Next-gen AI and Next-gen BS

[Shamus Young]

Next-gen AI and Next-gen BS

Lately I've been getting a lot (okay, a couple) of questions about artificial intelligence as it applies to gameplay. Is the new console generation going to give us better AI? Are these marketing claims true? What's going to change about computer players now that programmers have all this extra power?

Read Full Article

 

[VoidOfOne]

Yeah, never really thought much about AI until years later as a gamer. Hearing various segments on AI on NPR, I get the feeling that when better AI will be made, it most likely will not be from the games industry. Maybe from a gamer, but that's another story.

In any case, very good story. I still feel skeptical when someone mentions X will make games better, in one way shape or form. My X being the cloud, or cloud servers. I still don't fully comprehend the matter, and an explanation from anyone is welcomed.

 

[rofltehcat]

It is really sad that today's (combat) AI seems so frigging stupid. Sure, as you pointed out making them strong is easy but strong and smart are also completely different. Why does the majority of today's AI feel that bad? Did the publishers at one point just cut the AI budget in half or simply froze it while all the other costs exploded?
I just want 2005 AI back
I don't even expect it to be better than 2005 AI. But at least it'd be an improvement over today's titles.

Would you consider an AI in fast-paced RTS-games to be strategy AI or combat AI? They certainly have the capabilities to choose counter units and tech paths, so they'd be strategy AI, but on the other hand they can also go crazy on micro, so they need to be tuned down to human levels, control groups of units like a player would as opposed to controlling single units, maybe formations? Tactical AI?
Of course there are also people trying to write the strongest possible AI: example youtube video [http://www.youtube.com/watch?v=IKVFZ28ybQs]

As a side note I wonder why Thief doesn't seem to be able to handle more than 5 guys at once. Is it some exponential resource usage problem in the AI? Or is it because of the graphics/physics/whatever engine? Or another bug in the AI? Or some problem stemming from having the game on all different platforms (e.g. PS3 architecture and/or XBox360 limited RAM causing problems)?

 

[hermes]

Cool feature.

The thing with AI is that a lot of it was invented a long time ago. Not that there are no optimizations and small breakthroughs here and there, but the big branches of AI research have existed since the 60s to the 80s (which, for computer science, is pretty much prehistoric). New AI in games does not come from new algorithms, but from inventive ways to use old algorithms to create new effects. The amount of variables an entity can take into consideration before taking an action may increase, but the thought process is pretty much the same for most games.

The only exception to the rule, as Shamus said, is when the number of entities grows. Each entity has to go through the thought process for its own state, and that means more and more is spent on keeping tabs on the internal state of every entity in the game. It is truth that its peanuts compared with the power needed to render every entity, but there are optimizations to that too (like only working on rendering those that are visible) that don't apply to the AI. The clearest example on the last generation was on the behavior of NPCs in games, with things like Assassins Creed mobs patterns and scale not being possible in the PS2/XBOX era; and I can't wait to see what the new AC or the new Arkham can do with the extra power...

 

[hermes]

It is really sad that today's (combat) AI seems so frigging stupid. Sure, as you pointed out making them strong is easy but strong and smart are also completely different. Why does the majority of today's AI feel that bad? Did the publishers at one point just cut the AI budget in half or simply froze it while all the other costs exploded?
I just want 2005 AI back
I don't even expect it to be better than 2005 AI. But at least it'd be an improvement over today's titles.

Pretty much.

As I said in a previous post, most of the AI research was done during the 60s. That means the field has been quite frozen for many years; there is little that would be considered "new". That is the reason modern shooters' AI does not differentiate much from Half Life 1 AI (and calling it "AI" might even be a little far fetched, most of them are hardly considered "weak AIs"). In terms of programming, they are not that different. The only difference is in the scale...

Since investing in AI research is a dead end (so much so that games like FEAR and Rage used their AI engine as selling points), and most people would consider the current state of the art to be "good enough", there is no point in most studios spending resources on it, the same way no one spends money on making a new 3D and rendering engine when Unity and Unreal are a lot more cost effective.

 

[Bad Jim]

There's another way to categorise AI. AI that uses a few if/then statements to decide what to do, and AI that actually tries to calculate the best move.

The first kind is what nearly all current games use. It's cpu efficient but it is limited to the tricks that the programmer will teach it, and there are always exploitable holes in the logic.

The second kind is how chess AI works. It still makes mistakes, but it is much less exploitable and can potentially be much smarter than the programmer who designed it. The catch, of course, is that its' effectiveness depends on the cpu cycles you can feed it.

Galactic Civilizations 2 does actually attempt to analyse the game like this, and it is noted for its' smart AI. And yes, that AI still has many flaws, but it's also considered to be miles ahead of other 4x games.

 

[rofltehcat]

The second kind is how chess AI works. It still makes mistakes, but it is much less exploitable and can potentially be much smarter than the programmer who designed it. The catch, of course, is that its' effectiveness depends on the cpu cycles you can feed it.

Galactic Civilizations 2 does actually attempt to analyse the game like this, and it is noted for its' smart AI. And yes, that AI still has many flaws, but it's also considered to be miles ahead of other 4x games.

I like that idea because it surely can also be used to then choose one of the "less than optimal" options, making it seem more human.

 

[Rad Party God]

Yup, we rarely see '98 levels of AI (Half-Life, Unreal, Thief), let alone '05 levels of AI, wich was pretty much the pinnacle of FPS AI (Half-Life 2, Far Cry [that one is debatable], F.E.A.R., Crysis in 2007).

I really hate to generalize, but sometimes I get the feeling that the more the time passes, the dumber the AI gets.

Heck, speaking specifically of Thief, The Dark Mod, a fan made free game using Doom 3's engine, has a much better AI than the new Thief, heck, it let's you move more freely (rope arrows on any wooden surface), put down torches and candles by shooting water arrows indirectly to them (let the spreading water put them all out), you can literaly grab torches and light other torches and candles and a whole lot of other neat features.

And all that was made on the Doom 3 engine by people on their spare time.
And it's completely free [http://www.moddb.com/mods/the-dark-mod].

 

[Bad Jim]

The second kind is how chess AI works. It still makes mistakes, but it is much less exploitable and can potentially be much smarter than the programmer who designed it. The catch, of course, is that its' effectiveness depends on the cpu cycles you can feed it.

Galactic Civilizations 2 does actually attempt to analyse the game like this, and it is noted for its' smart AI. And yes, that AI still has many flaws, but it's also considered to be miles ahead of other 4x games.

I like that idea because it surely can also be used to then choose one of the "less than optimal" options, making it seem more human.

Lol no. It would then just seem completely retarded rather than slightly retarded. I said _try_ to calculate the best move, not _successfully_ calculate the best move.

Chess is a bit of a special case because a computer can see every possibility half a dozen moves or so ahead, while humans will make plenty of mistakes that are exploitable in half a dozen moves. It has no real strategy, it just doesn't need one.

Most games aren't like chess. But there can still be a lot of room for the AI to consider various courses of action. But the AI will not be able to consider everything in most games.

 

[WarpZone]

It's sad that this isn't the scummiest thing I've read about video game marketers in the past week.

Hell, it isn't even in the top ten.

This is a dark, depressing time to actually care about video games.

 

[DonTsetsi]

Of course there are also people trying to write the strongest possible AI: example youtube video [http://www.youtube.com/watch?v=IKVFZ28ybQs]

That AI is cheating. It uses information unavailable to a human player, namely, which zergling is being targeted by each attack.

 

[Olas]

I don't know much about AI and only have the barest experience with programming, but it always struck me as odd to hear people talk about how CPU speed will improve AI because it doesn't seem like AI should be that CPU intensive. It's just a bunch of instructions after all. Even if enemy AI is incredibly elaborate with thousands of lines of code it still shouldn't stack up next to graphics. It's nice to see I'm not crazy.

Side note, it always seemed unfair to me that in games like pong the same computer that controls the enemy paddle also controls the ball's trajectory.

It is really sad that today's (combat) AI seems so frigging stupid. Sure, as you pointed out making them strong is easy but strong and smart are also completely different. Why does the majority of today's AI feel that bad? Did the publishers at one point just cut the AI budget in half or simply froze it while all the other costs exploded?

They probably realized you can't show off AI in screenshots.

 

[Bad Jim]

Of course there are also people trying to write the strongest possible AI: example youtube video [http://www.youtube.com/watch?v=IKVFZ28ybQs]

That AI is cheating. It uses information unavailable to a human player, namely, which zergling is being targeted by each attack.

No, siege tanks follow predictable algorithms and the AI can calculate which zergling will be hit. If you were playing in a time warp or if you were like The Flash, you could do it too.

However, it's really just being fast, not all that clever. How does it modify its' production after scouting the enemy I wonder? How does it interpret a bunch of enemy units at the edge of its' vision? Does it realise, as humans would, that there are probably more units it can't see behind the fog of war? If it sees ten zealots on two seperate occasions, can it decide whether they are the same zealots?

 

[DonTsetsi]

Of course there are also people trying to write the strongest possible AI: example youtube video [http://www.youtube.com/watch?v=IKVFZ28ybQs]

That AI is cheating. It uses information unavailable to a human player, namely, which zergling is being targeted by each attack.

No, siege tanks follow predictable algorithms and the AI can calculate which zergling will be hit. If you were playing in a time warp or if you were like The Flash, you could do it too.

However, it's really just being fast, not all that clever. How does it modify its' production after scouting the enemy I wonder? How does it interpret a bunch of enemy units at the edge of its' vision? Does it realise, as humans would, that there are probably more units it can't see behind the fog of war? If it sees ten zealots on two seperate occasions, can it decide whether they are the same zealots?

Unless your enemy manually selects the targets. So, if 2 incredibly fast people were playing against each other, the tanks would still win.

 

[Arakasi]

"It's such a hard problem that in the early days programmers usually cheated and gave the AI extra resources to make up for the massive cognitive handicap."

And in the more complex strategy games, like Civ V, they still do.

I always thought it'd be easiest to just keep adding to the AI throughout development, programming from the most generalised situations possible to the most specific up until the point that it becomes pointless. They probably already do that.

I'd buy a full priced Civ V re-release if the entire AI system was redone, and all difficulty related to the AI's competency alone. But of course, that is a monumental task.

 

[Alpha Maeko]

The problem with AI, in my opinion, is that developers refuse to make the AI movements and action interact. If you're strafing sideways, you need to be less aware of things around you, because you're paying too much attention to not tripping while strafing (just an example).

As it is right now, AI is a dot on the map that moves anywhere it pleases with no navigational or awareness trouble (always locked dead center on your person, only missing if the accuracy RNG gods say so under specific circumstances, etc.)

They need to have a front, a back, and sides. Their legs are not wheels. They should actually have to turn in a direction before they can move in it. Strafing can be a thing, sure, but it needs to be a blind sideways walk, not a coldly calculated automatic movement at full speed that never causes pathing issues. (The AI should STRUGGLE to keep themselves situated and aware that the wall they just ran into will make them fall over, not just skate around on roller blades effortlessly in any direction with no consequence)

TLDR?

Put more work into AI movement and awareness. It shouldn't be perfect or effortless.

 

[Bad Jim]

It is really sad that today's (combat) AI seems so frigging stupid. Sure, as you pointed out making them strong is easy but strong and smart are also completely different. Why does the majority of today's AI feel that bad? Did the publishers at one point just cut the AI budget in half or simply froze it while all the other costs exploded?

They probably realized you can't show off AI in screenshots.

It's worse than that. Most people can't even properly evaluate the AI after they've been playing for weeks. A lot of it is about perception.

Ever played a 4X game like Civilization and found out, after spying, that the AI factions are all at about the same tech level, even the tiny ones that are barely surviving? Doesn't it annoy you that the game is obviously giving them tech at certain intervals rather than make them do proper research?

Well, suppose the AI plays by the rules. Should it trade tech? If the other faction is not a threat, then yes. It is also rational to sell even advanced tech to minor factions at prices they can afford, since nothing is lost and something is gained. The result of all this logically justifiable tech trading is that each faction ends up with the same tech.

Conversely, it is amazing how often people will get caught out by an AI with random behavior and assume it was being clever. I heard one little anecdote about a developer who made a pool game and got numerous complaints that the AI was too good at lining up its' next shot. But this was entirely down to chance, as the AI simply potted balls and had no code for lining up it's next shot.

Another thing about AI is that if you announce you have advanced AI, someone will invariably find a hilarous flaw and post a video on Youtube. A lot of people will, in fact. And making a sophisticated AI is no defense against this, because try as you might, someone will find it. Even full strength chess AIs have been caught making hilarious errors.

So if you know that most people cannot tell good AI from bad AI, and that announcing good AI is likely to backfire, why would you invest heavily in it?

 

[Artemis923]

We'll have to settle for if(), then () until we have true artificial intelligence.

Then we can just ask our robot overlords how it all works.