The Command & Conquer games alone have many examples of this, most evident in the first game, Tiberian Dawn. Some of the Aritificial Stupidities may overlap each other.
Tiberian Dawn has the most of them all and are is easiest to exploit. Some examples follow:
Wall Ignorance — The AI would not target walls even if they were built into its base. This means you can literally build a chain of cheap sandbags right into its base, build armed buildings there and block all the exits with walls so that their units can't get out.
Suicidal Overconfidence — The AI has a knack of fixating the first enemy it encounters until that enemy's defeated. This leads to almost total ignorance of anything else on the field, and will result in the AI sending loads and loads of units on suicidal runs against heavily fortified positions again and again until it runs out of resources.
When the AI sends all it's units into attack mode, the Harvesters attempt to attack, too. While this makes sense in RA 2 where half the harvesters have mounted machine guns, that's several years down the line.
Blind Shortcutting — When given a command (by a commanding player or by the game's internal mechanics) units will use shortest route possible, even if it means running into a trap of a thousand guns. And if they're shot, they don't fire back, either. This is can be either a nuisance or an exploitable flaw, depending on whether you're losing or winning. See this video (starting from 6:40) for a hilarious, satirical parody of this loophole, complete with Double Take, for a demonstration.
Also, the shortest distance is determined with algebra, not calculus; that is, it takes the distance from point A to point B, ignoring any obstacles, meaning a slightly farther tiberium patch will be ignored despite having to go much further to get to the "closer" one.
If the player has two Harvesters, one attempting to return to base and the other to go out and collect through the same narrow path, the units will sometimes meet, turn twice (each time continuing to block the other's progress)... And then center their orientation, and move right through each other... The last few seconds of the clip above demonstrates this.
Blind Harvester Replacement — An easy way to defeat the AI is to make them broke. To do that, you have to kill the AI's Harvesters which the AI will insist on replacing until runs out of credits to replace them... or, for that matter, anything else in its arsenal. At this point, the AI becomes a pushover and is wide open to attack.
If you side with Nod, you can take a Recon Bike and attack an enemy Harvester without losses. The Harvester is programmed to cease harvesting and attempt to run the Bike over as though it was a crushable infantryman, but since the Bike is a vehicle instead, it will fail to run the Bike over and just stand in front of the Bike like a sitting duck. It will remain fixated to the Bike and following the Bike everywhere, even if it means driving into a trap.
This might have been changed with a specific patch, but at least in some versions of the game it was perfectly possible to run over a bike with a harvester.
If you side with GDI you can do this by attacking the Harvester with a Rocket Soldier and then ordering that Rocket Soldier to escape back to base in an APC.
Targeting Fixation — As Nod, you could completely avoid GDI air strikes by leaving an infantryman in the north-east corner of the map. The AI would always target this one man instead of your army or base.
In general, the AI in Command & Conquer is purely scripted and doe not respont to the type, number or direction you attack it at all. Each map data tell the AI when to build which units, and which way they should take to your base. This explains why the exact same enemy unit compositions attack after specific time intervalls again and again, and why they keep getting blocked by simple sandbag walls.
Red Alert examples:
Pathfinding Stupidity — The AI would often decide that sending units as far as possible toward the north-western corner of the map, regardless of if there were enemy forces present, was an absolute priority. Cue a large multicoloured mass of AI units all moving to this one corner, ignoring each other completely in the process. This is most commonly seen in the V3.03 patch, but is also known to occur in all other versions.
The AI has a hopeless strategy in Skirmish games - it will typically build a base and send a single large attack at the player, after which it will cease most operation. The AI also builds its bases in a manner that contributes to the problem by making passages between buildings small and often crowded with infantry.
Target Fixation — Attack the AI's ore truck with so much as a single rifleman, and they will send their whole army to deal with it. On a spacious enough map, this could be used to draw the AI's army out of their base, or to prompt them to prematurely stage their Hopeless Strategy.
Take a map where the sides are separated by water. The AI will NEVER build a Naval Yard or Sub Pen.
Red Alert 2 examples:
Targeting Fixation — When using superweapons, the AI suffers a similar problem to the one encountered Tiberian Dawn, only it was based on buildings. The AI would always target any super weapons you had, then any war factories, then your naval yards. And it would be the most recently built one as well. So all you had to do to avoid superweapons was place a targeted building away from the base.
When you play against hard computer with superweapons, just build a warfactory right beside their base right when their superweapon is ready, they'll be stupid enough to nuke themselves. And sell the building as soon as the superweapon is launched
The Allied (As opposed to Soviet, not allied as in allied with the player) AI will also fixate when it comes to using its jets. It will always attack the first tank built until it is destroyed. You can exploit this by putting that tank at the back of your base behind heavy anti-air and the AI will keep wasting its jets over and over again.
Suicidal Overconfidence — Unless there's an enemy unit nearby, the pathfinder would always take the exact same path to the enemy base. This led to situations where the player could amass a gigantic wall of artillery pieces and have them auto-target a single small area in which all enemy units passed through, and enemy units would always blindly go through the massive killzone, never changing up their pathfinding at all.
Considering playing the Nuclear General in the Zero Hour expansion, upgrade to Neutron Bombs and have 4 Neutron Artillery cannons force-fire on the path the AI takes. Every five minutes or so, scoop up the empty vehicles with your own infantry. Instant tank army at a very low cost!
Generals (and ZH) has prelaid AI pathwork, if you look in the map editor. So this explains it. Made the SP/skirmish game far too easy if you knew the AI paths (or looked em up using the handy map editor included with the game!).
In Generals (without the Expansion Pack) the USA AI had very poor strategy (bear with the Unfortunate Implications). With their units thrown erratically at your base, they almost refused to build defenses, and could be defeated just because they went bankrupt from their stupidity.
The higher-difficulty USA AIs in Zero Hour also make a mistake that the lower-ones don't. The brutal AI doesn't seem to grasp the fact that the Avenger's an anti-air unit, and its anti-ground attack is a targetting laser that doesn't do any damage whatsoever.
Tiberium Wars examples:
Stone Walling — A base can become invincible against the AI with a good mixture of all usable kinds of turrets and aircraft for base defense. Then, after wearing the enemy's ability to attack, send in the aircraft to do bombing missions while selling off the turrets in place of more aircraft.
The unit AI in Tiberium Wars — and even more in Red Alert 3, which uses the same engine — was particularly vulnerable to long-ranged units because a unit under attack would not react if the enemy was outside of their acquisition range (contrast the behavior in e.g. Starcraft, where a unit would close in and counterattack if it could, or run if it couldn't). Air units were particularly vulnerable, since so many anti-air units had ridiculously long ranges; a flight of Twinblades with an inattentive player, for instance, could be taken out by a single Bullfrog without reacting in the slightest as long as it remained at range. Unlike most AI bugs, actual AI players were unaffected by this, since they would always treat an attacking unit as a threat (indeed, they had rather the opposite reaction, sending tons of units at the slightest provocation), but human-controlled units were vulnerable.
The AI in these two games finally fixed the Suicidal Overconfidence problem that has plagued CnC from its earliest days. The problem is, they turned things waaaay in the wrong direction. The AI is ridiculously cowardly, often turning back an assault force in mid-attack. It was especially prone to fleeing from base defenses even when it gravely outnumbered the defenders. The real problem was that they were always slightly too slow; they would almost always wait for the enemy to open fire before retreating, meaning their units would move into range, be shot at (but, critically, not have time to shoot back) once, retreat a short way, return to the front, be shot at, retreat... all while taking ever-increasing damage. In AI vs. AI skirmish maps (with the human player(s) set to observers), the battle might never end, with both sides becoming too worn down to break through the other's defenses.
Suicidal Overconfidence — The AI generally wasn't the brightest of the bunch. As a result, your rare allies would barely ever follow you to the area after the one you met them in, given that they survived against the respawning enemies while you killed the Officers (thus disabling respawns). The only allies that were able to follow you were Escort Mission targets, which in turn had the tendency to stand between you and the enemies. They also didn't follow you as much as mindlessly charging ahead after you caught up with them at a checkpoint or saved them from enemies.
The AI in Fire Emblem games tend to have very poor decision making skills. Archers will often go straight after mages, which are often the only units that can counterattack from that distance, and they can even do this if they need to be one square from your Lord to do so. Other enemies will see a line of five soldiers and will ALL choose to swarm either the one riding the dragon who kills things in one hit or the heavily armored one who doesn't take damage. And the bosses have this weird concept that the best way to fight the heroes is to stand perfectly still in their room until you're within range, even while the heroes start filing in around them. In most cases, unless they are scripted, the AI will most of the time choose to attack any units that gets into their attack range, even when they are going to do no damage on the target (like some cavalier with wimpy weapon attacking a knight, or a mage attacking a high-res bishop).
Further expounding on the boss problem, in the early section of the game may have only weapons with a single square of range, making killing them with archers a cakewalk. later on they'll get throwing weapons which have a two space range, but players can get longbows, which have three range, again making it easy to kill bosses with archers.
The whole suicidal attack on the mages thing is actually explainable. The AI is aiming to cause anyfatalities it can, no matter the cost to its own units. Keep in mind that winning a match in Fire Emblem is quite easy — winning with everyone alive is the rub. Given a 1% chance of an instant kill, and a 100 percent chance of doing half of the max hp of the target, the AI will go for the 1% instant kill chance every time and hope it gets lucky just to spite you. While most of the time you will laugh at the foolishness of the computer, when it finally manages to get that hit in...
To be more precise (at least in the GBA ones), the AI first targets any units it can deliver enough damage to kill in one hit, regardless of the likelihood to hit or the damage that it will receive, and if it can't kill anyone, then it goes for the one that it can deal the most damage to, again not counting likelihood to hit. The issue is that the people with the lowest defense are generally your magic users. They also happen to have ungodly dodge most of the time (at least for Sages), and are fully capable of one-shoting most units in the game once they've been trained. The GBA games do not take into account whether they'll be counter-attacked at all, leading to sending a Sage out in front of your army and let him destroy half the units on the map' being a legitimate strategy. And because of the way probability to hit works in this game, a 30% chance to hit is really more like a 10%, and that's the sort of hit rate opponents often get.
The stand still thing applies mostly to bosses who are on seize points, and if they were to move then you could rush right up and seize the gate/throne.
In an early stage in Radiant Dawn you are forced into using only two units: the Squishy Wizard Micaiah and the Black Knight. Since the enemies are there specifically to KILL Micaiah and the Black Knight is there specifically to PROTECT her, you'd think they'd send their oddly large force at her at once. Nope, they go two or three at a time and hack at the first living thing they see.
Even more fun is that enemies will always, always go straight for a unit that has no weapon equipped, as they (obviously) can't counterattack (or if laguz, not very well). You can easily get Micaiah to level 20 without a problem on that map by parking the Black Knight somewhere with his sword unequipped and letting the enemies flail pointlessly at him while Micaiah whittles them down from afar.
This has applications beyond power-leveling. Many Fire Emblem veterans know that the best way to save a mission that's going pear-shaped is to unequip your strongest character's weapon. Picture this: the enemy has three swordmasters standing next to a mission sensitive character, who has only a few HP left. In order to win the map, the enemy need only attack with a single unit. Yet if you move an armored unit up and unequip their weapon, any enemy unit within range will immediately abandon their attack on the almost-dead Lord and attack the armored unit instead, even if they can't damage it.
This is nicely averted with bosses who seem to stand still on the throne. So you carelessly move your Squishy Wizard up, planning to attack next round when the boss runs right up and kills your exposed wimps. Oops.
Then again, this makes it possible to slip past them and seize the throne for an automatic win by sending a disposable unit into their attack range at low HP. On some maps, it can be worth spending a deployment slot on an untrained rookie to have them serve as bait in this way.
Which extends to neutral/allied units as well. Particularly noticeable in 3-10, where, for example, the leader of the Crimean Knights, Geoffrey, will move right up to a Bishop and then not attack it, but then Astrid will take a shot with her longbow. (Thankfully, the longbow is very inaccurate, as this was one of those situations described below where the enemy was being spared for thieving purposes)
Neutral units fall into this, especially if they're the ones you have to talk to and recruit. For instance, when Erk shows up in the second story arc of Blazing Sword, he comes out of a village and polishes off two mounted units before you get to him. If you don't get to him immediately thereafter though, he'll run off and provoke a boss with high resistance and physical attack.
They also have a tendency to attack enemies that you don't want attacked just yet (usually because they have an item that can be stolen), throw themselves at enemies far more powerful than they are, and always make their moves in the exact same order, which means that on the rare occasions that one of them does have some sense of self-preservation, they'll often retreat when it's really not necessary. And they always retreat when it's time to heal, even if they're using a healing item. Even if it means leaving a plot-sensitive character exposed.
The Radiant Dawn AI also prioritizes units who have "rescued" someone. In theory this makes sense, since if you rescue someone, the rescuer loses half Skill and Speed, making them sitting ducks who can't hit the broad side of a barn. Except tanky units don't care (at least for survival), and more importantly, they still do that if the rescuer has the skill that nullifies the stat penalties, meaning Tibarn (Who starts strong and has that skill) can easily make all enemies flock to him and kill them on counterattacks, just by grabbing someone. Who needs Provoke?
The AI also has a tendency to ignore your equipment. Meaning that flying units will attack Mages with wind magic and armored units will attack Swordmasters(who, like Mages, have low defense and high avoid)with Armourslayers.
Then you get the units who are so outclassed by every player unit in range that it's literally impossible for them to actually do any damage (because they have a 0% chance to hit or 0 expected damage - or both) but insist on attacking anyway.
Then there's the behavior fans have termed "Matthis Syndrome", which involves recruitable enemies attacking the very characters who can recruit them. This often leads to absolutley ridiculous cases of Gameplay and Story Segregation, most notably with the character it's named after, who will willingly attack, and most likely kill his sister Lena, all the while calling out for her in his battle quote. He's the most famous example, but there are others, such as Astram having no qualms about attacking his lover Midia or Wendell openly attacking you despite saying he has "no love for war". Granted, some enemies in the series are smart enough not to attack their friends/relatives/lovers, but it's impossible to know which without risking it.
Chapter 2 of Genealogy of the Holy War has you dealing with Lachesis's three mediocre Paladin bodyguards which are very suidical when there are enemies in their ranges. These knights will surround their princess by default, so if you want to play defensively or don't want to risk having them killed off in order to obtain the Knight Ring, you have to keep Lachesis miles away from the incoming enemies.
The cardinal rules for Fire Emblem AI are simple: 1. Absent being scripted to stay put, a unit will attack if an enemy is in range even if it will both inflict zero damage and die on hte counter. 2. The AI hates you, the player, specifically, and will attempt to kill characters over actually completing its theoretical objectives.
The various X-COM games suffered from this as well... Apocalypse was new to real-time, reactive AI, so the following scenario is not at all unlikely: five troops kneeling (to present a smaller target) in an area filled with bookshelves. One of them is being attacked by a worm, and they just sit there and watch, including the one being attacked. The sixth one is trying to shoot it... while on the other side of a bookshelf. His rifle ran out of ammunition so he switched to his rocket launcher. And he was right up against a bookshelf. They actually got more intelligent when their brains were sucked out and replaced with green alien goo. In that same game, two were attacked and converted, switched to incendiary ammunition, spread out and accurately started a ring of fires in and around the other four who responded in a thoughtful and calm manner by shooting each other in the heads. Not even in the remote direction of the enemies.
The original X-Com and its sequel, Terror From the Deep, were turn-based, so it was only natural that your troops would just watch while the aliens shot at them (if the aliens were shooting it probably wasn't your turn). Characters did have a "Reaction" stat, which gave them a chance to shoot if an enemy moved in their line of sight when it wasn't their turn, but humans Reaction score tended to start so low that shooting 1 in every 4 times they see an alien is akin to lightning reflexes. They did, however, have the problem with characters - on either side - not taking the blast radius of their weapons into account when shooting. You never gave a rocket launcher to someone with a high Reaction score, or snuck up on an alien with a heavy weapon (aliens, in general, having much higher Reaction scores than humans).
The AI in the first (and likely the second) game has an interesting trigger in that their behavior will automatically switch to "all out attack" starting around turn 20. Once you know this, UFO missions become much easier, especially with UF Os that only have 1 entrance. Just sweep the area outside the UFO for any stragglers and then put all your soldiers in ambush positions with reserved time for reaction fire, all facing the UFO entrance. For better results, throw a smoke grenade or two to reduce the enemy visibility (and chance to kill one or two of your soldiers before being taken down by your firing squad).
Aliens manage to grenade themselves with depressing regularity. Is there not some alien sergeant shouting "When Mr Pin has been removed, Mr Grenade is not your friend"?
Panicing civilians will happily run into the middle of a fire fight, then run round in circles in some sort of weird suicide attempt
In Pikmin, the little creatures for which the game is named seem to be completely blind to the hazards around them. Whistle to your blue Pikmin, summoning them across water? Any reds that are too close will happily march straight across the water with them. Red Pikmin can't swim. Take a few Pikmin that you thought were all electricity-immune yellows to destroy some electric hazards, set them on the electricity, and oops. Apparently a white or two was in there as well. Too bad electricity's an insta-kill.
The enemies have a bit of this as well. A Firey Blowhog just keeps on breathing fire on an idle, heat-resistant red Pikmin.
Dear Lord, Civilization. This series has suffered from a great deal of this since its inception. It's gotten a lot better, but the AI leaders often make astoundingly poor decisions.
For instance, nuclear strikes against a fortified opponent or picking fights with someone with vastly superior technology.
Or refusing free money if they hate you enough. ("We would rather eat dung than accept your offer!")
The AI also doesn't know how to properly build up cities, instead relying on built-into-the-AI bonuses. While you wasted your time building up your economy, the AI is just blessed with an economy and can focus on military build-up. (You're probably going to have a technological edge, but see further down the page for answers to that issue...)
In addition to not knowing how to properly use their cities, the AI was usually a pretty bad judge of their inherent value. The ability to exchange cities via diplomacy (added in Civ 3) could be exploited to great effect, either to bleed an opponent dry simply by taking the same city over and over again, selling it back each time for massive profit, or merely as a way to dump a scrappy city that is doing you more harm than good. In either case, the computer is more than way too happy to take it off your hands.
A patch mostly removed this, so that AIs wouldn't include cities in trades any more, although they will include cities in peace treaties, and are still fine with you just giving them a city.
They also often give away every single city except one (their capital, or a random city) if they are losing in a war, as part of the peace agreement. If you are in a war against 3 opponents and they all surrender at the same time, be prepared for unhappiness.
In the Civ 3 Napoleonic Wars scenario it is possible to win on the very first turn, even on the hardest difficulty. As Britain or France, you can convince every independent AI to give you all of their cities save one each, in exchange for defensive pacts. Upon hitting the spacebar you then win a domination victory.
Sid Meier's Alpha Centauri can also be accounted for. Even on the higher levels of difficulty, the AI will often agree to obviously unfair city exchanges- like trading their densely-populated city in an area rich with resources, housing some Secret Projects, for some sea colony newly founded, lost deep in the ocean with next to no resources; They are also prone to declaring war on you if you have adopted a civic they hate (like green economy for Morgan), even if your faction is clearly overpowering theirs.
Another bit of stupidity in 'III: The AI only cares who declared war, not what caused war. So if you provoked your neighbor civ into declaring war by repeatedly trespassing on their land and spying on them, the other civ will be seen as the aggressor and you will suffer no diplomatic penalty. This allows for some delicious Wounded Gazelle Gambits: get a Mutual Defense Pact with the neighbor (let's say Greece) of the civ you want to fight (let's say it's Rome); trespass on Rome's land until they get Furious at you; run one espionage mission, which leads Rome to declare war on you; and then merrily invade as your ally Greece declares war on Rome too (because technically, Rome attacked you, not the other way around) and now the Romans are fighting a two-front war. Ta-ta, Caesar!
Beyond the Sword expansion pack of IV adds diplomatic features that the AI players tend to use extremely badly. For example, they will gladly cast their Apostolic Palace votes in favor of ending wars - even if they declared those wars themselves and had vastly superior armies.
In Civ V, the AI seems all too happy to repeatedly insult someone who A) is decades, if not centuries, more advanced than they are, and B) has already conquered several other civilizations. Then, they act surprised when they're next.
Also from V, there is already one report of an AI (specifically, Oda Nobunaga) sending their entire army against a city-state only for it to be routed by the city. This is made much more annoying for the guy who witnessed it because said AI was their teammate.
But nothing tops Civilization V's Worker units. When automated, they do whatever makes sense to them—for instance, if (say) Townsville is at zero growth, meaning it has exactly as much food as it needs to sustain its population, and it accomplished that by putting farms on some hillsides... the Workers will go on ahead and turn those hills into mines, resulting in instant death-by-starvation in the city. And if you tell them not to, well, you're bound to Automate them again, at which point they'll return to that hill and mine it, by golly!—because you may be the ruler of an empire, but only Workers know what a city actually needs! Thankfully, The "Workers leave pre-existing improvements" option from Civilization IV has been patched back in.
And not just usual improvements. Workers will occasionally remove incredibly expensive "great tile improvements" that take dozens of turns to have a go at building - only to replace them with a farm or a mine.
Also applies to the town border expansion by culture. It will sometimes ignore valuable luxury resources in favor of tiles with the same cost and only initially better output in resources, such as water tiles. This is somewhat deliberate as it means the settling patterns or the civ's people is slightly at odds with the prosperity of the civ as a whole, and "gold" can be used to claim tiles.
Workers are smart enough to run to your cities when land-based barbarians show up, but due to embarkation allowing them to cross oceans to work on the other cities in your empire, they'll go right ahead and embark into barbarian-infested waters. Also, AI civs you are at war with will occasionally charge at you with workers; this effect has been lessened with patches but there are still opportunities to grab unprotected workers here and there.
Many reviews of V have commented that the AI is quite poor at strategically placing its military units. Apparently the game's own AI hates the new stacking rules.
Related to that, the AI seems to suffer from the calculation of relative military strengths not taking concentration of force into account. One civilization could have a greater total amount of military strength on its side by way of having lots of low-tech units, while another side has less total military strength, but what strength it does have is centralized in a small number of higher tech units. This leads the AI leading the more primitive civilization into war with the more advanced one where their larger army will get picked apart piece by piece by the much smaller, better equipped, force.
And your Military Advisor calculated by the same means, urging you to make peace while you are wiping the floor with such an enemy.
The AI also fails at realizing that indirect combat units cannot capture towns, bringing little or no direct combat units to lay siege to a town. Pick off the direct combat units and you can decimate the rest with the town defense and a ranged unit stationed in it.
Finally, both the AI and military fail to take into account productive capacity—i.e., the ability to build units—and most especially, build a large number of units quickly. This capacity can be even more important than the number of units at the outbreak of war (real life history: World War II, USA and USSR). Even more frustratingly, humans are left in the dark on this issue; there is no easy way to directly and accurately assess the AI's industrial output unless you do a fair bit of espionage. If for whatever reason you cannot do that, you can only guesstimate production based on terrain improvements and city population, and even then these estimates can be wildly inaccurate.
Civilization V after researching trapping tech on the tech-tree which enables the building of trade-posts expect the enemy AI to build them on every freaking tile, meaning without some of the AI's Special bonuses on the higher difficulties expect them to be weak, bankrupt and technologically backwards past the 250 turn mark.
They do grow slightly with a certain tech, but so does any other tile improvement. They also provide one more gold during golden ages, which leads to them being better than most tile improvements early on, which is probably the source of this.
Another thing about the AI in V is how stubborn it tends to be. Imagine a Civ declares war on you, then you curbstomp their armies and take several of their cities and your armies are waiting just outside of another one of their cities and can take it easily. You'd imagine that when you give them demands for peace, they'd accept your conditions should they be reasonable. But no, anything other than an unconditional peace agreement (or one that doesn't favor you) is completely unacceptable. They won't even give you 1 gold if it means you will stop taking their cities. On the other hand, imagine the same situation, but they are demanding an insane amount of luxuries, resources, and gold from you to end the war. Or they demand such things even if they have yet to go into battle with one of your units.
China's Chu-Ko-Nu's can fire twice in the same turn, making them practically game-breaking in the hands of a human player. The AI manages to completely forget this, turning one of the best civs in the game into a complete joke.
No matter what civ they have, they will always go wide (making lots of small cities), even if they have a civ meant to go tall (making a few very big cities). Then with Brave New World, we have Venice, who literally cannot make more than one city as part of his gimmick. As you can imagine, this causes the AI to completely crap itself and never be a remotely competitive force in the game.
Not so much costly errors as wasting of real time, but sometimes units on roads will, within a single turn, repeatedly go back and forth on the road until who knows when.
This one is a little worse, especially since it affects you. Whenever you tell a unit to go somewhere more than one region away, it will plot a way to get there, taking into account the terrain it can cross vs. the terrain it have to encounter... but not allied borders until you get there. Allied borders may not be crossed without starting a war, and if a unit you just told to go somewhere without guiding it around borders may try for a little bit to found a way around, but will then forget everything it's learned and charge at the border again, only to be deflected. If you don't Cancel Move, it will waste many turns doing that.
The AI for the computer game Master of Magic was about as dumb as it got. It would do things like trade Great Drake (1200 research points) for Hell Hound (45 research points), keep huge armies around and then do nothing with them, load up elite units onto weak transport ships and then run them at powerful warfleets, and waste all its mana throwing firebolt after firebolt at a unit that was immune to fire. But for me, the crowning moment of stupid was when the following occurs. Freyja casts Nature's revenge, a spell that attacks all of a player's cities with an earthquake whenever that player casts a chaos or death spell. Jafar, who had taken some red books in this game, casts time stop, and while the rest of us are paralyzed, summons hellhounds about 60 times. When he lets go of the time stop, every single one of his cities is in ruins. The game needed to give itself triple on almost all relevant resources to keep up with a decent player to make it even somewhat competitive.
A lot of this was remedied with the 1.31 patch for Master of Magic. Some of it still applies though.
In Europa Universalis II, the AI does not know the army attrition rules. You can bring pretty much any enemy to his knees by letting him besiege one of your fortresses with his entire 60,000-man army while you send small forces to take over the rest of his territory. By the time you're done, attrition will have brought his army down to a size where you can take him easily, even if he started out 6 times your size.
Europa Universalis III has a bug that's survived years of expansions and constant patching known as the Naval AI Death Spiral. Every country has a limit of how many naval units it can support, and as it goes above that unit, every ship is has gets an exponentially-growing penalty to upkeep, meaning the country has to pay more for each additional ship. The limit depends on the number of provinces a country has. If it's near this limit and loses a few provinces in a war, then it can suddenly be above the limit. The AI doesn't know to disband ships when it can no longer afford them. This increased cost often leads to bankruptcy, which makes the country significantly weaker and even more vulnerable to neighbors and rebellion. It loses more and more land, incurring greater and greater penalties due to the size of its navy, eventually collapsing altogether.
While the diplomacy system is thankfully more advanced than in many other strategy games, it still has its weird moments, such as the far-off one-province minor ally of that big enemy you're currently fighting who has done nothing to aid their ally so far and would in fact be totally crushed if you actually bothered to move troops over there trying to blackmail you into giving up your vassals and paying them most of your treasury's content just so that they, and only they, remove themselves from this war.
In the Stronghold series, the easiest way to defeat a computer-controlled opponentis to take out their water wells and set their fortress on fire.Since the shape and arrangement of the enemy castle is predetermined, whenever a building is destroyed they would build it again on the exact same spot, while it was still on fire, and repeat until either the fire somehow died out or the enemy ran out of resources and gold.All of whis while all their soldiers and civilians die, without the enemy even thinking of moving them away.
In Sid Meier's Alpha Centauri, the AI doesn't check whether you have units in your allies' cities and so will launch absolutely stupid surprise attacks against you, by attacking a city that you don't own but do have units in.
In Star Trek: Armada the AI will continually send ships through a chokepoint even if you've got dozens of ships filling it full of bombs going off constantly. Thus racking up hundreds or thousands of kills. The AI will do it until it runs out of resources, or if they have infinite resources, until you get bored enough to kill them.
In the Star Trek: Starfleet Command series, the AI is all over the map with idiocy, especially ally AI. They won't follow orders, they refuse to move in close to enemies, they don't fire at enemies or only fire certain weapons, they cloak and decloak randomly. In the third game, if you're playing a difficult mission or one that takes a while to beat (such as Starbase Assault or Planetary Assault), it's almost a guarantee that one of your wingmen will self-destruct for no reason at all, even if it never took a hit during the battle.
The "planetary viceroy" AI in Master of Orion III, supposedly there to help run your empire for you and eliminate the need for micromanagement, is pretty much your main adversary for the duration of the game. It usually manages to hinder your efforts far more efficiently than the equally incompetent computer-controlled opponents.
In Supreme Commander, since the game has infinite resources, if you heavily fortify your base, the enemy will attack again and again until they are defeated, potentially losing tens of thousands of units—which you can then turn into resources. This is especially notable in some of the campaign missions, where the enemy would repeatedly attack with the same force, even though your base had gone from a tiny outpost to a huge, heavily shielded and fortified behemoth.
On island maps, the AI often builds a horde of low-tech land units with no regard to how they're supposed to get to the enemy base. Even on the hardest difficulty settings.
Even more annoying is that they'll often build loads of dropships, and then almost never use them.
It is also incredibly easy to win in the expansion pack Forged Alliance by getting an endless stream of aircraft built, and having them go straight into patrolling across the majority of the map, denying the AI valuable mass deposits. Note: fighting for, and defending mass deposits was made much more important in the expansion to encourage conflicts and reduce turtling. It is incredibly easy to screw over the AI this way, never giving him a chance to expand, whilst making your own economy unstoppable. Even the supposed "adaptable" AI never really catches onto the endless swarms of aircraft, producing only a token handful of anti-air units.
The AI in general is sub-par, even for strategy game AI. By default, it builds almost no Anti-Air, even when being attacked by air units, holds massive amounts of units in reserve (IE: Keeping about 50 tanks out in the middle of nowhere while it's base is being steamrolled) for no reason, waits on building Naval Forces until it has a massive (and useless, especially on ocean-based maps!) land force, and never seems to rely on artillery, tactical missiles, or strategic missiles (nukes) to crack bases, instead sending endless streams of troops that'll never even get a shot off.
It's also known to march its Commander (and losing your Commander means you INSTANTLY LOSE) into suicidal attacks, such as straight into defenses, into the middle of high-tier units, and at enemy Commanders. Having the enemy commander basically blow himself up and take your base down with him is incredibly frustrating.
The AI doesn't do well with mods that alter the resource system. Two of those are bundled with Forged Alliance: One where resource output is vastly increased (so you only need a few mass extractors and power plants) and one where only the ACU produces resources (but more than usual). They can also be combined. The AI still builds mass extractors all over the map and thereby wastes productive capacity. It also doesn't occur to it to just build a metric crapload of Experimentals ... like you do right now.
Age of Mythology island maps have produced a whole spread of dumbass moves the AI does automatically. As in, Norse opponents turn all their resource collectors into Heroes of Ragnarok without bothering to produce enough transports to attack. At this point, the nicest thing you can do is attack.
On the first skirmish-mode map in Dawn of War: Dark Crusade (called something like "Abandon All Hope"). The AI just doesn't seem to understand such things as "putting troops on the island in the middle will let them fire rocket launchers straight at the enemy command centre". And in any map, they don't get their turrets to fire at different targets, so an ethereal Necron wraith or a rally tough unit can soak up fire while everyone else blows up the turret.
A rather bizarre example occurs in Dawn of War 2. The enemies usually have enough sense to run away from a thrown satchel charge before it explodes. However, having detected a remote-controlled bomb (which is much more powerful then a satchel), they start shooting and hacking at it...while standing literally on top of it. Ludicrous Gibs ensue.
In Total Annihilation, the AI apparently does not understand the concept that land-based units can't cross water, and will make anti-nuke launchers without making the missiles for them to launch.
Also if you manage to choke only exit from enemy base with dragon teeth AI will build stuff until all space is filled with useless crap so that nothing can move anymore.
The AI's entire pattern is simply to build random things and send them to attack the nearest enemy unit. User mods somewhat improve it by tweaking the odds of building each unit, but mostly they just cheat horribly to make up for it.
The AI in the following game, Total Annihilation: Kingdoms, did the same. Since units increased in power every time they had a kill, and since the enemy would never attack walls, botttleneck death-zones of guard towers would reach nigh-impregnable status.
Starcraft's stairwells and narrow passages: small units generally didn't have much trouble, but ordering - say - a group of Dragoons up an elevated area always resulted in a few making it through, a few behind those staying in the way, and all the others deciding the passage was blocked and cheerfully going back and taking the ridiculously long route along the whole elevated area.
Goliaths are just as stupid as Dragoons, if not more so. That's because they both use the Hydralisk's pathfinding subroutine, but are physically larger units, will engage and follow enemies as if they are air units, even thought they cant fly, and don't automatically disengage enemy units that they can no longer follow or see.
Also, in multiplayer, try a Zerg Rush or similar strategy to knock out one computer opponent early on, but leave one of his buildings alive (preferably a support building that can't produce units) and set up an expansion there. The other computer-controlled enemies will focus almost exclusively on recapturing that base.
In a 1v1 against a computer, try sending in a peon to attack their mineral line. Once you get their attention, run. The computer's ENTIRE ECONOMY will chase you all over the map, leaving you free to harvest and build at your leisure. GG
Maybe Blizzard's just been too lazy to seriously reprogram their AI, but the Starcraft 2 beta STILL DID THIS.
They improved chokepoint pathfinding in Starcraft 2 to the point where you can/could cheat by ordering a unit into a potential enemy base location under the fog of war — if it took the long route this meant the main entrance was blocked, in other words, there was a terran player there.
The developers themselves admitted to taking shortcuts with the AI. In fact, the only reason they didn't make the game's Campaign Editor more flexible, was to hide the fact that the AI is not programmed to handle major changes to the game, as it operates entirely on assumptions.
Dungeon Keeper: Monsters who could not find a path to the treasure room to get paid, or could not find a path back to their lair, or a "helpful" AI that would put monsters across locked doors (that you shut to keep the monsters where you wanted them), etc.
The enemy keeper in Snuggledell frequently believes that he can win using only a level 1 fly and a level 1 imp.
The Nintendo Wars games are usually pretty reliable, but the first Advance Wars game overestimated the importance of supplies in the game (they're only essential in 100-turn epics where there's a high rate of attrition, air units or artillery at choke points). The AI would attack APCs almost exclusively, ignoring nearby units that would ream them in retaliation, or even the infantry capturing their HQ that will win the game next turn.
AI characters in the game were programmed to use their CO powers essentially as soon as they got access to them. Olaf's power mimicked the 'snow' weather effect, which lasted exactly one set of turns - it started on a CO's turn and stopped at the start of their next turn. An AI Olaf would use it even if it was already snowing and, yes, if it had only started snowing on Olaf's turn.
On particularly dodgy maps where an AI Colin is essentially handed an I Win button, he's been known to use his basic power, which gives him a 50% money boost...when his money is already at maximum. Black Hole Rising can only handle cash up to 999999.
The AI in the first game also tended to beeline its infantry towards your HQ, generally ignoring your units. They'll only rarely ever attack with their infantry and mechs.
Usually the computer will have a few units that will stay in one place and not move unless you send something into their range. In the earlier games they would immediately go charging after your unit and attack it, completely ignoring any other units nearby, making it possible to lure a strong unit into position with an infantry, then pound it with several other units that were lying in wait. They have gotten better in later games though - now if you have too many units lying in wait, the AI will actually realize it would be a bad idea to attack and stay away.
The AI will build a lander and load ground units into it, even if you block the port and it can't move.
One glaring example is the mission Test of Time in Advance Wars 2. It's a 14-day survival mission, where you have to hold a chokepoint against overwhelming odds... in theory. In practice, building absolutely nothing will result in the AI cluttering your base with units to the point where infantry can't reach the HQ because of other stuff already being there, thus giving you the win without trying. This happens reliably.
Dual Strike exploits this trope. The second fronts in the DS War Room maps are designed to be won much faster with Auto CO off, but you'll get more experience if you leave it on.
The PC version of the classic boardgame Axis & Allies routinely makes utterly brain-dead moves, especially in purchasing units. Any AI running the United Kingdom, for example, will routinely spend almost all of its resources on submarines, turn after turn, even if the Germans possess absolutely no ships to attack.
In Front Mission 3's AI always goes after whichever unit you put down first on the map; you can leave your first placed unit at the start point and venture out with your other three, and the AI will go crazy trying to kill your first unit. Oops.
Black & White 2 introduced army fighting to the series, but what with all the other complexities of the game the AI suffered a little - it was possible, faced with a vast opposing horde, to send your Creature to attack them, then have it run away into the countryside. The entire enemy force would give chase, leaving your own soldiers free to sack their cities.
In many of the Super Robot Wars games, the enemy AI for grunts tends to prioritize them toward the "weakest" unit within their attack radius (That, or Shoot the Medic First). However, "weakness" in SRW is relative: the units with the lowest armor and HP tend to be the units that can dodge anything you throw at them with the right pilot. Thus, the computer tends to waste time and ammo trying to hit a unit it simply has no chance of hitting, minus the intervention of the Random Number God.
Conversely, in some games they go for units with the higher HP. While that means they prioritize battleships (And if one falls it's a game over), it also means ignoring half-dead units and going after that healthy Super Robot who will take Scratch Damage and kill them on the counterattack.
The automatic combat setup for your units isn't too smart either. Small, crunchy units with Defense Support will gladly jump in to soak hits for massive, heavily-armored SuperRobots. Also, the automatic attack selection will pick the weakest attack the robot has that will destroy the enemy. This occasionally leads to a Gundam deciding to use its limited-ammo Vulcan over its completely free Beam Saber, or GaoGaiGar using energy on Broken Magnum when it could walk up and hit the enemy. Finally, recent games that have pilots automatically elect to dodge or defend when the oncoming attack would destroy them, but the evasion threshold for some pilots is oddly low - Real units have been known to try to reduce their chance to be hit from 5% to 3% rather than take a counterattack. All of these, mercifully, can be overridden by the player.
A good instance of Artificial Stupidity comes from the enemies with their MAP attacks. Normally a MAP would hit every enemy in its radius, but an enemy will refrain from using it on the off chance that one of their allied units would be hit. Makes many a boss easier. Except bosses with Friendly Fireproof MAP attacks (At least you get some of those too).
In the Total War games, the computer doesn't seem to grasp that positioning its troops within range of your archers or turrets is a bad idea.
This was mostly fixed in one patch for Rome: Total War, although the AI still fares badly against artillery or lots of missiles. However, the Rome and Medieval AI are notoriously awful - in Rome, sticking a couple phalanxes (pike walls that are virtually invincible from the front to non-phalanx units) at the edge of a bridge will result in the AI suicide-charging them. The battle AI also doesn't see a need to protect its flanks and is usually unfazed by the player taking his cavalry around the AI's back. Before Medieval II, enemy generals also enjoyed suicide-charging ahead of their armies. The AI logic must have been something like, 'I want a strong unit to charge. The general is the strongest unit. Therefore the general must charge.' What an Idiot.
In Medieval II, it is possible for heavily outnumbered enemy forces to retreat to the corner of the map...and then sit there, not moving the final few feet needed to end the battle and escape, while you pummel them to death with your ranged units.
That is, if you have enough ranged units to take him out. Otherwise, you are faced with a very dense enemy formation with no flanks or the rear. Good luck with that.
The campaign map AI is similarly problematic. In the original RTW it was prone to declaring war on the player even when in a much weaker position, and completely refusing all and any offers of peace even as it was being beaten black and blue. Sometimes it even declared war by using a vastly inferior force to attack a large army of yours. The incidence of this happening was decreased by later patches, but it still happens. It is also possible for enemy factions to ask you to become trading partners and declare war you during the same turn.
After much play, the AI is known to simply determine its enemies based on either trade routes or adjacent provinces, with some special coded wars set to happen. Of course, this results in the AI declaring war on its closest allies even if they are vastly more powerful once they have too many close provinces. It also means that you have no chance to maintain peace, ever, because you're always the mortal enemy of somebody.
Another favorite 'trick' of the AI was to decide to stab you in the back and then totally throw away the advantage of surprise by just attacking a ship or two rather than going for a general assault.
It's usually a strength of computers but the enemy AI often has trouble counting. It will attack provinces with a force that, if you put up any resistance at all, will win but no longer be sufficient to stop the conquered lands rebelling. With the games somewhat random selection of what units an uprising is awarded it is quite possible for a defender to come out of an initially successful enemy offensive twice as strong as when it started.
In Empire: Total War, the campaign AI in the unpatched version was unable to transport troops by ship. In a game that includes three continents, and dozens of islands. Thus the UK is invincible to non-player enemies, for example.
You'll often see things like the AI, should its regiments of musketeers be flanked, opt to have them shoot from the narrow side of their formation rather than wheel around to face your unit. It will also always send its cavalry units in a very wide flanking maneuver, even if your entire back and flanks are covered by pikemen, running all the way so its cavalry ends up exhausted before even joining the fray. There's also the infamous "melee bug", a holdover from older titles, which causes the AI to favor melee over shooting matches with its hybrid units (that is, units that can both shoot and duke it out), even if a shooting match would be in its complete advantage. It wasn't that noticeable in older titles, where such units were relatively rare, but in a game where almost all units are hybrids... You'll also often see it reforming its entire line within range of your muskets, and thus getting slaughtered. Finally, its equally infamous friendly fire: musketeers behind the main line will happily shoot at you through their friends, with predictable results. Note that this is also true of your units, which is why you should turn "fire at will" off for reserve units. The AI, however, doesn't. And don't get me started on artillery behaviors (and AI telepathic responses to yours)...
When the AI loses a certain number of men from its musket formations it will immediately order them to form square. The square formation is a counter cavalry tactic and it quarters your firepower. Not a good tactic when you're being shot by lines of infantry and there are no cavalry on the battlefield.
Not to mention that the campaign AI is still as idiotic as ever. Enemy factions are still notoriously hesistant to surrender, even when you've conquered all but one of their provinces and that last province is currently under siege. Since the AI is coded to essentially declare war on all its neighbours from the get-go, this means that you will be fighting wars pretty much constantly. This being so, it also results in the AI declaring war on their neighbours despite having no soldiers on their borders.
The AI in the original Medieval game had a fixed strategy for attacking castles, in which they plant their army just out of range of the towers, and then send one unit at a time to attack the gate. By the time they actually break through to where your army is waiting, they've often lost half their troops. Better yet, they tend to send their general in as soon as the gates are down, and if you manage to overwhelm the general, the rest of the army will run away. And then they'll try again the next year...
Another "fun" thing the AI likes to do in Medieval II (especially evident with the Mongolians and the Timurids) is siege your heavily-defended cities (i.e. towers fire cannonballs) and just stand there while your defensive fire slowly reduces their ranks until they run away. This is a Boring but Practical way to win these battles and not lose a single soldier. Or you can just constantly send a unit out, causing some of their units to move in for the kill... right in the range of your wall archers, and moving your unit back inside. The AI will fall for that every time. In Pre-Patch Rome, they will send the entire army won't even return to their positions, leaving them right in front of your towers.
The AI in Shogun 2 is a vast improvement over previous editions, but it still suffers from crippling quirks:
When defending a castle or fort, the AI is often hesitant to the point of suicide about not leaving the fort grounds. This is because being inside a fort grants a large morale bonus and the AI is unwilling to lose it voluntarily by sallying out to meet you. This often means a relatively tiny force of ashigaru archers (among the cheapest units in the game) can win sieges simply because the enemy does nothing but stand still and take volley after volley of fire.
The AI also likes to telegraph the state of it's affairs (sparing you some scouting/spying) with offers of peace treaties. In general, if an AI is offering peace at all, or better yet, with some cash bribery, they're very much on the ropes.
The expansion Fall of the Samurai has absolutely no idea what to do about artillery. It will march blindly into the face of cannons without much thought. This is mostly due to the AI being almost identical to the base Shogun 2 AI, which pretty much never needed to worry about long-range shelling.
Also something of a bit of Fridge Brilliance, since you're bringing contemporary weapons into a far less technologically advanced culture. They legitimately have no idea what to do about it or what to expect.
Age of Empires AI players are very inflexible. In one scenario with a King Lion right beside a computer player's town center, the computer player kept making its villagers try to build a house, ignoring the lion, until it ran itself out of food.
In another scenario where you start out on a tiny little island with two houses, some villagers, a scout and a transport, and you need to travel to the resource-laden mainland to set up your civilization, the AIs never leave their island. Makes it a ridiculously easy game to win.
Age of Empires II improved the AI. Then in the expansion they made it stupider again, apparently feeling they'd gone too far.
Although even in 2 they left a major blind spot in the AI. Computer controlled soldiers were programmed to not attack gates, only walls. So all you had to do was construct all your fortifications out of gates, and they'd never come close. Or if you wanted to really rub the AI's nose in its own idiocy, when it has its entire economy and/or production facilities behind walls, simply build one of your gates directly in front of its, and it will never be able to leave its own town.
Another flaw in the AI was the way a travelling army behaves. While moving in formation towards its destination (often quite slow due to the siege equipment), the AI would completely ignore any attacks and/or threats, allowing the player to destroy the AI's army without taking a scratch.
In the World War II RTS Company of Heroes, the AI is exceptionally dumb, sending basic units against tanks, sending infantry against machinegun emplacements without even flanking, and never upgrading. While you have tanks that could kill even the most advanced infantry with nothing even close to a sweat, the AI will still be sending out the basic unit with the basic gun.
It will normally only resort to this tactic if it's been cut off from fuel supplies and so can't produce better units. A much larger problem is its total inability to understand when it's best to breach obstacles rather then bypass them. This lets you funnel them into killzones if you block off an important bridge or other area with barbed wire and tank traps.
In fact, the AI can be so stupid at times that it will randomly mass every last one of its units on a destructable bridge if the game goes on for long enough, without even needing to place obstacles to encourage it.
Evident on a smaller scale when squad AI has a consistent habit of running out of cover to shoot, despite being ordered to stand behind a perfectly good wall.
This gets to the point where the cover AI will sometimes be ordered to use cover, and will use it, on the side of the wall where they're getting shot at from.
Reversed for the sniper, who always ran to the nearest cover (and never needs to) when told to take the shot. Fixed in latest patch.
Units in Battalion Wars and its sequel aren't that smart — of course, this may be to make sure manual control is more efficient. A glaring case, however, is the Battlestation in the final mission of Battalion Wars 2, as if it's AI controlled, it seems to have to be like two meters away from the Mining Spider before starting to attack it. According to a friend, it turns out the Battlestation attacks the guns that fire the weak lasers—something that the Heavy Tanks can fortunately take care of to save time — but you can't lock onto the guns yourself. Best part? You also get Fighters, which are far harder to control than the Battlestation which shouldn't require so much intelligence to use at all, but under player control, the Fighters can total the enemy air force that threatens anything else.
In Dominions 3, battlefield AI is not that smart, which is a problem since you can only control your units for the first 5 turns. It's very frustrating to watch your mages summon weak units one at a time on the complete opposite side of the battlefield from the fight, when a nice battle evocation would totally turn the tide. A mage, surrounded by bodyguards, may cast Fire Shield (a ring of fire surrounds you), killing his bodyguards, then die to an enemy charge. At least the AI on the computer side is equally stupid.
In Dune II, you can stop the AI's attacks on your base by building four sections of wall at just the right spot. The AI units that arrive to attack can't manage to find a way around it, and just sit there. As long as no player units approach, they sit still, and the enemy doesn't send out more attackers.
In another mission, the AI suddenly sends out a group of soldiers into an empty corner of the map for no reason at all, and they remain there, not moving, until the end of the mission.
Far more obviously in Dune II, the enemy will keep throwing units at your base defense turrets uselessly even as your actual troops are in the process of leveling their base.
Free Civ's computer players are extremely stubborn with "their" territory - build a city on any square they consider to be "theirs," and they'll raze the city - without any diplomacy scene or change in relationship. In fact, if you then attack one of their cities, they'll blame you for starting the war.
Bungie's Myth games are generally considered to have a decent AI, with one exception: ranged units do not check if friendly units are near their targets. So if you fail to micromanage your dwarves, they will gleefully chuck grenades into the middle of the melee.
The AI in Planet M.U.L.E. has some interesting trains of thought, particularly with respect to buying and selling:
Land auction starting at $120? don't bother. Land auction starting at $350? MUST HAVE!
Crystals selling for $70? SELL EVERYTHING! (for reference, it sells for $100 on average)
Do I have Smithore? SELL IT!
The Trope is actually justified in the 1983 version on the hardest difficulty setting, as the price and availability of worker M.U.L.E.s is based on how much Smithore the Store/town owns. Also, because it's a 12-turn game when you do this, the Pirates will ALWAYS show up to steal all of the Smithore/Chrystite at least once — you might as well sell it while you can!
Aw, does that player not have enough energy? is the store empty? well, tough, I'm gonna hoard my energy even though I'll lose more to spoiling than I would selling it to the player.
Did another player sell ONE UNIT OF ENERGY? SELL DOWN TO + 1 SURPLUS!
Do I have no surplus? BUY UP TO + 1 SURPLUS!
Compounding the issue (and inducing Fridge Logic) is that occasionally, players will gain money off of "Artificial Dumbness investments".
This is all particularly egregious when the original 1983 game's computer players were quite capable.
The AI on any level below hard in Rise of Nations is either extremely suicidal, extremely overconfident, or any combination thereof. Often, the AI nations will demand tribute of you and attack you if you refuse, never mind the fact that your military is making use of tanks, aircraft, and long-range missiles while they just discovered how to make gunpowder.
If an AI faction controls a shipyard, they will keep producing ships, over and over, until they tank their own economy so badly that their units desert in droves. In the first dwarf mission, there's a fairly small underground river that seems to have been built in as a shortcut between two (of many cities) in the depths. The orc AI will usually take the shipyard there and produce so many dragon boats that the game will run out of hexes to place them, so in order to beat the level you can just wait for the orcs to bankrupt themselves.
In the sequel, you can negotiate with rival wizards and trade them spells, resources or locations - and they will trade for a watchtower in your territory right next to a stack of dragons that can retake it at a moment's notice. Even better, you can give them a magic item with a serious drawback (like The Halfling's Ring which gives invisibility (which many high-level units can see through) but increases physical damage by 50%) and the wizard will always equip it.
Warzone 2100 has a very thoroughly implemented individual unit AI that micromanages an impressive number of tasks, such as peeling off from formation for repairs past a certain damage threshold and then returning as soon as repairs were complete, or driving around other units instead of into them. The AI is stone dumb. These features will, at best, frequently reduce a competent player to a snarling, hissing wreck. Vital units will peel off for repairs and decide that the best way to a nearby repair station is attempting to drive through an enemy wall of hardcrete and cannon towers, and units in formation will swerve randomly and drunkenly to attempt to avoid driving into the dozen other units, despite all supposedly going in the same direction at the same pace, causing an entire-army pileup in a perfectly empty field.
Also, The enemy AI would, in skirmish games, continue to build Mini pod rocket viper wheels (AKA Freaking joke) while you have lazers and Nexus bodies. In short, the enemy makes really sucky tanks, while you have the really powerfull. Curb-stomp ensues.
In Warcraft II, the AI tends to get stuck in whatever blocks their path.
Be it trees, rocks, buildings, the AI is a complete idiot
That's not even close to how bad the AI can be. It is entirely possible to win sea-based maps that are 5 or 6 comps vs 1 player by putting a tanker on top of the closest patch of oil and just leaving it there. The computer will not build a warship to destroy yours. They will not go to another patch. If they aren't specialized to produce air units, they will just send a bunch of tankers to sit around yours for the whole map, which results in them never getting oil for transports or more warships.
Warcraft III: While the AI does well most of the time, selecting a random hero at the setup of the game can really screw them up, as they will spend the rest of the game with a hero they don't use, in turn skipping the extremely important "get gold and experience from creeps" phase. Other times, they may decide to stop evolving their main building at level two, leaving them with only two heroes, no high-tier units and only the second level of upgrades. Finally, in perhaps the most literal interpretation of this trope, Night Elf computers sometimes manage to wall themselves inside their own base with no way of leaving (except by killing or uprooting a building, which they never do, or by killing trees, which they can't do except by accident).
Early rushes tend to leave them brain-dead as well. One particulary sad example is the "Friends" map where two bases start right next to each other. They will start fighting immediately, but the winner will somehow forget about their own gold mine, instead sending its workers to harvest the ex-neighbor's. And when that one dries up, they don't mine anything any longer.
While the AI in Galactic Civilizations is usually alarmingly smart, it occasionally suffers from minor glitches. For example, if a mega-event suddenly upgrades all the hitherto useless planets in its home system to class 7 or 8, it will periodically fail to realise that colonies = power and not start building colony ships instantaneously. It also has a bit of an issue with disabled victory conditions, meaning that in a huge game it can research the entire tech tree (which would normally give a tech victory), then leave its now useless research facilities sitting around costing them maintenance funds, rather than, say, replacing them with manufacturing or economic buildings.
Cossacks: European Wars: The AI of both your own units and the computers in the skirmish mood were never particular smart. The enemy AI was obsessed with building cheap units (Like mortars and mercenaries) and only using hit and run tactics. He would also quite happily spend hours using grapeshot against buildings, when it was specifically designed for infantry.
Enemy AI in Mech Commander 2 would always focus on the first target to aggro them and ignore everything else. This made it incredibly easy to beat much of the game; send a fast mech in to get their attention and then pull it back to the other side of your squad, and the AI would charge blindly after your 30 ton scout mech and not fire a shot at the 100 ton assault mechs you have laying into them. Salvage, scout, repeat.
In the Web GameWarfare 1917, the enemy will often use poison gas on your tanks even though they're completely immune to gas.
I'm not sure if Chick Chick Boom counts as a strategy game, but I'd say it fits, given it's a game based on war between two chicken species. But you'll eventually notice that your team's AI is this. On the other team, The Ai Is A Cheating Bastard.
In the Bitmap Brothers' strategy game Z, no matter what orders they were given, soldiers approaching artillery would be hit by the artillery at long range and react by retreating. After retreating for a while, they would resume following their orders to move towards the artillery, inevitably getting hit again. Since the artillery pieces had a minimum range, if the soldiers did not retreat they could easily get inside the minimum range and destroy the gun, but the AI blocked all attempts by the player to do this.
The unit AI in "Gangsters 2" was notoriously bad. When driving, your gangsters insisted on planning their route a few seconds in advance, meaning that clicking to have them turn a corner just before reaching it would often result in them driving past the corner, turning around, and coming back. When attacked by enemy gangsters, your gangsters would fire back even if there was no hope of them winning, and commanding them to retreat or take cover resulted in them spinning frantically on the spot as they turn to follow your command and then are overridden by the "fire back" rule.
The AI in Ascendancy was notoriously easy to beat even in a hostile galaxy (the game's version of a difficulty level, although all it does it make the enemies declare war on you and each other much more often without actually making them better). While the developers have since then released a "hard AI" patch, the AI's flaws are still glaringly obvious. It notoriously fails to take advantage of alien ruins (which give you a random tech when studied, and yes, you can get a top-level tech at the start of the game this way) and is terrible at colony management with colonies often lacking in industry. It's not much better at ship design either. It might put a large number of powerful guns on the ship but not enough power generators to actually allow the guns to fire. The AI's goal seems to be to expand constantly without regard for defensibility. Their colonies might have a few orbital shields and missile launchers but will fill the planet with farms and outposts (increase possible population), and tons of defenses, with maybe a low-level factory or two. Even the improved iOS version still has may of those issues. Yes, the developers have finally realized that all that micromanagement is boring but adding an "Auto-Manage" button is pointless if the planetary AI remains so dumb. You're better off without auto-management. Good luck if you tapped the button by accident and check back later to find all your cool buildings have been replaced by the above-mentioned nightmare. You're also better off not automating research.
Starbase Orion, an iOS port of Master of Orion, had terrible enemy AI at the start, although things have since been improved with patches. Diplomacy isn't an issue, as you're always at war with everyone else. However, until the latest major patch, the AI would notoriously send wave after wave of unarmed transports to your defended colonies. It has since stopped doing that. Unfortunately, the AI hasn't stopped building ("spamming" is a better term) said transports. It just keeps them all near its colonies, figuring they can somehow stop even a single enemy warship. All these resources could be spent building warships or improving the colonies.
Update: Latest patches have removed the "build tons of transports and hope for the best" strategy from the AI menu, and it now acts much smarter. Revamped diplomacy also means you're not always at war with everyone else.
To cite just one example from the Heroes of Might and Magic series (there are probably many more), the AI in the second game is programmed to always prioritize the destruction of troops with a ranged attack. You can rout a computer controlled army with a squad of 100 black dragons, knowing it will keep ordering its troops to charge towards the one halfling armed with a sling.
Where to begin with LEGO Rock Raiders for the PC? How about the fact that your Rock Raiders regularly cut corners across lava? Especially highlighted by the fact in one of the tutorial missions, Chief claims that "your Rock Raiders are very clever."
One mission very late into the campaign has some Rubble on the ground that's extremely far away from your base, which for no discernible reason the Rock Raiders continually place 200% priority on, ignoring ALL other tasks unless you've learned how to arrange the Priorities tab. For that matter, they never ever move out of the way of landslides either...
While the AI in Sword of the Stars tends to be fairly good, it does do glaringly bonehead things. For example, the game features several types of shields that absorb all damage of a given type and never going down. Deflectors block all ballistic and missile weapons, while Disruptors block all energy weapons. An AI-designed ship may only have energy weapons equipped but it will continue to pount your Disruptor-equipped ship from the front (the only area covered by the shield) and not even try to flank it. The only way to bring down these shields is with special Shield Breaker ballistic rounds, which the computer almost never uses due to their low damage.
The AI also doesn't quite know what to do about enemy cloaked ships if it doesn't have any Deep Scan-equipped ships in battle. While it's smarter than some other games in that it tries to fire at the spot where the now-cloaked ship was last seen (which means, if you cloak, you better move) and if the ship is able to fire while cloaked (meaning the very rare Advanced Cloak tech), it will target the area the fire is coming from. This may work fine with one enemy ship, but multiple ships have a tendency for one of the ships to move right into the spot where the cloaked ship was and stay there... while its allies are pounding the area with every weapon in their arsenal. Once the ship is destroyed, another one will likely take its place. Thus, your cloaked destroyer can be responsible for the destruction of several AI dreadnoughts.
Endless Space AI had some very easily exploitable flaws, most of which have been fixed in later patches. Probably the biggest one was that the game originally lacked a "cooldown" period after the end of a war, which would frequently lead to AI declaring war on one turn, agreeing to make peace the next turn, then declaring war again. This could be exploited by making peace with the AI and trading a large amount of resources for a system, new technology or Dust. Next turn they would declare war again and cancel the trade agreement, giving you back the resources you gave them. At one point it was even possible to get the AI that only posessed a few systems to give them all up in exchange for resources, causing them to be removed from game (that got patched pretty quickly).
While the above mentioned strategy no longer works, it's still possible to exploit the AI's stupidity in some cases. If an Ai you are allied with or just have good standing towards is getting close to winning an economic victory, just offer some valuable resources in exhange for an amount of Dust per turn that is equal to their income per turn. They will usually happily accept, even if it prevents them from winning and is likely to make you reach the economic victory instead.
Scavengers in Battle Zone 1998 were infamous for running over the player when he gets out of his tank, causing an instant game-over. Units sometimes got confused inside bases or would take long meandering paths to distant objectives, and units often spend more time trying to dodge rapid-fire shots rather than shoot back. Some of the issues were alleviated when the game was updated and re-released as a Freeware Game by the head programmer in 2011.
In Battlezone II, the newly introduced treaded units often had issues dealing with the huge amount of inertia the vehicles possessed, which sometimes resulted in them going flying off cliffs into water, which damages treaded vehicles. Pathfinding issues from the first game were compounded by bases (particularly the ISDF's) becoming more densely packed, which meant that factories usually had to be build at the front of the base in order to prevent units from piling up against the wall of a power generator or gun tower.