He said, among other things, the goal of hte AI is not to beat the player but to give the player a strong experience.... I think that's bullshit, the AI should do everything possible to kick my ***. I'm not having much fun unless it's really challenging and my victory is only by a narrow margin.|||The AI should always try to beat me, otherwise, what's the point really?
The best campaign missions in FA were those were you were under attack right from the beginning, outnumbered, outgunned. I think it would be nice to have an adaptable AI. If you are doing well the AI gets more resources (and additional units in the campaign).|||IMO, a strong experience should be a fight for my life!
RTS gamers don't particularly want to watch pretty explosions and special effects - I thought the world learnt this when EA took control of the CNC franchise (God rest it's soul).|||I would like to know Sorians opinion on this.
Also, I wonder if he is talking about Skirmish AI or Campaign AI, which in GPG games has historically been scripted.|||FunkOff|||Quote:|||Difficulty can be annoying.
Ever played demon soul ?|||Nerdfish|||Speaking of Cybran campaign missions. I think the mission with the crazy engineer swar was fun and very entertaining. It was a very unusual scenario and you really had to adjust to it. And I liked how the AI sends some engies around the back later. I had to recapture a bunch of buildings during my first playthrough. :)|||It is fun to face weird strategies like that. XD|||I agree the AI needs to try it's best to stomp someone into oblivion. I agree that the AI needs to be fun to play, though that could be accomplished with other means, but the "hardest" setting for the AI needs to provide as much of a challenge as possible.
I think it might be fun if the AI could choose unorthodox strategies in skirmish mode, like the engy rush, and try to adapt different strategies together. In other words I want AIs to act like humans.
Of course, we are dealing with today's CPUs so I suppose that might not be possible right now...|||TehClawEternal|||lol yes, RAEG quitting AI XD|||challenge is part of the experience. I think CT's point is the focus of AI design, isn't to kick FunkOff's *** as hard as it can, but to give the kinds of experience that all players want (difficult, relaxing, epic. or something else) There won't be just one level of AI so no one has to worry. Ragequitting AI might exist, it won't be difficult to code, lol|||FunkOff|||On the topic of rage-quiting or GG'ing AIs; UT3's AI try some really ballsy tactics in the last minutes of a game. Greed with the Titan mutator is incredibly lulzy because the AI loves hoarding kills and then unleashes a team of walking tanks with rapid-fire rocket launchers.
If you're in a losing position, there's really not much preventing you from trying a high-risk ploy. Worst case scenario is that it backfires and you confirm that you were going to lose.
I'd love to see an RTS AI look at its cards, realise it can't win conventionally, and then try to throw a curve-ball.|||I agree with both FunkOff's view and CT's view. I want the AI to try its damn best to steamroll me, but I don't want it to be in a BS or unfair way.
Using Supcom 1 as a reference: they shouldn't get nukes 5 minutes into the first campaign mission without the player's knowledge, but then it could be a challenge to get your antinuke up as fast as possible if the player is told from the start.
My opinion is that you should make the AI fun and fair first, and then throw in a "motherf**cking hardcore mode" for those who like a challenge. Just jack up whatever you come up with for Hard mode with insane multipliers and a few surprise units or something, you don't have to go scripting an entire new difficulty (although I'd love to see that).
I'm not talking about something like Supcom 2's Hard difficulty either, I mean something that will kick my *** 30 times from sunday until I perfect my strategies and even then, I should narrowly avoid certain death. I think that right there will enhance the campaign's longevity, just by being a sort of "challenge mode" because you just can't quite beat that last mission even after the umpteenth time. Who's with me on this?
WARNING: GIANT POST AHEAD __________________________________________________________________________________
(Yes, my post is so long awesome it needs a divider)
I think this thread is the perfect place to pose a problem that I see with campaign AIs, and even AIs in general.
When it comes down to a player versus the AI, the AI almost always needs some sort of a large bonus in multiple areas to stand up to even a mediocre RTS player. So, in a narrow example, for every 30 tanks the player may build, the AI probably has 50 or more. Now what happens here is that makes it ridiculous for the player who actually wants to try any sort of strategy that requires fair-fighting without a lot of stockpiling of units while the AI smashes theirs against your PD wall, or abusing the AI in some way, such as kiting it to death with long range experimentals. Thanks to this, on any sort of challenging difficulty, the player usually doesn't even bother creating an army until they're absolutely sure it can wreck the enemy, because any fair-and-square army could never win against what the AI has to begin with.
This is my opinion, so I'm sure some would completely disagree with it, but I think there needs to be some sort of encouragement to actually send out units early on and fight battles and not just turtle up. Artifical incentives such as saying "destroy the nuke launcher next to your base within 10 minutes or your face (and base) goes splat" work alright, but that's a bit too constrained and only encourages the player to focus on that singular goal and nothing else. I'd like to see the campaign maps require more tactical thinking where the game evolves more dynamically because of what the player does and how the AI reacts to it. The problem is figuring out how to accomplish this.
First, the player needs an incentive to move out and build armies from the start, but he should still be allowed to do whatever he wants if he so chooses. What could this incentive be?
Second, the player should be able to use what he builds against the AI in some productive way, without having to overcome impossible odds, but while still keeping the game at least somewhat challenging and fun. How could this be done?
And Third, on the harder difficulties the AI should act in an unpredictable manner, or at the very least a manner where the player cannot effectively prepare for every tactic it has available to use. Nothing BS like nukes or artillery rain, but things like a transport drop in an area where you have weak defenses or a bomber snipe on your power or something. I don't think that these should be sudden and crippling, so the player should get at least some warning to prepare, but not enough that they can go. "oh, he's doing strategy X, I'll get counter Y in spot Z and forget about it" Something more along the lines of "Oh **** he's doing strategy X, what do I have that can stop it on hand, and can I build enough of counter Y in time to respond!?"
My super-long-wall-of-text-idea of how to answer all three questions:
Have the AI try high-risk strategies fairly often, that the player for one reason or another knows about. Teaching the player in the first few levels of the game to scout often could help this (although far from a perfect solution) or perhaps adding a small ally-AI that performs this action itself. Think of the first mission in Forged Alliance, and picture the ally bases sending out scout planes once in a while. You can fudge it a bit and let the ally-AI know exactly where to scout in this situation.
An example of this could be the AI moving some builders and a token force (maybe relative to what the player has?) of troops to a location near the player and starting an expansion base to threaten the player. If they have units to respond with, they can attack the expansion base and prevent the AI player from gaining a foothold, encouraging the player to build units pre-emptively. As an extra bonus, thwarting the AI's plan to set up a base is both fun because you get to blow **** up, and satisfying because you feel you accomplished something worthwhile. Do NOT let things such as this become tedious for a player, however. They should not happen very often, or in some way stay interesting through even a long game.
The AI should also have targets of opportunity the player is able to exploit if he chooses a strategy that involves building lots of units. Something such as a lightly defended base containing the power generators to power the AI's shields come to mind. One critical thing is that these bases should not be acting in a vacuum as far as the AI is concerned; it should still try to defend them to an extent, but it shouldn't use all 300 of its tanks/aircraft/experimentals to beat back 50 of the player's tanks/aircraft/experimentals (At least not on the lower difficulties. But on OMGWTFBBQ difficulty I want them to do absolutely everything in their power to stop me).
The big key to this is getting the AI to analyze what the player is doing at least on a basic level and respond in a somewhat smart manner. The work Sorian has done on the Supcom 2 AI is miles farther in this direction than the AI in any other game I have ever seen, period, so I think it can definitely be done.
I'd write a TL;DR section, but even that would still be TL;DR
.|||AI should have a cheese mode.|||Destroyer224|||This is a longtime debate, between an AI that "Plays to Win" and an AI that is designed to provide a good game experience. The latter doesn't mean that the AI isn't trying to defeat the player obviously, it just means that first and foremost the intent is to provide a fun and interesting gameplay experience.An example of the latter system is some of the different Leaders in Civilization 4, if people here have played that. Montezuma is a bloodthirsty, rabid, dog. No matter what you do, there's a very high probability that he's going to declare war on you, because he's just that aggressive. Now maybe that aggressiveness isn't the best plan for the AI, maybe a war isn't necessarily in his best interest, but he does it anyway because that's that's how his personality was coded. Just because the AI was designed for a fun gameplay experience doesn't mean that it's not going to curbstomp you.
RTS games generally lend themselves far more to an AI that plays to win, though, so I'm curious to see exactly what they're planning here.|||I think if they're gonna have an extra-extra hard AI campaign situation thing, they should make it multiplayer, at least. That way, not only does a player have to devise the perfect strategy, they have to perfect teamwork too in order to stand a chance. The survival defense/attack custom maps made for FA were such a big hit because they challenged players to do both of these.|||Just got done listening to the podcast and I have to say, I agree with Chris to a point.
The goal of skirmish AI should be to make it fun, but also to make it not stupid. I am a firm believer of making the AI as intelligent as possible and giving players the ability to adjust the difficulty of the AI as they see fit.
We should still give them some default options (ie easy, normal, hard) even if it means the normal AI has to have a handicap to make it accessible to the average player. But, we also need to give them the ability to fine tune the difficulty with options like economy cheating levels, full map LOS, or even a limit on APM.
This would give us the ability to offer an AI that is fun to play against for the average player, and has the killer instinct that more hardcore players like.|||sorian|||I completely agree!
And maybe add in "cheese" builds into the AI and have them pop up randomly. :)|||FunkOff
No comments:
Post a Comment