Yes I've read that NNs are good in pattern recogntion etc! But I was thinking that also they would be used for determinating monsters' "Life expectancy" let me explain : a monster can evaluate its life expectancy basing on his characteristics and on the player to fight. I think it's a good think for monster re-evaluating this life expectancy (measured in second for example) basing its evaluation also on its experience (so maybe with NNs work) and, for example, learn from it! What do you think?

And what about artificial neural networks?

Don't. Just don't.

Don't. Just don't.

Ok, i trust you, but they seem so fascinating and suitable for evaluating monsters' life expectancy (basing also on their experience)! I'll try to study in deep how they work.. I'm at the beginning of video game AI study and I love it, but there are so many informations and design methods that it isn't easy taking decisions about it!

Ok, i trust you, but they seem so fascinating and suitable for evaluating monsters' life expectancy (basing also on their experience)! I'll try to study in deep how they work.. I'm at the beginning of video game AI study and I love it, but there are so many informations and design methods that it isn't easy taking decisions about it!

Since it is such a big field, you should heed the warnings and not waste time in irrelevant methods like neural networks and genetic programming. If you want to use your time well, learn about finite state machines, behavior trees and other scripting options, planners and utility-based systems, because all of those things are truly useful when programming game AI.

Since it is such a big field, you should heed the warnings and not waste time in irrelevant methods like neural networks and genetic programming. If you want to use your time well, learn about finite state machines, behavior trees and other scripting options, planners and utility-based systems, because all of those things are truly useful when programming game AI.

Understood. Thanks again!

NN and Genetics will just slow you down, since all they will provide you is what you're going to do during gameplay testing anyway i.e., tweaking your algorithm's variables and structure. You're smarter already than the best NNs and Genetics out there, so they will provide you almost no benefit. Also, HIGHLY recommend getting Dave Mark's book on behavioral mathematics for game AI (link in his signature), almost all of your answers are contained in there

NN and Genetics will just slow you down, since all they will provide you is what you're going to do during gameplay testing anyway i.e., tweaking your algorithm's variables and structure. You're smarter already than the best NNs and Genetics out there, so they will provide you almost no benefit. Also, HIGHLY recommend getting Dave Mark's book on behavioral mathematics for game AI (link in his signature), almost all of your answers are contained in there

Thank you, I'm purchasing that book on amazon :-)

You're smarter already than the best NNs and Genetics out there

not waste time in irrelevant methods like neural networks and genetic programming.

I don't think I agree with these sentiments, but the advice is good. Neural networks and genetic algorithms are among the most promising subjects in AI research. They are the only techniques with a good likelihood of producing real emergent behavior or machine learning. However, most implementations *are* unpredictable, training intensive, and unlikely to provide a whole lot of benefit in a typical rpg.
I would say if you want to look at something more advanced than finite state machines or scripts, the most fertile ground would be production systems. These are kind of (not really) like a large collection of interrelated if-then-else statements. The idea is to build up a collection of productions (sometimes called rules) that represent all of the 'mental' considerations of the AI.
In my opinion, this type of representation is much more authentic than a real-time plastic method like a GA or NN. Consider, during the course of a battle, it is pretty unlikely that a monster will be learning and incorporating a whole lot. It seems a bit more realistic to say that a monster or enemy has a lot of knowledge and experience (in the form of productions) that it brings into the battle, but that set of knowledge doesn't necessarily change a whole lot.

I am SOOOO not saying anything here.

I don't think I agree with these sentiments, but the advice is good. Neural networks and genetic algorithms are among the most promising subjects in AI research. They are the only techniques with a good likelihood of producing real emergent behavior or machine learning. However, most implementations *are* unpredictable, training intensive, and unlikely to provide a whole lot of benefit in a typical rpg.
I would say if you want to look at something more advanced than finite state machines or scripts, the most fertile ground would be production systems. These are kind of (not really) like a large collection of interrelated if-then-else statements. The idea is to build up a collection of productions (sometimes called rules) that represent all of the 'mental' considerations of the AI.
In my opinion, this type of representation is much more authentic than a real-time plastic method like a GA or NN. Consider, during the course of a battle, it is pretty unlikely that a monster will be learning and incorporating a whole lot. It seems a bit more realistic to say that a monster or enemy has a lot of knowledge and experience (in the form of productions) that it brings into the battle, but that set of knowledge doesn't necessarily change a whole lot.

Thank you for your post. Different ideas are only good things!