The Video Game Industry

It's not about playing well or playing poorly. It's about being disruptive to the single-player experience. Playing solo vs. playing co-op are extremely different experiences, with different sets of incentives and strategies. Part of the reason escort missions are so noxious is because you spend hours cultivating solo strategies and suddenly there is a second person here and now all of your muscle memory and honed strategies are in the shitter.

It's for this exact reason that The Last of Us made the innovative decision to have Ellie be invisible to enemies so that she can't fuck up a stealth game that is balanced around solo play. The plot needs her to be with you, but this was not made to be a co-op game.

I can definitely see benefit to games which are designed from the ground up to be this sort of co-op with AI partners, such as Kingdom Hearts. Or co-op games that give you an AI partner if you don't have a human to play with, like Resident Evil 5.

But Escort Missions will probably continue to suck.

Edited by TobiasDrake on Mar 18th 2024 at 10:06:07 AM

Yeah, escort missions sound like the most pointless thing to put AI into. The point of an escort is to be as little an obstacle as possible to the player, and it's not some kind of hyper-advanced AI that will make that a thing. If anything, removing as much of the AI as possible is the best thing to do.

The thing is you can train the AI on "being as little as a disruption as posible" as well, rather then just "play the game well" it all depends on how you set up the rules for it.... give it rewards for not getting hit, and keeping some what close to the players advances (nothing is worse then escorts who are slower then a run but faster then a walk), punish it whenever it gets into the players crosshairs, and so on, and rather then making some killing machine you are going to make something that is a much more belivable duck and dodge then a script.... could probaly even get it to do better because it could be proactive about avoiding fire.

But honestly I don't get the focus on escort missions, with how many games are mandatory co-op these days, I would just like some reasonable bots to fill in the spare player slots when there is only 2 of us out of a party of 4.... Or I don't know just something that makes the loby feel more complete then just running bosco in a game like DRG.

Edited by Imca on Mar 19th 2024 at 6:03:24 PM

Bioshock 2 had the right idea in making Eleanor Lamb summonable via Plasmid near the end of the game. So it was essentially your choice if you wanted an overpowered ally to help you out or not.

Edited by RainingMetal on Mar 19th 2024 at 9:49:45 AM

Escort missions will always be the bane of my fucking existence, like I'm sure you can't make something that involves escorting a easily killed, useless, slow-as-shit, braindead half-wit NPC who has no goddamn bearing on the plot beyond that one mission and has all the self-preservation instinct of a suicidally depressed toddler while 100000 enemies tries to kill you and that NPC, anything other than a low point in any game its in

Fuck escort missions

Edited by Cordite-455 on Mar 19th 2024 at 11:27:47 PM

Hence the "easily killed" part not really being a thing nowadays. Unless the devs are sadistic of course.

Being "not easily killed" doesn't matter if I'm gonna be carrying a useless lump of code who only serves to block my line of sight to point A to point B while a fuckton of enemies try and kill the shit out of me and him for the next 30 minutes (because I keep failing), not fun

I'd rather defend a spot for a few minutes and not worry about carrying a slow ass lump of flesh, at least I don't have to turn around every 5 seconds to check if the idiot is following me properly and not being stuck in the walls

Edited by Cordite-455 on Mar 19th 2024 at 11:33:13 PM

One would think that they could program an AI to hide and stay out of the players way. Some of these escort missions would be a nightmare in real life.

If the goal is to minimize the burden of the escortee's presence, programming an AI to try not to be a burden is never going to be as effective as quality-of-life hacks to the game logic to prevent the escortee from being a burden.

The issue with the application of AI is the same issue that often comes up with the tech industry: We have not identified a problem to which AI is a feasible solution. Techbros have worked very hard to create a solution, and are now trying to come up with problems to fit it into.

It's a fad, not unlike motion controls, virtual reality, or procedural generation. Big Tech made this neat toy that they think is really cool and are now trying to figure out how to apply it to gaming, but it's never going to be as effective as just writing dialogue and scripting NPCs.

They aren't really trying to use AI to improve gaming; They're trying to use gaming as a vehicle to play with AI. And that's fine, but it's unlikely that it's ever going to become anything more than a niche interest.

Edited by TobiasDrake on Mar 19th 2024 at 10:10:51 AM

Eexcept we have come up with problems it is a feasible solution too, and have been using it on those problems for years... Google search has been an AI since the 90s, and the feild has existed since the 70s.

The problem is that now you have two camps of people, one of whom pretends it is the solution to every problem and one of whom pretends it is the solution to no problems and both are extremely wrong.

Like it or not "I wish the computer was smarter" is a feedback that players have been giving for years, the problem is that when you use pure scripting to make smarter bots they end up coming across as unfair instead since the computer inherently knows what your doing and can react to it without seeing it.... this is why its not done even though it has been something trivial to do since the 2010s, and you can find plenty of development blogs talking about that... your stuck between the rock and the hard place of "Computer either pretendss it doesn't know what your doing dispite having a constant screen peek" which makes the computer seem incredibly stupid since an actual human opponent will learn what your doing and adapt, or "Computer acknowledges what your doing, but becomes an unfair opponent that plays with wallhacks"

To get the middle ground of "smart enough to be fun, dumb enough to not be unfair" machine learning is a pretty valid application.

Edited by Imca on Mar 20th 2024 at 1:21:56 AM

I do think AI will find more and more uses over time, but the way it's being forced into everything is ridiculous.

If AI is able to convincingly play games, you won't know if the individual you're playing with is real or not. A bot or not. There will be a lot of legitimately good players accused of having the AI play for them. There's already been reports of women accused of cheating, like South Korea's Geguri.

I do think it's worth noting that people have tried approaches that approximate autonomous AI and the result is usually hilariously broken games.

There's a Lord of the Rings text adventure that's completely notorious because all the characters being more or less autonomous means the game frequently breaks itself within like 10 moves because Gandalf ran off and got killed by a warg. Bethesda's Radiant AI system had to be deliberately crippled because the game world would wildly reshape itself because all the guards in Cyrodiil would go to war with the monsters and one side or the other would win decisively. Including killing off a lot of important NPCs hours before you get to meet them.

This is funny and it can be fun to play that LOTR game and see how far off course it gets, but it also gets incredibly frustrating and annoying in the long run because it makes the game nearly impossible to beat due to factors you have literally no influence over.

It's like the idea of letting NPCs have a chatbot run them. It's a fun idea and in a full sandbox environment, it could be interesting, but eventually, it'd make players really annoyed that the characters just don't get on with it and are just playing an endless game of "yes and" with you. Games with NPCs that have too much dialogue that isn't interesting train players out of talking to them.

There are use cases for AI in video games, but it's probably not the obvious ones.

The thing is those are using randomized scripted agents, its ironically a bit easier to contain them when your using an actual learning agent, since you can give rewards for things like "staying alive" and "sticking in this area where the player can find you"

But also as you noted, the problem there is overuse.... you don't have to use this for everything... and I wouldn't use it for NPC in a role playing game... Its better for things like opponents in strategy games where right now our options are either "The computer cheats its ass off to stay competitive with a human player" or "The computer is stupid as shit".... most strategy games don't even just stop at screen peeking level of cheating either, they actually give the AI resource hacks and the like, and this is a common complaint from players....

But the other option is with the level of complexity in a strategy game the computer breaks its economy in the first 10 minutes, and becomes impotent for the rest of the game.

I know these ones from experience, but given how often I hear FPS players complain about the stupidity of bots, I do have to imagine you could get good use there too.

Its not an all or nothing type deal, its noting the places where application is usefull, then applying it to those cases.... just like any other tool we have..... We don't use screwdrivers to drive nails, but likewise we don't just discard there use utility for screws because they cant drive nails.

Edited by Imca on Mar 20th 2024 at 1:35:40 AM

Feels like we're talking about AI in two different senses. Games have had AI for decades, governing the behavior of enemies and NPCs. We seem to be talking specifically about generative AI, such as having dialogue created on the fly via a language model rather than scripted ahead of time.

The idea is that generative AI should make worlds feel more lifelike and immersive. Once you've talked to the guards in Skyrim a few times, you've exhausted all of their random voice lines, so talking to them is no longer interesting. But the guards aren't there to be "interesting characters"; they serve a very limited mechanical function.

Ultimately, games need to pick a core experience that they're going for. Heroic fantasy needs the player character to feel powerful and like they're the ones with agency over the outcome. If your AI-driven NPCs independently figure out how to solve the world's problems... or run off and kill themselves when you aren't looking, that core experience no longer works.

Perversely, AI needs to be at least somewhat dumb. If the NPCs who act as your allies and support characters are too competent, your presence feels unnecessary. If the enemies get so good at tactics that fighting them is frustratingly difficult, the game is no longer fun to play.

Edited by Fighteer on Mar 19th 2024 at 12:38:05 PM

IMO, the best uses are in the realm of "the player shouldn't be able to tell there's an AI involved at all" type stuff.

Fallout 3 is actually a good example of the latter thing, seemingly by design. The final big battle sequence of the game, it's genuinely really hard to even get any kills because the Brotherhood is doing everything for you. It's weird and unpleasant and doesn't work.

Edited by Zendervai on Mar 19th 2024 at 12:38:40 PM

Generative AI can have value in games designed as simulations rather than scripted experiences. Imagine going into a world, finding a random NPC, and deciding to have a chat about the weather. One thing leads to another, you're bonding over drinks, and then you decide to go off and liberate Elbonia from the evil aliens.

That's the sort of emergent gameplay that simply can't be designed, and which generative AI would be ideally suited for. But you run the risk of a homogenized experience that has no defined objectives and just sort of peters out once you hit the limits of what the AI can do.

Edited by Fighteer on Mar 19th 2024 at 12:43:02 PM

Once again I am going to have to disagree in the sense that it really depends on the game....

I want to be able to play something like Stelaris, and have the computer be able to use something other then just spamming more ships at me without breaking its economy with the systems upon systems upon systems that make the games foundation....

Or even on a smaller scale, XCOM where the aliens don't just run out into my guns

You want opponents that are still beatable and not godlike, but "The AI is dumb" is a fundamental problem with strategy games right now, the entire point of them is to out think the opponent.... but this is cut into once you realize that the computers tactics are simple brute force 9 time out of 10.

Edited by Imca on Mar 20th 2024 at 1:50:09 AM

Honestly, I've never thought that feedback to be what people mean they want. Players don't actually want good AI. They want an AI that can lose believably and entertainingly. More than that, they believe that as they get better, they should be equally challenged and that's not how difficulty or skill works. Things get easier the better you are at them, and when you've played the same game or series for decades, you're not going to experience the game in a way that will make you play badly or go through a learning curve.

Edited by Resileafs on Mar 19th 2024 at 1:08:25 PM

Would be nice to play Pavlov without the pick-your-poison choice of either really dumb as shit bots or the near-inevitable run-in with toxic live players.

I don't mean that AI must be "shit its pants at the starting line" dumb. I mean that it can't be so smart that it makes the game frustratingly difficult for the average player.

Where that line is varies by game (and by player), but unleashing Deepmind on newbie Starcraft players wouldn't just be a waste of resources, it would guarantee that they'd quit the game almost immediately.

Strategy Games specifically are a case where smarter AI actually is needed.

Most strategy games suffer from being too complex for the AI to play even competently, and virtually all strategy game A.I.s get around this by cheating. This has negative effects on gameplay too.

Let's take the game Empire Earth for example. In Empire Earth, the AI is very dumb but cheats allow it to get free resources and such.

Problem is, logistical strategy is part of the game's point. Against human players things like trying to cut off their access to resources and such is part of the game's depth. If the AI can just cheat resources in then you're left with a more shallow experience even if it's technically "difficult".

If the AI actually understood the game better, it would add more mechanical depth as the AI would actually be engaging with the game's deeper mechanics and not just bypassing them.

The Civilization series is another particular example of this. It almost feels as if Firaxis intentionally sabotages its AI so that the player experience will be consistent from game to game, but the metagame against human players is immensely different because they're actually trying to win.

And of course it makes difficulty higher by giving the AI more resources instead of making it smarter.

Since we're on the topic of strategy games and macro-based AI. I recently bought Disgaea 7 and found out that it and it's predecessor has a macro system built-in. Allow the player to program the AI of their units (consequentially they also added an auto battle system as well).

I think it's rather rare since the only other instances of it I know of were in two final fantasy games (xii and xiv) and in the latter case it is deliberately crippled because of botting issues.

Edited by MorningStar1337 on Mar 19th 2024 at 10:26:55 AM

It's not just a matter of difficulty necessarily, but of Fake Difficulty. The AI in a game like Civilization is very bad at the game but can be made into a difficult opponent by allowing it to cheat.

Problem is that there actually is a significant difference between an opponent that's good at the game, and one that's bad but able to cheat frequently. And the latter can result in a more shallow experience.

I will say in fairness that you often do want AI to be less intelligent than humans for a variety of reasons. Most players want to win, after all, even if they want to be challenged. And most players also would rather the winner be a human even if it's not them.

To use an example, if you have a match in a multiplayer game with both humans and bots, the humans will often preform an inversion of Gang Up on the Human in order to ensure that the victor is an actual human player.

Edited by Protagonist506 on Mar 19th 2024 at 10:28:36 AM

Also because the AI tends to be kind of an asshole. Why yes I'll build/not build stuff just to satisfy your agendas kindly Civs.