How Should Beyond Earth Handle AI Difficulty Scaling?

[stealth_nsk]

There's nothing preventing CBE from having an ages mechanic. They are perfectly within limits to invent some. They could even just grade the tech web, so each ring outwards is the equivalent of an age.

If they implement ages, they will have to give some bonuses to them. IMHO, it would destroy the whole "tech web" idea. Also, it would look a little bit strange if "later" technology will be used as prerequisite fot "earlier".

 

[Quintillus]

I'd like to see multiple options for tuning the AI rather than just one slider. In GalCiv2, you can tune both how advanced of algorithms the AI uses, as well as factors such as resource production bonus and aggressiveness (and have it be different for each opponent). Even in Civ3, it's pretty easy to fire up the editor and change either the production bonus, initial starting unit bonus, or unit support bonus to make a semi-custom level. I'd like to see something like that (without having to go in to a programming language to do it) in Beyond Earth. Have an option for scaling the resource production, but have a different option for starting bonuses. That would also be great for those of us who really fit best between two of the pegs on the difficulty scale - if I'm between Monarch and Emperor, I can tweak it a bit so the AI gets some, but not all, of the Emperor bonuses, and it's a better fit to my preferred difficulty level.

I'm a bit tired of hearing that excuse. I don't agree that good AI is simply a function of available CPU cycles... my experience is that such a claim is a crutch of inadequate AI programmers. I know that AI programming is hard, and it's true that pathfinding is a challenging computational problem, but when AI units jump in the water rather than engaging the enemy, this isn't a computational problem. It's a problem of inadequate algorithms. Anyone can beg for more CPU cycles to throw at a problem; a clever engineer finds a way to solve the problem with the resources he's been given.

Even if it were true that Firaxis had reached the pinnacle of AI design (which they haven't) and the limiting factor in a 1UPT system truly was processing power, then designing a game that your AI can't play competently within proscribed computational limits is poor game design. Either the AI programmers are incompetent, or the game designers are. Which do you choose?

I agree with this assessment. AI is an important part of a strategy game, and some do it better than others. Firaxis may have made the job harder for themselves by going 1 UPT, but whether the AI could competently handle 1 UPT should have been part of the decision as to whether to use 1 UPT in the first place. It's certainly possible that Firaxis didn't realize that 1 UPT would make AI as much more difficult as it appears to have done. But considering that the AI hasn't been fantastic in earlier games, either, I think it's only part putting themselves in a hard place, and in part not having adequate AI expertise/resources to really deliver a stellar AI.

I don't know how the AI was divvied up for Civ5, but in Civ3, only one guy programmed the entire AI. Considering that, it's actually not bad, but it definitely has its shortcomings. Civ4's AI was better, but I recall that also had a pretty small team. I don't know if Civ5 had a larger or better AI team, but if it did, it wasn't enough to get a better end product than Civ4's AI.

An increased focus on multiplayer is not a valid excuse for adding parts to the game that the AI can't handle well, because most Civ5 gaming is still going to be single player.

By comparison, Stardock generally has a reputation as making turn-based strategy games with notably above-average AI for the industry. Brad Wardell, their CEO and main AI programmer, has mentioned that one of the major factors in whether they include a feature in the game is whether the AI will be able to use it well. They do a much better job at avoiding the "Waiting for AI" at the end of turns than Firaxis, as well. And with regards to not having enough CPU cycles? Firaxis's AI is still single-threaded in Civ1 -> Civ5 (in Civ5, the graphics handling is multi-threaded, but not the AI, unless this changed in one of the expansions). Stardock has been using a multithreaded AI single at least GalCivII in 2006 (roughly the same time as Civ4's release).

There are probably other, more niche developers who are just as good of an example of getting an above average AI. But Firaxis indeed is not at the pinnacle of strategy AI. And to be fair, neither is perhaps the biggest comparison point in strategy, Sega. I haven't played the latest Total War games, but from the ones I have played, I'd put Firaxis ahead of Sega in AI. That's more a sorry state for Sega than a crowning achievement for Firaxis, however.

 

[stealth_nsk]

I agree with this assessment. AI is an important part of a strategy game, and some do it better than others. Firaxis may have made the job harder for themselves by going 1 UPT, but whether the AI could competently handle 1 UPT should have been part of the decision as to whether to use 1 UPT in the first place.

That's incorrect goal. Correct goal is to provide player a challenge. Civ5 tactics give player challenge even though it's "skill vs. numbers". Civ3/4 tactics have no challenge.

I don't know how the AI was divvied up for Civ5, but in Civ3, only one guy programmed the entire AI. Considering that, it's actually not bad, but it definitely has its shortcomings. Civ4's AI was better, but I recall that also had a pretty small team. I don't know if Civ5 had a larger or better AI team, but if it did, it wasn't enough to get a better end product than Civ4's AI.

Civ3/4 AI never faced real tactical combat. And if you think it's good in strategy...
In Civ3 try the following thing. Once you at war with a civilization on another continent, try to remove units from one city on the other side of your territory. You'll see your enemy boarding troops and going to attack this "unprotected" city. Once they are nearly where, move units back in and remove units from another city - on the other side. You know how AI will react, aren't you?

 

[Lorizael]

I think for many people it will be even more cheating than the current situation. Also, it's difficult to define, which land belongs to whom (except for capital).

It wouldn't be based on which land belongs to whom. It would be based on who peels back the fog of war. If an AI on a high difficulty uncovers new territory, they're likely to find valuable extra resources. If a human on a high difficulty does so, they'd encounter the same terrain but with less valuable resources, or perhaps dangerous resources such as the miasma we've heard nothing about.

 

[Fabio1701]

There likely will be some kind of grading along the tech web: the change in unit appearance over time demands sucha mechanic, even if it is not explicitly named and namable by players.

 

[stealth_nsk]

There likely will be some kind of grading along the tech web: the change in unit appearance over time demands sucha mechanic, even if it is not explicitly named and namable by players.

The change is unit appearance is actually unit upgrades and its explicitly named. It's like this:

We decided on a different way to make your army customisable, and also feed back into your affinity, your cultural identity. So you have a catalogue of generic unit types that will upgrade as you level up. As your dominant affinity goes higher you'll be able to stack these guys with perks, with special abilities that are themed to that. Your marines can start out as normal marines, and then you play a little while and they level up to level two marines and you give them the anti-alien perk instead of the anti-city perk, because that's the kind of army you want to field. Then they have more abilities beyond that, so your stock types upgrade throughout the game.

So yes, they are tied to progress somehow, but it's not necessary tech progress. Also it could be tracked for each individual unit separately - we don't know anything for sure. Probably this thing could be used for general progress tracking, but not necessary.

 

[theadder]

I'm a bit tired of hearing that excuse. I don't agree that good AI is simply a function of available CPU cycles... my experience is that such a claim is a crutch of inadequate AI programmers. I know that AI programming is hard, and it's true that pathfinding is a challenging computational problem, but when AI units jump in the water rather than engaging the enemy, this isn't a computational problem. It's a problem of inadequate algorithms. Anyone can beg for more CPU cycles to throw at a problem; a clever engineer finds a way to solve the problem with the resources he's been given.

Even if it were true that Firaxis had reached the pinnacle of AI design (which they haven't) and the limiting factor in a 1UPT system truly was processing power, then designing a game that your AI can't play competently within proscribed computational limits is poor game design. Either the AI programmers are incompetent, or the game designers are. Which do you choose?

I agree with this completely. The game could be considerably better if Firaxis focused on the AI in this way.

The game ought to be designed around the constraints they apply to AI development. It clearly doesn't get much development time.

 

[grant2004]

I'm a bit concerned by what they've been saying about the different affinities when it comes to AI behavior. It sounds like as you move deeper into an affinity, the combat style of your units changes. This sounds like a big challenge for the AI, that it has to be capable of playing the combat style well for the starting units, each of these three affinities, and any intermediate state as it's progressing to an affinity.

Maybe the combat styles aren't as different as I'm reading from the initial interviews, but the supremacy combat style, which I remember someone saying required "finesse" to arrange units for positional bonuses, that seems like it would be a big challenge for the current AI.

The reason I bring this up is that if they follow the same handicapping strategy of making it easy for the AI to produce huge numbers of units, that's not going to be as effective for all of the different affinities. A carpet of doom could make it very challenging to position units to get those bonuses.

The purity units, which sound like they use a brute force sort of strategy seem like a much easier play style for the AI to handle.

I hope that when the game launches we don't see that the AI is better at certain affinities than others.

 

[stealth_nsk]

For those who want to learn a little bit more about the AI and how it applies to Civilization, a couple of links:
http://www.youtube.com/watch?v=IJcuQQ1eWWI - that's speech about the AI by Soren Johnson, AI programmer for both Civ 3 and 4, lead developer of Civ 4.
http://jonshaferondesign.com/2012/09/10/the-recipe-for-good-ai/ - that's a post by Jon Shafer, original lead of Civ 5. He also have some additional info further in his blog.

 

[Kissamies]

The worst kinds of cheating are the ones where the AI is basically omniscient and can see everything you do regardless of whether they have visibility in your territory. And that's pretty common.

Agreed. The worst immersion breaking moments that I can remember with Civilization AI have been when I have caught it doing this. AI bonuses I can accept, but this omniscense feels like blatant cheating.

 

[Lancor]

Boni and perhaps even mali for the AI are fine and can be in line with the immersion. The same way as I like to imagine their faults, stupidity or stubborness to relate to real people. If everyone would be a godly strategist and without emotions, the world would look very different.

 

[Jman5]

Civ5 tactic is much more complex than previous SoD, so the AI is actually near limits of modern hardware and significant improvements are not possible. So, yes, AI bonuses are needed and they should be big.

I also there should be less initial bonuses and more long-lasting ones. I would even suggest some gradually increasing bonuses. However this would suit Civ more CBE, which doesn't have epochs.

In my opinion, many of the shortfalls of the AI could be circumvented by simply altering play style to a more optimal tactic. A good example for civ 5 could be adjustments to its science strategy. Currently, a good player will often set up their capital as "science city" where he will use a garden + national epic and then use science specialists in the city to generate great scientists. Then he will plant a few of them around the city because you're also putting all the science modifiers in that city. Then you just have to adjust some building priority to emphasize science, bulb the rest, and suddenly you could start scaling back some of the built in cheats.

The problem with Civ 5's AI isn't necessarily the amount of algorithms. The problem is that it's been programed to play badly. In my opinion this has less to do with the code and more to do with the coder's understanding of the game itself. He doesn't understand advanced strategy the way a high level player would, so he makes the AI do things which are actually kind of dumb. Instead of carefully manipulating your specialist choice so you're pumping out only great scientists, you've got a random assortment of specialists that are constantly being readjusted. So you wind up getting random great people.

What I would love is if Firaxis brought in some good quality players and had their AI team pick their brains about decision making. Then have them observe the results and critique it.

 

[stealth_nsk]

In my opinion, many of the shortfalls of the AI could be circumvented by simply altering play style to a more optimal tactic. A good example for civ 5 could be adjustments to its science strategy. Currently, a good player will often set up their capital as "science city" where he will use a garden + national epic and then use science specialists in the city to generate great scientists. Then he will plant a few of them around the city because you're also putting all the science modifiers in that city. Then you just have to adjust some building priority to emphasize science, bulb the rest, and suddenly you could start scaling back some of the built in cheats.

The problem with Civ 5's AI isn't necessarily the amount of algorithms. The problem is that it's been programed to play badly. In my opinion this has less to do with the code and more to do with the coder's understanding of the game itself. He doesn't understand advanced strategy the way a high level player would, so he makes the AI do things which are actually kind of dumb. Instead of carefully manipulating your specialist choice so you're pumping out only great scientists, you've got a random assortment of specialists that are constantly being readjusted. So you wind up getting random great people.

What I would love is if Firaxis brought in some good quality players and had their AI team pick their brains about decision making. Then have them observe the results and critique it.

Actually good player is usually bad game designer. Your post is a good example
The problem with great people is not AI, the problem is in balance. The strategy of building great scientists only is imbalance exploit - the thing which AI should avoid.

 

[Mustakrakish]

In my opinion, many of the shortfalls of the AI could be circumvented by simply altering play style to a more optimal tactic. A good example for civ 5 could be adjustments to its science strategy. Currently, a good player will often set up their capital as "science city" where he will use a garden + national epic and then use science specialists in the city to generate great scientists. Then he will plant a few of them around the city because you're also putting all the science modifiers in that city. Then you just have to adjust some building priority to emphasize science, bulb the rest, and suddenly you could start scaling back some of the built in cheats.

The problem with Civ 5's AI isn't necessarily the amount of algorithms. The problem is that it's been programed to play badly. In my opinion this has less to do with the code and more to do with the coder's understanding of the game itself. He doesn't understand advanced strategy the way a high level player would, so he makes the AI do things which are actually kind of dumb. Instead of carefully manipulating your specialist choice so you're pumping out only great scientists, you've got a random assortment of specialists that are constantly being readjusted. So you wind up getting random great people.

What I would love is if Firaxis brought in some good quality players and had their AI team pick their brains about decision making. Then have them observe the results and critique it.

All behold the mighty high level players of Civilization, bow before them for they are magnificent and no coders and designers can ever reach their level!

 

[Kid R]

For those who want to learn a little bit more about the AI and how it applies to Civilization, a couple of links:
http://www.youtube.com/watch?v=IJcuQQ1eWWI - that's speech about the AI by Soren Johnson, AI programmer for both Civ 3 and 4, lead developer of Civ 4.
http://jonshaferondesign.com/2012/09/10/the-recipe-for-good-ai/ - that's a post by Jon Shafer, original lead of Civ 5. He also have some additional info further in his blog.

Jon's blog is interesting where he says designing the game and designing the AI to play the game are almost the same thing.

In designing a game you're putting yourself in the head of the player, trying to imagine the decisions they'll have to make, where those will lead, what long term strategies they'll be able to put together, and so on. You *have* to imagine all that stuff - to decide it if it all might be fun, which is the whole point of a game. And in imagining it you're effectively outlining in your head the structure of an AI to play the game.

One conclusion is, a poorly designed game is likely to be unfun, and is likely to have poor AI, both for exactly the same reason

 

[stealth_nsk]

Jon's blog is interesting where he says designing the game and designing the AI to play the game are almost the same thing.

In designing a game you're putting yourself in the head of the player, trying to imagine the decisions they'll have to make, where those will lead, what long term strategies they'll be able to put together, and so on. You *have* to imagine all that stuff - to decide it if it all might be fun, which is the whole point of a game. And in imagining it you're effectively outlining in your head the structure of an AI to play the game.

One conclusion is, a poorly designed game is likely to be unfun, and is likely to have poor AI, both for exactly the same reason

Well, that's quite a small part of the post
The main idea there (and in Soren Johnson video) is - AI shouldn't try to act like human. It doesn't have to be smart. The goals of the AI are:
- Keep challenge for the player (actually the AI is just one of the game systems).
- Do not break immersion by making too dumb actions.

 

[bhavv]

One thing I'd like is for the difficulty to increase as it did in Civ IV, but without also handicapping the player at the same time with things like city maintenance and research time.

 

[Windsor]

Actually, I'd most prefer the AI to NOT start off with initial bonuses but instead continual bonuses.

If you don't give the AI initial bonuses then the human will just cripple the AI at the start.


I think it's really hard to say what bonuses the AI in CBE needs since I know so little of the actual game mechanics. In Civ5 however I would have preferred if the AI got straight up combat bonuses (a strength 10 unit would be strength 12) at the higher levels. It's a lot more in-your-face than the economic bonuses the AI does get, but also a lot more directed at where the AI actually needs help. The AI in Civ5 doesn't need more units, it needs better units.

 

[Windsor]

One thing I'd like is for the difficulty to increase as it did in Civ IV, but without also handicapping the player at the same time with things like city maintenance and research time.

Deity players in Civ4 already blasts through the tech-tree. So the increased cost is really needed.

 

[Matthew.]

IMO enemy combat units should be given overall buffs, increasing on difficulty. Civ 5 went with moar units, which often harmed the AI more than helped. Carpet of Doom meant scenarios where the AI would have 15 archers surrounding a city, and a poor lone pike off in the distance unable to reach the city.

Stronger units rather than moar works better. Against another AI nothing changes (presumably; AI vs. AI should theoretically work out like a MOBA where more or less at a stalemate) but for the player it increases challenge on the tactical level. It keeps the battlefield relatively clean, which is ideal for 1UPT, and helps the AI out without feeling overly cheap.

I think it is also subtle enough where it doesn't fall into the "blatant cheats" category. If an entire battle took a combination of 20 attacks, but that same battle on a higher difficulty would have taken, say, 28 attacks, the player isn't going to notice the difference as much. In Civ 5 terms that may be the equivalent of a few extra archer attacks, perhaps actually losing that fortified pike on the front lines.