Automoderated users, Autopatrolled users, Bureaucrats, Comment administrators, Confirmed users, Moderators, Rollbackers, Administrators
213,561
edits
Artificial Brilliance: Difference between revisions
took a first stab at sorting the sections while I was here to fix a link
m (update links) |
(took a first stab at sorting the sections while I was here to fix a link) |
||
| (7 intermediate revisions by 4 users not shown) | |||
Line 4:
Artificial Brilliance is, quite simply, the ability of the computer characters to make the player think "Hey, these guys are actually pretty smart!" It occurs when the A.I.'s freeform actions, based on real-time decisions, result in behavior that seems, frankly, brilliant. It can be anything from an enemy that manages to outwit and outmanuever the player on the player's own terms and by the player's own rules, to an NPC ally who manages to assist and even save the player in ways that are unexpectedly helpful and seemingly smart. Just being able to react quickly and enter button combinations flawlessly doesn't count, computers are [[Computers Are Fast|naturally good at that]]; in fact, toning down a computer player's inhuman speed can be a facet of Artificial Brilliance.
The polar opposite of [[Artificial Stupidity]], when the A.I. makes unbelievably bad decisions that make the player think it's a complete moron. That doesn't mean
Of course, it's a balancing act between an AI that is bad at the game and an AI that is too good at the game. The trick is allowing the AI to make human like mistakes while also allowing it to make human like brilliance. After all, in a first person shooter, the AI isn't really playing the game in the same way a human does. They don't actually have a mouse/keyboard to manipulate or have to watch a monitor. Thus it's an easy task to make an AI that always knows where you are and can hit you perfectly; it's not so easy to make an AI that can act like it doesn't know where you are and can act like it has reflexes. And as some of the examples below note (see the Half-Life example about greatest threat), sometimes being smart makes it dumb.
Line 11:
{{examples}}
== Game Shows ==▼
* Most recently,{{when}} IBM took another shot at human vs. computer, with Watson starring on ''[[Jeopardy!]]''. The machine can answer the questions almost as well as a good human player, and (more importantly) its [[Computers Are Fast|perfect timing on the buzzer]] gives it a nigh-unbeatable advantage. It beat the two highest-ranked human champions [[Curb Stomp Battle|by a three-to-one margin]].▼
** In doing so, however, Watson exposed the ''limits'' of his brilliance more than once, giving wrong answers that no human would guess.▼
*** In an instantly classic moment, a Final Jeopardy! question in the category of "U.S. Airports" wanted to know the city with two airports, one named after a World War II hero and one named after a World War II battle. The answer was Chicago, but Watson answered with "What is Toronto?", prompting much hilarity about how apparently Toronto had been annexed with nobody noticing. This was caused by a slight oversight in Watson's programming that intentionally undervalued the category name since the category doesn't always have much to do with finding the answer. Since the clue didn't mention a country, Watson didn't rule Canadian cities out of its answer. In addition, according to the developers it didn't really understand the question (it didn't get the connection between World War II and airports)... so it guessed.▼
*** In the "Literary APB" category, the clue wanted to know the person responsible for the murder of Severus Snape -- "he'd be easier to find if we could just name him!" Watson didn't know and didn't buzz in. His top three choices were Harry Potter, Albus Dumbledore, and Voldemort. Watson had very obviously figured out that the answer had something to do with the Harry Potter franchise, but didn't understand that the category was exclusively villains (being unable to make the link that APBs are only issued for criminals), and couldn't catch the allusion to He-Who-Must-Not-Be-Named.▼
** In [[Shocking Swerve]], Watson was defeated shortly thereafter... by a Congressman. [http://bits.blogs.nytimes.com/2011/03/07/watson-vs-humans-score-one-for-congress/ Rep. Holt (D-NJ)] to be precise.▼
==
* Computers have gotten really, really good at chess.
* They haven't gotten quite as good at [[Go]], but the average beginner will still lose most or all of their first 50 games against it.
==
* Older ''[[Yu-Gi-Oh!]]'' games relied on psychic AI and other gimmicks to provide any difficulty, as otherwise the AI was very poor. In ''Stardust Accelerator'' and ''Reverse of Arcadia", however, the AI actually takes time to consider the possibility of every possible move and the consequences in order to decide the best action. The AI will take note of things such as what cards will provide better advantage, whether they can subvert the Tribute process for certain cards and monsters, and which ways they can take control of the situation, to name a few.▼
=== 4X ===
* ''[[Galactic Civilizations]]'' is known for this. See especially [http://www.computerandvideogames.com/article.php?id=161570&site=pcg this] [[After Action Report]], which goes so far as to cause Artificial ''[[Fridge Brilliance]]''. In the mid-to-end-game, the player was up against an alliance of a warmonger, a diplomat and an average race. The warmonger could have crushed him easily, but didn't. In fact, he even used his massive empire's voting power to punch through a galactic motion for peace against all three other races. The reason for this uncharacteristic behaviour: If the player had been defeated, it would have instantly resulted in an Alliance Victory, meaning a victory primary for the diplomat race that engineered the alliance. The goal of the warmonger faction was to keep the player alive until it was strong enough to break from the alliance and conquer both of its former partners at once - and the diplomat even anticipated this betrayal and built up his forces explicitly to be strong against his own ally. At which point the player happened to play "[[Dark Horse Victory|lucky third party]]" by achieving a [[Ascend to a Higher Plane of Existence|Technological Victory]].▼
** And if [[Word of God]] is to be believed, this is another one of those things where the AI was never intended to do anything even close to what happened.▼
** Back in the first Galciv, one of the people who wrote the A.I. was repeatedly curbstomped by his own creation.▼
** It's telling that the "Normal" difficulty setting has the AI's economy working at reduced power, with the mechanically equal setting known as "Tough".▼
** Galciv's spiritual predecessor, ''[[Master of Orion]]'', also has some very clever AI. The [[Harder Than Hard|Impossible difficulty]] really earns its name.▼
* ''[[Sword of the Stars]]'' AI can recognise your weapon loadouts and adapt by using the appropriate counters.▼
** The battle AI leaves something to be desired, though. Enemy ships armed with only ballistic and missile weapons will keep hammering away at your deflectors, which cover only the front of your ship, even though they have no chance in hell at penetrating the shield, instead of flanking. Ditto for energy-armed enemies against disruptors. In fact, the battles pretty much boil down to "get close to the enemy and keep firing until either side is dead". Retreat is not an option, unless the ship in question is not a strict military vessel.▼
*** The AI will also never explicitly target your ships' turrets or specific sections. It'll fire at whatever's closest. This gives the player doing this an advantage.▼
▲* Older ''[[Yu-Gi-Oh!]]'' games relied on psychic AI and other gimmicks to provide any difficulty, as otherwise the AI was very poor. In ''Stardust Accelerator'' and ''Reverse of Arcadia
* For Wizard of the Coast's ''[[Magic: The Gathering]]'' video game Duel of the Planeswalkers the programmers had a very difficult task in programming an AI that could deal with all the rule-changing cards of the game itself (and there are possibly more of them than in any other cardgame because of Magics "Golden Rule", which states that the text of a card takes priority before the rules themselve). Seemingly they succeeded.
** This refers mostly to the ''second'' game by that name and publisher. While the first game's AI was a remarkable achievement for its time, there was no way to provide it with hints on how to play specific decks, rendering a number of monsters crippled in the single-player game (their difficulty was instead balanced though life handicapping and chance of [[AI Roulette]]). (It also tended to cast any playable spell as soon as possible, and assume you have no castable spells, a greater weakness given the longer games 5th Edition tended towards.)
* [[Poker Night At the Inventory]] uses an AI system that not only considers each character's hand, but also factors this in with each character's unique personality and abilities. Thus Tycho and the Heavy will play cautiously and strategically, Strong Bad will usually attempt an overconfident bluff, and Max will play erratically to reflect his short attention span and poor grasp on the rules of the game ([[The Computer Is a Cheating Bastard|and possibly use his future vision power to cheat]]. Yes, this game even applies that trope in a justified manner.)
=== Fighting Game ===
▲== Fighting Game ==
* Many fans of ''[[Super Smash Bros Brawl]]'' have argued that the level 9 AI is capable of learning from human players. Players often point to examples of a certain characters' AI altering their recovery strategy after a human player has repeatedly used this character. Though this has been proven false<ref>replays in Brawl are text files, not video files, containing a list of all actions the player took and all random outcomes, but no AI actions. The AI plays out a replay match the same way it would a live match. Therefore, the AI cannot change over time, or else the replay files with AI in them would desync over time. Also, more recently, the AI has been decoded and shown that no learning capability exists.</ref> [[Urban Legend of Zelda|but the rumor still persists.]]
** The stock AI can also be replaced with more effective AI. Among various projects, one Japanese hacker is working on an Ice Climber AI that never messes up its [[Game Breaker|chaingrab infinites]], something that is possible for
* A similar rumor exists for ''[[Samurai Shodown]] 5''. The AI on higher difficulties will alter tactics in response to the player.
* ''[[Sonic Battle]]'': If you just spam a special attack to beat an opponent and they respawn, they'll change the special moves they use- since each player is granted immunity to the type of special attack they aren't using to attack, spamming one type of attack will lead the opponent to swap specials to defend against yours. This also runs in reverse, with the computers changing their specials to get by your immunity.
Line 33 ⟶ 51:
=== First
* The first fan-made "bots" for ''[[Quake (series)|Quake]]'' were a major leap in terms of A.I., being able to imitate the movements and behavior of a player-like character (as opposed to the simplistic "move in a straight line and shoot" enemies of the time).
* ''[[Unreal]]'' was the first FPS game to use "bot-like" A.I. for its single player enemies, leading to the creation of the memorable Skaarj, a video game adversary that could dodge and strafe while firing, jump from platform to platform, intelligently pursue the player throughout the entire level, and otherwise move and behave just like the player character instead of (as was the norm at the time) a simple turret-like enemy wandering aimlessly in semi-random directions while periodically pausing to fire at the player's direction.
Line 43 ⟶ 61:
** Also, the Combine gunship, which were programmed to shoot at "the greatest threat". It exceeded the dev team's expectations by shooting down incoming missiles, rather than at the player.
*** Leading to the sad case of [[Artificial Stupidity]] in which they would never fire at the player as long as a (smarter AND faster) missile was in the air. Any number of gunships could easily be defeated at no risk as long as the player could keep a missile flying in circles until it could hit one from behind, then launch another or find cover immediately. Partially averted {{spoiler|on higher difficulty levels; when you're up against a duo of Gunships, each of which requires more than half-a-dozen hits to take down, and they keep shooting down your rockets, it gives the enemies on the ground ample opportunity to ruin your day.}}
* Raven Software is well known for pioneering the use of friendly NPC A.I. squads in FPS games, beginning in ''[[Star Trek Elite Force]]'' and continuing on in ''[[Soldier of Fortune]] 2'' and ''[[Quake 4|Quake IV]]''. Their games often feature friendly A.I. squads of several [[NPC
** The friendly A.I.'s ability to follow the player in most Raven games without getting lost or stuck is usually due to the use of strictly linear [[No Sidepaths, No Exploration, No Freedom]] levels, but it's still impressive given the time it was made.
* ''[[First Encounter Assault Recon|F.E.A.R.]]'' was widely praised for the A.I. of its Replica Soldiers, touted as the best enemy A.I. ever at the time of its release. Enemies would work in squads, provide covering fire for each other, advance under covering fire, fire from behind cover intelligently, and even use cover to flank and circle around the player instead of charging straight for him and making themselves an open target in the process. The A.I. also had extremely high mobility allowing it to exploit the entire battlefield instead of limiting itself to a single small area, and could also dive through windows, vault over objects, and crawl under obstacles, allowing it to access every area of the level the player could.
** The ''F.E.A.R.'' A.I. is actually a good example of emergent behavior, which is programmer-speak for "we didn't program it specifically to do that, but for some random reason it does it anyway, and it's really, really cool that it does!". More accurately, complex behavior stemming from simple rules. Specifically, the A.I. is programmed for a limited number of simple behaviors: moving in coordinated squads, providing covering fire, seeking cover, and repositioning itself based on the player's movement and position. The A.I. isn't actually programmed to flank or circle behind the player, but its tendency to seek cover and reposition itself based on the player's movements results in flanking and circling behaviors occurring naturally without "conscious" effort on the A.I.'s part (mostly due to the A.I.'s high mobility combined with its preference for seeking lateral cover rather than charging the player directly).
** Also of note is that of the level design. The levels were designed to help facilitate the AI in action. You will find that there's generally two ways to get to any position thus always allowing the AI to find a path. Of course, this also gives the player the impression that levels are bigger than they are so it was a win-win.
* ''[[STALKER]]: Shadow of Chernobyl'' features a complex Artificial Life ecosystem featuring both wildlife and human [[NPC
* The enemy AI for the Legendary difficulty level in the ''[[Halo]]'' games is notoriously brutal, but still holds back in order to provide a reasonably playable game. One of the easter-egg Skulls in ''[[Halo]] 2'' (named "Whuppopotamus" in the game files) corrects this, allowing enemies to discern the wibbly outline of a cloaked foe, hear guns reloading, and generally behave as though they were capable human players. Meep.
** Exacerbated by the [[Artificial Stupidity]] of the friendly AI.
Line 57 ⟶ 75:
* ''[[Turok (series)|Turok 2: Seeds of Evil]]'' had the Endtrail enemy, which use cover whenever possible, move around to make itself harder to target, chase you a bit and use hit & run tactics. The other enemies of the game weren't so bright, however.
* The AI for the offline mode for ''[[Team Fortress 2]]'' has its periods of [[Artificial Stupidity|fail]] from time to time, but at least it knows that when on defense, spam [[Macross Missile Massacre|Soldiers]] and [[Stuff Blowing Up|Demomen]].
** The Bot AI has gotten remarkably better in recent months. Engineers will check suspicious behavior around their buildings, and snipers will now be smart enough to back off while firing their SMG instead of charging enemies. If spies are spotted, expect a thorough round of spychecking from the bots. Medics will also wait until their patient is incurring damage to engage an ubercharge, and scouts will know to use their pistols to take out sentry guns outside their effective range. Bot pyros are also quite good with using the airblast secondary attack on their flamethrowers to clear points and reflect projectiles.
* ''[[Star Wars]]: [[Republic Commando]]'' was notable for having exceedingly good party AI. If you told one to take up a sniping position, they ''would'' get there unless there was a positively absurd amount of stuff in their way and no intermediate cover, and would last quite long in such a position without assistance. If you got shot down (which at times happened frequently), your team would either cover each other as they attempted to heal you, or set up a proper defense and wait out the threat depending on your orders.
* The AI was massively improved in ''[[Halo: Reach]]''. Apparently, they were programmed to behave like real players. When a grenade falls near them, they will engage Armor Lock if they have it. They'll take cover, dive away from bullets, and basically play like you would.
* The Stalker from ''[[Dead Space 2]]'' is designed to behave similarly to the velociraptors from Jurassic Park, meaning they will attempt to flank the player while another peeks around cover within your field of view in order to distract you. They can be [[Demonic Spiders|very effective]].
* The first time you play [[
* ''[[Crysis (series)|Crysis]]'' and ''Crysis 2'' may have their occasional hiccups, but the enemy AI in both games shows impressive levels of intelligence. Enemies will suppress you and flank your position, call in re-enforcements and sweep the area for you if you try to hide. In Crysis 2, a highly alert enemy will notice a cloaked player if they are close enough (though by then, it is typically too late) and they will throw grenades at your last known position, or where they think you may have gone. Enemy troops will even track what direction you're going in; if you run toward cover while the enemy can see you, cloak, and then change direction, the enemy will continue firing along your original path and try to suppress the general area it thinks you were running toward.
* ''[[Command
=== Flight Simulation Game ===
* While [[Ace Combat]]'s AI isn't known to be the smartest, more recent games have demonstrated better AI abilities, such as better response from teammates to orders (Pixy (''ACZ''), for example, behaves much better than the rest of Wardog Sq (''[[AC 5]]''). combined, to say nothing of Shamrock (''[[AC 6]]'')). A very good example of better enemy AI is in ''Zero'', where [[Ace Pilot]] squadrons that come in several missions will constantly flank you and keep your wingman busy while they move in for the kill. They move so well-coordinated that one reviewer notes that "They put the [[That One Boss|Yellow Squadron]] (of Ace Combat 04) to shame."
** Notably, the Ace squadrons almost always outnumber you 2 to 1, and they use their numbers to their advantage. One plane might act as bait while another tries to shoot you down, and in particularly annoying furballs, ''three'' planes will come after you while the fourth keeps your wingman busy. The one squadron that doesn't outnumber you (Gelb, which only has two planes, as do you) is in vastly superior planes to what you're normally flying at that point in the game.
===
▲* ''[[Galactic Civilizations]]'' is known for this. See especially [http://www.computerandvideogames.com/article.php?id=161570&site=pcg this] [[After Action Report]], which goes so far as to cause Artificial ''[[Fridge Brilliance]]''. In the mid-to-end-game, the player was up against an alliance of a warmonger, a diplomat and an average race. The warmonger could have crushed him easily, but didn't. In fact, he even used his massive empire's voting power to punch through a galactic motion for peace against all three other races. The reason for this uncharacteristic behaviour: If the player had been defeated, it would have instantly resulted in an Alliance Victory, meaning a victory primary for the diplomat race that engineered the alliance. The goal of the warmonger faction was to keep the player alive until it was strong enough to break from the alliance and conquer both of its former partners at once - and the diplomat even anticipated this betrayal and built up his forces explicitly to be strong against his own ally. At which point the player happened to play "[[Dark Horse Victory|lucky third party]]" by achieving a [[Ascend to a Higher Plane of Existence|Technological Victory]].
▲** And if [[Word of God]] is to be believed, this is another one of those things where the AI was never intended to do anything even close to what happened.
▲** Back in the first Galciv, one of the people who wrote the A.I. was repeatedly curbstomped by his own creation.
▲** It's telling that the "Normal" difficulty setting has the AI's economy working at reduced power, with the mechanically equal setting known as "Tough".
▲** Galciv's spiritual predecessor, ''[[Master of Orion]]'', also has some very clever AI. The [[Harder Than Hard|Impossible difficulty]] really earns its name.
▲* ''[[Sword of the Stars]]'' AI can recognise your weapon loadouts and adapt by using the appropriate counters.
▲** The battle AI leaves something to be desired, though. Enemy ships armed with only ballistic and missile weapons will keep hammering away at your deflectors, which cover only the front of your ship, even though they have no chance in hell at penetrating the shield, instead of flanking. Ditto for energy-armed enemies against disruptors. In fact, the battles pretty much boil down to "get close to the enemy and keep firing until either side is dead". Retreat is not an option, unless the ship in question is not a strict military vessel.
▲*** The AI will also never explicitly target your ships' turrets or specific sections. It'll fire at whatever's closest. This gives the player doing this an advantage.
▲== Game Shows ==
▲* Most recently, IBM took another shot at human vs. computer, with Watson starring on ''[[Jeopardy!]]''. The machine can answer the questions almost as well as a good human player, and (more importantly) its [[Computers Are Fast|perfect timing on the buzzer]] gives it a nigh-unbeatable advantage. It beat the two highest-ranked human champions [[Curb Stomp Battle|by a three-to-one margin]].
▲** In doing so, however, Watson exposed the ''limits'' of his brilliance more than once, giving wrong answers that no human would guess.
▲*** In an instantly classic moment, a Final Jeopardy! question in the category of "U.S. Airports" wanted to know the city with two airports, one named after a World War II hero and one named after a World War II battle. The answer was Chicago, but Watson answered with "What is Toronto?", prompting much hilarity about how apparently Toronto had been annexed with nobody noticing. This was caused by a slight oversight in Watson's programming that intentionally undervalued the category name since the category doesn't always have much to do with finding the answer. Since the clue didn't mention a country, Watson didn't rule Canadian cities out of its answer. In addition, according to the developers it didn't really understand the question (it didn't get the connection between World War II and airports)... so it guessed.
▲*** In the "Literary APB" category, the clue wanted to know the person responsible for the murder of Severus Snape -- "he'd be easier to find if we could just name him!" Watson didn't know and didn't buzz in. His top three choices were Harry Potter, Albus Dumbledore, and Voldemort. Watson had very obviously figured out that the answer had something to do with the Harry Potter franchise, but didn't understand that the category was exclusively villains (being unable to make the link that APBs are only issued for criminals), and couldn't catch the allusion to He-Who-Must-Not-Be-Named.
▲** In [[Shocking Swerve]], Watson was defeated shortly thereafter... by a Congressman. [http://bits.blogs.nytimes.com/2011/03/07/watson-vs-humans-score-one-for-congress/ Rep. Holt (D-NJ)] to be precise.
== General ==▼
* The "It's Thinking" advertising for the Dreamcast was all about this (in theory). For a football game the ad would say e.g. "It knows you like to go for it when it's 4th and goal. It's thinking".▼
* A non-video game example: In the climax of ''The Two Towers'', the enemy Orcs were programed to make the best tactical decisions possible against the tree-folk. Apparently, they figured out that '''running away''' was the best thing to do, and so the programmers had to keep dumbing them down until they stood their ground and got slaughtered like good little bad guys.▼
* Some web apps are getting very clever at interpretation. This troper once made a Facebook update consiting solely of the phrase "Gorram Reavers", and FB grouped it in with a bunch of friends' comments talking about [[Firefly]].▼
* Google for something like [[Memento|"that film that runs backwards"]] or [[Children of Men|"that film with no babies"]].▼
▲== Maze Game ==
* ''[[Pac-Man]]'' is perhaps the earliest example of "smart" A.I. in a video game, as each of the four ghosts was programmed with a specific "personality" and navigating/tracking style that allowed them to make decisions on how to pursue the player character, rather than simply moving at random or in a linear line. The aggressive Blinky was programmed to chase Pac-Man, the cunning Pinky was programmed to corner Pac-Man (by heading toward a spot a few steps ahead of him), the wired Inky was programmed to follow Blinky and try to aid, and the scared-witted Clyde was programmed to run away when he got close to Pac-Man. Ms. Pac-Man made the ghosts move randomly for the first 7 seconds so that players couldn't learn and re-use an optimal path. More [http://www.atariage.com/forums/topic/68707-pac-man-ghost-ai-question/page__st__25?s=4746e308fcf60d785d27114d52b5bbb4 here].
=== Mecha Game ===
* AI behavior in ''[[Armored Core]]'' isn't the best, granted, but in Silent Line, you are able to train an AI pilot of your own. The AI starts out very stupidly, but as you pilot the AI's assigned mech, the AI starts emulating your combat behavior, both good and bad. With enough practice, the AI would even be able to accurately aim rockets, something that even the best human players have difficulty to do. The AI opponents in the Arena also improve with repeated combat exposure.
=== Racing Games ===
* In ''[[Mario Kart]] DS'', the AI actually seems to know that if you put a banana peel on the loop-the-loop, [[Hilarity Ensues]].
* ''[[Mario Kart]] 7'' upgrades the AI's tactics with items. Now they will usually drag items behind them for a while (shells, banana peels, etc.) before using them so that players can't make them crash with their own items too easily. Should you get in the line of sight of an AI racer? They won't hesitate to use their items on you.
=== MMORPGs ===
* ''[[World of Warcraft]]'' generally has fairly limited AI opponents with only a handful of abilities. The encounter commonly referred to as Faction Champions in the Argent Coliseum, however, is quite impressive - it's meant to mimic a huge [[PvP]] battle, as each enemy has access to most of the spells you'd expect a player of the same class and specialization to use, the healers are quick to remove status effects and heal injured comrades, and the group will often gang up on one player and then switch to a completely different one.
** And [[Older Than They Think|before the Argent Coliseum]], we had [[That One Boss|Princess Delrissa's party]] in Magister's Terrace, and the Tier 0.5 arena battle in Blackrock Depths.
* ''[[Guild Wars]]'' still does have some obvious flaws with the AI, namely that they'll attack enemies rather than trying to run away from them, and that the heroes often need you to tell them how to use the more complex builds, but the specs they ''do'' have, they ''know'' how to use. It also helps that if you ping a target, they immediately focus fire whereas all that does to players is say "Attack here".
** Still even some individual behaviours are nice - Herta uses the Ebon Hawk-Stoning combo, the interrupt/domination henchmen are actually ''really'' good at interrupting (helps they have godlike reflexes), [[Minion Master]] heroes will actually ''heal'' their minions if you have them use something like Karei's Healing Circle or Heal Area, etc.
* ''[[
** A refined version of Sleeper AI package is currently being used by the Sansha's Nation incursion fleets, and is introduced in the eponymous "Incursion" expansion in Jan. 2011. On top of the features provided by the Sleeper AI, the Sansha's incursion ships are equipped with specialized behavioral rules base on ship classes and roles (stealth bombers targeting larger ships for torpedoes, electronic warfare ships pinning down smaller support ships, battleships focusing fire on single targets, etc). As many regular combat mission runners severely underestimated the competence of Sansha's incursion fleet and ran into them unprepared, the galaxy map with the "Ships Destroyed within 24 Hours" filter lit up like a christmas tree for more than a month, making the developers, and especially the in-house economist, ''very, very'' happy.
=== Platformers ===
* [[Rock Man 4 Minus Infinity]]
** If you use Rush during the Snatchman battle, the doppelganger will ''use it against you.'' Yes, you can end up healing the ''boss'' when you're low on health! To ''really'' rub it in, it ''still uses your weapon energy.''
** More brilliance: all the Robot Masters change their patterns to dodge Rush Cannon if you fire it at them. Some even [[Hoist by His Own Petard|turn it against you!]]
=== Programming Game ===
* ''Battlecode'' is a competitive game for programmers in which they must program their own AI for their robots. When the game actually begins, players have no control over the robots; they act based on the player programmed AI.
=== Puzzle Game ===
* One of the most notorious puzzles in ''[[The
* This is the reason [[Puzzle Quest]] has such a notorious reputation for cheating. The computer doesn't make mistakes. It doesn't miss 4 in a rows, and knows how to set them up for optimal follow-ups. Combined with observation bias and some less than perfect playing on the human's part, and it seems to be cheating.
=== Real Time Strategy ===
* ''[[AI War Fleet Command|AI War: Fleet Command]]'', a seemingly innocuous indie 2D space RTS made by a single person. But that doesn't stop the AI from actually understanding flanking tactics, creating distractions for the human player, utilizing hit and run warfare...
* Despite how stupid the AI Bots are in ''[[League of Legends]]'', there's some
** Ryze, Annie, and Trundle bot go positive. Trundle is able to put his obstacle in the ''absolute WORST'' location possible, whereas players will often mess up. Ryze Bot will often harass you with Rune Prison (sometimes ''right next to a turret''), and Annie bot is well aware that her passive gives every fifth spell a stun. You'll notice that she'll sometimes use her Disintegrate to last-hit minions...however there is a visual warning that says her stun is ready. Immediately, she stops last-hitting minions with Disintegrate and starts to go ''right'' for you. She even knows the oldest trick in the book when it comes to Annie - saving her stun and using it when she summons Tibbers.
* The AI in the original ''[[Supreme Commander]]'' and its expansion Forged Alliance is far from perfect, and in fact often actively cripples its own chances of success, but higher levels are excellent at early-game harassment and have nearly perfect base layout and economy management. The true brilliance, however, is that the Adaptive AI will occasionally emulate player tactics it has observed to be effective, and will use them against the player.
=== Roguelike ===
* This is the point of the short roguelike ''Smart Kobold''. Your character's a nigh-indestructible melee-fighting death machine (i.e. a typical late-game player character) with no ranged attacks, so the second you set foot in their caves, they grab all their weaponry, valuables, and babies, and start running (setting traps as they go) into a large room from which they can easily snipe you to death. If you somehow manage to corner one of their mages or archers, there's a good chance he'll snap his wand or bow in half to deprive you of the ranged attack you'll need to win. Furthermore, even when you ''do'' get a ranged attack, they'll do their best to stay at maximum range - or in the "blind spot" between the eight directions you can shoot in, like [[Nethack]]'s unicorns.
=== Role
* In ''[[Crisis Core]]: [[Final Fantasy VII]]'' you will find yourself frustrated by enemies who chain their attacks such that there's almost no time to dodge, or that rolling to dodge one attack leads to Zack entering recovery frames in time for another to land.
* This is why your party members are useful in ''[[Final Fantasy XIII]]''. Synergists remove debuffs and prioritize buffs that are more effective against whatever you're fighting. Ravagers focus on whatever a Commando's doing and try to help him/her out, focusing chain attacks or spamming area-of-effect spells. Saboteurs focus on [[Standard Status Effects|debuffs]] the enemy is weak to, [[Status Buff Dispel|remove]] enemy [[Status Buffs]] efficiently, and prioritize based on what everybody else is doing. Medics remove debuffs quickly and select spells based on how injuries are distributed. Synergists prioritize the most helpful buffs for the situation. Sentinels are really effective at drawing and surviving enemy attacks.
Line 157 ⟶ 149:
** Not only that, they use smart responses to your moves. For example, they'll stop using attack moves if you start using Bide.
* The final boss of ''[[Lunar Silver Star Story Complete|Lunar Silver Star Harmony]]'' is a ''nightmare'' because of this. Good luck keeping Jessica alive, because he knows to [[Shoot the Medic First]], for one.
* ''[[Golden Sun: Dark Dawn]]'' shows a case of this with the [[That One Boss|Ancient Devil's]] [[That One Attack|Demon Sign]]. He is capable of judging which of your members is most damaging to him either by pure damage, healing potential or some combination thereof, and [[Brainwashed and Crazy|control]] that character, [[Non-Lethal KO|provided they're on the front line and conscious]]. Also, said character, once controlled, is highly prone to [[This Is Gonna Suck|using the party's standby Djinn for summons against you]]. His number-one target for Demon Sign is [[Game Breaker|Sveta]], of course (don't use Beast
* ''[[Xenoblade]]'' has a battle system that relies on well-timed, structured combos of special abilities, and often relies on two or even all three party members using skills in concert. The player can only control one at a time, but fortunately, your AI partners are smart. Each has their own AI, uses their skills at the best time and position they can, and tries to fulfill their proper role: for instance, your tank will switch targets to draw aggro away from other characters, while squishier characters will stop using skills for a few seconds to let that happen.
* The first four games of [[The Elder Scrolls]] had AI that was at best laughable, since they would just dead-zone you and try meleeing (if you could tell they're doing that; sometimes the sprites wouldn't show them attacking) or use up all their magicka in the first ten seconds and then stand in front of you swinging their weapons, while saying they fought mudcrabs better than you or calling you an n'wah. However, ''[[Skyrim]]'' features much better AI. Enemies might flee into another room to get help to fight you, archers will switch to a dagger when they're in melee range of you, and enemy casters are ''annoying'' because they'll use frost spells on you to reduce your stamina and deny power-attacks And mages, when fighting other mages, will spam lightning on you to drain your magicka. Oh, and some of the higher-level Draugrs, who're armed with Shouts, will disarm you and force you to pick up your weapon while they get free shots. It's most prevalent to archer-player characters. The enemies know it's harder to hit a moving target, and will sometimes notice you're aming at them and will strafe, sometimes they even wait until after you fired to sidestep and force you to miss. They also know the game averts the [[No Arc in Archery]] and will often shoot arrows from seemingly impossible angles where an archer trying to counter-attack would miss due to an object in the way.
** Players have also reported dragons use PC-tactics against ''players''. They're well aware that sometimes, people will sometimes [[Gang Up on the Human]] and land in a spot where you're forced to go into the aggro radius of other monsters who will then weaken you while the dragon breathes on you. Another observed tactic is that a dragon may attack a monster and then fly around the player, causing the creature to follow, approach the player, and attack him/her.
** NPCs that can cast magic on themselves do so in appropriate contexts. Powerful vampires, for example, are quick to cast invisibility on themselves before running away.
Line 167 ⟶ 159:
*** Note: Little girls can hire bandits as well. So yes, a 10 year old can hire criminals to kill you.
** Hostile archers will constantly move back and forth if you're armed with spells or bows. They won't move at all if you're running toward them to attack in melee.
* ''[[Mass Effect 2]]'' has pretty good AI, especially on Insanity level. Enemies "leapfrog" each other, use powers to flush you out, and flank you.
** ''[[Mass Effect 2]]'' has got NOTHING on ''[[Mass Effect 3]]''. From the demo alone, the game has displayed ridiculously impressive AI who use their powers and abilities and strength in numbers to overwhelm and flank you at an incredible rate. Its got probably the best tactical AI since ''[[F.E.A.R.]]'' or ''[[Crysis (series)|Crysis]]''.
*** See [[Mass Effect 3|the trope page]] for a bulleted list of brilliant tactics, exhibited by Cerberus units alone.
* Indie RPG ''[[Fortune Summoners]]'' features extremely good friendly and enemy AI. Pretty much every enemy in the game up to and including ''the very first slime you encounter'' will bob and weave around your strokes and teach you very early on that mashing attack is very much not an option in this game. On the other hand, the meek little [[Staff Chick]] you pick up early on will lead you on with her easily-terrified demeanor, then proceed to effortlessly dance right through a cloud of enemies while keeping you healed.
** Heaven forbid you try Vs. mode. One could be forgiven for thinking ''[http://www.youtube.com/watch?v=2SgEdA_fpzc your own main character]'' got hijacked by [[That One Boss/Other Games/Atlus|Atlus]] or something.
=== Simulation Game ===
* ''[[Creatures]]'' was notable for its use of a genetically coded Artificial Life system, allowing you to breed creatures called Norns, mixing their DNA and resulting in evolving behaviors unpredictable to the original programmers.
** Unfortunately, there was [[
*** And [http://creatures.wikia.com/wiki/Wall-bonk Wallbonking].
** Also, [http://creatures.wikia.com/wiki/Socrates Socrates], a norn genetically engineered to be incapable of learning from experience, actually did better than the normal norns. Their ability to learn isn't as good as their instincts.
Line 186 ⟶ 178:
=== Sports Game ===
* A glitch in the NES ''Lunar Ball'' game was recently found and utilized in a Tool Assisted Superplay. The entire TAS in question was done not by a human but by a bot that was specifically designed to find the most frame-efficient way to finish each board using what was known about the game, taking into account that deliberately performing a shot that pockets no balls reduces the number of frames spent waiting for the bonus to be added to the score. [[Crowning Moment of Awesome|The bot stumbled across the glitch by mere luck and decided "hey, this works better".]] [http://tasvideos.org/1565M.html This is the TAS in question].
* Almost paradoxically, happens in Pro Evolution Soccer a lot on the easier difficulty settings: not only will the AI let the player take his time while practicing basic skills and passing, it will also purposely decide to defend or attack according to how well the player is learning. It's most notable on a beginner's first run, as any player who has mastered these skills will obviously notice all loopholes.
=== Stealth Based Game ===
* ''[[Thief]]'' is notable for implementing a complex sensory system for its A.I. characters, allowing guards to be aware of and respond to environmental factors such as noise, lighting, movement, and shadows. While [[Zero Punctuation|Yahtzee]] was playing it, he accidentally jumped out and then leapt back in while a guard was watching him. Instead of the guard just moving on, he yelled "Don't think you can just jump back into the shadows, boy!" which, Yahtzee said, "Surprised both myself and [[Potty Failure|my dry cleaner."]]
* Every [[Metal Gear Solid]] game after (and including) 2 has had some MASSIVELY improved AI. When walking on metal floors, an enemy will hear it if a) they are close enough or b) if it is loud enough. If a soldier is downed and another person sees the bloodspot or finds the corpse, they will automatically kick it up to maximum security and a new guard will be sent to the area. If a guard is knocked out, they will kick it up to maximum security. If all guards have been killed, the commander will notice after a while and send a heavy-duty team to investigate. During Alert Mode, if there are blood trails on the floor leading into lockers and/or closets, the enemies will either throw a grenade into that area, or have each soldier comb each nook and crannie of the area and search the lockers (not to mention that they can HEAR YOUR HEARTBEATS and your breathing if you stand too close to the locker door!). If you are hiding behind cover, some guards will provide suppressing fire while some others will flank around you. Enemies also can have the uncanny ability to spot discolorations in the atmosphere or a floating gun if you are using stealth camo.
Line 199 ⟶ 191:
=== Turn
* The enemy AI in ''[[Final Fantasy Tactics]]'' can be quite good at times. Its main flaws are that it usually doesn't have good skills or equipment to work with, and it can't plan its actions multiple turns in advance, so its brilliance is limited to what the best action within the current AT list is (including spells/abilities currently charging). Fixing these is the purpose of one of the more notable [[Game Mod
** To give examples, the AI properly understands elemental absorption for healing its allies, attacking MP (or restoring it) to affect charging spells, and even ''bouncing spells off of Reflect to increase range''. That last one is a very rare situation even for human teams. It also performs extremely esoteric actions such as placing a doomed unit (e.g. a fatal spell is charging on it that it can't escape) in front of a dead ally Mime solely because if the player uses a Phoenix Down to revive their Mime, their Mime will mimic the Phoenix Down and resurrect the enemy unit.
** An unprepared player may be caught off guard as the AI even knows how to exploit the game's projectile rules. For example, range weapon-equipped characters can shoot enemies that are too close to be targeted by shooting at enemies standing behind their intended targets; this can even be done with longbows, but requires a major elevation difference between the attacker and intended target! The computer will occasionally mess up these attempts though- not surprising considering how hard they can be to predict.
* Its spiritual predecessor, ''[[Tactics Ogre]]'', also has the same. You'll ''hate'' that they averted the [[No Arc in Archery]] trope when you're fighting an uphill battle, and love it when you realize you can do the same. The PSP remake also takes this up to eleven, where the enemies know to focus fire on low-health targets, targets that will take a lot of damage such as low-levels and squishies, and to [[Shoot the Medic First]]. You can still exploit this by sending someone with no equipment out to act as a decoy. They also will send knights with rampant aura (Which prevents you from moving further than them) to body-block you and just stand there with Phalanx on (reduces damage by 90%).
* The AI in ''[[Fire Emblem Jugdral|Fire Emblem: Thracia 776]]'' is dangerously smart. For instance, if you steal a mook's only weapon, he will run to the nearest weapon shop (if there's any that sells weapons he can use), and will actually ''buy the best weapon he can use and attack you with it next turn''.
** Enemy flying units will prefer to attack your ground troops from whatever range that denies you from counterattacking(if possible) and then use their remaining move spaces to fly away into terrain where you cannot retaliate, [[Spiteful AI|effectively denying you the chance to finish them off]].
*** Similarly in [[Fire Emblem Elibe]] and ''[[Fire Emblem:
* In 1981, and then again in 1982, Douglas Lenat [http://www.newyorker.com/reporting/2009/05/11/090511fa_fact_gladwell?currentPage=all tested his learning program], [[wikipedia:Eurisko|Eurisko]], in a [[Traveller]]: Trillion Credit Squadron tournament. Eurisko [[Level Grinding|simulated thousands of battles]], [[Loophole Abuse|found unconventional ship configurations and methods]], and defeated all comers. Twice. In a row. Even with notable rule changes.
** Eurisko could have done it a third time, but Lenat decided to retire it from the tournament, since if the program had won a 3rd time, it would be the last such tournament.
=== Wide Open Sandbox ===
* [[Bully]] Has some. Prefects don't only chase you if you break a rule, they chase other kids who break rules.
* ''[[Dead Rising]]'' was famous for having [[Artificial Stupidity|incredibly stupid]] survivor AI. In ''[[Dead Rising 2]]'', they're much, ''much'' better. They follow you intelligently and are very good with weapons. If a zombie jumps onto you, they even [[Big Damn Heroes|attack the zombie in order to save you]]!
* The Rapid Response military ships in ''[[X (video game)|X3 Albion Prelude]]'' are some of the smartest enemies in the game. When they detect one of their race's stations are under attack, they'll jump their ships in and curb-stomp the enemy with [[Macross Missile Massacre|heavy missile spam]] fired from 80 kilometers away. When one of their capital ships starts to take heavy damage, it'll use its jump drive to jump to safety and go repair.
▲* The "It's Thinking" advertising for the Dreamcast was all about this (in theory). For a football game the ad would say e.g. "It knows you like to go for it when it's 4th and goal. It's thinking".
▲* A non-video game example: In the climax of ''The Two Towers'', the enemy Orcs were programed to make the best tactical decisions possible against the tree-folk. Apparently, they figured out that '''running away''' was the best thing to do, and so the programmers had to keep dumbing them down until they stood their ground and got slaughtered like good little bad guys.
▲* Some web apps are getting very clever at interpretation. This troper once made a Facebook update
▲* Google for something like [[Memento|"that film that runs backwards"]] or [[Children of Men|"that film with no babies"]].{{context}}
{{reflist}}
| |||