Many people seem to be interested in improved graphics when new games are released these days. Perhaps I am a weirdo, but I am more interested in the A.I. To me, games already have been looking good enough. Even AOE 3 (released almost about 10 years ago) looks good to me. But the A.I. of AOE 3 is very very stupid. Even with the hardest settings, the A.I. does not get smarter, it just seems to have more resources to send more units. As an analogy, playing against the normal A.I. is like fighting with a brainless stupid zombie, and playing against the hardest A.I is like fighting with 100 brainless stupid zombies. I lose to 100 zombies, but not because they are smart.
Of course, these days, most players play online games with other human players. But to me, other human players are too good. And when you are playing with human players, you cannot pause and return to the game at your leisure time. So, I prefer playing against the computer.
I’m pretty sure they are improving the AI in general. Also, the Forgotten Empires team packaged an improved AI with their AoE2HD if memory serves, so I don’t see why they wouldn’t do the same here.
Hopefully with improved matchmaking you will be able to play against people at the same skill level as yourself and win about 50% of your games! Otherwise, let’s hope the AI is getting an overhaul.
Yeah, the AI will definitely have to be improved. I can easily thrash Hard/Hardest AI in the original AoE/RoR, and I’m nowhere near good. I hope they make it give us a real challenge
My thought would be to alter AI based on your in game level of experience. Say your a novice and your level of experience equates to sucking rocks. So, on your next campaign the AI would be adjusted to level “suck rocks + 2”. Still giving you some competitive edge. :#
I remember the A.I. in origional aoe1 always liked to either go mass priest or mass catapults scatered around their whole base depending on their faction
@DutchAmanda said:
I remember the A.I. in origional aoe1 always liked to either go mass priest or mass catapults scatered around their whole base depending on their faction
This is still the case with UPatch, I hope they change it for DE. I faced a wololo spam just yesterday.
@Otani64 said:
If nothing changes to an AI that is 20 years old that would be odd …
Agreed. I’m sure there’s plenty of ways they can improve things. Personally I much prefer Harder AI opponents to be smarter rather than just having artificial advantages. The latter approach just sucks the fun out of it for me, whereas an opponent that alters his force composition during the game to counteract what I’ve been using, tries to attack from multiple directions or tries to bait me into doing something stupid is much more interesting.
I would love an AI at the level of some other RTSs that I may not be allowed to name here. I don’t know if naming other games is frowned upon. But there are some butt kicking AI out there and I always wanted AoE to have than kind of AI to battle against. Something that a friend an I can play against and be challenged!
@Shaikhu86 said:
I would love an AI at the level of some other RTSs that I may not be allowed to name here. I don’t know if naming other games is frowned upon. But there are some butt kicking AI out there and I always wanted AoE to have than kind of AI to battle against. Something that a friend an I can play against and be challenged!
I don’t think it’s frowned upon. Multiple people have referred to other AoE/SC games in various arguments around the forums. There’s nothing wrong with constructive criticism, or referring to newer games - AoE:DE is, after all, still being developed - I’m sure the devs appreciate knowing what we’d like.
@Shaikhu86 said:
I would love an AI at the level of some other RTSs that I may not be allowed to name here. I don’t know if naming other games is frowned upon. But there are some butt kicking AI out there and I always wanted AoE to have than kind of AI to battle against. Something that a friend an I can play against and be challenged!
I don’t think it’s frowned upon. Multiple people have referred to other AoE/SC games in various arguments around the forums. There’s nothing wrong with constructive criticism, or referring to newer games - AoE:DE is, after all, still being developed - I’m sure the devs appreciate knowing what we’d like.
In that case, StarCraft II! The AI on Brutal is impressive. I hope AoE ends up being a bigger challenge!
I think people really sort of underestimate the work that needs to go into AI and why AI in games is so fundamentally difficult
AI is complicated. I mean sure we have machines that can play Go and chess. But you might have noticed those computers are basically the size of an apartment complex. They’re also in turn based games where if hte AI takes 1-2 minutes per turn that isn’t considered ‘omg why is this taking so long’. they’re also 1v1 games as opposed to 1 vs many. With 1 vs many you run into again ‘why is this computer taking so long’ problems.
AI finalization timelines. So while you’re working on the overall framework of the AI, note that actually filling that brain with data is hard. The other part is that with game development, lots of things aren’t done until like the very end. So 1 month before you are to ship, you might be waiting for some giant thing to get implemented. Not to mention all the ongoing ‘balance tweaking’ which can cause your AI to have a seizure or to act dumb because it hasn’t taken into account how archers were nerfed in the latest build.
Your brain does a lot of stuff very quickly that seems easy but is super hard for a computer. As an example think of a walking humanoid robot. The reason most robot makers dont do humanoids is because to actually make a robot walk is incredibly complicated. I mean you don’t remember spending 2 years crawling on the floor chewing on the floorboard trying to figure out how to walk as an infant. You just think “well walking is easy I do it all the time”. But all the things that go into that are hugely complicated that we just internalize and automate. The way we play games is the same. we internalize strategies. We see patterns. But we can’t really ‘describe’ them. Its just “The Ai shouldn’t have done that”.
The AI guys aren’t in the back twiddling their thumbs. AI is hard work and requires a lot of trade offs between CPU cycles, how long things take to process, etc. If AI was ‘easy’ everyone would be doing it.
There is a Japanese expression that may fit perfectly in this situation: “ãã“ã‚’ãªã‚“ã¨ã‹”. It means kind of “I know it is difficult, but still, please do something with it”.
Since I am not a good player myself, I do not want the A.I. to be a very clever and strong player in a way that strong human players are, but I just want it to be “not dumb”. For example, in AOE3, I can defeat the hardest A.I. if the map is of small islands, because the A.I. never leaves the starting-point island, even though there is a big expansion land near the island.
A lot of times, the villagers are just standing doing nothing, armies move in a weird direction when it come to me, the A.I. keeps trying to build new buildings right near where my large army is destroying its base, etc, etc. I cannot remember all, but there were many stupid things.
I am not an expert, but these days, aren’t Microsoft and other companies focusing on A.I. and machine learning or whatever? I wish Microsoft could use its games to advance their research on A.I.
@Satoru said:
I think people really sort of underestimate the work that needs to go into AI and why AI in games is so fundamentally difficult
AI is complicated. I mean sure we have machines that can play Go and chess. But you might have noticed those computers are basically the size of an apartment complex. They’re also in turn based games where if hte AI takes 1-2 minutes per turn that isn’t considered ‘omg why is this taking so long’. they’re also 1v1 games as opposed to 1 vs many. With 1 vs many you run into again ‘why is this computer taking so long’ problems.
AI finalization timelines. So while you’re working on the overall framework of the AI, note that actually filling that brain with data is hard. The other part is that with game development, lots of things aren’t done until like the very end. So 1 month before you are to ship, you might be waiting for some giant thing to get implemented. Not to mention all the ongoing ‘balance tweaking’ which can cause your AI to have a seizure or to act dumb because it hasn’t taken into account how archers were nerfed in the latest build.
Your brain does a lot of stuff very quickly that seems easy but is super hard for a computer. As an example think of a walking humanoid robot. The reason most robot makers dont do humanoids is because to actually make a robot walk is incredibly complicated. I mean you don’t remember spending 2 years crawling on the floor chewing on the floorboard trying to figure out how to walk as an infant. You just think “well walking is easy I do it all the time”. But all the things that go into that are hugely complicated that we just internalize and automate. The way we play games is the same. we internalize strategies. We see patterns. But we can’t really ‘describe’ them. Its just “The Ai shouldn’t have done that”.
The AI guys aren’t in the back twiddling their thumbs. AI is hard work and requires a lot of trade offs between CPU cycles, how long things take to process, etc. If AI was ‘easy’ everyone would be doing it.
You are wrong in multiple levels, AI simulation is in fact extremely easy for almost any computer nowadays , in fact even your smartphone can run it. When mentioning “computers in size of apartman complex” i suppose you were talking about the ones used to train neural networks , but thats still irrelevant since a NN would need to be trained only once and running a trained neural network is still , quite very easy.
Only problem with AI is the time it takes to code it , which shouldnt be an issue to microsoft. There is no point of mentioning cpu cycles as core logic of games almost never use more than 10 percent of a single thread , not the mention AI can quite easily be multithreaded.
Render calls BY FAR take more cpu cycles than simulating even the most complex AI does
@LoserLoner said:
There is a Japanese expression that may fit perfectly in this situation: “ãã“ã‚’ãªã‚“ã¨ã‹”. It means kind of “I know it is difficult, but still, please do something with it”.
Since I am not a good player myself, I do not want the A.I. to be a very clever and strong player in a way that strong human players are, but I just want it to be “not dumb”. For example, in AOE3, I can defeat the hardest A.I. if the map is of small islands, because the A.I. never leaves the starting-point island, even though there is a big expansion land near the island.
A lot of times, the villagers are just standing doing nothing, armies move in a weird direction when it come to me, the A.I. keeps trying to build new buildings right near where my large army is destroying its base, etc, etc. I cannot remember all, but there were many stupid things.
I am not an expert, but these days, aren’t Microsoft and other companies focusing on A.I. and machine learning or whatever? I wish Microsoft could use its games to advance their research on A.I.
Machine learning , more precisely deep learning is the focus of all tech companies nowadays. The sheer potential it holds is unpredictably high , it would be fun to fight aganist a NN instead of hand coded AI. There is a starcraft 2 NN called DeepMind if you are interested in the topic
I don’t agree with Satoru. Referring to Deepmind they even have to slow down it’s decision making when it comes to StarCraft 1. A human player can only do one action at a time while the computer would be able to controle every unit separately but at the same time, which makes it massively unfair against a human. The problem to make a good AI (referring to this particular item of AI) is not it’s slowness but it’s speed.
I have briefly heard that the DeepMind team who made the A.I. Go player started working on StarCraft 2. Is it publicly available already so I could test it?
Also, is the strength of DeepMind adjustable? I want a non-dumb A.I., but not a super strong, undefeatable A.I. like the one that defeated Go world champions. Basically, I want an A.I. that behaves like a novice human player. (Even the hardest A.I. of AOE 3 behaved stupidly.)
Again stop comparing systems like Deepmind that are computer clusters the size of a small apartment complex
Your computer is literally a billion times less powerful than what Deepmind uses
The ‘junkiest’ AlphaGo setup is 48 cpus wide with 1 gpu.
The one that you read about in the news is a custom ASIC made by Google. It cascaded 50 of these custom ASICs.
Your PC does not even come close to this kind of power. Yes if you have 50 custom ASICs distributed for the sole purpose of defeating a single player
Deepmind hasn’t even started ‘beating’ starcraft yet. They have the API in place but that’s it. Previous competitions for actual Starcraft AI were to put it mildly ‘underwhelming’
Deepmind does not need to ‘down scale’ to play Starcraft.
Again this is the giant misconceptions around AI. People think its magically easy because we’re beating human players in GO and Chess. Without understanding the context around them. Yes beating a Go player is super impressive and few thought such a feat would be possible so soon (most were predicting 20 years out). That doesn’t translate into somehow a game AI going to trounce you on a CPU setup that is a billion times less powerful
