The new forums will be named Coin Return (based on the most recent vote)! You can check on the status and timeline of the transition to the new forums here.
The Guiding Principles and New Rules document is now in effect.
Artificial Intelligence. No, intelligence is not the right word...
SkyEye
Registered User regular
Registered User regular
What do I want from the next-generation? Better graphics and sound would be welcome, but the law of diminishing return says the same leaps won't impress as much next time around. No, if there's one area that games could use improvement on, it's the AI.
Let's look at Halo. One of its many accolades is intelligent enemy AI. Each of the enemy races has its own way about things, and they certainly make a nice challenge on the higher difficulties. But then you look at the marines who accompany you and start to think, is humanity in 2552 really worth saving?
And strategy games. There are chess programs capable of grand-master level gameplay. But a lot of strategy games rely on cheap tricks instead of actually intelligent opponents. Advance Wars has war room maps in which you start with a third of your enemy's properties which you can still clear in 30 days. And there's a battle in FF Tactics which you can handily win without casualties despite the cumulative enemy level total being about 20--twenty levels higher than yours.
Maybe I just pick games with poorly done AI. Anyone here found AI comparable to that of a human with even intermediate skill?
Let's look at Halo. One of its many accolades is intelligent enemy AI. Each of the enemy races has its own way about things, and they certainly make a nice challenge on the higher difficulties. But then you look at the marines who accompany you and start to think, is humanity in 2552 really worth saving?
And strategy games. There are chess programs capable of grand-master level gameplay. But a lot of strategy games rely on cheap tricks instead of actually intelligent opponents. Advance Wars has war room maps in which you start with a third of your enemy's properties which you can still clear in 30 days. And there's a battle in FF Tactics which you can handily win without casualties despite the cumulative enemy level total being about 20--twenty levels higher than yours.
Maybe I just pick games with poorly done AI. Anyone here found AI comparable to that of a human with even intermediate skill?
SkyEye on
0
Posts
ArmA 2. Lacking in certain areas, but in others is very good.
Since we have graphics and online connectivity out of the way, AI seems like the next big leap for gaming. You're in luck.
I remember an article about that. The player was the only person left against an Alliance of three other mega-civs. The reason he wasn't destroyed is because one the AIs was aggessive and didn't want an 'Allied' victory, while the other industrial civ knew this, and was building ships to combat his inevitable betrayal. The player managed to sneak in a tech victory. I think that was the jist of it.
I don't think so. Most of the advancements we've seen in gaming thus far have been due to brute force rather than ingenuity - now we have enough CPU power to render 16 colors, now 256 colors, now thousands. 1000 polys, now 10,000, now 100,000. AI isn't something you can brute force so easily. You're not talking about simply displaying the world, you're talking about replicating our own thoughts and feelings. There's a whole field of computing/science based around this that, while making incredible progress, is still a slow and steady process.
Look at Dwarf Fortress, the game with some of the most incredible AI I've seen. Dwarves have individual wants and needs, and when they want something they go and get it. They prioritize. They get sad when family members and friends die. Things have a lasting effect upon them. And yet it's still full of really stupid, erratic behavior, stuff that shows a computer is obviously a computer behind the wheel. And whether from sloppy programming or not, it still slows computers to a crawl (of course it does, computing water/physics/pathing/AI for hundreds of creatures).
Of course it will improve in the future, but it won't be a generational hallmark as in the past (2D, 3D).
Having a complex AI that anticipates problems ingame and is able to account for this will be seen as 'cheating' by new players who dont quite have a grasp of the finer details of the game.
Games with a difficult learning curve but once that learning curve have issues with the game starting too hard and ending too easy.
The real goal I think of developers will be to create a system that detects and plays up to the skill of the player without the apparent 'rubber banding' that people complain so bitterly about.
Admittedly we have come a very long way from say hot spots to flanking, but even in games with good enemy AI and party AI you get alot of bizarre 'retarded' play that befuddles the player.
So I dont believe more complex AI is the next big thing, rather a more perceptive understanding of the player skill based off ingame performance, that allows the game to on the fly compensate for initial skill and learned skill of the player to make a more rewarding experience throughout.
In some ways that's good though. You want simulated, otherwise ethical issues abound.
I think it will get better, it has been if you really look at it. I mean even back in the psone days, they could only run X number (three I think) different routines per group of zombies, even though they had seven or eight different ones. Brute force fixed that.
It really is all about the faking it though, even if you have a group of smart enemies doing some real strategy unless there's some window into that they can seem just as stupid. Though the quirks and ghosts in the machine that pop up are I think one of the most fascinating sections of development to look at. Stuff like the way those gunships in Half Life ‘learned’ to shoot down incoming missiles.
Okay, let me rephrase that: am I the only one thought that that friendlies in the Halo series, particularly the later games, didn't suck horribly, unlike the other 98% of friendly FPS fodder?
Is it just that FPS have historically abysmal friendly AI even when their enemies are start? I mean, Half-Life wowed me with its enemy AI. Maybe I'm just waxing nostalgically, but there were times when the Black Ops forced really wowed me throwing grenades around corners, taking cover, etc. But the friendly AI in both it and its sequel (along with the enemy AI in the sequel, unless it was scripted) was stupid as a brick. Seriously, those fuckers were so incompetent, I wonder, "Christ, maybe the Combine's only crime to the Universe is failing to kill off humanity faster."
It's probably just because they're physically tougher in Halo (especially stuff like Friendly elites), but given that they can actually drive vehicles from one point to the other puts their ahead of the curve, sadly. And let's not forget what qualifies as 'AI' in the Battlefield games...I mean, sure, you're not supposed to play alone, but a few combatants who weren't amazingly stupid would be helpful in rounding out team numbers.
Best option really.
Disgaea, [Franchise Name here Tactics], Fire Emblem and Advanced Wars all manage to be fun despite the lack of any real AI.
That's because winning is fun. It'd be easy to make an AI that plays perfectly, but then they'd do what you do. Huddle up in a corner and then mob the first person to come close.
Of course, that assumes that I utterly destroy the opposing side. If there is a recurring character who gets his ass kicked, comes back using the same tactics and gets his ass kicked again I start to feel let down. But generally, even if they don't get a tactics upgrade, they at least get a power upgrade.
I doubt we'll see any real advancement in video game AI until the advent of organic computers or real limited AI
You have a very poor understanding of AI if its nothing more then if/thens. AI is nowhere near that simple. You have pathfinding algorithms, goal based algorithms (what do I do at this moment? reload? run for cover?), group algorithms (I am grouped with this guy I'll distract, you flank).
AI is advancing and there are some test games which use neural networks to monitor how players (humans) play. Then use the information gained to improve the AI. So as people play the AI improves and adapts to how humans fight. The problem right now is that it takes a lot of time to implement such systems. Its not that they are advanced, its that people just accept current game AI as "that is how it is" or "AI is always stupid". You dont need organic computers for AI. We have the technology right now, we just need to improve the algorithms speed and overall design and we can have very challenging AI.
We don't even know if real AI or organic computers will ever be possible.
Real AI isn't needed anyway.
The current "fake" AI in games coule be 100s of times better than it is now. It's mostly a matter of how much effort (time/money) they put into it. It's just that dimishing returns also apply to AI. Sure you could make the AI seem way smarter, but only a small percentage of the audience would really appreciate the extra effort, and you can't show great AI in a screenshot.
EDIT: Or offer to let some hopelessly outgunned mook live if he'll deactivate security and give me his pass ID.
To the solsovly, Chessmaster games. I find it hard to believe that Advance Wars AI can't think 5 turns ahead when Chessmaster planned out your demise from setup since what, 20 years ago?
To Aroduc, tactical RPGs may have that problem, but not Advance Wars-games with resources. It's like RTS; if a player refuses to expand, he or she or it will get steamrolled.
This.
Guess what the appropriate action is when I cut down your 3 bandit buddies in a single swing each or walk by you wearing equipment that is not only worth more than the GDP of your entire kingdom but is also glowing? Hint: It's not to attack me mindlessly. :P
It's one of the things I always loved about sitting around a table playing DnD with friends. When we were outnumbered 2-1 and we killed half of the attackers in the first round the rest would scatter while swearing/screaming.
I always wondered how Samson killed 1000 Philistines with a club. You'd think after the first 400 corpses piled up, they'd run away.
It isn't a question of "can the game-controlled friendlies and enemies be smarter?" because they can. It is a question of "what am I willing to make simpler to make the AI better?" Look at the slow-down suffered in, say, Shadow Complex when you fall too quickly or in Fallout 3 when you have too many tin cans on screen simultaneously. Now, if the console can't handle everything that the game is throwing at it already, it obviously would be brought to its knees even quicker if it is worrying about what Suzy McGhoul is interested in today instead of just parking her ass in the bar.
There is a reason that Dwarf Fortress looks shitty, and it is not because the authors suck at or don't care about graphics (at least not completely).
You may not be able to show it off in screenshots, but if your Bethesda you sure can make people gush about it in every review. :P
But I agree, NPC AI can be much better. In fact I hope one day the NPCs of the future make our current NPCs look about as life like as their 16-bit counterparts of the 90's.
When it comes to whether or not it's worth it to make the AI better seems to depend on the game really. For RTSs, TBSs, Sports games, and maybe even RPGs I would certainly love to see vast improvements and I think a lot of fans would too. For platformers, FPSs*, and hack 'n slash games it doesn't matter too much to me.
*depends on whether it's supposed to be an action FPS or tactical FPS.
Ok so, this wasn't me, but it's one of my favorite GalCiv stories of all time. It's absolutely gorgeous.
As you may know, the AI before declaring war will check their military strength against whatever they know about your military strength. However, this check is only to the immediate. As in, if you have an ally, they won't check the ally's military at all.
So this player used this to their advantage. You see, they basically had, for reasons inexplicable, the entire fucking galaxy wanting to beat their ass in. They also were a race of about 3 weak planets and one old busted down frigate. They're like 1939 Poland in space. Or like Canada without the rocks, trees, and syrup. And the galaxy wanted to fuck them up something good. Their political discussion consisted of "YOU WON'T GIVE US MONEY FOR NO REASON? THEN DIE!" from everyone else. And why not? Their military was a German Shepard with a space suit on, so the Battlecruisers with super-lasers figured they could take them. Of course one race didn't hate them. This race was the Drengians.
For those of you who don't know, the Drengians are the meanest, baddest, most fearsome military in the history of the galaxy pretty much in every game. The Drengians are like what the Klingons with the fleet strength of the Cylons would be. But better.
Anyways, the Drengians loved these fuckers. Maybe they thought they were cute or something, but these guys had one ally, and that ally had a few Death stars lying around in the garage.
Anyways, I'm sure you can see where this is going with the anecdote I gave at the beginning. Race after race would want to wipe out these ass fucks, check the military tables, and then declare war and send a massive fleet to take out Spot, the defensive hound. Then the Drengians would see the war and fleet and go "NOT OUR LITTLE FRIENDS!" and come down like the hand of god onto whatever foolish race tried to wipe this guy out. Of course, since this guy didn't actually, you know, fight, they didn't get any new planets or anything like that, but the Drengians got to blow shit up, so they were happy, and the rest of the galaxy saw these wars and figured they needed to wipe out this stupid fuck of a race that just killed off all their friends, and hey look their military is Rover!
Rinse, Repeat.
AI
Space Empires V has a surrender option and a demand surrender option.
After countless games of SEV, I'm still entirely unsure as to why these exist. Once I was in a massive game that by the end only had three empires remaining. I had about 2/3rds of the sectors under my control, and had spent outrageous amounts of materials and time blockading the fuck out of my borders. It would take forever to break through, and I had a drone defensive grid as well as some retaliatory "cruise missile" style drones(also used to punch through any defense they have).
Ok this isn't so bad. Sure, it'd kick their ass to attack, but if they had enough construction planets they could. And I turned the AI all the fuck the way up. Well, I get bored, and of course take out one drone group breaking through their borders and bombing one races shit without any actual risk to any fleets. Then I start sending in the dropships and a few.... "peacekeeping" fleets. So at this point I control almost 5/6ths of the galaxy. I figure considering the obvious way the war is going they're fucked, and send a message to one opposition race, being benevolent as I am, and demand their surrender. They refuse.
So I take them down to one planet. Another message. Nothing.
Well then the drones took them out. Bastards.
Next race! Same goddamn thing I controlled every sector, had them down to one planet, which was blockaded and surrounded and there was literally nothing they could possibly do to make any difference. There was literally 0% chance, unless I decided to lose, that they would be able to cause any damage to me.
Seriously why even have the surrender option.
I read it was the jawbone of a donkey?
If it's like Civ4, then they will never surrender to you if they hate you enough.
Right, and they hate you more if you are at war with them.
Pathfinding can be very easy. If you are working on a platform that has some sort of grid system (like Defense Grid) the A* pathfinding algorithm can handle that extremely quickly. A* is used in many games including Dwarf Fortress. Now A* and many other algorithms are dependent on either a position system that has a single set of coordinates or some form of node graph. A node graph would be used in a 3d world where you can move all over the place. The idea being every x units you place a node where an AI is able to walk. This is why in MMO's you see MOBs sometimes make a dart in the wrong direction to go around an obstacle. It simply follows a grid of points on the map that you can't see to move around. To make a pathfinding algorithm that isn't dependent on some form of node graph is extremely complex. FEAR with its AI has nodes everywhere marking locations where an AI can jump down or leap over for cover. It is very difficult to parse information from a purely mathematical standpoint.
But, yea, most people here are way too ambitious in their goals. I'd settle for pathfinding working consistently in most games:
http://www.youtube.com/watch?v=lw9G-8gL5o0
Here's an article highlighting a possible solution, if you're interested and technically versed in the subject.
(This comes from a developers perspective by the way. It is interesting to see what you guys have to say about it.)
We've had the ability to create an intelligent opponents for some time. In fact, a simple bit of fuzzy logic (well thought out and well tested of course) is usually way more than you ever need. However, most enemies you come across are none to smart. This is on purpose. Playing an intelligent AI is really no fun.
I do agree somewhat, that AI could be improved upon in that better techniques aren't always used.
State-machines are still heavily prevalent for trying to do complex behaviors.
In general, the rule of thumb is not to create something bordering intelligence, but to allow the player to project whatever behavior they think the enemy should be expressing onto the enemy. For instance, in Halo 2, in a lot of situations you would walk into a room and there would be an elite and a bunch of grunts. The elite would howl in rage and point at you and the grunts would charge. When you see this you think, oh, that is the leader and he is giving orders or some such. This is merely theatrical.
As is most game programming actually. Haha.
Care to elaborate?