AI in games

[Darkpriest]

Recently I watched how a fairly fast developing OpenAI managed to get more than a basic grasp at a co-op game - DotA 2.

 

Last year, they've shown one hero AI, being able to beat consistently with high rate top human players. Now, while still with significant restrictions, they managed to make 5 AI bots co-operate in a team game and beat 5 men team of really good players (retired pros).

 

I wonder, how could this affect gaming. I mean, we often hear complains how AI is bad in this or that game, etc., but think of a scenario, where AI will beat average player 99.9999999999999% times, and the best human player 90% of the time. Where is the limit for an enjoyable, yet challenging AI in games, which would be playing by the same rules and with same resources, and where the good AI becomes too good for a game to be fun (no one wants loosing all the time, even though you know the conditions are fair)

 

https://www.joindota.com/en/news/72530-openai-stomped-first-but-there-is-still-hope-for-the-human-race

[HoonDing]

How about AI that adapts to the player's IQ.

[ShadySands]

 

I don't think it will be an issue as long there are different difficulty settings. 

[SonicMage117]

So many variables when talking about such a topic because different genres move them in so many different ways.

 

We have games like Infinity Engine games which feature some of the most horrible a.i in existence, we can see it in examples such as pathfinding (probably why combat is limited to limited format) and then we have great a.i in games like Alien: Isolation which take the a.i to anothe level in greatness. Where literally everyone a different experience as the Alien learns from players on an individual basis.

 

Some of the highest quality and most technical a.i though, can be found in high speed fighting games such as anime 2D fighters or anything like Tekken 7. Ex. You can take a turn and cast a spell, the enemy on his turn will counter it with a remefied toxin which is faster reaction time than most players will ever have - I suppose that's exactly why Evo has some of today's most competetive matches, players are so used to top-tier a.i that fighting humans on the sam level or higher is just second nature at that point.

 

A.I that learns isn't really that much of a financial investment but it's time consuming and takes so much playtesting to round out. If a company cares enough, it will show and as we consumers are often smart, we are able to tell which games show light on this. I just wish that Crpg's had good or decent a.i for once instead of being like The Sims and using other things to give difficulty to players in cheap ways.

 

Other examples of bad a.i are often found in twin-stick shooters, games like Serious Sam series, and the like. These games heavily rely on swarms at the expense of a.i, when enemies are programmed to run at you with no second thought, I find it takes away from the overall fun factor as immerseness is taken away.

 

I have alot more to say but not enough time so I'll just leave that for now. My point being that you'd figure you'd see the best a.i in rpg's which encourage strategy but it's really just the opposite.

[Darkpriest]

 

I don't think it will be an issue as long there are different difficulty settings. 

 

How would you decide what's on given difficulty level if you want to maintain same ruleset? How would you teach AI to be less efficient and handicap itself on purpose? Or would you rather keep "AI" to the level of fallble scripts?

[Katphood]

I want npc's with schedules, goals and needs, accompanied by an intelligent A.I that can create different situations.

 

 

Nevermind, I think I woke up Todd and Peter.

 

 

F.E.A.R had one of the best A.I in games, the bots in Unreal Tournoment were nice, too.

[ShadySands]

I'm not an expert on AI or on the game mentioned in the article but I have a hard time believing there'd be no way to limit the AI's effectiveness or even for the AI to limit its own effectiveness given OpenAI's goal of safe AI. The comments of the article (sorry I didn't do any digging myself) already mention that the AI had some constraints like reaction time so I'd imagine you can build upon that. Also, I'd imagine that once the technology progresses further you'd not see SkyNet level AI in your games but rather something designed to give the player a meaningful (definitions may vary) challenge.

[injurai]

How about AI that adapts to the player's IQ.

 

The Rick & Morty game is going to be a cpu melter.

[Fenixp]

F.E.A.R had one of the best A.I in games, the bots in Unreal Tournoment were nice, too.

Yet FEAR's AI wasn't actually particularly clever, it was simply cleverly scripted within constraints of rather linear levels.

Heck, people tend to laud Half-Life's AI, yet it was quite literally a simple state machine.

 

Good AI isn't one that'll defeat players consistently - on the contrary, good AI is one which'll provide decent challenge while being fun to play against.

 

With that in mind, OpenAI uses machine learning to 'grow' - and I don't think it necessarily needs to learn to beat all players that play against it, but rather to be 'fun', if being 'fun' can be put into a set of rules (like unpredictability, win:lose ration on set difficulties etc.)

[injurai]

If you let the soldier's in fear away from their spawn points they break, but the plus side is they are all well tuned for where your supposed to encounter them.

[SonicMage117]

The best A.I is one that convinces you that it could be a human playing alongside you. I don't think being "unfair" really has anything to do with it. If you play against a human in real life who has more experience and wins, some may view that as "unfair", while others with tougher skin will take that as a learning experience (as they should).

 

That said, progressive nature in A.I is also key to being believable, it's okay for smart a.i to make dumb mistakes....And it's logical. Seeing an a.i learn from mistakes isn't as rare as many think, it's the behavioral cortex in which it applies that process in a given situation correctly which is a rare thing.

 

I think that's why Alien: Isolation's award winning A.I is well deserved. It's pretty much a shining example of all the right things in what an excelling a.i should be ajd how it should behave. Despite a few glitches here and there - which have nothing to do with the a.i but collision detection instead - you'll see how it bends without breaking.

 

I think alot of people though, don't know how to separate animation and world/world intraction from a.i. Most people don't know that world interaction with cpu has nothing to do with a.i, for it is a whole different thing altogether.

 

Also, another reason why Alien: Isolation is well deserved is bfor the simple fact that it reacts in real-time. None of it was linear or scripted event and that's a hard feat in itself.

Edited August 7, 2018 by SonicMage117

[Katphood]

Good AI isn't one that'll defeat players consistently - on the contrary, good AI is one which'll provide decent challenge while being fun to play against.

 

Which is what F.E.A.R did?! It's a shooter so their aim is to defeat you. In whatever way it was implemented, it ended up being fun.

 

I think there are other games which do what you described. Haven't played it yet but isn't Rimworld aiming for such a think?

Edited August 7, 2018 by Katphood

[ShadySands]

It's a shooter so their aim is to defeat you.

This made me think about this topic some more, well this and watching the TNG episode where Data plays the Strategema master. The goal of the AI doesn't need to be victory in every case, maybe the goal is just to counter whatever you do. Might be more difficult in certain types of games.

[Darkpriest]

Well, i guess it can depend on the setup. It can be set to be annoying and counter you to delay your progress, or it can be setup to be the most efficient and try to defeat you as fast as possible

[Katphood]

 

It's a shooter so their aim is to defeat you.

This made me think about this topic some more, 

 

 

Of course it did,

I'm a visionary...

[Fenixp]

Which is what F.E.A.R did?! It's a shooter so their aim is to defeat you. In whatever way it was implemented, it ended up being fun.

Well yes, that was precisely my point. Well not that their aim was to defeat you, because ultimately, it wasn't - if AI in FEAR was designed to defeat you, player character would be dead upon turning a corner. It was cleverly designed to be fun to play against. Edited August 8, 2018 by Fenixp

[Gorth]

You can argue whether it was good or not, but dang, some of those old SSI games on the Commodore 64 impressed with their AI opponents at the time. Whether small or grand strategy. Bear in mind, there was NOT a lot of giant shoulders to stand on at the time to look for ideas...

[SonicMage117]

I was thinking today, what would it take for I.E styled games to have decent a.i for once? I mean, we have great a.i in Age Of Empires 2, decent a.i in Starcraft... it's unscripted at least. I often wish I.E and I.E style crpg's were treated with thr same respect in a.i as other games, instead of trying to emulate a game of chess with walls to break.

 

[Hulk'O'Saurus]

Any game is set in rules and limitations. The speed at which AI can operate, or rather, expert systems, which 'AI' in games and everything else we have essentially is, insures that said AI will always defeat humans at any game provided it is sufficiently developed, the player is not statistically overpowered - as with RPGs or player-only stun-lock gizmos, or the AI is not nerfed, for instance, as in having slow, very well telegraphed movements like in DOOM. The most fun is to emulate human/animal-like behaviour, and therefore have limitations to which we can adapt. 

[Althernai]

Recently I watched how a fairly fast developing OpenAI managed to get more than a basic grasp at a co-op game - DotA 2.

 

Last year, they've shown one hero AI, being able to beat consistently with high rate top human players. Now, while still with significant restrictions, they managed to make 5 AI bots co-operate in a team game and beat 5 men team of really good players (retired pros).

 

I wonder, how could this affect gaming. I mean, we often hear complains how AI is bad in this or that game, etc., but think of a scenario, where AI will beat average player 99.9999999999999% times, and the best human player 90% of the time. Where is the limit for an enjoyable, yet challenging AI in games, which would be playing by the same rules and with same resources, and where the good AI becomes too good for a game to be fun (no one wants loosing all the time, even though you know the conditions are fair)

 

It's an interesting question, but keep in mind that this is not relevant for gaming on your local hardware and is not likely to become so anytime in the near future. Here's a slightly more technical article on the DotA2 bots:

 

 

The bots learned to play Dota 2 through playing hundreds of years of matches against itself, previous versions of itself, and preprogrammed scripted bots. For last year's bot, the training was done on Microsoft's Azure platform with some 60,000 processor cores; this time around, OpenAI is using 128,000 cores on Google's Cloud Platform. The bots learn the game from scratch: initial versions will just wander aimlessly and at random as the game plays itself out. As thousands upon thousands of games are played, it figures out which actions will improve its chance of winning.

 

 

If you follow the link in the quote, it will take you to a site with detailed descriptions of the hardware where it can be found that in addition to the 128,000 CPU cores, they also used 256 GPUs (Nvidia P100s). So yes, the AI can beat humans... but it requires a decent-sized server farm to do so. Unless you play at a high enough level to get into e-sports, it will be years or perhaps even decades before you can personally play against this level of AI. What it might do (as AlphaGo did for Go to some extent) is teach human beings that some tricks are more useful than they were previously known to be and thus change play styles.

[213374U]

I was thinking today, what would it take for I.E styled games to have decent a.i for once? I mean, we have great a.i in Age Of Empires 2, decent a.i in Starcraft... it's unscripted at least. I often wish I.E and I.E style crpg's were treated with thr same respect in a.i as other games, instead of trying to emulate a game of chess with walls to break.

 

What would it take? A handful of dedicated developers.

 

SCS does what the guy in the video showed, and then some (not knocking his work, just stating facts). The new scripts governing enemy behavior were expanded from several hundred lines each originally to several thousand. All it took was a physics professor writing an entirely new set of AI scripts in his spare time, based on a third-party piece of code developed by professional programmers as asked by one of the original designers of BG2 (detectable spells).

 

The added complexity makes for a really enjoyable experience—if you have sufficient game knowledge to cope with it. Otherwise, it's a fairly miserable time. And so we come to one of the problems with good AI: most people won't spend enough time with a game to appreciate good AI, or derive enjoyment from constantly facing an AI opponent that is simply a better player.

 

What fraction of fans view SCS as a must? How many even know about it? And this is BG2, one of the most popular and highly rated CRPGs of all time.

 

TL;DR it's just not worth it for the most part