The "General Games Chat" forum, which includes Retro Game Reviews, has been archived and is now read-only. You cannot post here or create a new thread or review on this forum.
Well, in the drought of good topics recently that aren’t console wars (even I am beginning to tire of them now) I thought I’d write about artificial intelligence.
In many developer’s eyes, artificial intelligence is the ultimate creation. Its something that everyone wishes to make, and the code is worth millions. So what is it? Well basically, AI gives the computer the ability to make its own decisions, learn from mistakes, build on itself and more importantly have its own opinions and feelings. AI plays a large part in Sony’s plan for the future, as its main selling point is the Emotion Engine, and its supposed ability to simulate AI.
But ask yourself this: Is AI already upon us? In a way, yes. As I write this, I am witnessing a very simple form of this. Word 2000 has the ability to learn from its mistakes. For example, the first time you type “PSone” into word, it will decapitalise “S”. If you correct it a few times, it no longer does this. Word is learning that you don’t like this, and adapting. OK, let’s make a list of AI features:
1. Learn from mistakes
2. Make decisions
3. Adapt
4. Build on itself
5. Have it’s own feelings and opinions
Right then; scratch 1. What about 2? Well, decision-making routines have been in games for yonks. Let’s look at Tekken. According to its situation, Tekken will choose between high, low, hand, kick, flying, special and combo attacks whilst you are fighting. It assesses the situation then makes a decision and executes it. This is a very basic routine as don’t forget it is running on a PlayStation here, which is essentially a 66mhz. Let’s look at a higher form. In games which has multiple routes such as SSX on the PS2 or Speed Freaks on the PSone, the computer has to decide which route to take based on: Risk according to mistake parameters and advantage gained in relation to power-ups. If the computer decides to go for it, it will. Games that do not have a fixed route such as kart racers have to make very complex path finding mechanisms every second, which is a testament to modern technology. OK, decision making’s current highest level: The First Person Shooter. When we play a FPS we usually stay to a basic plan, such as: Keep moving, shoot people when opportunity arises, kill Sniper quick before he kills you, and stay near the BFG. We are constantly running these thoughts through out head and we make split-second decisions that eventually boil down to a list of commands. Forwards or back, left or right, shoot, jump or change weapon. It’s the same with a FPS AI routine. The computer assesses how many people are near, threat imposed, success probability and other things such as terrain and power-ups. It then generates commands. These routines are very sensitive however and that’s why movement can be erratic.
OK, now we’ve covered decisions, let’s look at adaptation. Adapting in videogames is also not a new thing. The PSone game Speed Freaks supposedly learnt your driving styles and adapted to your driving styles. I never really noticed this but the reviewers said they could. Essentially, this means that if the computer recognises that you aggressively barge people out of the way, it will stay away from you. If it notices you are targeting a particular character, it becomes aggressive itself. The computer analyses your actions then basically recompiles its AI routine to compensate. Therefore, each game will be different. Of course, for this to be any good you need a sophisticated piece of hardware, and this is where the nest generation console comes in. Now, I don’t know about any GameCube games with any noteworthy AI, so let’s take an upcoming PS2 title, which I think is also a PC game as well. Conflict Zone is a real time strategy much like Command and Conquer that is being developed by the Paris based firm MASA (Mathématiques Appliquées). You probably would have heard of them unless you are involved in military warfare but they are experts in AI. Early reports have shown that this game is staggeringly intelligent, and it is really noticeable. Every time you play it’s a completely new experience because the game actually adapts to the situation and changes its game plan to suit. It can predict what you are doing and counteract it. For example, if one of the troops strays into a minefield and gets killed, the unit remembers that and asks you for an alternative destination next time. Another example is you can give a set of instructions to a unit and it then begins to think for itself, carrying out your orders via the best route possible. However, if it’s a new recruit don’t expect it to last long, and more experienced commandos are less likely to attempt a stupid bravado stunt.
As for 4 and 5, well, I think at the moment we can only dream of things. An AI routine that builds on itself effectively learns new information and develops new skills. A computer that can learn is a truly outstanding achievement, as its abilities will know no limits. The ultimate AI though, is AI that thinks, AI that forms opinions, and AI that has its own opinions. If we develop a program like this we can truly say that it is alive, and imagine the fun we could have playing against that.
In many developer’s eyes, artificial intelligence is the ultimate creation. Its something that everyone wishes to make, and the code is worth millions. So what is it? Well basically, AI gives the computer the ability to make its own decisions, learn from mistakes, build on itself and more importantly have its own opinions and feelings. AI plays a large part in Sony’s plan for the future, as its main selling point is the Emotion Engine, and its supposed ability to simulate AI.
But ask yourself this: Is AI already upon us? In a way, yes. As I write this, I am witnessing a very simple form of this. Word 2000 has the ability to learn from its mistakes. For example, the first time you type “PSone” into word, it will decapitalise “S”. If you correct it a few times, it no longer does this. Word is learning that you don’t like this, and adapting. OK, let’s make a list of AI features:
1. Learn from mistakes
2. Make decisions
3. Adapt
4. Build on itself
5. Have it’s own feelings and opinions
Right then; scratch 1. What about 2? Well, decision-making routines have been in games for yonks. Let’s look at Tekken. According to its situation, Tekken will choose between high, low, hand, kick, flying, special and combo attacks whilst you are fighting. It assesses the situation then makes a decision and executes it. This is a very basic routine as don’t forget it is running on a PlayStation here, which is essentially a 66mhz. Let’s look at a higher form. In games which has multiple routes such as SSX on the PS2 or Speed Freaks on the PSone, the computer has to decide which route to take based on: Risk according to mistake parameters and advantage gained in relation to power-ups. If the computer decides to go for it, it will. Games that do not have a fixed route such as kart racers have to make very complex path finding mechanisms every second, which is a testament to modern technology. OK, decision making’s current highest level: The First Person Shooter. When we play a FPS we usually stay to a basic plan, such as: Keep moving, shoot people when opportunity arises, kill Sniper quick before he kills you, and stay near the BFG. We are constantly running these thoughts through out head and we make split-second decisions that eventually boil down to a list of commands. Forwards or back, left or right, shoot, jump or change weapon. It’s the same with a FPS AI routine. The computer assesses how many people are near, threat imposed, success probability and other things such as terrain and power-ups. It then generates commands. These routines are very sensitive however and that’s why movement can be erratic.
OK, now we’ve covered decisions, let’s look at adaptation. Adapting in videogames is also not a new thing. The PSone game Speed Freaks supposedly learnt your driving styles and adapted to your driving styles. I never really noticed this but the reviewers said they could. Essentially, this means that if the computer recognises that you aggressively barge people out of the way, it will stay away from you. If it notices you are targeting a particular character, it becomes aggressive itself. The computer analyses your actions then basically recompiles its AI routine to compensate. Therefore, each game will be different. Of course, for this to be any good you need a sophisticated piece of hardware, and this is where the nest generation console comes in. Now, I don’t know about any GameCube games with any noteworthy AI, so let’s take an upcoming PS2 title, which I think is also a PC game as well. Conflict Zone is a real time strategy much like Command and Conquer that is being developed by the Paris based firm MASA (Mathématiques Appliquées). You probably would have heard of them unless you are involved in military warfare but they are experts in AI. Early reports have shown that this game is staggeringly intelligent, and it is really noticeable. Every time you play it’s a completely new experience because the game actually adapts to the situation and changes its game plan to suit. It can predict what you are doing and counteract it. For example, if one of the troops strays into a minefield and gets killed, the unit remembers that and asks you for an alternative destination next time. Another example is you can give a set of instructions to a unit and it then begins to think for itself, carrying out your orders via the best route possible. However, if it’s a new recruit don’t expect it to last long, and more experienced commandos are less likely to attempt a stupid bravado stunt.
As for 4 and 5, well, I think at the moment we can only dream of things. An AI routine that builds on itself effectively learns new information and develops new skills. A computer that can learn is a truly outstanding achievement, as its abilities will know no limits. The ultimate AI though, is AI that thinks, AI that forms opinions, and AI that has its own opinions. If we develop a program like this we can truly say that it is alive, and imagine the fun we could have playing against that.
Page:
Well done Turbo, you won a game for that. Wasn't it you who was gagging for GT 3?
Hm, AI just has different algarythms programmed into it. The more variety there is, the more 'real' they are. What we realy need to do is program AI that can program itself!
It may seem weird, but thats what humans do. they 'think' for themselves. We make up new things each time.
If AI could be told to follow a certain set of 'rules' to program new tactics into itself, we'd be on the way to good enemies!
Unfortunately, this is around 5 years away :(
It may seem weird, but thats what humans do. they 'think' for themselves. We make up new things each time.
If AI could be told to follow a certain set of 'rules' to program new tactics into itself, we'd be on the way to good enemies!
Unfortunately, this is around 5 years away :(
Me neither...
Ah, but bots in shoot em ups aren't really AI. They don't think, they just act. The idea of 'tactics' isn't there - they're just made better by increasing accuracy.
Notice how perfect sims always run down the center of corridors.
Notice how they always remain in a doorway when they open it.
I could go on, but you get the idea.
I've always thought that being 'good' at games, was the ability to find these loopholes in the AI, and exploit them. Maximise your advantages.
On various games, you've always compteted with computer opponents who were better than you - but were any of them better tacticians than you? No, very unlikely.
Decent AI is still a long, long way off. Unfortunately.
Notice how perfect sims always run down the center of corridors.
Notice how they always remain in a doorway when they open it.
I could go on, but you get the idea.
I've always thought that being 'good' at games, was the ability to find these loopholes in the AI, and exploit them. Maximise your advantages.
On various games, you've always compteted with computer opponents who were better than you - but were any of them better tacticians than you? No, very unlikely.
Decent AI is still a long, long way off. Unfortunately.
Your Honour wrote:
However, I don't want to see AI get too good. Whats the
> point in playing a gane when the enemy is so good you never have a
> chance? It needs to be realistic, especially in driving games. You
> want to see the opposition try to hard and mess up a corner etc,
> otherwise it's unrealistic and boring.
I know exactly what you mean, does anyone else think the perfect sims in perfect dark are beyond human? i mean they never miss, if perfect accuracy is a human trait, i must have been in the wrong queue when they were handing it out.
However, I don't want to see AI get too good. Whats the
> point in playing a gane when the enemy is so good you never have a
> chance? It needs to be realistic, especially in driving games. You
> want to see the opposition try to hard and mess up a corner etc,
> otherwise it's unrealistic and boring.
I know exactly what you mean, does anyone else think the perfect sims in perfect dark are beyond human? i mean they never miss, if perfect accuracy is a human trait, i must have been in the wrong queue when they were handing it out.
I really ought to post a well informed reply on AI, but for some reason I just don't feel the desire to...
Yeah, well done Robin! 'Bout time!
Dringo wrote:
> Really well maybe it should!
Hint hint!
Do you think they got the hint?
> Really well maybe it should!
Hint hint!
Do you think they got the hint?
Well, I wouldn't say "about time too" but it was just in the nick to get GT3!
I think I'm going to get some anti-deflatory tablets delivered to my house now!
I think I'm going to get some anti-deflatory tablets delivered to my house now!
Congrats on your GAD win turbonutter, about time too!
Page: