There is enough of material in this subject for months-worth of typing and debating.
But - ultimately - and to make it short, the right answer to your question in the title is: No. (it's way more complex than just "no", but yeah, it comes down to that).
* If you feel like reading more... well there's a bit more on that below, but feel free to be content with the answer just being No *
A more comprehensible way to put it other than just say "no" would be to say that there's been no A.I. coders out there who 'dared' making their A.I. "smart enough". Because in A.I. coding, from what I've often read, the 'toughest' part isn't the coding language, but it is to "dumb down" the A.I. enough so that it doesn't murder you with light-speed decision-making and laser-perfect aim-bot headshot you the moment one pixel from your head's hitbox shows up in its line of sight. If the coders don't do the "dumbing down" part, A.I. would essentially not even have to avoid you, per se, because you'd be dead already. The real topic really is about the coders themselves and their apparent incapacity through their coding knowledge and/or their actual will to do so in coding A.I. that acts intelligently ("believably human"-like, making decisions and doing actions that are 'plausible' if it was a real person) while keeping a certain balance for pure game-play purposes. In short, the A.I. NEEDS to be somewhat stupid for most gamers, otherwise no one would play video games. Now, of course, this is an exaggerated statement. But it's the gist of the idea. If freakin' Goombas were to form group tactics against Mario no one would have ever beaten World 1-1 back in 1985 and the franchise wouldn't have lasted a single year. That's also why demons and zombie soldiers in DOOM back in the 90s weren't coded to aim-bot with perfect precision, and instead the difficulty settings mostly increased numbers and spawn points to compensate. I could also mention, for example, how in the first Halo game the "Elite" Covenants are significantly better-coded than the "Grunts", and that's fine (because it 'fits' with their respective context in the lore, their species and their actual chain of command in that universe; the Elite are... well, elite troopers and are intelligent, while the Grunts are easily afraid, pretty dumb but can overwhelm you with higher numbers to compensate). There's many examples like that out there that do actually make sense (that in itself could be its own subject: the distinction of A.I. coding for specific types of enemies in games where it makes sense to code enemies differently than one another).
As I mentioned, the subject is complex. There's a lot to consider with it comes to the actual goal of the game. Taking my DOOM example above, zombies cannot and should not 'normally' be intelligent, so of course their A.I. is designed as such, so there's design decisions that goes along with the actual type of game you're making, and what 'x' enemies' should do versus 'y' enemies later on in other levels, et cetera. To use your Fallout 4 example, now in that game there's A.I. for a good bunch of different situations, conditions and enemy types (giant cockroach, or Follower, or wasteland bandit, or A.I. for a scripted scene... there's a lot to consider), but when it comes down to the shooting part (Combat A.I., specifically) then it's definitely not the best example out there.
But let's pretend a moment that in Fallout 4 basically all the bandits in the Wasteland would actually be very intelligent. They would use cover and NOT pop-up their head every 7 seconds to make sure you could head-shot them. Let's say they would actually use the environment to enter buildings (if present in the vicinity) and go up the 2nd floor, or even up on the roof to have the high ground advantage (which would have been done intentionally as part of their A.I. system) and take out a weapon that they had on them that works better at long range (a sniper rifle, shot from the roof tops; again would be part of their A.I.) and doing all that while maybe communicating their decisions to nearby bandits while telling the others to flank you on ground level, while the others recieveing the order respond to their leader and as they do you'd actually see them flanking you, maybe even crouching their way to you carefully (even if it was just to mimic their "knowledge" that their footsteps sounds might reveal their positions to you). Let's say that most A.I. (at least when it comes to human character A.Is, because from this point on we could start talking about "alien" character A.Is in other types of games, "mutants" or other types of creatures A.Is, and so on) in Fallout 4 and in most games this decade (at least) would have A.I. like that. Would it be fun? I suppose that's the ultimate question. It would be fun to see A.I. act intelligently (especially in group), but how much of it would actually be balanced? At that point, what would be YOUR chances as the gamer here? The A.I. must have "their chance", sure. Now, they don't have to always just stand there and basically walk towards you as you proceed to make multiple bullet holes in their skull (as it is today and as it has been for the better parts of the past 25 years of video gaming history). So the real problem comes down to how "intelligently" is the A.I. being coded, and how much of that whole A.I. system are the coders willing to 'dumb it down' to balance it.
Now it is THAT whole 'streamlining' process of "A.I. intelligence VS end result balance for gameplay" that has been the issue for a solid 90% of all existing video games with enemies in them, since two decades or so. I DO understand of course why A.I. could not have possibly been 'too intelligent' (and really didn't NEED to be) anyway in old school gaming (first few generations, basically) until around the 32-bit years finally arrived. However, starting at a certain point (around 3D-acclerated PC gaming, PS1 / Saturn / 3DO years) there's an unmistakable pattern in A.I. "progression" up until this very day. The pattern usually consists of increasing A.I. aiming (or aiming 'speed'), the Health of enemies and the damage (or "DPS") done to YOU from them on increased difficulty levels of almost every games there is. There's an extremely low number of games where increasing difficulty levels will actually result in BETTER overall A.I. in decision making, movement (taking cover, staying in cover longer, etc), group A.I. decision making and so on. The 'pattern' I speak of for A.I. is present to this day, most developers are going in with a very simplified, straightforward methodology to give the 'impression' of increased challenge through simply increasing your enemies HP, and damage (instead of making the A.I. progressively better overall). I can already think of that game that probably nobody ever heard of, heck I'm not even sure if I'll get its name right, it's something along the lines of The Legend of the Princess' Wild Breath, or something like that. Anyway, in that game there's a "Game+" mode, and instead of making most enemies more intelligent all it does is increase spawn points, change some enemy types around, and add HP Regeneration to all enemies in the game just like that, magic stuff. Taking a wonderful game I myself do love as an example here: Borderlands 2... well I absolutely adore that game, but the Ultimate Vault Hunter mode (difficulty setting #3) is a very good way to take your gamers for complete idiots and show how little effort went into creating something fresh and new with their artificial intelligence. All the difficulty does is force you into one specific gameplay mechanic to Slag opponents first, along with giving enemies HP Regeneration, there it is again... now THAT is the pattern I'm talking about.
So anyway, I could keep on going with this. It's a complex subject, but I'll repeat my main point again to conclude: No.
Facebook
Twitter