Stronger AI's in the game?

[marsh rivet]

Hi, is there a possibility of developing an AI stronger than "expert" for high-intensity play? I know that the bots do pursue different strategies (breaking, continent capture, and the like) but after a while single-player games can become a bit predictable.

I've also noticed that there's a much less flavored approach to dealing with bots that I've seen in high-level tourneys (thanks to Pete for demonstrating this in many tournament and FFA runs). I really do like to use single-player mode for practice, as well as because I often have to remain away from the screen during the game. And at best, I'd love for it to feel like I'm playing other players. (That said, I do understand how difficult it is to create a bot that feels like a player, but more aggressive/defensive ones might add more spice to the mix).

The short of it is, is the dev team planning on creating a stronger and/or more aggressive AI for non-scenario maps? If so, I think having a stronger AI (called something like epic level?) would definitely add a lot of flavor to the game.

~ Cosmic Fury

[granite tree]

AI has come so far in recent years, in terms of games, art, and other applications.

It could well be time to reconsider upgrading Risk’s AI, to give a renaissance to singleplayer and scenario modes.

[marsh rivet]

Thank you very much for your swift and thoughtful response!

I am figuring much the same thing. I'm not asking for unbeatable AI's, because that crosses the fine line between challenging and demoralizing. But, I really do want one that forces players to slow down and think seriously about its presence in either single-player or multiplayer games.

[solemn rivet]

Essentially: survey top 100 GMs (and hope for 40% to respond): What factors do you consider every turn? What factors make you consider to attempt a kill? Take a continent? Break a continent? Which dice roll outcomes in middle of turn make you reconsider. Strategies/paths/limits can be built into a code to make a new Master AI profile, it’s just a matter of understanding (and coding) a set of behaviors to emulate a player stronger than Expert.

[wooden shore]

I think the biggest challenge when creating stronger AI in a game like risk, is you would want them to behave like a human, just a week into the game I was able to read expert AIs so well I could win games without attack other players myself most of the game, I'd just use the bot as my personal hit man, this would need to be changed as bots, depending on circumstance, are both way to passive and way too aggressive. They need to be more "human" for it to be an enjoyable experience imo

[marsh rivet]

In short, they're too predictable? If that is your case, I agree there too. All I ever need to do is bottleneck any area with any decent force and it'll almost never get attacked. This applies especially to capitals.

[wooden shore]

But here's the problem, they SHOULD be predictable, at least to an extent. Once you get experienced enough it's somewhat easy to predict what even the best players will do. Bots should be somewhat predictable, like people, but not so predictable that it's genuinely hard to even LOSE against them

[granite tree]

Currently the All vs One and One vs All scenarios, where you’re outnumbered up to 10x against an expert bot, are a good illustration of their difficulty relative to a novice. Against most novices, these scenarios would be almost impossible, but with the bot in this position, it’s relatively straightforward for Grandmasters to win, regardless of luck.

While it would be expensive to create superhuman bots better than top Grandmasters, creating bots with a comparable skill level to an average human player (intermediate/expert), would mean that at least the bot could win games from a strong starting position, or where the bot gets lucky.

Currently singleplayer is rarely used by medium or high level players, since beating the bots is so trivial, but if they were even half decent, there’d be far more room for singleplayer to grow and become popular again

[pastel haven]

Design a bot that collaborate with another bot and target only the human’s in the game.

[granite tree]

For singleplayer, that could be a very simple way to increase difficulty.

The inverse version where the bots attack each other, may also have some value in other contexts where humans are competing, and don’t want to waste troops hitting the disconnected players, but don’t want the bots to get too numerous

[pastel haven]

It’s fun in singleplayer to let the bots eliminate each other, You can win a 6p singleplayer by only eliminate the last bot. The bot i like the most is the big stack bot that crushes all troops that comes on his path, it goes all over the board. Watch out if he trades in his cards in progressive .

[granite tree]

On the right maps in singleplayer, it’s possible with 4 Expert bots and zombies, to let all the bots eliminate themselves, without making any attacks

[wooden shore]

Something else I've noticed is that bots don't try to take better continents, they often stick to the ones they have more troops in, last night I played one that had taken over a continent connected four different ways to other continents, and it got wiped because of course it did, even though it had enough troops to wipe the adjacent continent and take 4 troops per round while only holding one entrance

[nocturne vale]

It’s not that easy to develop bots in risk. The reason why bots can be so strong in chess is because in chess, there is always a couple good moves, no matter who they’re playing against

The way Risk is made, a good move against one opponent is a horrible move against another. Other people in risk have a big impact on the game, in progressive somebody could always just recklessly attack all your different positions, destroying you and themselves in the process, in order to make the perfect risk bot, you’d have to teach it to somehow be able to use diplomacy to its advantage. This is even more important in Fixed, when all that matters is diplomacy and continents/territory bonuses

[granite tree]

It’s undoubtedly very difficult to make perfect Risk AI, but there’s a huge amount of room to improve the current ones that play at a novice level, without needing to even half-way approach perfection.

There’s also been huge advances in AI this year, across many different fields e.g. check out this prize-winning piece of art created by AI. As the tech advances, it becomes more and more possible for Risk bots to be improved using existing tech

[stuck copper]

I very much doubt that SMG will make any notable improvement to bot gameplay, but as other people have said we absolutely have the technology to make a bot of any level, including ones that would beat human players.

As an aside here I don't ever play capital conquest since bots literally do not care about capitals and will just throw the game to whoever goes next by reinforcing off of it

[pastel haven]

This week i have seen bots defending there capitals and even take over capitals and defend it, it where expert bots. It was funny to see, that the bot attacked my opponents capital and succeeded,

[granite tree]

Depends on the bot personality more than the difficulty. There are personalities of Beginner bot that defend caps, as well as personalities of Expert bots that abandon them almost immediately.

[coral thorn]

L

[low plover]

I find that One Vs All and other scenarios have more to do with exploiting bot idiosyncrasies than actual "skill". When it is impossible to hold a reinforcement advantage before your initial army is spent, any human, even a dumb one, would easily win by just fortifying all the random 10's and 15's to the borders

[granite tree]

Here you start with 318 against the bot’s 600 stack.

If the green opponent was a Grandmaster, they’d have the game won it 1-2 turns

Even if they were a novice, your best hope would be to survive enough turns for a couple of card trades, but since it’s fixed, the novice would almost always eventually win