Game AI - Funtelligence - Extra Credits 

Extra Credits Why is there 12 comments i only see 3(incluoding my own in both numbers)

Extra Credits Please make a video about PMCs ( Private Military Contractors)

Give us the instructions how to make microwave burrito please :3

The RandomQueso that's a great idea

Extra Credits I Would like to see a Game in the Future that uses real, Self-learning A.I. Imagine For example a Game like Battlefield where a "Black Knight" A.I. Joins every few Games and changes the way the Players have to behave in that Match by simply being to Strong For one player to Deal with Or a new Game Mode: the player-Horde against the A.I. Ninja A good example is the bossfight Mode from Robocraft, where one player can Build an Robot with 10 Times the strength of the Enemy players

Thank you! We are all incredibly excited about it! The news was a welcome morale boost. (For reference: time.com/4890205/best-youtube-channels)

they were no1 on their "15 youtube channels you should watch" list

link, please?

the competition isnt that hard when fucking jenna marbles is in fourth place. Still, congratulations. I just thought a show of this quality would deserve a higher award than just this.

Yeah, I thought a lot of Time's choices were questionable too. There's a ton of channels (like Game Theorists) that should've been mentioned before vlog crap

Why not both

An AI that is super good will frustrate the player, leading to him, eventually (be either on purpose or not), to find ways to put the AI in a situation where it was not mean to be, in turn, leaving the AI to look as it was poor, that of course, if the player doesn't leave the game first. In the end, the player will remember the encounter, not because it was memorable but because it was bad; as such, you want to challenge the player, entertain him, not show him who is best.

+Ghost Tactician Every "I wanna be the X" game ever.

Unless your Elon Musk, then the job of the ai is to destroy the best!

Not really, even if your NPC is unbeatable by lore, it should not affect the way the encounter is played off, if you are forcing the player to fight you should still provide a memorable encounter, take the first boss of Megaman X or Metal Gear Rising, both of them are "unbeatable" and yet you still have a win condition, and both battles serve a purpose, but if you're just going to dust off the player on the first hit you may as well do it in a cutscene, otherwise the player may feel cheated. Also by entertaining, I don't mean easy, I mean it has to provide fun to the player, it can be simple or complex, easy or hard, but not just plain unfair. If your NPC is supposed to be the best of the best of the best, of course, it makes sense to make him challenging, and that's the thing, there's a clear difference between testing the skill of a player and abusing your systems to make a battle artificially hard. As an example, the last boss of AceCombat:AH is supposed to be a very skilled pilot, so skilled in fact that it will teleport to the back of the player most of the time, making the battle artificially hard and unfair regardless of your skill level.

But sometimes you assasinate 2 drunks and the third just walks right past not caring at sll

LoTR Builders as long as you don't offer them black braier mead they will like you:D

But are they really... truly... IMMERSIVE?

I usually assassinate someone, preferrably a bandit, and when another bandit stops and notices the corpse, I kill that bandit, too. :)

Not as ... IMERSIVE as this booty mod there are many like it but this one's the new one

Derpaboop Derp "predict"

Don't forget, an AI can react in 1 frame, probably 0 frames. It's more likely they're sometimes told to wait for you to do something and respond instantly as though they had predicted it.

Mortal Kombat 2 is notorious for this type of AI

What if... *YOU* Are the AI‽ (Duhn-duhn-DUHN!)

I hate the Smash 4 AI exactly for that. All those frame-perfect reactions actually kill the fun. Mind-game strategies like baiting are of no use, all you can do is go for safe moves, punish their moves, or hope they decide not to dodge your attacks.

Isn't that why the Extra Credits to-do stack is a mile tall?

Timothy McLean Its more like a concept than a number. Because its infinite.

Somewhere James is building a Taj Mahal out of individual pages for script ideas for episodes he wants to do.

This need more UPVOTES! XD

It looks like there are a lot of ppl who'd like to see less mourderous enemies and less friendly friends. A companion leaving the party becouse you didn't engage with them? A regular trooper joining your epic quest out of desperation? I'd love to see that in a big production one day.

Yeah, more chance of non-player characters changing opinions and even sides, expectantly or not as expectantly.

One place where i think machine learning would excel is rougelikes, because the ai would be different for every game too, whle also consistnelty making each run much interesting

Machine learning has a few more specific uses, one is to learn the players patterns so it can predict and become harder, so both the player and the ai become better with time, making the game always challenging. The second one is so the player cant exploit anything in game, haveing an ai assess the players spam choices to see if there is an issue and stop the them from doing it so the players are forced to always find different and creative ways to play the game. I'd say those are the main uses for it

I can say for sure machine learning is not good for all game. If not very carefully crafted, a machine learning AI could become impossible to beat, which totally counter the purpose of an AI. There are here to be beaten. Also, most game have no use of machine learning just because, as said in the video, sometime a simple AI is more than enough. Take a racing game for exemple, an AI in a racing game need to follow the optimal trajectory, and go just fast enough to provide a challenge to the player. Yes you could use machine learning so the AI could learn the optimal trajectory, or you could just program the trajectory directly, which will take less time and ressources.

Also, its decided whether to use mechine learning or not depending on the mechanics in the game, depending on how you designd a game it can become harmful to the gameplay, i dont quite remember the name of the game but there was one that you played as a god and you interacted with the world through a chosen avatar, that you commanded and slowly it learned on how to automatically act when an event ocurred, but towards the middle of the game there were so many commands and changes on player interaction that the ai got confused and started to ruin the game. It is a very dangerous to choose to have machine learning in a game, specially because it takes too long to see how it turns out that when its finished, you dont have time to go back and start again if it goes wrong

Nic Fishman that sound like its not happening in the next 5 years, hardware requirements would probably too high

András Fogarasi This is actually a big issue with AIs for board games like chess. It is most fun to play these games against an opponent with a similar skill level whose thought processes can be understood and counterplayed against. Therefore, having the most powerful AI is actually not a benefit, as AIs play chess differently from humans (because they do tree searches rather than heuristics). Often, chess programs will simply give the AI less time to run and call that the "easier" difficulty setting, which doesn't make the AI moves any more understandable. I am hoping that the AI architecture used in AlphaGo can change this. High level Go players have reported that AlphaGo has a distinct "style" of play - that it has a "personality". And its value and policy network architecture definitely _feels_ like how humans learn these games.

well said!

like chess reaching perfect is impossible. you will never be able to compute every move in chess

Rad¡o Head basically any companion ai except Elizabeth

Rad¡o Head nothing specific, but when you have to lead someone from point A to point B as a quest, and the AI is so bad that that they end up walking into walls and can't get through doorways.

Probably any escort quests. Just pick one.

Dice AI

rome 2 on launch

Have you looked at "Hello neighbor?" It boasts a Learning Ai as the only enemy of the game who's trying to defend his house while you try to break in. He'll place bear traps, lock or barricade doors, and set all manner of DIY traps based on what paths you've taken in the past. Really cool project.

The recent Lord of the Rings game and the old QuestWorld game did similar things too, though those were more like the games analyzing your style and creating a duplicate of the player character based on what it figured out.

I remember reading that alien isolation had AI that learned from your playstyle. anyway it is not a futuristic idea, just go on facebook and look at the ads you get, probably ads for stuff you have searched for earlier this month. it is not hardwired to say "he searched for shoes so show him NIKE", that would be insanity given the amount of search words and synonyms you can use

Fuck, I just wrote a long comment about that, and now I find yours:D

Curtis Jones Ohh, and are you annoyed by the "I give up" npc line, they run away but start fighting again, and the fight doesn't end:D

How complex and in what will the AI be complex in is completely dependable on what genre and type of gameplay you are going for with the game, if not thought thoroughly it can destroy the gameplay

The two cases you described are not complex situations, they are referring to the kind of AI the game uses when the player interacts with NPC's, a schedule or interaction between NPC's is not always that complex (it may look like it is, but most of the time is not).

Didn't Spore (2008) have a feature in the tribal stage where two ai would talk if they were near each other?

Geoffrey Perrin A? why not multiple

It's hard as fuck but go for it. Study up on mathematics (particularly calculus and algebra). Then go for discrete math and algorithms. Programming is also a must at some point. Then you can get directly into the meat of it.

I'm actually already a graduate of computer science and game design from my university, it's just that I can't make AI work my first job, i'm not prepared as most of my work hasn't required much AI, so I don't have any AI work to show on a portfolio (working to resolve that)

DemonGrenade274 I disagree. You certainly will need such knowledge for game development in general, but not for AI in particular. Like this video says, AI isn't as difficult or scary as you think. The scarier the prospect of creating that AI, the worse that AI will almost certainly be, especially for a Monster Hunter or Dark Souls style game. There will be no machine learning or neural networks here (I feel I should note that AI in games is very different from AI in education or research). You can always make an AI simpler by mimicking real choice with a decision tree and a bit of RNG for flavor, for most games no one will know the difference. In terms of decision making, it mostly just comes down to a tree of decisions that it makes periodically. If player behind me, do tail swipe. If player in front and close, do bite. If player in front and far, breath fire. That kind of thing. If you wanted to add a bit of variety in, you simply roll an RNG and make a choice based on that. If player behind me and rolled 0, tail swipe; if player behind me and rolled 1, tail slam. Actually implementing the attacks or movements goes beyond AI itself (though it is notable "Video Game AI" is not a full job on its own, it is simply part of being a game dev :P) Regardless, the key thing to remember is that, in general, what you really want is an AI that _appears_ to be smart. You don't want to try to actually make it smart, it will usually just end up a lot dumber for the effort. Oh, and I would be remiss not to mention Ellie from The Last of Us and Elizabeth from Bioshock Infinite as good examples of AI that can appear smart in a variety of ways, and yet are still far simpler than their choices seem. I think the Bioshock one even has a video of the devs discussing it somewhere... I'll leave you to look for it on your own though.

Geoffrey Perrin I'd love to do computer science and chemistry and zoology when I graduate ( I know it's not the year 1500 where that would be really possible even computer science) but I feel like the public doesn't see science and I can't explain it and I've been wanting to make a game where the ai re able to build and learn a kind of mid evil based game and it's kind of a experiment to see if a ai can out class a human at building a successful civilization.

A good question. I guess they put in a random number generator that triggers skripts that make the NPC miss or screw up a dodge.

A term I have heard regarding shooters is: "Designing a perfect AI is easy, making it beatable and realistic is the hard part." I can why it would be the case. The perfect playing AI always hits a headshot the moment it can see the player, but that is hardly fair or fun. Delays and variables need to be a part of it.

As someone who's dipped their toe into MUGEN a time or two, I would dearly love the answer to this question.

It might have been on this channel or another similar one, but I heard that in the Uncharted games the AI is programmed to always miss you for a specific amount of time after you come out of cover.

A very interesting question. After all if the AI in a shooter game actualy worked at 100% of it's capasity then every enemy would literaly BE an aimbot.

Hi Kabloosh! Currently there is not any formal method of guaranteeing that a viewer-suggested topic will be featured on Extra Credits (the way that Extra History topics can be suggested and voted on). Most people just tweet/email/message us about their ideas anyway. We do read them all!

Extra Credits I know it is ironic to reply to this comment with a request but I have an interesting topic. How do fanbases affect games. For example Minecraft and undertale

ok, thanks for answering!

doggo mcwoof I doubt it. Plenty of other channels have already covered that topic TO DEATH exactly because of Minecraft and Undertale.

Zach Fischer yaaasssssss Minecraft and undertale.

emotional can be had at even really basic AI.... I had a guard in Hl1 survive a encounter against a hostile Marine and well I did what I could to escort said guard to safely (or a ladder). ((the security guard even whit there crappy pistil could still kill a marine in a 1 vs 1 if the fight was on mid range mostly becos the guard had better aim)).. the short range and the marine won all the time.

Add finite state machines to that list. That's another common approach. I wouldn't be surprised if there are game AI systems which use more powerful models, like pushdown automata, too.

Parts of GalCiv2 AI worked like a chess computer, using the free CPU power during players turn to calculate turns ahead and choose optimal outcomes. So the longer the player took their turn/more powerful CPU would mean more better play by the AI opponents.

So what AI technics have you discovered?

they never said that complex ai is bad. They said complex ai doesn't nessisarily mean better. In some circumstances, yes, complex ai is better.

I just considered that because I am far less engaged in a game when I have a general sense of what the AI is going to do, which is the exact opposite purpose of AI as they stated. Even when I play an FPS I'd rather not know what would happen, but instead make assumptions that could be wildly off base and improvise when that occurs.

Not necessarily complex AI but many different AIs. The worst thing for an RTS is for every faction/side to behave the same regardless of how complex that behaviour is.

Not exactly complex, but it does need a good learning pattern ai, it doesn't need to make complex decisions, but it needs to read the players movements and move accordingly, forcing the player to change the strategy used.

Stephen von Schulmann you got a point. More complexity in the cpu make the game Harder and more fun to discover. Fighting a lvl 9 fox would be very borring if you could read every reaction

Vaidas Šukaukas I think that in order for an AI to appear smart you don't really need a lot of decision paths, just don't make them DUMB. Like you said, team mates standing in the way, or in Crysis (I am playing it right now so it's a pretty specific example that I could remember) when you shoot enemies with a silencer they will just stand there instead of looking for cover

Johannes Berger lol

@@yoavmor9002 lol

nah

James has been a part of Rainmaker Games, Snow Cannon Games and was the CEO of Divide By Zero games. I don't know about the rest of the crew.

I see the point you are trying to make here - behaviour should be more natural and unpredictable. Yet all games from Halo to Splinter cell are all about devising a plan and then putting it to action. If you throw that grenade and the enemy runs for cover only 50% of the time then you don't get that satisfaction of being a strategic thinker. Same with guys in stealth games always annoyingly voicing their thoughts. I see many ways AI could improve while not becoming more unintuitive, reacting more diversely to more player actions is one of them. They see you carrying a shotgun - they will try to keep their distance, you broke into their compounds unseen armed with a tranquilizer gun - they will set up traps, you walked into the women's restroom - they will berate you for being a pervert. It's not in unpredictability, it's in variety, because games are solid mechanics to be conquered with diverse skills.

There is no one definitive answer. The Mario games would not benefit from unintuitive AI, in fact, they would get worse by making the AI less predictable.

Well, that´s why he said over time. If you can´t learn an AI, then it´s no better than an random number generator, because unpredictable decission mean it rarely does the best decissions. It also wouldn´t make the NPCs more lifelike, but less. In real life you learn to predict people, too. That what makes you connect with them. Someone who is completely unpredictable isn´t a good character but a psychopath. It would get even worse if those were mission critical and on your side. Remember every escort mission you ever had with a bad companion AI. Now imagine those missions with completely unpredictable companions.

That's a flawed concept for FPS's if you've actually been in any kind of fight. Plans never survive first contact, and enemies being predictable is absolutely immersion-breaking.

An AI can be unpredictable and still have reasonable decisions. It's no the same as making random decisions.

In fairness I think attacks like that play a big part in the "makeing interesting problems for the player" part of an AI's job in a game. While it would be frustrating to die or get screwed over by that inescapable attack the first time, on every subsequent encounter with that enemy you know they have that ability. At that point the game shifts gear from "spot the tell and react" to "figure out how to bait the AI into doing it's less anoying attacks". Heck as a SoulsBorne veteran I can say that for my own part knowing how to bait the attacks I personaly find easiest to deal with is 80% of how I beat most bosses (and even most reasonably hard enemies) in such games.

theyre only inescapable because you put yourself in the wrong position at the wrong time. one of the things i love about MH is that every single attack is completely avoidable, and its completely fair that i got smacked by it. they are hard games. very hard. but theyre just too fair.

justdontask3 While I agree, to be fair the AI is also a lot less predictable than in older games both in who they target and in choosing attacks. I think most of us run into those situations where they home in on one player or spam or chain attacks at some point. For me that was Deviljho, it followed me around a few minutes spamming its head-swipe, with the occasional boulder or body-check. I survived several close calls while the other guys sat and watched.XD

Why

ikr? it irritates me for some reason and i don't know why.