@Satoru said:
@IronParsley212 said:
You are wrong in multiple levels, AI simulation is in fact extremely easy for almost any computer nowadays
You seem to be confused that ‘making’ a simulation is the equivalent of making a GOOD AI. These are entirely different problems. Running a simulation is not the problem. Its running an AI that humans think is making good rational decisions. And within a timeframe that is acceptable. And with totally limited resources on what could be a Pc that resembles a toaster
Making a GOOD AI is creating the AI, while simulating the AI is running that AI. I did not say anything about complexity of this AI, it can far exceed even the best players decision making abilities. AI algorithms, especially on a game where AI has direct access to data (No image processing required) , takes very few CPU resources. Unless we are talking about intel pentium level processors.
Only problem with AI is the time it takes to code it , which shouldnt be an issue to microsoft. There is no point of mentioning cpu cycles as core logic of games almost never use more than 10 percent of a single thread , not the mention AI can quite easily be multithreaded.
I’m not sure where you’re pulling this ‘10 % of a single thread’ from. I can assure you something like Alphago or watson is not using '10 % ’ of the custom ASIC or server racks.
3 years of game development experiance, its not that long but i still have pretty good understanding of how hardware resources are utilized in games. And AI is not part of the problem.
(There are sub-problems that are connected to AI, such as Pathfinding, which can put a serious load on CPU if not coded properly, but thats another topic)
Alphago and Watson are Neural Networks they are not hardcoded AI. No game, as far as i know, utilizes Neural Networks for their AI. And it would be extremely stupid do to so right now as Deep Learning is a field that is way too dynamic, the most effective and frequently used function is in fact merely 2 years old, being found in 2015.
You are comparing two, completely different forms of AI. And you are trying to win an argument thats about hardcoded AI by making referances to NN AI.
The thing that is on the “size of an apartman complex” is the hardware that _creates _ the neural network. After it is successfully created, neural network itself can be run on any computer with ease. An easy way to understand this situation is comparing rendering movies to watching movies on your PC. It takes hundreds of computers to render a movie and it takes months-long time, yet your computer can easily watch even 20 movies at same time.
Render calls BY FAR take more cpu cycles than simulating even the most complex AI does
This is actually part of the problem. Because rendering now sucks up more cycles there is less cycles for the AI to process the environment, meaning it has less resources to work with
This is not really part of the problem, Render calls are only done in main thread, AI can be run in any thread, and considering every CPU now has more than single thread, bottleneck certainly is going to be on the main thread.
AI hasnt been a hardware resouce problem in past 5 years, and im talking about AI that are far more complicated that like of which is going to be put on this game. It is not a problem and i have no idea why you are making it a problem.