How can new console improve on games compared to this gen?

[Shifty Geezer]

Pacman like all games is set in a world based on rules. I am sure the logic puzzles can be extended without making the game impossible to beat. Its rules are much simpler than chess and yet chess is still possible to beat against a computer that plays by the rules.

The AI of computer chess is seriously dumbed down for many players who'd never beat it otherwise.

There's only so much you can do with Pacman regards ghost movement - how do you propose the ghost AI can be improved in a way which is fair and challenging and different to how it is? We can extend game complexity by having different behaviours from different ghosts etc., but the AI level is fundamentally the same with a few tweaks to parameters. I don't see any reason for it to be different. I don't see any reason to change AI behaviour in the likes of current shooters either (based on my moderate experience). Grunts offer suitable challenge while being eminently beatable. Maybe there's the issue of dumb AI every now and then that sees troops walking into walls or standing around, but those are bugs rather than AI design faults.

 

[owen]

Game A.I. is not about perfectly emulating a "computer", its about emulating "human" behaviour. Playing grandmaster chess against a computer is not the same as playing grandmaster chess against a human gm.

I imagine that the way to improve a 30+ year old game is yet undiscovered. Most grunts will run into your shooting line everytime and are less interesting that human opponents and the fact that they are less interesting means that there is room for improvement in the A.I. to give the enemy more human traits.

 

[Shifty Geezer]

If you want the AI to change, you must have ideas of behaviour you'd like to see different. You don't need to know how to program state machines to have an idea of how you'd like enemies to act. eg. Take cover when shooting, try and flank. In Pacman, if the Ghost's intention is to capture Pacman, then AI would have them all gravitate towards the player instead of wander aimlessly about. That'd make the game impossible. Alternatively, you could have random or patterned movements with a chase when a line of sight or proximity fix is made, with a break-off after a time to give the player a chance to escape. That'd be 'dumb' yet better for enjoyment. And we what currently have.

 

[owen]

You are taking the wrong approach to what I am saying. I am not trying to tell you how it can be improved / or what tricks to use to hack some lame pattern based path finding that has been done better in starcraft 2. There are tons of hacks and designed levels with choke points.

What I am saying I want next gen A.I. to be as good as online multiplayer against "human" opponents. impossible?

 

[Shifty Geezer]

Probably not, but the mechanics of online and solo games are different and don't warrant smart AI. In a 32 v 32 game or similar offline (SW:Battlefronts), what you want would work. With suitable AI, you could play alongside 31 human-level AI bots. Of course, for a lot of gamers those bots would play far better than they and leave them wandering around fairly aimlessly. In a game like uncharted where you outnumbered 5:1 or worse, bots having human intelligence would be no fun whatsoever, so there's no reason in pursuing those developments. Given developers can provide human-level opponents by providing online play, that's a much cheaper option for them, and a checklist feature. So I wouldn't be surprised if AI doesn't progress outside of RPGs.

 

[ERP]

This has come up before.
The problem with AI that consistently beats human opponents, especially in 1 on 1 games is players simply assume it's cheating.
Any AI in an FPS like game will have absolute perfect knowledge of the map, will be able to run the exact shortest distance between two points, and have perfect eyesight and aim.

You can add a degree of randomness to this, but it still feels "cheap".

I could recount stories of people building flight sim AI's in the1980's that were close to unbeatable, and having to dumb them down.

The goal of good game AI is to challenge a player, but let him feel he has a degree of control.

There aren't a lot of new interesting ways to do game AI that haven't been tried, state machines are still in broad use because they are predictable and that gives the player a feeling of control in the environment.

But other mechanisms have been used, what amount to weighted decision trees or neural networks have been used, but they are really hard for a designer to control behavior or debug bad behavior.

There is also the people are stupid problem, people do dumb things, if an AI does them it's viewed as dumb AI and a bug.

IME what people want when they ask for more human AI is more variation in behavior from game to game, that could be done today, it's not a hardware problem, it's just hard to create the behaviors.

 

[Arwin]

Would HD PacMan be more fun if the ghosts had more processor intensive AI?

Hackman II is certainly better than the original. I think it's from something like 1988 ...

 

[joker454]

Game A.I. is not about perfectly emulating a "computer", its about emulating "human" behaviour.

Many years ago when I was still in the biz I had proposed the idea of using all those hours that q/a take play testing the games as a basis for ai. Basically record how they each handle certain situations and add that to a database of ai on how to resolve situations. The theory was that after weeks of play testing the game the human q/a testers will have played thru the different scenarios in the game in many different ways, likely ways that the creators didn't even think of. So I had pitched the idea of tracking that data as a means to make computer ai seem more human and unpredictable. I hope someone would try something like that as I never got around to it back in the day. It is more brute force I guess, but the good thing is that it was mostly a storage issue and ultimately fairly light on cpu.

 

[tuna]

Game A.I. is not about perfectly emulating a "computer", its about emulating "human" behaviour. Playing grandmaster chess against a computer is not the same as playing grandmaster chess against a human gm.

No, game A.I. is about making a fun experience for the player. For some games, that might be done by mimicking human behavior, but for some games it might be done by having very deterministic and simple states.

 

[owen]

Many years ago when I was still in the biz I had proposed the idea of using all those hours that q/a take play testing the games as a basis for ai. Basically record how they each handle certain situations and add that to a database of ai on how to resolve situations. The theory was that after weeks of play testing the game the human q/a testers will have played thru the different scenarios in the game in many different ways, likely ways that the creators didn't even think of. So I had pitched the idea of tracking that data as a means to make computer ai seem more human and unpredictable. I hope someone would try something like that as I never got around to it back in the day. It is more brute force I guess, but the good thing is that it was mostly a storage issue and ultimately fairly light on cpu.

Well its good to know that some people have new ideas to old problems.

I don't think its a difficult problem. It just needs some fresh eyes. I don't want next get to be filled with physic simulations (angry birds) and tunnel shooter. that is all.

 

[willardjuice]

I don't think its a difficult problem.

lol.

 

[fearsomepirate]

What I am saying I want next gen A.I. to be as good as online multiplayer against "human" opponents. impossible?

No, you don't. You think you do, but if AI were actually that good, it would mean that first-person shooters would not be able to throw more than one enemy at a time at you, since two enemies with human-like AI should be able to act in concert and out-maneuver you nearly every single time, and three would be unbeatable. I'd like to see more realistic materials and destructibility, and console versions of Battlefield to have 64 players.

As it is now, we've come pretty close to an upper bound in the quality of assets that are economically feasible. So I would expect a lot of the improvements to be on the less cost-intensive side. I expect floating-point RGB to be standard next gen, better overall IQ, better shadows, less obvious LOD, fewer polys removed from the initial high-quality artists' meshes and so on.

 

[kache]

No, you don't. You think you do, but if AI were actually that good, it would mean that first-person shooters would not be able to throw more than one enemy at a time at you, since two enemies with human-like AI should be able to act in concert and out-maneuver you nearly every single time, and three would be unbeatable. I'd like to see more realistic materials and destructibility, and console versions of Battlefield to have 64 players.

Wrong.
As I've said a lot of times in the past, when the enemies become smarter than you, you get stronger, the same way the more stupid were the bosses in the past, the more HP/atk/stats in general they had.
Basically, the balance is between equipment/stats and inyelligence. If the enemy is smarter, you get better equipment, if the enemy is stronger (but stupider) you win with your brain.

 

[Silenti]

I must disagree with the "you don't want better AI" crowd. As Shifty stated I do not want them being human multiplayer champion smart. I want to learn their flaws and take advantage of them. This is also one of the main reason I prefer slower, as opposed to twitch, FPS and prefer campaign to multi-player. The problem arises when you have encountered the all the available AI tactics in the first 20% of a given game and the only way to make the game more difficult is to make the AI enemies be ridiculous shots, both in accuracy and how fast they react to your presence, and the usual increases to their health combined with a drop in yours. Take HALO 2. One of the most frustrating elements I have ever encountered was the jackal snipers, 1 shot 1 kill. That made certain portions of that game intensely frustrating. I would much prefer a smarter AI to the bog-standard version of tougher enemies from above.

 

[milk]

I think AI needs to improve more on the behaviroal level then on the tactical one. They can still be dumb fighters, but they at least should move about, and act more like humans do.

 

[kache]

Personally I disagree with the whole "make them more human" argument.
I really liked a proof of concept a dev made: basically he gave guards a certain awareness field (visual limited at a certain degree, and limited by light available, auditory limited at a certain range), gave them a certain comunications delay (the walkie talkie) and a certain lag for other things (reactions, shooting, moving, ecc) then let them develop autonomously. The results were very interesting.
Machine efficiency is awesome. It's a waste not to use it.

 

[owen]

There are clearly two multiple forms of A.I. I am not talking about the "logic puzzle A.I.". I am talking about A.I. for local bots in shooters and various other open world games. If you like dumb A.I. well thats good for you but some of us want a deeper expirence.

If you login to online multiplayer MW3 at level one you are going to get killed because you do not know the maps or the guns or the camping spots. The more you play the game the better you get (you develop strategy) and soon you will be able to kill even the smartest opponent with the best gun. This is why online multiplayer is so addictive, it takes practice. I want this for single player.

I am not sure why so many people are arguing against it. Its not impossible, we just have not figured out how to do it.

 

[Shifty Geezer]

The more you play the game the better you get (you develop strategy) and soon you will be able to kill even the smartest opponent with the best gun.

That's not true for a great many players. Many turn up, get shot, get shot lots more, get spawn killed, etc. Very few develop to be able to compete with the higher levels on a level footing. That's why many avoid online games. This gen saw my first expereince with online competitive gaming, and it's been the most frustrating experience of my gaming life. There's a huge barrier to entry that isn't fun at all when playing superior opponents, which is why I'd advocated better matchmaking or game balancing. Different people want different things from games. AFAICS most expert online players don't want a challenge but would rather play noobs and mince them relentlessly, exploiting the 'dumb intelligence' of the opposition. We don't have specific figures on the gaming interests of players, so we can't really say one way or another, but I think your belief that people value clever, challenging opposition is misplaced.

I am not sure why so many people are arguing against it. Its not impossible, we just have not figured out how to do it.

People have figured out how to do it. They just don't want to because it breaks the game in the solo experience. Why are you so keen to have smart AI when you can just play online instead? That's way easier to implement.

 

[kache]

People have figured out how to do it. They just don't want to because it breaks the game in the solo experience. Why are you so keen to have smart AI when you can just play online instead? That's way easier to implement.

Because humans are not perfectly efficient. A non-human, smart IA would provide an incredibly fascinating challenge, and would be different from what multiplayer could give, thus making for a much more interesting single player experience.

 

[owen]

That's not true for a great many players. Many turn up, get shot, get shot lots more, get spawn killed, etc. Very few develop to be able to compete with the higher levels on a level footing. That's why many avoid online games. This gen saw my first expereince with online competitive gaming, and it's been the most frustrating experience of my gaming life. There's a huge barrier to entry that isn't fun at all when playing superior opponents, which is why I'd advocated better matchmaking or game balancing. Different people want different things from games. AFAICS most expert online players don't want a challenge but would rather play noobs and mince them relentlessly, exploiting the 'dumb intelligence' of the opposition. We don't have specific figures on the gaming interests of players, so we can't really say one way or another, but I think your belief that people value clever, challenging opposition is misplaced.

Well then shifty you are not the type of person that values challenging A.I. Not use arguing something that doesnt appeal to you. I play fighting games and just started MW3 online this generation. I need my games to challenge me and I need to feel that I am getting better at them. We could start a whole new thread about the topic.

Online multiplayer is VERY limited. I have clocked 10 solid days playing the same 5 levels over and over. I have killed 100,000 players on the same levels with the same guns. Its a big waste of time and has absolutely no story elements but its still fun.

People have figured out how to do it. They just don't want to because it breaks the game in the solo experience. Why are you so keen to have smart AI when you can just play online instead? That's way easier to implement.

Point me to the game.