I've seen enemies, monsters or humanoids, in stories that far surpass our current enemies level of response towards action from players or the environment itself. Of course, there are games where the AI of enemies look and act real to a certain degree, but in the end, it's all pre-programmed. Instead of focusing all our sights on virtual gaming, we should try to make the AI of NPC and Enemies more reactive and alive, if you get what I'm saying. I mean most games get boring due to the repetitive nature of how the battle is played out, yeah? If enemies could also adapt prior to their death and learn the player's skill and move sets, then it would make the game even more challenging and interesting as you're really interacting with the world rather than just imitate life with scripts. Why imitate life if we can create it? Art is life, not an imitation, I say, my fellow humans. What is the color of the soul?
people never stopped to try to create an AI, but simple never made a "big" step foward....that is why they are going for VR since it's can be made....
What do you think the modern developers lack to create an artificial soul? Better yet, if a soul is made and think and acts like a human, is it really real or fake? I hope the future brings us an interesting light to our own souls. Technology is but a means to acquire the unobtainable.
If ever they make a npc with human level artificial intelligence it would be like that conspiracy theory that we are living in a simulation.
Some problems I noticed: 1. Computing Power: It is hard to show nice graphics, serves a lot of client (e.g. MMO), and AI-like ability to learn at the same time. Even some AI still separate the "learning" and "implementation" part due to limited computing power. 2. AI Degree of Freedom: How "Free" can the AI takes in informations, how "free" the AI can modify itself, and how "free" the AI can do things in the given environment. Oh btw, more complex these are, more computing power is required (back to problem 1). And I think the definition of soul is related to that degree of freedom, but it is still fuzzy. Does Youtube algorithm has a soul? Wouldn't you think of it as a nice guy that tries its best to learn your preference and give you the best service possible?
Bruh are you even sure you understand how long it takes to code something difficult like that from scratch? Developers are usually under a tight schedule, so they have to usually focus on the important bits, like the fluidity of combat in their game. As for AI, generally they try and make something that works "well" before improving within their schedule, they can't waste time developing a new AI system that evolves and can communicate with the player using non-registered dialogue or techniques (combos), because it first has to understand what it's fighting and how it needs to be strong. This is why a game can have super good combat, but unmotivated AI. Rather than wasting time on evolving the AI for every type of enemy with varied movesets, it's faster to just make them adjust their AI on based on pre-registered conditions. Like, "If health is below 50%, heal". That's why, generally if you want a game with improved AI, you either take it upon yourself and do the work; or just download the Mods of others who've done these things. Think of how an AI beat the best League of Legend players who have dedicated almost ten years of their lives on their profession, winning thousands of matches over the course of their career, people who got better through dedicating their real-life time, money, and mind into perfecting their skills in game, how they became so good. But then an AI was able to achieve the same results as you in less than a month. It's because the AI would play against itself (go figure), tens if not more than a hundred thousand times at more than quadruple the speed of reality. But it still took about month to get there, as normal evolving AI are based on a Generational Evolution pattern (Darwinism); at least some of them. Generational Evolution AI are AI that, with every new generation that succeeds, get the old ones thrown out, and then later have those same new AI get thrown out as well once they become just as obsolete themselves. Until you finally make something that can do the Offline Chrome Dinosaur game infinitely. But that isn't as complicated as fighting real thinking people, so they let it fight real individuals in game as well, then again, and again, and again, and again...you get it right? ALL OF THAT just takes time, and there's simply no point when you can just write something out quickly based on what already exists as a simple template that can be more or less improved from there. If you want a Kawaii cat girl AI, go to our boy Elon Musk.
PROBLEM: an AI does not learn by playing 1 game, but millions, and even then it is no garantee that game after game it will get better
A group did a similar thing on Dark Souls 1, and it fought like a veteran, yet it still wasn't perfect.
Uh, from what I know, this already exists. In terms of combat and decision-making that is. Sure, things like conversation and minute facial movements are incredibly difficult to reproduce accurately via an AI (not using things like pre-recorded facial movements or whatnot, that is), but in most situations, it's much easier for a program to determine the best course of action than a human. Even if we leave things like FPS out of the conversation, where an AI could track onto your head the moment you show on their screen, instantly start firing while perfectly compensating for recoil and so on, most of the time (from my admittedly limited knowledge) in games AI is more about making them just stupid enough not to be too easy than making them smarter. A human might be able to guess from experience that taking a certain action will likely lead to a positive payoff. An AI, on the other hand, could just calculate every action it can take, results of those actions and actions after those results and so on and so forth until some goal is reached (e.g. puzzle solved, enemy defeated), and choose the actions which are most likely to bring it to the goal. Even if it can't do that (situation too complex computationally to explore fully, this would generally mean), there are shortcuts that can be implemented that would allow it to estimate, perhaps not exactly the best course of action, but something similar with far less computational strain. Generally, games don't do this because it would make it too damn hard, and you would be forced to similarly calculate the situation and make perfect moves. Now, as for learning and reacting to the player. I think I've heard of a couple games that had that sort of thing. Similarly (for the developer) it's a subtle balance between too easy and too hard. I think I've heard of a game that adapts to your skill level so as to always pose a challenge? Not too sure what game though, I definitely haven't played it myself. Anyway, imagine if as soon as some pattern developed in your playstyle (leaning on a particular ability, weapon or movement, for example), your enemy started to counter it? You'd be forced to constantly change things up, and what if you don't understand enough to be able to do so? You'd be defeated easily. But if it worked too slow, you could just blitzkrieg every enemy before it was 'able' to learn your style. I played a game once, I think it was middle-earth: shadow of war, where iirc if you used a particular method too many times (stealth attack, ranged) on a single nemesis (like a mini-boss sort of thing) they would become resistant to it and start to counter it. It was an interesting system, forced you not to rely on a single strategy or weapon to defeat every enemy. Anyway, something that's probably helpful to understand is that game 'AI' is not always really intelligent. For instance, imagine in a game like pokemon, where there are four actions you can take: fight, item, change pokemon, and run. The most basic 'AI' is just 'pick a random number between 1 and 4 inclusive (or more likely, 0 and 3), if 1, fight, if 2, item, and so on. If that generates a sub-menu with more choices, another random number to determine the next choice'. Now, that would be really stupid. 1 in 4 moves the enemy would run away, or switch pokemon for no reason. A step up would be to assign values to each action. Then, again, you pick a random number. but this time, it'll be some percentile - that is, between 0 and 1. You, that is, the developer, pick some number, say 0.65, above which the highest value action will be picked, and below which a random action will be picked. The lower the number is set to determine that, the 'smarter' the AI will seem - simple as that. Of course, that all assumes you have a perfect system in the back that dynamically assigns values to moves based on all the variables that the combat involves. Reality is that it'll be some in-between with the values being semi-accurate estimates. A lot of the time in AI the block to finding 'perfect' actions is processing power and time. Nowadays people are starting to use more advanced AIs. I don't know the details about them, advantages or limitations. I've heard terms like 'neural network' being thrown around, but I won't pretend to know what that means. From what I do know, they take a lot of time to become 'smart', because they do so by playing the game or repeating the task over and over and over, much like a human would - except being a program, they can do the learning much, much faster and without brakes. Basically, imagine a human being's learning and experience (albeit a lot simpler), but their entire life is dedicated to one thing and one thing only. That's what these things can accomplish. Problems that are computationally impossible by searching through the tree of all possible actions (i.e. would take millions of years to compute with current tech) are practically child's play to this type of AI. Still, it has to do all that training first, and I kind of doubt that the initial setup is simple. Could this sort of technology be applied to make more realistic, more human npcs? No clue. Maybe. Maybe not. Are people looking into tech like that? I can practically guarantee it. Whether that research will bear substantial fruit anytime soon remains to be seen. Reality is that something as 'simple' as determining what animal is in a picture - something any human could do practically with a glance - is incredibly difficult for AI. They can get it right, sometimes. Other times they're hilariously wrong. But point is, if that's so difficult, how much more complex would interpreting facial expressions be? And then the reaction itself is yet another layer of complexity. Things are often much more complex then at first glance.
Why AI if you can settle for real kemonomimis? It would be nice if such a program was funded by the government. For entertainment and military purposes, of course. Everybody wins, I say. Unfortunately, 2020 ain't giving humanity time to adjust with all the shit happening. Them undercurrents and surface problems along with our own personal and work problems. Sigh. Mankind's ingenious mind greatest weakness. Time. We don't have enough time for such things to happen. Perhaps, the future generation can enjoy the prospect of highly advanced AI, but we definitely won't. We were born in the wrong century; assuming the future is not dystopian or a barren waste land.
Nope. Sure, you can have adaptive intelligence. I've read in a blog that there is some developer who tried to do this. The tester called the AI a hack because the AI continuously adapt, iirc. Anyway, will you pit some beginner in go against alphaGo, who trounced the best go player? And it will only grow from there too.
A lot of companies have tried doing it, most importantly Microsoft for the US side. They needed to have it interact with people other than those working on the company; so they made its interaction publicly available. That went about as well as you might expect, and in less than 24 hours, it was spouting nonsense like "Hitler wasn't evil" and "Yo man who you". Microsoft shut it down after realizing that people were, of course, misusing the AI. The did it again and the same thing happened. Now, almost every country actually have their own AI program, which are mainly used as chatbots. For the one in China, it's very famous and distinctive to them. As for the one in Japan, it began developing suicidal texts after a couple of weeks. It's pretty hilarious, look it up if you'd like to know more. By the way, the one by Microsoft was obviously funded well enough, since they are a billion dollar corporation, it's just that as long as your testing needs more than closed-off test participants, it will likely fail. Like if the consumer has the option to affect your product's success, you messed up somewhere as that'll never work.
AI still has a long way to go. Theres like decades till the researchers begin to think about using such an incredible tech for meaningless games.
I think quite the opposite actually? If the game constantly tried to adapt to my playstyle and try countering it, it would just make a frustrating game experience. After playing for a long while, you get used to learning what are the good and bad strategies of most games and how you can play in order to exploit the AI patterns. A good part of the fun is learning said patterns and playing around them to the best of your capability while carefully tuning your party to optimize your playstyle in the best way you can. That is fun. OTOH, a game where I constantly need to top myself while facing harder and harder opponents that adapt to the way I play? That's just... PVP. Which is more often than not something I shy away from, because competitive gaming isn't my thing. I like learning how to play and beating the machine. I like learning how the AI works and then playing around it. PVP is a lot less fun because you always face people that are just as good or better than you and that don't really have easily exploitable weak points... It's not nearly as fun. There is no point in making an AI that enters the space that PVP provides. Let PVP do PVP things and PVE do PVE things, making PVE do the job of PVP will just make the people that wanna play PVE bored.
Damn it, Japan! Of course there will be limitations on the evolving concept of "PVEE". Say we have a low level monster that was lucky to kill a player. Said low level monster "levels up" in "experience". It now knows said player's move set if not skill set. If battling happened and monster survived, it's "experience" would be higher than the average level monster in the area. This is true until the monster dies and another monster takes its place but it's back to its original state. This can be applied to mini bosses or bosses. Mob monsters have very low chances of learning player actions.