Tales of the Rampant Coyote

Adventures in Indie Gaming!

The Joys of Crappy AI, and the Importance of Obviousness

Posted by Rampant Coyote on October 9, 2013

Read most articles about game AI, and you’ll see a call for improvement. “Why can’t we have better AI?”

As it turns out – and this is an open secret in the game dev community – players often don’t like it. That’s why. A recent article, Artificial Intelligence?, on MMORPG.com, discusses one failed attempt to make more interesting AI in City of Heroes.

My personal feeling is that it kinda comes down to not letting the player know you are letting him or her win. You can have neither incompetent AI, nor AI that’s clearly throwing the game. Players want AI that gives them a run for their money, enough challenge to make things interesting. Most of all, I think players want AI that – for lack of a better word – emotes intelligence, even when it is stupid to do so. It’s not enough to have AI that has very sophisticated problem-solving strategies… it needs to project this decision-making so the player can see it at work, if the end-result doesn’t make it obvious. In other words, the players need feedback to recognize the AI at work – and to therefore predict and adapt to the AI.

And of course, as suggested in the article, players want familiar patterns that are easy to respond to. The power mentioned in the article was – with the attempted AI response – not really a damaging power. It was a control power – it forced AI to respond by moving to range. Which is actually pretty cool, so long as its costs are effectively balanced. But players didn’t like it, according to the article, because it broke existing patterns of gameplay. And that’s certainly true.


Having played City of Heroes, I do recognize that this sort of thing can task-overload a player whose primary abilities are going toe-to-toe against bad guys.  Frankly, to effectively use that kind of control power requires a different perspective – both figuratively and literally. You cannot effectively play “crowd control” when you are in literally at the center of the fight. Plus, it’s very hard to keep track of that when you are playing the kind of game that City of Heroes gave to tankers and scrappers. It was hard to focus mentally, and hard to use from a mechanical and UI standpoint. So I don’t know if it can entirely be blamed on players preferring crappy AI or no-brainer classic MMO patterns.

Another element may simply be that the faster the action and the more players that need to coordinate with each other, the simpler the decision tree needs to be for the players. At least for most players. The more experienced they are, of course, the broader and deeper the decision tree can be. On the other hand, in a single-player, turn-based game with a top-down perspective giving you a perfect or near-perfect view of the battlefield, it’s okay to demand the players be a little more clever to keep up.

The bottom line is, of course, that simply making the AI “smarter” isn’t the answer. The AI has to fit the game, and it’s all about making the game more compelling to the players. How that is achieved really depends on the game.


Filed Under: Design, Programming - Comments: 7 Comments to Read

  • Stephen said,

    Yeah, I think “players that prefer the trinity also prefer dumb AI” is a huge oversimplification. In this specific example, the player outcry was much more to do with the change to AI changing the implicit contract of the archetype. The fire tanker (and really, most tankers) game was very much “the more enemies you can get to stand next to you, the better.” Suddenly adding on, “but your best power will make none of the enemies stand next to you,” was a huge dissonance in that plan.

    And it was a very specific tactic to make the AI smarter against. Virtually all the tankers had auras, and all of them could only fight you if you stood next to them. Why is the AI smart enough to decide not to stand in the fire but not smart enough to avoid all the other devastating melee-range effects? In fact, Taunt specifically overrode the desire to stay at range for everything else, so why can you taunt someone to let you hit them in the face with a fire sword but not to stand in the fire next to you?

    I think the takeaway is that you can only have smarter AI if they players are given tools to still succeed in the face of it. Nobody’s going to like, “we realized we needed to nerf you, so now the AI is smart enough to constantly avoid the core strategy of your build (and only your build).” AI that acts more like a person is probably something players enjoy, if the mechanics allow them to feel successful against it: I doubt any player would prefer beating a stupid AI to a smarter one. But AI that makes you feel less effective (especially of the kind designed explicitly to nerf a particular build unless they have help from a friend), is never going to fly because it makes the game less fun.

  • Maklak said,

    I was just playing freeciv and the AI in that game is pretty frustrating. Then I saw this article just popped up. I’ve played Civilization 1 and got pretty good at it and generally like 4X games, but freeciv is another matter entirely. It seems that no matter what I do, it keeps declaring war on me for no apparent reason, wouldn’t accept a cease-fire even if I throw lots of techs into the deal and eventually I end up at war with everyone. And even one of those AIs can be frustrating as they keep sending ships that patrol the shore and kill my workers. I got used to having peace treaties and even trading in civilisations games so this is very frustrating.

    Sorry, I had to get this out of my system. Anyway, I’ve seen some articles in the past that the AI shouldn’t use an optimal strategy, but be fun to play against. Which is even harder to do.

  • alanm said,

    I think he’s spot on that the Trinity nonsense is a holdover from crappy AI. I’ve never understood how people can keep on enjoying dumb AI. How many times can you herd mobs into a static fire patch before it stops being fun? For me, maybe two or three times. Clearly this is why I’ve never gotten into today’s MMORPGs.

  • Jeromai said,

    I think the outrage around that change was more a reaction to a well-used strategy suddenly not working and making a good skill ineffectual.

    If fire tanks had a mass immobilize they could also build for, along with the burn patch, there would probably have been less furor. It was more that they were now at the mercy of having to be partnered with someone else, just to bring their power back to a level they were used to, which was galling.

    Conversely, if the smarter AI only affected certain factions of enemies (maybe new factions of more ‘difficult’ mobs), there may have been less protest too. (though players would still have sought the path of least resistance to their goals.)

  • McTeddy said,

    AlanM, People prefer dumb AI because it SEEMS fair. Enemies that telegraph their actions, and are easily predictable… the player understands what is happening.
    The problem is that our brains have a hard to time with accepting that we missed something. If an enemy crept around the whole level and backstabbed me… my brain will yell “CHEATING AI TELEPORTED BEHIND ME!”

    For proof of people’s refusal to believe things… look up the new X-Com. There are number of complaints that the AI cheats on the higher difficulties… that the game reduces your accuracy while boosting enemy accuracy.
    But in truth, it’s the exact opposite. Easy and Normal cheat in the player’s favor. Player’s feel strong, smart and in control which makes the game feel “Fair”.
    The human brain SUCKS at seeing reality. It’s all skewed by our opinions, emotions, and our egos.

    If my AI experience has taught me one thing… it’s that appearance is everything. The AI is nothing more than a system to provide the player with emotions.
    It doesn’t matter whether the AI is smart or dumb or cheater or fair-player, all that matters is that the AI entertains and immerses the player.

  • alanm said,

    McTeddy… AI that denies an actor the ability to react to a situation that’s destroying it doesn’t seem very fair to me.

    Now, if the AI knew that the fire patch was busy killing it but made some trade-offs based on it’s motivation, other tactical options and so on… that seems fair to me.

    Like I said, this is just me. XCom is a great example – the old one. Aliens would cooperate to flank you (even through destructible terrain), use their superior sensory, weaponry and psionic options ruthlessly… and flee and hide when they got near death. They certainly maneuvered out of fire patches. I haven’t heard many complaints about that AI.

  • McTeddy said,

    You really haven’t heard people complain about the enemies killing them on the first step off the ship? Or about hidden enemies killing them from off-screen? The fact that the aliens ALWAYS know where you are?
    I seem to recall ALOT of people being frustrated with the original AI*.

    The truth about AI is summarized in your own quote… “What SEEMS fair to you”. Good AI isn’t actually smart and trying to win… but it is designed to look fair and look like it’s trying to win… while secretly going easy on you.

    Standing in fire is definitely a bad thing… but that’s only because you are seeing it. If it was out of your vision, you’d never know it happened.
    This is the problem with “Smart AI” the player didn’t see the enemy circle around behind him, take his time aiming, and head-shot.
    The only thing the player saw was that an enemy was suddenly behind him and perfectly aimed for the head. Clearly… cheating AI with an enemy spawning behind him.

    * I love the game too… one of my favorites. But I remember a TON of complaints and even alot of personal swearing at it.