Artificial Intelligence

Community General Discussion
#1

So a while back I was watching a series on the development of the Halo games (I’d say whose it was but I seem to remember a thread that was locked for naming a you-tuber, so I better not). Any way, they mentioned that when halo Reach was made Bungie decided to forego intelligent AI in favour of fitting more enemies on screen. So my question to the community is…

Would you rather have large scale encounters against lots of fairly basic AI, or would you rather have smaller encounters with much smarter enemies?
It’s difficult to have both, since they are each quite taxing on the hardware, so we’ll assume they are mutually exclusive.

My thesis is that it is more fun to play against a small number of genuinely intelligent enemies than it is to play against a large number of predictable enemy types. Large scale battles are fun, but they get repetitive after a while, so I don’t believe that they should be used all the time, rather they should be used sparingly when it matters, and the majority of battles should be small encounters. The difficulty would be the same, but for different reasons. Too often in a large battle I feel like I’m playing a bullet hell shooter, when I’d prefer to be playing something closer to first person chess. Halo CE seemed to encourage audacious and intelligent gameplay (IMO), but Halo Reach and especially Halo 5 seem to just hand you some weapons and tell you to go off and fight twenty to thirty Covenant, with the emphasise on strafing and fighting from cover rather than flanking and quick thinking. I played through Halo 5 on legendary recently, and I completed the first mission without dying once, having never played the game on legendary, simply because the AI didn’t have the foresight to flank me while I was engaged, or to advance while I was on low health or reloading. These were the kinds of things that made Halo CE on legendary a frantic and constant fight. Too many times I felt like I was taking a breather during the battle.

But that’s my two cents on the issue. I want to hear what you have to say. Do you prefer the large battles of the more recent titles, or would you rather the fights to be smaller and against more intelligent foes?

TL,DR: Quantity or Quality?

#2

That’s interesting, because I remember the enemies in Halo Reach to be quite smart–sometimes annoyingly so. Sure, the Grunts would just run towards you with reckless abandon. But the Elites, Brutes, and Hunters were quite intelligent. I would question where this youtuber got this information about AI intelligence in Halo Reach being sacrificed for the number of AI units that could be supported on screen.

#3

> 2533274817408735;2:
> I would question where this youtuber got this information about AI intelligence in Halo Reach being sacrificed for the number of AI units that could be supported on screen.

I thought the exact same. Definitely would need to see the original source on this one until I accept it as true.

This sounds a bit fishy because how do you measure “AI intelligence” to begin with? I guess you could consider it as the complexity of their decision tree, but I question how well that actually matches the player’s perception of the AI.

The whole point of “intelligence” seems kind of moot anyway, because the only real purpose of game AI is to be engaging. And how engaging the AI is doesn’t just come from how complex it is, but in fact one could argue that the greatest deciding factor (as usual), is how smartly the designer creates encounters using that AI. For example, I don’t doubt that the AI of Halo CE may be very rudimentary by today’s standards, but that doesn’t change the fact that for me CE still has some of my favorite encounters. And that’s because the actual complexity of the AI is only a tiny part of what makes an engaging encounter. Level design, enemy placement, health, weapon damage, projectile speed, movement speed (the last for for both player and enemies) all contribute into the feeling of the encounter.

So, when it comes to the question:

> 2535431563534182;1:
> Would you rather have large scale encounters against lots of fairly basic AI, or would you rather have smaller encounters with much smarter enemies?

I don’t think this is well-posed, first of all, because “smartness” is not well defined (I’d prefer “simple” vs. “complex”). Secondly, because I don’t think either of these things are really that significant factors in creating interesting encounters.

However, I think both approaches have their advantages (e.g. lots of enemies with simple AI, or few enemies with complex AI). Lots of simple enemies gives the designer much more freedom in designing encounters. For example, you can imagine that if the AI in a game was so complex that any given encounter could only afford one enemy. That would significantly restrict what kinds of encounters could be designed, and after a while it would probably all start to feel kind of samey. On the other hand, putting more into AI complexity at the expense of enemy numbers could allow for more interesting movement patterns for the enemies, which might give more depth to encounters in a very abstract way.

So, my real verdict: I don’t think it really matters. The complexity of AI behavior will never be the deciding factor in how engaging the encounters are. Unless either complexity or enemy count is significantly restricted, a good designer will be able to work either option.

#4

If you had a few smart enemies, that would make gameplay more intense, a rather large amount of dumb enemies would be more difficult due to strength in numbers. An example of this is the Unggoy gathering up in war zone or having a large mass of grunts in Legendary campaign. So I would preferably have dumber enemies than smart.

#5

Personally, I prefer quality to quantity–but I disagree with the Reach assessment. I played through the Campaign, started at Normal and worked my way up to LASO. Once I had about 4 skulls on, the only way I could kill an Elite was to assassinate it–battlefield chess indeed!

#6

So you are asking for either a game like Halo 2 or like Halo 5. Halo 2 had somewhat small encounters but with smart ai (plus the lack of health on legendary and the fast time to kill did not help). Halo 5 has a somewhat enemy overpopulation, (looking at you, Unconfirmed, with your countless crawlers). Out of the two games though Halo 5 was by far the easiest (on Legendary).

Overall though I wan’t a game where the AI is in between. Not too many enemies on screen at once, and not overly smart to give me a headache like H2. I think Reach did it quite well, so a game with the AI in that region would fit my liking.

#7

> 2535461041207054;5:
> …but I disagree with the Reach assessment. I played through the Campaign, started at Normal and worked my way up to LASO. Once I had about 4 skulls on, the only way I could kill an Elite was to assassinate it–battlefield chess indeed!

Fair enough, it’s been a while since I’ve played Reach so I might be remembering it wrong. :slight_smile:

#8

> 2533274825830455;3:
> I don’t think this is well-posed, first of all, because “smartness” is not well defined (I’d prefer “simple” vs. “complex”). Secondly, because I don’t think either of these things are really that significant factors in creating interesting encounters.
>
> However, I think both approaches have their advantages (e.g. lots of enemies with simple AI, or few enemies with complex AI). Lots of simple enemies gives the designer much more freedom in designing encounters. For example, you can imagine that if the AI in a game was so complex that any given encounter could only afford one enemy. That would significantly restrict what kinds of encounters could be designed, and after a while it would probably all start to feel kind of samey. On the other hand, putting more into AI complexity at the expense of enemy numbers could allow for more interesting movement patterns for the enemies, which might give more depth to encounters in a very abstract way.
>
> So, my real verdict: I don’t think it really matters. The complexity of AI behavior will never be the deciding factor in how engaging the encounters are. Unless either complexity or enemy count is significantly restricted, a good designer will be able to work either option.

A comparison would be Halo 4 and 5 (I think)
Example:
In (4) Shutdown, (video example) while going back to a hallway, a Knight Commander with an incineration cannon appears. Instead of a one on one battle, with the player taking cover and using long range tactics, 2 crawlers also appear next to the cover you would want to use. This creates a distraction. Usually 2 crawlers are an easy target, but now you have to deal these 2 crawlers while the knight is either throwing a pulse grenade (yes these are dangerous) or shoots you. Your decisions rely on defeating the crawlers being aggressive towards the knight while 4 pillars(cover), a knight, and 2 crawlers create an encounter that is simple yet interesting.

in (5) Glassed, (video example) this time there are more enemies.In the beginning, all of the Promethean forces are concentrated in the center, with a few near the edges. In order to counter them, you have to go around the “bowl” and use what ever options you find. The left side has snipers turrets, and the right has turret controls and mid range weapons to use. The A.I amount is greater, but the ways you encounter them can vary.

Granted these examples are a bit random, and have their own issues, but I think it’s a good comparison between simple (many A.I) and complex (fewer A.I), and how both can be interesting.

#9

> 2535442569875751;8:
> A comparison would be Halo 4 and 5 (I think)
> Example:
> In (4) Shutdown, (video example) while going back to a hallway, a Knight Commander with an incineration cannon appears. Instead of a one on one battle, with the player taking cover and using long range tactics, 2 crawlers also appear next to the cover you would want to use. This creates a distraction. Usually 2 crawlers are an easy target, but now you have to deal these 2 crawlers while the knight is either throwing a pulse grenade (yes these are dangerous) or shoots you. Your decisions rely on defeating the crawlers being aggressive towards the knight while 4 pillars(cover), a knight, and 2 crawlers create an encounter that is simple yet interesting.
>
> in (5) Glassed, (video example) this time there are more enemies.In the beginning, all of the Promethean forces are concentrated in the center, with a few near the edges. In order to counter them, you have to go around the “bowl” and use what ever options you find. The left side has snipers turrets, and the right has turret controls and mid range weapons to use. The A.I amount is greater, but the ways you encounter them can vary.
>
> Granted these examples are a bit random, and have their own issues, but I think it’s a good comparison between simple (many A.I) and complex (fewer A.I), and how both can be interesting.

Eh, I don’t think either of those demonstrates particularly complex AI behavior. As is often the case with minor enemies in Halo, the Crawlers have (at least on the surface) very simple behavior, which consists of running to one position, shooting, then running to another position, and repeating this until they’re dead. The Soldier AI reminds me a bit of the usualElite AI, which contains a bit of dodge maneuvers and cover tactics. Then there’s the Knight in your first example, which generally just ends up chilling behind one of the columns until the player comes.

With all that said, I doubt AI complexity is something you can glean from watching a couple of video clips. The player can really only look at the number of different maneuvers a single enemy can make. However, for example, I’d struggle to say what level of awareness each of the enemy units have about each other, i.e., to what extent they’re all just kind of doing their own thing, and to what extent there are some large scale teamwork mechanics behind the scenes guiding their behavior.

Either way, I can’t consider that first as example of really complex AI, because neither the Crawlers, nor the Knight, do a lot. The Crawlers just charge, and the Knight chills behind the column until the player engages, at which point it goes through the usual attack and dodge maneuvers.

#10

> 2533274825830455;9:
> > 2535442569875751;8:
> >
>
> Eh, I don’t think either of those demonstrates particularly complex AI behavior. As is often the case with minor enemies in Halo, the Crawlers have (at least on the surface) very simple behavior, which consists of running to one position, shooting, then running to another position, and repeating this until they’re dead. The Soldier AI reminds me a bit of the usualElite AI, which contains a bit of dodge maneuvers and cover tactics. Then there’s the Knight in your first example, which generally just ends up chilling behind one of the columns until the player comes.
>
> With all that said, I doubt AI complexity is something you can glean from watching a couple of video clips. The player can really only look at the number of different maneuvers a single enemy can make. However, for example, I’d struggle to say what level of awareness each of the enemy units have about each other, i.e., to what extent they’re all just kind of doing their own thing, and to what extent there are some large scale teamwork mechanics behind the scenes guiding their behavior.
>
> Either way, I can’t consider that first as example of really complex AI, because neither the Crawlers, nor the Knight, do a lot. The Crawlers just charge, and the Knight chills behind the column until the player engages, at which point it goes through the usual attack and dodge maneuvers.

Good point, though I’m sure that crawlers in 4 could use the environment to climb walls and shoot you from there, unlike 5 where they are more static
I use the videos as a visual reference, mainly to check if I remembered correctly.
I probably used the wrong terminology, I guess should be saying enemy placement rather than AI.

#11

“Smarter” AI can mean a lot of different things. Some, like behaving strategically, are a challenge to simulate… while others, like flawless aimbots with bottomless clips and relentless reinforcements, are easier. Yes, I would prefer the former to the latter, but as you pointed out, hardware limitations are a thing. That said, I keep imagining my dream of a solo firefight experience that I could build myself in forge, and maybe even decide for myself the number, lethality, fragility, and intelligence of my enemies. Then 343 wouldn’t have to worry about a one-size-fits-all approach, I wouldn’t have to compete against teammates for kills, and the only way I’d be taxing my console is with the object limit.

#12

I mean technically, it wouldn’t even be realistic for it to be a straight trade-off either.
Think of the bystander effect, there’s a similar concept for battle/war but not sure of the terminology…
But basically, 100 enemies are going to be way more confident & just charge without thinking, whereas a group of 10 are going to think more tactically since they’re more vulnerable.
So just scale the AI to be smarter when it can, and bring it back when it’s not able to (note: not that I’m implying that’s necessarily an easy thing to do).

One thing I will say is that Reach definitely required way more tactics than H5.
Not sure if that has anything to do with the programming though, the levels just don’t support it.
Despite adding these “hidden passages” and stuff, H5 remains an extremely linear experience.
It does a lot to create the illusion that you have options & scale, but the alternatives usually aren’t very viable.
With reach, certain places had a ton of different ways you could approach each scenario, and you had to be careful about alerting other enemies to your location. With H5, it’s like they all have psychic communication or something, I find it very boring.

#13

Definitely quality. Quantity gets boring.

#14

Depends on the game.

Halo? Quality, smarter AI is more fun to fight.

A game like Dynasty Warriors? Quantity is more fun.