If anyone could answer me on this: Does any form of example exist, where a game company made a learning system which took information from players as they were playing, and slowly ramped up difficulty based around the way a player plays?
I cannot think of any examples from the top of my mind, but it seems that these days, companies like Valve would be able too implement a "learner" system on random competitive games, and use it too create a bot which acts like a player.
[QUOTE=JoeSkylynx;52864215]If anyone could answer me on this: Does any form of example exist, where a game company made a learning system which took information from players as they were playing, and slowly ramped up difficulty based around the way a player plays?
I cannot think of any examples from the top of my mind, but it seems that these days, companies like Valve would be able too implement a "learner" system on random competitive games, and use it too create a bot which acts like a player.[/QUOTE]
Don't the Amiibo in Smash 4 function like that? I'm pretty sure they mimic your fighting style, and if you do stuff like taunt after every kill, they learn to taunt after kills, or if you always go for meteor smashes, they go for meteor smashes more often.
I can't think of any other examples at the moment, though.
[QUOTE=JoeSkylynx;52864215]If anyone could answer me on this: Does any form of example exist, where a game company made a learning system which took information from players as they were playing, and slowly ramped up difficulty based around the way a player plays?
I cannot think of any examples from the top of my mind, but it seems that these days, companies like Valve would be able too implement a "learner" system on random competitive games, and use it too create a bot which acts like a player.[/QUOTE]
the drivatars in forza learn from your driving style, so if you're more aggressive on the track your drivatars starts becoming more aggressive too
There are plenty of games that adjust the difficulty based on your performance. Resident Evil 4 is one of the better known examples.
[QUOTE=TheGoodDoctorF;52864274]There are plenty of games that adjust the difficulty based on your performance. Resident Evil 4 is one of the better known examples.[/QUOTE]
Does it do it in an interesting way or does it just go "You haven't died in a while, have some more enemies"?
I don't believe any game uses a neural network to do this though.
[QUOTE=JoeSkylynx;52864215]If anyone could answer me on this: Does any form of example exist, where a game company made a learning system which took information from players as they were playing, and slowly ramped up difficulty based around the way a player plays?
I cannot think of any examples from the top of my mind, but it seems that these days, companies like Valve would be able too implement a "learner" system on random competitive games, and use it too create a bot which acts like a player.[/QUOTE]
I mean the AI Director in Left 4 Dead (2) sorta does something like this. It ramps up or chills out on difficulty based on how the players are doing. I don't think it's overly sophisticated, though.
[QUOTE=Karmah;52864337]I don't believe any game uses a neural network to do this though.[/QUOTE]
Neural networks like this are relatively new. We'll probably see the kind of game he's describing pretty soon.
[QUOTE=Karmah;52864337]I don't believe any game uses a neural network to do this though.[/QUOTE]
I'm fairly certain some of EA's titles use pretty basic ones.
Neural networks aren't some kind of space-time continuum machine, they're a pretty simple, effective, and cool algorithms that don't take up many resources while they're not in either data collection/processing/backprop mode.
Just because you haven't played a game that has really crazy good AI doesn't mean you haven't played against a neural network yet. It just means you haven't played against any real advanced ones.
[editline]7th November 2017[/editline]
I design GAN's and other nns as a hobby, it's really not that complex once you understand what's going on.
[img]https://www.pyimagesearch.com/wp-content/uploads/2016/08/simple_neural_network_header.jpg?width=600[/img]
This image explains exactly how they function. This is essentially all a neural network is. Every circle and line represents only a singular number.
[editline]7th November 2017[/editline]
[QUOTE=TurtleeyFP;52864452]Neural networks like this are relatively new. We'll probably see the kind of game he's describing pretty soon.[/QUOTE]
The video is using the most basic form of neural network, which has been in use for about a decade now. Since then we've developed significantly more advanced nns, such as GANs, which are hilarious imo.
[QUOTE=JoeSkylynx;52864215]If anyone could answer me on this: Does any form of example exist, where a game company made a learning system which took information from players as they were playing, and slowly ramped up difficulty based around the way a player plays?
I cannot think of any examples from the top of my mind, but it seems that these days, companies like Valve would be able too implement a "learner" system on random competitive games, and use it too create a bot which acts like a player.[/QUOTE]
A lot of games claim to do this, they don't do it through neural networks or machine learning.
Both techniques require sample sizes in excess of tens of thousands of trials to even get borderline acceptable results.
Any games that say they 'learn from the player' or 'adjust to you your playstyle' aren't using AI, but are either
1) Turning the difficulty up if it detects you're doing well (unreal tournament bots)
2) Has pre-scripted alterations that it makes based on certain actions you perform (I think MGS5 does this where if you headshot a lot enemies wear helmets etc)
3) Observes your current state and adjusts it based on directions from the developer (HL2 ammo crates always give you ammo you need)
4) The developer is lying and you just *think* the game is changing to beat you.
None of these are really AI, more pre-defined rules to make the game change its behaviour if it detects you doing certain things. None of are really learning since the responses are already pre-programmed in.
Didn't alien isolation have an ai that sort of learned from your gameplay style?
[QUOTE=omarfr;52864814]Didn't alien isolation have an ai that sort of learned from your gameplay style?[/QUOTE]
I believe it was in vein of "fool me once" at most since if you used a certain way to escape it one time, it'd check there one if the next times it hunts you. Or better yet, it already knows you're there and it's waiting for you to come out for the kill.
[QUOTE=JoeSkylynx;52864215]If anyone could answer me on this: Does any form of example exist, where a game company made a learning system which took information from players as they were playing, and slowly ramped up difficulty based around the way a player plays?
I cannot think of any examples from the top of my mind, but it seems that these days, companies like Valve would be able too implement a "learner" system on random competitive games, and use it too create a bot which acts like a player.[/QUOTE]
There's this old gem:
[t]http://i.imgur.com/dx7sVXj.jpg[/t]
[QUOTE=squids_eye;52864332]Does it do it in an interesting way or does it just go "You haven't died in a while, have some more enemies"?[/QUOTE]
They adjust drop rates, drop types and also change how enemies behave depending on how well/bad you're doing.
If you're doing well the enemies will drop less ammo and healing items; but will drop more money. They will also run towards you a lot quicker, try to flank you and also use throwing weapons better.
If you're doing poorly they will do the opposite, so enemies are a lot slower/straight up stupid; also it will remove enemies from certain gauntlet areas.
[QUOTE=JoeSkylynx;52864215]If anyone could answer me on this: Does any form of example exist, where a game company made a learning system which took information from players as they were playing, and slowly ramped up difficulty based around the way a player plays?
I cannot think of any examples from the top of my mind, but it seems that these days, companies like Valve would be able too implement a "learner" system on random competitive games, and use it too create a bot which acts like a player.[/QUOTE]
MGSV was supposed to work like this, but it I don't feel that it got properly implemented.
The idea was that the enemy soldiers would take note of what tactics you used and react accordingly. Pull off too many headshots and the enemies will wear more helmets, go in at night every mission and they start wearing NVG's..
The method for countering this was to attack their shipments of supplies so they wouldn't have the resources to counter specific methods of attack, but at least in my experience it really does fuck all and just costs a lot of resources. So in my game all the enemies are so equipped that stealth is no longer a reliable tactic.
[QUOTE=JoeSkylynx;52864215]If anyone could answer me on this: Does any form of example exist, where a game company made a learning system which took information from players as they were playing, and slowly ramped up difficulty based around the way a player plays?
I cannot think of any examples from the top of my mind, but it seems that these days, companies like Valve would be able too implement a "learner" system on random competitive games, and use it too create a bot which acts like a player.[/QUOTE]
The earliest example which is like this is Zanac (1986) by Compile. The better you're doing, the more enemy ships appear. The type of enemy is also dependent on your equipped gun. Of course this isn't really learning, but I don't know of any earlier game which attempts a similar idea.
There is a neural network style space invaders game where the enemies develop skills and abilities to counter the player as each level/stage is completed. I forgot the name but I think it was a university project.
I found out about a game called [I]Creatures[/I] that was first released back in 1996. I haven't played it myself, but the level of simulation for the in-game creatures seems quite sophisticated, if [URL="https://en.wikipedia.org/wiki/Creatures_(video_game_series)#Species"]the Wikipedia page[/URL] is anything to go by:
[QUOTE]In the games, the Norns have their own simulated biochemistry, haploid genetics, and neural network to serve as a brain. From Creatures 2 onwards, Norns also possess organs determined by their genetics; mutation and injury can render these defunct, with typically fatal consequences. Each of these has grown somewhat over the course of the series, allowing the creatures to become more complex and intelligent.[/QUOTE]
I loved Creatures when I was a kid. Didn't really know what I was doing at the time, but it was fun
[QUOTE=JoeSkylynx;52864215]If anyone could answer me on this: Does any form of example exist, where a game company made a learning system which took information from players as they were playing, and slowly ramped up difficulty based around the way a player plays?
I cannot think of any examples from the top of my mind, but it seems that these days, companies like Valve would be able too implement a "learner" system on random competitive games, and use it too create a bot which acts like a player.[/QUOTE]
MGS V had a simple system where enemies would act and be equipped differently depending on playstyle. (Lots of headshots, and helmets would be more common)
There's also [URL="https://www.washingtonpost.com/news/comic-riffs/wp/2017/09/22/in-the-smart-and-slick-new-game-echo-you-become-your-own-worst-enemy/"]Echo[/URL], which just came out, where the whole game is basically based around enemies copying your behavior. It's nothing particularly advanced though, bots just aren't allowed to do things like vaulting over obstacles if you haven't done them yourself.
edit: didn't see MGS had been mentioned twice already
Sorry, you need to Log In to post a reply to this thread.
