Why Doesn’t Artificial Intelligence Have Enough Power to Always Defeat Human Players?
- This topic has 25 replies, 1 voice, and was last updated 1 year, 11 months ago by
.
-
Posts
-
I understand why it might be difficult to create an AI that can outperform human players in Age of Empires II, as it could be a discouraging experience for all participants, especially the most experienced ones.
However, I have noticed that human players display behaviours that AI programs should be able to replicate better than them.
For example, when using archers and town centers, people will run them up while taking fire, whereas competent human players would stay just out of range to prevent the villagers from returning to work.
Additionally, why don’t AI micro to coordinate their arrow shots?
Not only should AI be able to perfectly micro, but it should also be able to complete more actions per minute as they don’t get distracted as easily as people and can multitask more efficiently.
If AI can beat master chess players and has been able to do so for a long time, then it should be advanced enough to always beat human players in AOE2.
The in game AIs run on a very specific built in language, which is clunky and doesn’t let you attach other methods like machine learning etc, whereas these more sophisticated AIs for other games are running on an external thing and just plugging their inputs into the game (but they also take millions of $$$ to build)
It’s like saying a chess board itself doesn’t play chess to the level of a sophisticated chess AI
A lot of the macro gets abused in the midgame.
They’re also atrocious in dark age, where if they miss a boar or deer, they’ll put 12 villagers on berries and make 5 farms.
If the AI was less abusable in terms of the economy and made less terrible strategic decisions, like running foot archers into castle fire and building castles in the back of its base, we might reach a point where the AI is close to unbeatable.
We can’t underestimate how difficult it is to explain to a non-sentient robot what makes a castle position valuable.
How it blocks the enemy’s advance into the economy, is it on a hill, how far is it from other hills, does it defend a valuable resource, does it deny the opponent a section of economy.
There’s so much that goes into just castle placement and there usually is never a truly optimal castle position, so there needs to be a failsafe such that the AI doesn’t get decision anxiety and end up doing nothing.
You might want to check out the google AI for SC2.
From what I recall it’s not able to beat top players but is competitive.
AI development isn’t as easy as you are making it sound.
Complexity.
How do you get an AI that can be godly with both Huns and Goths?
Two polar opposite civ’s, both requiring very different play styles.
How does an AI recognize things like a scout rush, Drush, castle drop, fast castle, etc.,
Unit counters would probably be the easiest thing to program in.
And then you also have to deal with the Civ’s.
For example, playing a Goth?
Letting them reach Imp is a bad idea, but how do you program that to an AI?
Versus whatever it might do against the Koreans?
Why haven’t we solved world hunger, I mean, just give everybody food?
Things are rarely as simple as they sound on paper
I think that devs goal wasn’t to make the AI unbeatable.
They wanted to make it human-alike.
That’s why it doesn’t cheat (as other AIs do) and it doesn’t micro as much as it could.
In theory they could make it control each unit separately, but it would feel weird.
Other reason is that it’s not a very modern AI implementation.
It’s based on scripts/decision trees rather than machine learning.
I think you might have an incomplete understanding of machine learning.
Games like chess are actually way way simpler than games like Age of Empires.
In chess, there actually aren’t a ton of possibilities for how the game plays out, since you’re only operating on a 8 by 8 board, and each piece can only move in certain ways.
So even though its hard for the computer to learn, with more computing power and advanced ML techniques, it can do so.
Famously, Go is a much harder board game that computers have managed to solve, where its a lot harder because the board is 20 by 20 (i might have gotten the board dimensions wrong) and there are way more combinations, but the computer has managed to solve it.
Now think about AOE.
The map is huge, with way more tiles than on a chessboard.
But another added difficulty is that this game is called “partial information.” In chess, both players can see all the moves, so in some sense, a computer can calculate all the possibilities of what the opponent is thinking.
But in AOE, you need to scout your opponent then adapt based on what you see, so you can imagine that with all of the different features of the game, the complexity is much much bigger than that of chess.
This note about partial information is also why poker is difficult to solve, since you can’t see your opponents cards
Chess is a much simpler game.
AOE2 is a much more difficult game for AI than chess.
Look up Alphastar, the difficulty of the problem is explained better there.
Current technology would allow for an AI that could challenge even the best players without cheating, but as you’ve guessed this wouldn’t really improve the play experience for most players, and is not worth the massive effort and expense.
The AI isn’t intended to be as difficult as possible, but to create appealing player experience.
I can’t prove it, but from my experience playing against the AI I thinks its macro declines when it’s fighting, to better imitate a human player.
There some low-hanging fruit that could be picked, like preventing the AI from donating so many units to static defense, which could collectively constitute a higher AI difficulty setting.
What would come after extreme?
AI could absolutely be better than humans in literally any and all video games.
The reason they aren’t is because there is not nearly enough interest to spend the time and money to develop the AI to that level — not to mention the added processing power to run the AI in addition to running the game.
It makes no sense to improve AI for games that don’t have a permanently large fan base like chess has.
It’s not worth it financially to dump that much money just so the AI can beat all players instead of only beating 90% of them if the game will be either forgotten or updated (necessitating a complete AI re-evaluation) within a few years.
Chess hasn’t changed in centuries, and interest has increased — not decreased — since Deep Blue.
Take it from a chess player: no one will spend the time or money to develop video game AI to the same level as today’s chess engines.
AI archers hitting the first building in sight ( almost always it’s a house) is something which never happens with any elo player
AI is basically mot intelligence at all when it comes to Aoe2.
AI is a long set of if else conditions.
If you want to write all that, then go ahead.
Such a strong AI may also fry your computer because of the heavy computing.
The AI already does things humanly impossible like instant reactions while defending and having absolutely zero idle time.
But the total number of variables in aoe2, especially maps like arabia, nomad etc are so large that I feel like a strong AI would be a heavy resource consuming AI
The ai in aoe2 is not a ai, it’s a script.
The ai that beats everyone in chess is not an ai, it is a neural network.
The last one gives very impressive results but needs massive amount of data and processing power.
The AI isn’t designed for victory.
It’s designed to set checkpoints for a player’s progression and emulate a few real online strategies.
An AI designed for victory would cheese you to absolute death with insane unit microing.
- You must be logged in to reply to this topic.
