The AI Complaint

[Atrocities]

Over the last few years the number one complain about SEIV seems to have been the AI. With all do honesty I would like to counter the arguement that the AI in SEIV is weak. It is not. In fact it is quite good AI given the limitations of AI scripting in general for games like these.

It is next to impossible to simulate a true human response in an AI but SEIV does a wonderful job of simulating it.

Granted, the political aspect of the game is, in my humble opinion, one of the weakest parts of the game over all, and this does factor into how the AI works overall. Never the less, the AI, when properly configured, can be quite challenging for a single player.

Now I have played many games over the years and I have to say that the AI in this game does a far better job than most games in this genra. Games such as Rebellion and Birth of the Federation.

Compare SEIV AI to games like MOO, MOO2, and MOO3 and you can see that SEIV AI is on equal footing but has one huge advantage over the others, its extremely moddable.

That being said is without a doubt the shining star of Space Empires IV AI. It can be heavily modded and or customized.

The game allows for bonuses and difficulty settings that also enhance the AI's ability, albeit by cheating, to challenge a human player.

Over all, asided from the political aspect of the game, the AI is quite good and does a very decent job of challenging a player when poperly configured.

So the next time someone says that the AI in this game is weak, just remind them to give the AI max difficulty and bonus, while they take 50% on all charateristics for themselves. Then lets see who has the last laugh.

"All your bases belong to us."

[Kamog]

The A.I. is difficult enough for me. I have never won against high difficulty and high bonus TDM empires. Especially the Tesellates, they always clobber me. Even if I use all sorts of gamey tactics, and take advantage of the A.I. with unfair trades, etc.

[Q]

I agree with you that the game against the AI can be both fun and challenging. Otherwise I wouldn't continue to play this after all the years!!
It is more that in some situation the AI is just stupidly unflexible and you can't change that because it is hard-coded. Just a few examples of potentially fatal flaws:
- The AI will never scrap a facility to build a vitally needed facility like space port or space yard after it lost a colony with these facilities.
- The AI will not react to rebellions/unhappiness of the population.
- The AI will not attack the enemy core systems, only the contested border systems.
- The AI will send colony ship after colony ship to planets they won't reach alive or within reasonable time and at the same time ignore nearer empty planets (due to lower values).

That is my hope for SE V that we can improve the AI and make it more flexible.

[Emperor's Child]

Nevermind the fact that they keep pumping fleets and ships into the same ambush zone time and time again. I'd hope that SEV will allow the AI to tag a site as "enemy ambush" similiar to how it does mines, with a minimum size of fleet needed to enter and maybe a counter to remove the ambush tag after so many turns. It should then only send a fleet of ships into the ambush zone, but will prevent the single ships from getting easily snapped up.

[Ragnarok-X]

Atrocities said:
Compare SEIV AI to games like MOO, MOO2, and MOO3 and you can see that SEIV AI is on equal footing but has one huge advantage over the others, its extremely moddable.

I disagree. I have played MOO2 for about a few hundreds of hours and the AI is pretty awesome, compared to the AI of SE4 it is, in fact, superior. Both in combat, economics and especially diplomatics.
Same goes for MOO3, though i havent played that quite as long, most probably on a 2-3 douzen hours.

[El_Phil]

Q said:
- The AI will not react to rebellions/unhappiness of the population.
- The AI will not attack the enemy core systems, only the contested border systems.
- The AI will send colony ship after colony ship to planets they won't reach alive or within reasonable time and at the same time ignore nearer empty planets (due to lower values).

The first point is worse than that. After switching sides to a near defeat enemy I chucked troops on almost every world to calm the riots and make the jubilant. After a while I swapped back to the original side. A while later I swapped back again (what can I say, for some reason I really wanted to prolong that game ) and most of the troops had vanished and half the empire was rioting. Again.

The endless colony ships is somehow worse, streams of colony ships will continue to try and colonise unreachable systems. Its worse in say STM where supplies are tighter as the colony ships will run out of supplies and slowly limp through your space trying to reach an already colonised planet.


Having said all that the AI, especially TDM, can be a challenge if you steer clear of gamey tactics, fight strategically and don't maxi-min your empire.

[Atrocities]

They are drastically differant games and the MOO AI cannot not be modded as easly as that of SEIV. Also SEIV for how it was made, and how long it has endured, has surpassed MOO2, and 3 popularity by a considerable amount.

[Fyron]

The moddability of SEIV AI is extremely limited. You can not affect how the AI operates, as the actual algorithms are hidden from the user. All you can do is tweak the parameters. You can fiddle with designs and build queues, but the AI still behaves in exactly the same manner. Even TDM AIs are just as terrible as the stock AI in most regards. They make all of the same behavioral mistakes as the stock AI, suffer from the same limitations. Certainly, the TDM authors have done some great things to try to make up for the shortcomings of the AI, but they could not do enough due to hard code limitations. Aaron has a lot of work cut out for him to make a semi-decent AI for SEV...

The MOO2 AI is better designed than the SEIV AI. It still isn't all that good, but definitely better. Of course, it did not have to be designed to be adaptable to a wide variety of data modifications, so it could be optimized for the game better.

[ToxicSlurpee]

The problem here is the "A" part. It's artificial, which makes it not at all intelligent. A human can outsmart an AI opponent very easily by just watching to see what it does. In an abstract situation such as a strategy game, the only option that an AI programmer has is to give the AI a bonus to provide a challenge.

I think that's why some people say the AI in SEIV sucks. Mind you, it's horrid in most other games as well. The thing is, it gets a bonus dependant on difficulty. One human ship is worth more than one AI ship in many cases, as the AI is probably predictable.

Until we create artificial sentience, AI will have to do as is. When that isn't satisfactory, we'll simply have to play each other.

[geoschmo]

ToxicSlurpee said:
Until we create artificial sentience, AI will have to do as is. When that isn't satisfactory, we'll simply have to play each other.

I agree it's very difficult but I think it should be possible. The problem is a limitation of using algorithms itself. You quantify the variables, determine the best possible outcome, the most efficent means of achieving that outcome, and follow those steps. You continue on that course until the variables change and then you stop and reevaluate. It's all very logical, but isn't at all how most humans operate.

We often times ignore important pieces of data, make incorrect assumptions, reach incorrect conclusions, and don't always do what is the best in any circumstance. This intruduces an aspect of unpredictability that will sometimes result in success in spite of poor planning.

I think an AI could approximate this without having true sentience. Instead of always taking the "best choice" for instance, it could be programmed to rate various options and randomly choose one from all options that are at least within a certain percentage points of the best one.

I am sure it would be hard to code such a thing though. The key would be for it to remember what it's done in the past and what other players do in response, and factor that into it's future decision making. Other wise you'd have a bot that carreened from decision point to decision point with no clear stragetgy and constantly changing course.