More Intelligent Enemies

[obscurus]

Eventually you learn it's really easy to say "forget the rules!" but much harder to make new ones that actually work.

 

 

Rubbish. You just have to ditch your preconceptions about what a computer game can or should be. If all people thought the way you do, there would never be any innovation and we'd all still be chasing antelopes with rocks and digging for tubers.

 

The easy way is the boring way - it has been done before. Take the hard path, it is infinately more rewarding in the end.

 

 

I am all for adding unpredictability to AI, and I can't think of any technical reason why it can't be done. It would create a completely new gameplay dynamic, one that I look forward to eventually experienceing, though I don't expect it from TDM. But then I'm not an AI coder, so I guess I'll just have to take what I can get...

[Ishtvan]

Haven't we been saying we plan to add more unpredictability to the AI all along? The hard part is mostly balancing, like balancing the fun you get from having AI that act more human with the frustration you get from planning something out for 5 minutes, then getting caught and failing because one of the AI happened to look over their shoulder, or because they suddenly change their patrol route. Remember that in first person, there is no leeway where you can see the AI and they can't see you. The whole time you're watching the AI, they can look at you too, so randomness can be very unforgiving.

 

As realistic as it is, no likes making a plan based on careful observation and then getting caught due to random chance. I'm not saying there should be no randomness, we just have to be careful with it.

[NeonsStyle]

Haven't we been saying we plan to add more unpredictability to the AI all along? The hard part is mostly balancing, like balancing the fun you get from having AI that act more human with the frustration you get from planning something out for 5 minutes, then getting caught and failing because one of the AI happened to look over their shoulder, or because they suddenly change their patrol route. Remember that in first person, there is no leeway where you can see the AI and they can't see you. The whole time you're watching the AI, they can look at you too, so randomness can be very unforgiving.

 

As realistic as it is, no likes making a plan based on careful observation and then getting caught due to random chance. I'm not saying there should be no randomness, we just have to be careful with it.

 

I agree with that completely. It would be as frustrating as playing against a human, even a bad one. However, that is exactly the point. AI are meant in a simplified way, to simulate a guard, and as such, you need to be on your toes all the time with an element of unpredictability in his framework. Sure, you can carefully craft your plan, but you also need to develop a degree of flexability, so as to make changes to it at a moments notice if/when he turns his head.

 

In that framework, it rewards you for accurately behaving theif like. In this game, thief has to be quick witted, and agile. So it shouldn't be any problem for a human.

 

Venus

[Domarius]

Strawman argument

Hope you feel better now. Have a re-read after you've cooled off and you'll see that I didn't say innovation was impossible. And try to be more polite. Your unessecary rudeness has reminded me why I don't frequent these boards anymore.

 

Venus, you have good intent, but you're not saying anything new. "Game theory" doesn't tell us to make dumb AI, but we see this in games anyway because there are practical limiations on development time, budget, and technology.

 

Of course we are working towards making the AI as human and as interesting to play as possible. Our AI actually has some of the things you're suggesting.

 

Yes humans are random and aren't entirely predictable, but I don't think humans are really that unpredictable when they're going about their routine. No one is consitantly vigilant forever. If someone picks up a book and sits down to read it, after a short time you can assume there's a good chance they'll be there for a little while - and not going to suddenly start looking left and right like they expect to be attacked at any moment.

I remember someone here who had experience in the army posted about how people in real life on patrol, pretty much have the same predictablitity as a Thief AI. They might stop and look out a window and stuff like that, but not much variation. Imagine doing it all day every day. You're not going to keep looking behind you randomly.

Of course once you've alerted someone, it's a different story. They will be looking around randomly. And we are planning to do this also.