condition to trigger every possible branch decision and because of the nature of the model itself.
Then we have complex machine learning models, which is an entire different kind of worms. To the point that machine learning tech is inherently alien and produces mathematically optimal behaviors (in average and in theory) which implies also decisions no sane person would make. Because machines don’t think like people. Open Starcraft 2, which is basically a showcase of academic machine learning research, is a good example. Because of this unexplainable behavior, models need to be combined to bias the system to make not optimal choices but understandable ones. This is the paradigm of Total War games, they being the saga that defined a good state of the art AI, as started in Shogun one, where they coded RNAs to model unit behaviors and expert systems to rule higher level tactical choices where they translated the Sun Tzu war strategies to computational rules. Now they use an insanely complex hierarchy of optimization models, decision trees, genetic algorithms, montecarlo algorithms, RNAs and God knows what else to make the most complex AI in gaming I think ever. And still oh boy when they fail, they fail hard.
Point is, you cannot judge AI just in the vacuum. Of course it should ultimately be satisfying to play or otherwise is a failure. And usually the best AIs in this regard are the ones that cheat the most. (a good bad example here is XCOM which basically forces enemy misses after a low probability hit, and hard codes intentional AI fails in order to not frustrate the players), but this goes to enemies not firing at you when you are not looking at them to forcing reloads when the player is vulnerable or enemies putting themselves intentionally out of cover…
PP AI is not perfect; it is just as good or just a bit better than the AI any other turn based strategy game I know. Is certainly not a broken AI (again looking at you Civilization VI), and is also not perfect. It makes mistakes, some quite bad. But still is not easy to cheese (I’m looking at you Xenonauts), and is not so gimmicky and artificial that you realize you are playing with a stupid algorithm that forces a constant rate of engagement by spreading out enemy encounters (I’m looking at you XCOM and XCOM2). And is also mostly consistent and challenging.
To me PP may not have a revolutionary AI, and certainly it does not innovate and does not try to do anything particularly new. But it has a pretty dammed well put together AI.
