Increased difficulty?

[visage]

So, my games with SiS have been, in order:

1. On normal; restarted after a few turns on the grounds that having one's first colony revolting over and over again isn't so great.
2. On normal with Yoral; win via diplomatic victory, because apparently everyone loves you on normal difficulty.
3. On hard with Imperials; win via "diplomatic" victory by conquering half the map.
4. On brutal with the Orthin; win via "diplomatic" victory by conquering half the map.

The last game in particular was a walk in the park, which seems very wrong.

Are there more difficulty levels planned? Are there plans to bump up the difficulty? ...or should I be looking for modding information?

[sven]

So, my games with SiS have been, in order:

1. On normal; restarted after a few turns on the grounds that having one's first colony revolting over and over again isn't so great.
2. On normal with Yoral; win via diplomatic victory, because apparently everyone loves you on normal difficulty.
3. On hard with Imperials; win via "diplomatic" victory by conquering half the map.
4. On brutal with the Orthin; win via "diplomatic" victory by conquering half the map.

The last game in particular was a walk in the park, which seems very wrong.

What sized maps have you been playing on? In my experience, in addition to RNG issues, large maps are often much easier, in practice, than smaller ones.

More generally: brutal is not yet as brutal as I want it to be. This should change soonish.

[zolobolo]

Also had an easy run on Normal so far. Too easy. The AI has never declared war on me on this difficulty level and when I attacked it, they rarely tried to muster a coordinated attack. Most issues I have seen the AI having was related to range and fending off Raiders. The later particularly seem to keep the Normal AI at bay, which is an ok thing in the beginning but is bad mid and late game when the Reaiders are not much of a threat for the Player but the AI still keeps playing cat and mouse with them.

Played all the races expect Imperials on Normal, all games on the largest galaxy size with the default number of stars and distance between planets

Tried one game on brutal, AI next to me declared war, but couldn't reach my colonies after I took out its forward outpost and that pretty much put and end to that empire as I was able to slowly out-build it.

[visage]

What sized maps have you been playing on?

I've been playing on 'normal' maps.

large maps are often much easier, in practice, than smaller ones.

Huh. I'd been thinking of taking a larger map out for a spin in hopes of getting deeper into the tech tree. I guess I'll put that theory on hold for now.

More generally: brutal is not yet as brutal as I want it to be. This should change soonish.

...and presumably the combat rebalancing I've seen you mention will help with that, as well.

Thanks for the reply!

[SilasOfBorg]

I've only ever completed one game of SiS - usually ran into one bug or another, or couldn't play consistently enough to not have various bits of the game change underneath me, or whatever. On release I put an evening aside and played a full game.

So, my one full experience on normal as Orthin (medium map) was that the AI expanded reasonably well (except for the humans, who may have simply been screwed by starting in a corner hememd in by marauders) but wasn't very adept at either offense or defense, in a strategic sense. The Yoral in particular developed quite well and could have smashed a couple Marauder bases in their area quite easily, but didn't. I was nice to them, gave them stuff, and had them eating out of my hand for the rest of the game (they even voted for me at the end, which I like -- they win too, right?).

Normal currently feels "about right" though I'd be happy if the strategic AI were better. DeepBlue has an API for that, right?

[bjg]

I've only ever completed one game of SiS - usually ran into one bug or another...

Most bugs allow to ignore them by clicking "ignore" (optionally after uploading the report). I've just finished a very long 500 stars game.

[rattle]

A question to Sven and on topic:
Does the hard and brutal AI cheat as in gets a resource multiplicator or some base income?

I played a large map versus 8 brutal AIs and I allied with half the guys while they annihilated each other. In the end the Phidi controlled 40% and I as Yoral 30% of the map and the Phidi cranked out Dread Stars like mad while I could afford 2 from savings.

In the other game on brutal I played a small map where one AI player tried to conquer my worlds within the first 30-40 turns and it got Ion Cannons and shields (one of the dinosaur races). That's a bit fast unless there's hidden magic at work.

I do agree that large maps gives you time to turtle up compared to small ones and are quite easy to win.

[newanilamo]

The game is very easy to play (once you familiarise yourself with basics) on every difficulty level, especially on brutal!
I see two reasons why it is so. First but not main is marauders factions because for player it is very easy to stay at peace with them. And AI players always fighting them (very poorly) spending all resources and exhausting themselves. (And on brutal difficulty level this is most common situation)
Second but primary reason is behaviour of AI players. They simply cannot plan for couple of turns forward. Very primitive strategy. I'm sure there will be a lot of improvements but will they ever get close to human player? This great, long awaited game surely needs multiplayer! Can't wait when its done!

[spitfire_ch]

Just ask DeepMind nicely to train a deep neural network for the game. It was able to beat the best Go player, so it should be able to beat us in SiS They already have trained it for some old Atari arcade games (just by "looking" at the screen, not by seeing the code or by being taught the game mechanics) and is now learning to play Starcraft II. I would really wonder how it would fare with a 4x game.

[zolobolo]

Just ask DeepMind nicely to train a deep neural network for the game. It was able to beat the best Go player, so it should be able to beat us in SiS They already have trained it for some old Atari arcade games (just by "looking" at the screen, not by seeing the code or by being taught the game mechanics) and is now learning to play Starcraft II. I would really wonder how it would fare with a 4x game.

Probably poorly, as it would need to understand the elements it is working with which it cannot do even though it mimics the human learning process very well.

Contrary to Go, or even starcraft, a 4x game typically contains much more moving parts. They are like a puzzle riddled with unknown variables depending on map size, number of enemies, and the strategy and tactics these can follow.

When it is playing Go, it has full information availability and quality. The only real variable is what the next move of a single enemy, set out for a known victory condition will be.

[spitfire_ch]

Just ask DeepMind nicely to train a deep neural network for the game. It was able to beat the best Go player, so it should be able to beat us in SiS They already have trained it for some old Atari arcade games (just by "looking" at the screen, not by seeing the code or by being taught the game mechanics) and is now learning to play Starcraft II. I would really wonder how it would fare with a 4x game.

Probably poorly, as it would need to understand the elements it is working with which it cannot do even though it mimics the human learning process very well.

Contrary to Go, or even starcraft, a 4x game typically contains much more moving parts. They are like a puzzle riddled with unknown variables depending on map size, number of enemies, and the strategy and tactics these can follow.

When it is playing Go, it has full information availability and quality. The only real variable is what the next move of a single enemy, set out for a known victory condition will be.

Yes, you are probably right. Yet, in theory it should be able to learn to play 4x games, as it uses the same learning methods as a human mind. But that might involve learning everything a human does and gaining experiences similar to humans, so that it would finally be able to really understand the concepts behind the variables. That would be a true AI, though, and certainly decades a way and kind of an overkill to play a game

Maybe a deep neural net could still be used for aspects of the AI. E.g. it could learn how fast it should colonize new planets and which kinds of planets (or maybe it could be able to learn tactical combat), while other aspects would still be covered by an expert system.

[zolobolo]

Yes, you are probably right. Yet, in theory it should be able to learn to play 4x games, as it uses the same learning methods as a human mind. But that might involve learning everything a human does and gaining experiences similar to humans, so that it would finally be able to really understand the concepts behind the variables. That would be a true AI, though, and certainly decades a way and kind of an overkill to play a game

Maybe a deep neural net could still be used for aspects of the AI. E.g. it could learn how fast it should colonize new planets and which kinds of planets (or maybe it could be able to learn tactical combat), while other aspects would still be covered by an expert system.

Interesting idea: If it is able to learn like a human does that mean that it can eventually understand the concepts behind the mechanics as well? Does it need to?
I would argue that an AI is already successful when it can solve the problems it has been created for, but this largely depends on our definition of AI which there is no real consensus. The current AI is not created for 4X level of complexity that is why I think it would fail. Not that a competent AI is not possible, it just has not been created yet for this purpose. Tactical combat maybe, yes.

The way I see it, humans come from a different angle then our artificial intelligence. The "original" does not aim to deliver a perfect, good or even decent decision. Just a decision of any sort but promptly and based on vague and largely missing information at its disposal. Our AIs on the other hand are goal driven. Engineered to find the perfect solution, and thus they fail if they fall short of this goal.
The AI which plays chess as the 5th best player in the world is considered a failure and this is largely due to the fact that it is practically useless for anything else

[username]

On normal; restarted after a few turns on the grounds that having one's first colony revolting over and over again isn't so great.

Wait, what? Your colony revolted? What did you DO to them?

[*]On normal with Yoral; win via diplomatic victory, because apparently everyone loves you on normal difficulty.

Playing as the Yoral does seem to remove one of your largest threats. These guys always, always, declare war on me on first contact, if they're the first guys I meet.

[*]On brutal with the Orthin; win via "diplomatic" victory by conquering half the map.

Well, by the 3rd game, you've learned the ropes, while as far as I can tell, the AI is a bit derpy. You know exactly the right number and type of forces to take out each kind of random navigational hazard and how to build a functioning colony. The AI knows none of these things. I have seen them make some very confused-looking attacks on monster-inhabited planets. These kinds of roadblocks cost them a lot of time and money that you don't. By your second game, you know exactly the kind of ship you need to build to take out even heavily Space Harpy infested systems, while the AI is still milling in confusion around them.

[wjmackenzie]

I am finding large maps easier as well.

Currently playing a 500 star brutal, sparse star density, frequent pirate raids, rare inhabitals, each faction in play, with myself as imperials. So far, at stardate 1390, i have yet to encounter any other other empires, only one pirate raid (which was turned away by simply giving them 50% of the money i had at the moment they got to the colony)....

Hell, i havent even spotted another empires scouts at the edges of my territory (about 25% of the map at this point), and I have scouts all along the borders with an 8 parsec scanning range...

At least i have no distractions from building many many outposts while trying to find the aliens...

[visage]

On normal; restarted after a few turns on the grounds that having one's first colony revolting over and over again isn't so great.

Wait, what? Your colony revolted? What did you DO to them?

My first colony, as distinct from my homeworld. I'd found this really nice planet to colonize that already had some sentients present that I clearly should incorporate into my fledgling empire...


In other news, alas, it seems that difficulty has not been improved during my absence. Perhaps I'll look into those modding suggestions to nerf the marauders.