I think we’re getting into the ever-present, ever-frustrating topic of corporate influence on game development. Maybe it’s a copout, but I can see how “Good AI” development can be pricey and thereby unappealing to developers (specifically development firms) driven by profit. As much as I would argue that games are someone’s or a group of people’s works of art, I do recognize a significant difference between artists in the traditional sense and game developers: money. Even the most famous of traditional artists starved, often surviving only on their love for what they did. Please do correct me if I’m wrong, but I at least have the impression that there are few if any starving artists in the game development community who would have enough passion and resources to invest the time and money in developing better AI, not knowing beyond a shadow of a doubt that it would make them rich.
However, the above is my belief related to the development of a perfect (or close to perfect) single artificial entity, a bot. Because of corporate interests and the easy alternatives that Twitchdoctor pointed out, I don’t think we will see development companies focusing on making the bots in their games ‘think’ rather than simply giving them more health, stronger weapons, better aim, and of course, more grenades. Twitchdoctor’s post (Good AI, Bad AI) presents a powerful alternative to adjusting the bots though—changing the conditions of the game. As in Twitchdoctor’s example of Thief, the conditions of the game can be changed to accommodate difficulty increase and substitute for (or at least distract from) imperfect AI.
So maybe there will never be an independent developer or development firm passionate enough to build the perfect bot, but there already have been productive alternatives that distract from this. The achievement systems used by Xbox and World of Warcraft are two successful examples.
Xbox requires its users to create an avatar which has a corresponding gamerscore, and each achievement programmed into the various Xbox games will improve this overarching score. In some games, achievements are granted for such simple tasks as beating the game on the various difficulty levels, but in other games, achievements are awarded for players’ added efforts to create the specific scenarios that the achievements call for. For example, an achievement titled “Fear the Pink Mist” is awarded to Halo 3 players who use a specific weapon (and a difficult one to use, at that) to kill 5 enemies without dying themselves. This achievement requires players to think and change the way they play the game to earn it. While Halo 3 has the traditional “Bad AI” classifiable as what Twitchdoctor calls “Dumb Difficulty Substitutes for Smarts,” the reasonably extensive list of achievements somewhat distracts from this, and if a player is really seeking a challenge, he/she should play Halo 3 on its highest difficulty setting while pursuing all of the achievements. The “Dumb Difficulty Substitutes for Smarts” is far less noticeable under these conditions.
In World of Warcraft, players create characters who each have an achievement score, starting at 0. There is a vast list of achievements viewable easily while in game. Some are basic, awarded for leveling up, some are more difficult, awarded for killing bosses in dungeons. But the real interesting ones are the ones awarded for killing bosses in dungeons in very specific ways. Of many, the following one stands out as the best example: there is a 10 or 25 player dungeon boss who has 3 separately killable lieutenants. Each lieutenant gives the major boss an increase in health and damage done, and has its own unique ability which is easily dealt with separately. Killing all of the lieutenants one by one, then separately killing the major boss is achievable even for a ragtag group of noobs. Leaving the lieutenants alive however, changes not only the health of the major boss, but the dynamic of the entire group. Leaving all three lieutenants alive requires players to deal with the main boss’ health increase, the special abilities and damage done by all three lieutenants, and the general chaos of this being all together, and this awards an achievement called “The Twilight Zone,” a challenge that even some of the most elite and experienced groups don’t have the coordination for. World of Warcraft has no difficulty setting to change, and by offering an optional achievement system with appropriate rewards, players can effectively choose their own difficulty. Even though these boss encounters are extremely scripted and repetitive, they present optional challenges that require player adaptability.
In summary, I think the achievement system is a great way for games to move away from, or rather function in spite of “Bad AI.” From a development standpoint, it’s a relatively cost-efficient way of using what the game already has in ways that require extra thought, extra time, and extra skill on the part of the player. Even if the AI isn’t perfect, well-developed, well-thought out, and difficult achievements can cause the player to have to out-think the game—for players, that will at least create a perception that the game is trying to out-think them, thus also creating the perception of a sophisticated AI.
—waltersthegreat
Intelligently Artificial 
I’m making a mental note to myself to bring up the larger issue of games and art at a later date, something you mention in the first paragraph.
This larger question of achievement systems is a really useful idea to throw into the mix. I’m not familiar with the Xbox one, but Steam has a similar system for individual games that are sold and patched exclusively through Steam (like Empire: Total War). One major benefit of this system is that it encourages players to see gaming competition in another way. You are obviously still competing against people (to get more achievements than the next person, etc.) but actually gaining the achievements is often competing against yourself. Some of them, as you noted, require players to move out of what may be their more familiar, more comfortable, play styles. It is also a carrot rather than a stick approach; not forcing people to play in an alternative style, but giving them something to shoot for if they do.
The only difficulty I see in terms of connecting this with AI issues is that while it might contribute to a player’s expectations about the complexity of the game as a whole, but not necessarily the AI? I know that I’ve been saying they are related, but perhaps sometimes they aren’t. For example, isn’t it likely that for many of the achievements, a player will see them as externally imposed constraints? So not as challenges that emerge organically from within the game, but an extra set that come from without? I dunno, I’m in two minds about this. The Halo example would, I agree, tend to produce a different conception of the challenge level of the game which might influence perception of the AI.
I’d agree with the idea that achievement systems are external constraints on the player rather than an improved AI. The player makes a choice follow those particular criteria, the AI does not cleverly force the player to ‘try for the chieve’.
If the achievement system could be considered an AI, I would say it’s a Dumb Difficulty Substitutes for Smarts kind of AI. Taking the Sartharion Twilight Zone example that WalterstheGreat made, Sartharion does not get more intelligent, the lieutenants do not give Sartharion the ability to consider hitting someone besides the plate-covered tank in front of him. It makes the boss stronger and the player weaker.
What if instead, when Waltersthegreat walks into Sartharion’s chamber, Sartharion swears under his breath and chooses to throw everything he has got at Waltersthegreat’s group, rather than the group deciding to give Sartharion the chance. I think that would be interesting, however somewhat ego boasting to the player when a boss judges the player’s group a serious threat.