I started a new game with the visibility cheat on to see how the AI was doing. By turn 61, one faction had suicided itself another had somehow lost both of its cities. The third wasn't doing too badly with 3 cities, but it was extremely far behind in pretty much everything. I still had fun - the 60-odd turns were great as I put together some small armies, leveled my heroes and sovereign, started my little kingdom in a secluded valley, and even banded my heroes together to meet a threat greater than any one of them could have faced alone.
Quick list of problems:
City placement is non-optimal. When they can see a 5/3 2 tiles away, they pick a 4/3?
AI armies had no idea what their limits were and frequently killed themselves off on absurd fights (level 1 sovereign vs Hoarder Spider Army, 3 times). One of the AI heroes had 5 wounds on her by turn 60 after killing herself at least twice to the same rock spider. [edit: it's been pointed out that this may be because of incorrect labeling of monster difficulty, or possibly a result of charting a course into the fog of war, not deliberate suicide on the part of the AI]
Outpost placement is non-optimal. They'll place an outpost for a resource they can't harvest yet, and miss harvest-able resources nearby that could have been claimed with the same outpost if it was placed better
Initial build queue selection could use improvement, it wastes a lot of turns early on. Later, when rushing, might have been losing turns by not putting anything in the queue, but I can't tell that for sure
Don't intelligently manage unrest. Yes, the unrest mechanic is still in its infancy, but as far as I can tell they ignore taxes and leave them at normal.
Doesn't build improvements sometimes - Verga created an outpost specifically to get a fire shard but didn't build a shrine on it for 31 turns. All of the AI's started with a gold deposit near their capitol, but, by turn 61, not a single one had built a mine on it. I had mining by the mid-20's, so they should have by then.
Here's the full story:
Map details:
Small, Temperate
3 opponents (2 empire, 1 kingdom, I was kingdom)
Challenging Difficulty
Moderate Monster Frequency (I figured this would be easier on the AI)
Fast Tech Pacing
Dense Magic
Dense Resources
Only reloaded once when the graphics went haywire until turn 61, when I reloaded trying to see what had happened to Magnar and GildenTurn 1Lord Markin - started on a 4/3, 2 tiles from a 5/3. Founds a city on the 4/3 and it becomes a 5/3.Warlord Verga - started on a 4/2, 1 tile from a 4/3. Founds city on the 4/2 and it becomes a 5/2.Magnar - started on a 4/3. Founds a city on the 4/3 and it becomes a 5/3.On challenging they get a bonus grain? Totally ok with them cheating, but I was surprisedNot a single sovereign moved from their city, all had remaining move.
I have a start in the woods by a grain, with a 5/3 2 tiles away. Great start.Turn 3Lord Markin: Recruited a hero, not building anything, just moved one space from his Capitol Warlord Verga: Also moved just one space from his capitol, building a scoutMagnar: Moved just one space from his capitol, building a scoutTurn 5Lord Markin: Has cast Tireless March (challenging AI?), hero was immobilized by a rock spider (I'd lol but it's Lisbeth, she's weak)Warlord Verga: Ignored the hero near his start, but gained level 2Magnar: Recruited a heroBy this time, I have my sovereign and a hero at level 2 and close to level 3, and am 2 turns from producing a pioneer. I think they're all 7-8 turns from getting the Scout. They're operating at normal taxes (38% unrest) and, judging from how much gildar that's making me, they're exchanging almost 30% of their research and production for 1 gildar a turn. That's a dumb decision.Turn 7Lord Markin: Killed some bandits, is nearing level 2Warlord Verga: ExploringMagnar: Has somehow moved his hero into the same square as a bandit unit without a fight occurring, but otherwise has done nothingMy pioneer is done and moving towards a 4/3 that I spotted. They're 6, 6, and 9 turns from having a scoutTurn 9Lord Markin: Still level 1, close to 2. Exploring very slowlyWarlord Verga: ExploringMagnar: Hero got killed by the bandit army it was on top of, but Magnar avenged them (Still level 1, close to 2)Turn 11Lord Markin: Started a fight with and lost to a Hoarder Spider's Army!!!! At level 1, alone?? He did 20 damage though. But really? Attacking an army with 141 attack at level 1?Warlord Verga: Hit level 3, 5 turns from producing the ScoutMagnar: Still 2 turns from producing a scout, still level 1. Hero just finished immobilized, but hasn't moved yetI've founded my second city and recruited a second, level 3 hero. My first hero and sovereign are both level 3 as well, though my sovereign has lost both bandit units he started withTurn 13Lord Markin: Finished the scout, and Markin came off immobilized. Lisbeth has been sitting next to the rock spider that killed her for a turn or two now, at 14/19 hpWarlord Verga: Finished the scout also. Must have rushed production or something. Scout is sitting in the capitol city though.Magnar: Has finished the scout and reached level 2, Hero just came off immobilizedMy second pioneer is done and headed to claim an iron ore and 2 shards nearby, as the second city site near me is guarded by some Knights of Asok that I'm not ready to faceTurn 15Lord Markin: Lisbeth lost to the rock spider again. Producing a tower of domination. Markin staying in the city, healing.Warlord Verga: Scout was off exploring, must have rushed a pioneerMagnar: Scout and other hero sent off exploringOne of my heroes was immobilized by a sand golem, the other reached level 4 killing itTurn 17Lord Markin: Rushed the tower, rushed a pioneer, but Markin (11/32), hero (5/19), and pioneer are all sitting in the city, probably healingWarlord Verga: Finally recruited the hero near his start location, has sent the pioneer off in a direction I don't think he's scouted yetMagnar: Magnar is blocked by an army of 3 fire shrills and a level 5 empire hero. I keep waiting for him to suicide against the shrills. His starting location has 2 ways out, one is blocked, the other is being explored by the hero / scout (in separate armies). It was a crappy start location to be honest.Turn 20Lord Markin: Pioneer moved out of the city solo, but just one space, and to a location right next to a mite army. Managed not to get killedWarlord Verga: Used the pioneer to grab a nearby fire shard, and recruited a level 3 hero. But, 2 turns later, he still hasn't started a shrine on the fire shard that I can seeMagnar: Did suicide against the fire shrills, but managed to take 2 and a half of the 3 with him. Why did he send his pioneer that way?
[URL=http://imageshack.us/photo/my-images/804/fallenenchantress133600.jpg/][/URL]
His hero also walked right by a lost library (later went back for it, losing a turn)All 3 now appear to be rushing pioneers. It looks like I can tell when they rush by the "no construction" I see in their cities. Does that mean that, after they rush, they don't start something immediately though? If not they might be losing a turn of construction. I can't tell though.Taking stock:At turn 20, I have 2 cities and 2 outposts, a level 4 sovereign, a level 4 and a level 3 hero. Both my cities are halfway through workshopsIn contrast, my strongest opponent has 1 city, 1 outpost, a level 3 sovereign, a level 3 hero, and a level 1 hero. And a scout.Turn 22Gilden chose a poor location for its first outpost. Moving it 2 tiles would have brought more resources under his control
Magnar has just spend the last 2-3 turns not producing anything and, as far as I can tell, not rushing anything. His hero did turn back and get the lost library though, but wasted at least a turn doing itTurn 30Lord Markin: Put up a second outpost, again in a non-optimal position.
Warlord Verga: Either hasn't built that fire shrine or had it destroyed by the widow spider. Close to founding a second city, the best available is a 3/4 or a 4/3Magnar: After heroically slaying the last fire shrike, decided a Hoarder Spider Army would be easy pickings and is recuperating in his capitol.
Turn 35Every single one of them started with a gold deposit next to their capitol. By turn 35 with fast research, none had built a mine on itTurn 40Warlord Verga: Founded his second city, it has 3/4. Not sure if it was on a 3/4 or a 2/4 and got a +1 grainI was about to found my 4th city in the last of the fertile land near my start position, having brought together all 3 of my heroes to kill those damned knights. I control 4 shards, while Magnar has 1 but Verga hasn't even bothered to build a shrine on the one he used an outpost to take over. Markin placed his outposts such that he just missed the shards, apparently he didn't want them.Magnar got his second city at turn 44On turn 45 Lisbethe and a pioneer attached the same Hoarder Spider Army that killed Markin a while ago with predictable results (although she was level 2). On turn 47 Rahelle the Scorpion (Verga) and a Black Widow shared the same tileOn turn 51 Verga finally started construction on that fire shrine (31 turns later) and got his 3rd city, on a 3/3
By turn 61 Magnar had somehow lost both of it's cities (no idea how, as Magnar was immobilized at the site of his former capitol)
And Gilden had been destroyed! Lord Markin tried attacking a group far too powerful for him (again) and died, somehow his capitol was gone as well. I didn't see it fall and couldn't go back far enough to see why it did
To be fair, you're using a cheat to see everything, so you know the dangers before you go and all the stuff that's out there, there's no need to explore or keep close to your city to defend. As a result, you can't possible compare your progress to theirs. Take for example, outpost placement, if he doesn't see the other resource, then he's not going to be able to tell where that "optimal" spot would be. That isn't to say that they can't improve, like avoiding fights that you can't win (assuming that they CAN avoid it). I do think that it would be better if they could scout a bit better before committing to building stuff, but I fear that it would end up making them vulnerable to being rushed by players. It's a tough balance.
Is it a cheat to see everything though? Or just AI visibility?
( I would agree that either way, this could influence your own progress: takes a bit of the random/risk factor out of the equation)
I agree that my progress was improved by being able to see everything, but, even more than that, I had a much better start location than any of the AIs in this fight. I probably should have been in the lead at that point. The main point of this post is to shine a light on unintelligent things the AI was doing, not point out how I did better (looking back now, I probably should not have even included my progress, and I won't in future AI posts).
We can't make excuses for the AI though. This game is progressing extremely well - the magic is epic, faction traits mean something, heroes are legendary, etc. If I had to rate the game (minus AI) right now it's easily an 8 heading towards a 10. It's a lot of fun. But the AI is sitting solidly at 2/10 and hasn't gotten a whole lot (visibly) better in the months since the Beta started.
Verga was on a 4/2 standing next to a 4/3 at the beginning, but the AI seems to be hard coded to found the initial city on the spot where it starts. It wouldn't have even wasted a turn moving there, as, after founding the city, it ended turn with 1 movement point left. And Markin was 2 tiles from a 5/3 (2 tiles is in his FOV) but founded on a 4/3. The 5/3 was well within his initial FOV. I've never coded AI before, but this seems like an easy fix? Choose the best yield within 2 tiles, settle there.
As for the outpost placement - same thing. The resources locations were clearly known to the AI, as 2 were between his city and the outpost and 1 was one tile further away, also inside the FOV of the pioneer when he made the outpost. If the AI isn't able to reevaluate placement as new information becomes available, it's doing something wrong. If it doesn't prioritize resources that it can use, it's also doing something wrong. And if it doesn't use the resources that it captures (that fire shard above), it's doing something even more wrong.
A few other things that occurred to me since last night:
The suicidal assaults on the Hoarder Spider - what's the chance that the AI was simply moving into the fog of war and accidentally picked the Hoarder Spider's location as its destination? How likely is it that the AI even has the option to cancel an "attack" like that? With the fog of war not receding until after units have moved (a bug in my opinion), it could have landed on that spider by accident. 3 times.
Kalin correctly points out that information (sight) is power - at harder difficulty levels, does the AI get a sight cheat? Maybe at challenging it gets +1 to units, +2 to outposts, and +3 to cities. Double that for hard, and give it universal visibility at ridiculous.
Each AI had 3+ enchantments out (Tireless March, Burning Blade, something on the city). I have to believe that some of those are ++ for challenging difficulty, but does it have to pay mana upkeep for them? If it did and didn't have a huge mana bonus for challenging, it was probably bleeding mana every turn. That would be more of a handicap than a bonus.
I really don't see why you found it necessary to call out Brad in your otherwise very interesting post. There's still a lot of suck in the Beta, not just the AI.
I think what we're seeing is the reason why AI development is usually left for last and most games ship with a braindead (and heavily scripted) AI.
The game we're playing now is going to be immensely different than the game at release (for example, we know that city development is getting completely redone, probably the tech tree too, and who knows what else). Yet it's still going to be Brad's AI playing that game. The challenge he's set himself is really huge, not because he's a terrible AI programmer, but because the AI is actually going to have to be able to play the game in whatever form it eventually takes (mods and all).
Yes, the AI currently does some terrible things (and it definitely cannot handle early-game adversity). But it's much smarter than it was in the last Beta. And putting the AI through its paces is part of what this Beta process is about. On that point, this post should be quite useful for Brad.
I also think you're underestimating the way a seemingly small change can effect the performance of the AI. There's a thread somewhere in these forums discussing the difficulty the AI has with building units intelligently because the unit costs in the XML are all out of whack. That sort of decision making is transparent to the player during the play-through, but it has repercussions throughout the game.
If the AI doesn't build quality units to back up its low-level heroes, they're obviously going to die a lot. But if it does do that (And at some point soon, I feel like it will), it will start wiping out the monsters around it and leveling up its heroes. Its pioneers will be able to travel safely and its cities, outposts, and improvements wont under constant threat.
I called him out because, as you say, "The challenge he's set himself is really huge". The AI challenge is huge, daunting, and incredibly important to the final state of the game. As CEO, he's probably working 50-60 hours a week just managing the company. If the AI team is just him, the game is going to suffer for that despite his proven skill in AI coding. I'm not 100% sure, but I believe he did GalCiv? That game had exceptional AI. but was he Stardock's CEO when he did it?
I apologize if I came of as disparaging his coding ability - I'm just questioning his time availability.
Frogboy is the founder/owner/ceo of stardock. So, he was indeed the ceo when he did the ai of the other games. Honestly, the AI for this type of game is very difficult. The possibilities are so varied that it is a big challenge. There is work to do, but right now the ai is doing a pretty good job at some things. As the balance gets worked out and the AI priorities get set correctly you'll see an improvement. The things they are doing wrong are not so much a defect in the AI coding, but the way the AI sees the world.
For example. You may know that a hoarder spider army is really a dumb target for a level 1 sovereign, and the AI might know that a army with a rating of deadly is a dumb target, but what if that army is marked as weak, and the AI has a 20% chance of attacking it. Is it really the AI's fault that the army is mis-identified or is it following rules that aren't being adhered to by other game mechanics. I'm not saying that's what happened here, but you can see how a simple balance error in a xml file can cause the AI to do "dumb" things.
In any case, your analysis was very good and should be helpful to Brad in his AI development. I would suggest you attach a save game as well, so he can debug the game you played.
AI coding is a very specialized skill set. What was the last game with GOOD AI that you've played?
You make an interesting post but calling out Brad is exactly the reason his participation has been waning.
Most of the things you mention are due to the values in the game's XML. I suspect one of the modders can verify this. You don't think the game hard codes what spells it should cast when do you?
It's a good post but the editorial is off putting.
That's a great point about the save games, next time I'll keep saves of the turns where weird things happen. Planning on doing this again as soon as I have a few hours free.
And I'll skip the editorial
What I was trying to say is that while the AI in Beta 3 is poor, it's a vast improvement from Beta 2. And I think that there are some small changes (and, as Murtea points out, bugs) that are significantly hampering our perceived performance of the AI.
If the AI was stagnant, or not as ambitious in scope, I could understand you thinking that Stardock hasn't committed the resources it needs. But since the AI is clearly improving from beta-to-beta, and since we're still a LOOOONG way away from FE going Gold, it's a bit ridiculous to call Brad out.
Also it distracts from what is otherwise a useful beta report.
If I were you I would delete that bit from your OP. Just because you made an error in judgement, doesn't mean you can't fix it. You might not need to apologise like I did when I went a bit too far in a thread, but I think it would show good character to take into account what sweatyboatman said, and remove the calling out bit.
No problem doing that at all. I feel I wasn't quite calling out, I was asking a question, but it's clearly caused controversy that has detracted from the points in the main post, and I apologize to any I've upset / annoyed. I'm a huge fan of Stardock and their games (even going as far recently to see if they were a publicly help company so I could invest) and didn't intend to insult them.
Yeah, sometimes you intend the right thing but don't get the best reception. Life is like that sometimes. And you might not know that Brad is mostly taking a break from the FE forums until Beta 4, well for his own reasons, but there was some critical stuff he didn't like. If that hadn't happened maybe people wouldn't have worried so much about what you put in the OP. But I realise you were talking about time availability rather than technical expertise. Even though it was not your intention to cause controversy, well done for taking notice of suggestions and editing the OP! karma for you!
wow...thats something really unique in the internet.
Your AI observation is very interesting. Thx.
Wow, I completely missed any / all of those posts StevenAus - in that light, what I wrote is extremely controversial
Not everyone has the time to read all the forums. And I know you didn't intend to insult anyone. Just that Brad (Frogboy) is feeling a bit under fire at the moment, for his own reasons, and you kind of stepped in without full knowledge. Don't worry, I've done it too (stepped in a situation without full knowledge).
There are many great features available to you once you register, including:
Sign in or Create Account
