TotalBiscuit thinks the AI is good. and i'm now willing to believe it...

Which you seemed to agree with, when you said this:

The difference is I did not include the word "situations." I said "changed agendas." Without a grasp of situations as such, there's no way for the AI to imagine and progress-check and update "agenda change x" without repeating the same blind process by way of a new superimposed pseudo-strategic conditional, the "check if values have improved after implementing the previous agenda change and opt to change the agenda again" conditional, which is never going to approach "on the fly" without devolving into bloated random thrashing.

You can program AI to do a lot of things (not anything, of course), but the more complex AI becomes, the more it's dependent on the exact game balance and values. That's significant. If developers release some balance patch, they want to be able to change only couple of values in the AI to be in sync with the patch. If they have to reprogram the AI for each patch (which would surely involves a lot of testing), this would hurt patching schedule a lot. Say, instead of monthly patches we'll get patches every 3-5 months (with the same amount of changes). Eventually this would lead to much less developed game in the end of its cycle.
Exactly, agreed. Which is why we can't expect a solution to "the system's optimal strategy for the human is too template-able," a-la Community Patch, within the first three years.

I think we can see that VI is trying to lay groundwork for what Patch accomplished, which is specialization-punishment (amenities as a placeholder for city needs), but without closing off expansions that would toss the whole delicate balance out.
 
Last edited by a moderator:
I have heard plenty of complaints that the AI is easy to beat if you take the optimal path because the AI doesn't do that. But if they did program an AI to do this there would be a huuuuge outcry from people. The amount of hate Civ5 copped because the AI was "too gamey" was insane. People said it felt like the AI was playing a game, rather than being a "realisitic" Civilization and it broke immersion.
Now imagine you had a Civ that ploughed straight to a Science Victory, taking the optimal build to get there. The rage would be crazy.

Personally, I'd like to see an additional AI setting "Game Player" or somesuch. Where the AI is programmed to be more "gamey". To analyse their starting location, their neighbours, etc and then choose a target win condition based on that. Then chase that goal, rather than following their bits and pieces method.
Totally immersion breaking, but it offers something different for people who are more interested in the game, rather than a role play or a casual experience.
 
Personally, I'd like to see an additional AI setting "Game Player" or somesuch. Where the AI is programmed to be more "gamey". To analyse their starting location, their neighbours, etc and then choose a target win condition based on that. Then chase that goal, rather than following their bits and pieces method.
Totally immersion breaking, but it offers something different for people who are more interested in the game, rather than a role play or a casual experience.

As an option, yeah. But with more funny quips etc as they berate you. I think part of the issue with V's "gameyness" is that the face of the AI's intentions is ultra serious dry leaders, who do a semi good job of looking annoyed in a game where they don't really do other emotions. It just gets real depressing when the denouncements get on a roll.
I guess I love the immersion aspects of the game. I enjoyed that I'd usually build an alliance in IV which would survive the likelihood of me winning, cos I'd built a good relationship with that leader; and as the attacks increased they were someone I could count on to stand by me. It's just a game, they aren't real lol... but if you play one game of Civ for up to 80 hours as I often have, the experience is much more holistic with a bit of levity :)
 
This is basically the great summary of what I think of AI as well, provided by reddit user monkeyfetus: https://redd.it/57n82v

I hope he doesn't mind me quoting his post as I think it describes AI issue exceptionally well:
That's the big deal here. For me, I want to see an AI that appears to have a purpose and a plan, and that's what the AI seems to lack, not in the diplomatic sense, but in a small scale strategic sense. Like, I see Germany spamming missionaries from two different religions at the same City, and I wonder, is it actually trying to accomplishing something, or just saying "I have faith, buying missionaries is what you do when you have faith. Now I have missionaries. Spreading religion is what you do when you have missionaries."

Likewise, when you see an AI settling a city that doesn't do anything for them, obviously it's not as big a deal in Civ 6 because there's no penalties for too many cities, but it still makes you wonder whether there was any objective beyond "It's time to build another city".

Real players identify opportunities and goals, then perform actions to take advantage of those opportunities and form plans to achieve those goals. They see a cluster of luxuries or a strategic resources that they have the opportunity to claim for their empire, and they settle them. If I want to forward settle, I escort a settler, plant the city, and prioritize aggressive border expansion, walls, and an encampment. If I need a coastal production city, I search for the best possible location, plant a city, and prioritize improvements, a harbor district, and growth/production.

Unfortunately, as I understand it, Civ AI is a system of behavioral weights and and biases that goes around making each decision completely separately, based on an analysis of current needs, weighted by leader priorities. It's easy to program and tweak from a design perspective and adaptable when it's in game, but also discontinuous, incapable of intent or planning, and as a result, very erratic and apparently illogical in its decision-making.
To add to this, AI needs to have some agenda behind his decisions and it needs to be consistent. It might be still "weak" as those decisions might be far from optimal, but they will be consistent. Afraid it was not the case with CiV and by the looks of it, it is not going in that direction in CiVI...
 
Yeah I can confirm what monkeyfetus says as well. It doesn't just appear like the decision it makes are not goal orientated, goal orientated behavior is actually not in the code (outside of some of the military/diplomacy stuff), and it's very unlikely that civ 6 will include much more of that. There are some empire building 'strategies' that the AI can end up pursuing, but they're based on rather minimal data, and don't do much more than slightly increase the weights to favor say gold buildings. These in-the-moment weight based systems can still be surprisingly powerful if tweaked correctly, but it will definitely have a power ceiling, and lacks a certain immersiveness.

In almost every single case where players think they observe some clever AI plays, its just their imagination, the current AI doesn't do any of that. It isn't forward settling to do harm to you, it just scored that spot the highest. It isn't deciding to rush a great walls to prevent another AI from having it, it just liked the weights on it.
To some degree even actually competent AIs, like chess AIs are the same way. But they at least do take the amount of information you expect into account and have much more of a concept of the future.
 
I think a large part of the problem is that the AI generally only does decisions based on current situation, while being poor at planning for the future, but that's also a solvable problem.

Take the case of an AI declaring war and not having the troops to effectively wage it. That's an error of planning. If it instead changed an internal AI state to "intending to have war with this civ" which again changes motivations to increase military spending, then once military is strong enough it starts its warfare (whether sneakily or openly depending on circumstance).

It would always need to re-evaluate it's plan each turn though (like if someone attacks him before he gets his war started), so not saying it's easy. But given enough ability to decide both short term and long term plans, it should be possible to have the AI try to pursue both in a way that makes it challenging.
 
^^

The AI in Civ5 does pretty much exactly that actually. You can argue that it doesn't do it well, but it definitely does do it.
 
I am very late to the whole Youtube/twitch thingie.. That being said, I watched TB a few years back, not ever watched again. Having seen you say he said he thinks this AI is good, based on the press release makes understand my decision was 100% right. There is not enough data to come to that conclusion.
 
I'd love it if AI's would have a "goal" each era. Might be as simple as:

Ancient Era: found 2 cities
Classic era: Conquer neighbour
Mediëval era: Gold focus (goal: bunch of trade routes)
Renaissance era: Culture focus (goal: specific wonders)
Industrial era: Work towards culture victory

Basically force the AI to have the same goal for an extended period instead of switching long term goals too often.

Anyways I'll judge the Civ6 AI tonight in the AI battle royale.
 
The more you force the AI to do things for 'flavour' or 'personality', the worse you make it at doing the basic things well. This holds for any level of programming skill you have.

The more priorities you give it, the less well it can focus on each.
 
From people who didn't have to code a game or AI from scratch, and who took years to code their enhancements

Inaccurate in terms of time, and irrelevant otherwise. If a few civilians could pull it off, so can Firaxis. As many above are pointing out, it's most likely a business decision on Firaxis' part. I don't even knock them for it. But don't tell me they couldn't have made a better AI with the time and resources they had.
 
"But the problem is, the player is just a bit better at those things, it will always be the case because we haven't actually invented an artificial intelligence that is more intelligent then humans yet."

I dont believe that is so true, especially for the cases where there are a specific number of defined choices. Take for example chess, If you evaluate every possibility and choose accordingly, it will definitely outmatch any human player (given enough computational time of course). The problem here is you need to evaluate still to many possibilities quickly converging to infinity as the number of possibilities increases. I dont know if they are scripting the AI in that way but there are several global optimization methods that evaluate only a specific number of possibilities and through each iteration it can go better choices (if not the best). If an AI is scripted that way, I am pretty sure it will outmatch the human if the computational time given enough (of course it may make the waiting times enormously long).

so I would like to also have another difficulty setting which we can sacrifice in terms of turn times.
 
One thing I have not seen mentioned here is the lack of AI 'algorithms' that CHANGE with difficulty level. They need to add this instead of just the old AI 'cheats' mechanic. The 'cheat' mechanic giving them boosts is fine, but at each increasing level of difficulty, better 'algorithms' should be enabled so the AI actually gets 'smarter' as you raise the difficulty level.

I also hope these 'algorithms' are exposed to the modders so they can evolve with the game rather than just better 'settings' getting applied.
That is a very bad idea.

It is hard enough to make a somewhat decent AI, why waste time thinking up ways to cripple it. (instead of just giving it a penalty)
Basically they should make their best possible AI, and then give it boosts/penalties.
 
That is a very bad idea.

It is hard enough to make a somewhat decent AI, why waste time thinking up ways to cripple it. (instead of just giving it a penalty)
Basically they should make their best possible AI, and then give it boosts/penalties.

pretty much. why make multiple algorithms of the same thing, when you can just make it once and aim for best, and either 'give bonuses' if it's not enough, or 'insert logic reductions' if it's actually good. The latter case would be for below Prince generally.
 
pretty much. why make multiple algorithms of the same thing, when you can just make it once and aim for best, and either 'give bonuses' if it's not enough, or 'insert logic reductions' if it's actually good. The latter case would be for below Prince generally.
Well, there is one thing that could be changed for higher difficulties, which is reduce the effect of agendas and personalities. After all, those influence the way AI behaves not based on trying to do the best job they can but their predetermined personality. However it's debatable if we really want that because it makes AI personality less interesting.
 
Well, there is one thing that could be changed for higher difficulties, which is reduce the effect of agendas and personalities. After all, those influence the way AI behaves not based on trying to do the best job they can but their predetermined personality. However it's debatable if we really want that because it makes AI personality less interesting.

Well the predetermined agendas seem designed to help the AIs maximize their uniques..so it might actually be bad to cut them out.
pretty much. why make multiple algorithms of the same thing, when you can just make it once and aim for best, and either 'give bonuses' if it's not enough, or 'insert logic reductions' if it's actually good. The latter case would be for below Prince generally.
I think it is definitely better to put in AI penalties/human bonuses for lower levels rather than "logic reductions"... penalties/bonuses are easier to scale.
 
As an option, yeah. But with more funny quips etc as they berate you. I think part of the issue with V's "gameyness" is that the face of the AI's intentions is ultra serious dry leaders, who do a semi good job of looking annoyed in a game where they don't really do other emotions. It just gets real depressing when the denouncements get on a roll.
I guess I love the immersion aspects of the game. I enjoyed that I'd usually build an alliance in IV which would survive the likelihood of me winning, cos I'd built a good relationship with that leader; and as the attacks increased they were someone I could count on to stand by me. It's just a game, they aren't real lol... but if you play one game of Civ for up to 80 hours as I often have, the experience is much more holistic with a bit of levity :)
I agree 100%. If you are going to do it, make it different. Break the immersive barrier, but make it FUN doing it.
Have your normal difficulties where they play as more immersive, and act more "rationally" and talk properly. Then have a difficulty that is specifically turns them into game players. They act more like gamers, their quotes are more light hearted, etc.

When you get a run of denoucements, have the first talk like he's actually serious about it. Then his allies make quips and comments. While the ones who are just following up might say "Sorry, peer pressure." or "It's not that I have anything against you, but it's clear no-one else around here likes you. If I'm going to get anywhere I can't be directly associated with you... But it's only because everybody hates you."

When you declare war on Rome and he's already at war he can quip things like "Et tu, <insert your leader name>".
 
I agree 100%. If you are going to do it, make it different. Break the immersive barrier, but make it FUN doing it.
Have your normal difficulties where they play as more immersive, and act more "rationally" and talk properly. Then have a difficulty that is specifically turns them into game players. They act more like gamers, their quotes are more light hearted, etc.

When you get a run of denoucements, have the first talk like he's actually serious about it. Then his allies make quips and comments. While the ones who are just following up might say "Sorry, peer pressure." or "It's not that I have anything against you, but it's clear no-one else around here likes you. If I'm going to get anywhere I can't be directly associated with you... But it's only because everybody hates you."

When you declare war on Rome and he's already at war he can quip things like "Et tu, <insert your leader name>".

Yes!! That would be awesome :D
I mean I love playing board games to win, and don't enjoy it as much if others aren't doing their best to win either. At the same time though, the friendliness and banter (generally! lol) remains. It's a big part of why board games are making a resurgence against computer games :D Making the AI more witty or light doesn't replace humans at all of course; but I think I'd enjoy what you have described far more than V's in built...sourness.
 
Top Bottom