When will we finally get an AI that can challenge us in games without resorting to cheating?
Video games are like sports. You can't create a perfect AI for video games because the human factor plays too strong a factor.
>>330212697
What about for strategy games like Civ 5 or Xcom?
>>330212754
I think resource based games are more fair because the Human and the AI are given equal amounts of resources. Or at least they should be.
>>330212954
Resource based games regularly cheat to keep up with the human.
>>330212635
>without resorting to cheating.
Where do you draw the line of cheating though?
>How does AI hear without cheating?
>How does AI anticipate recoil without cheating?
>How does AI change tactics after analysing an opponents tactics without cheating?
>>330213053
They get the same amount of information that the human player gets.
>>330212635
About same time when AI creates whole game plot on fly from developers instructions.
>>330212635
If AI wants to really win, it needs to predict and take risks (understand what it's doing) instead of only relying on instant reaction times and huge preprogrammed memory.
>Baduk
Whoooooo gives a fuck? Ever since the phone ai beat a 4th Dan Baduk has been a damned joke.
>>330213091
So what if there's a gunshot and the AI is just within the range of the noise? Do they auto detect it as long as they are in range, even though a human might not be able to detect the slight noise? Maybe each AI gets a regenerated "hearing factor" that dictates how soft of a noise they can hear?, I mean how do you handle this?
>>330213151
Currently GO is not a solved game so the AI is taking risk to bet the player. Or at least calculated risk.
>>330213209
Yes. The AI should respond to any stimuli that a normal human player would.
>>330213209
Fuzzy logic
>>330213005
The alternative to an adjustable resource advantage is adjustable AI intelligence, which is kind of unreasonable. The real problem with games like Civ is that the AI is too stupid to pose a challenge without considerable advantages.
>>330213225
GO perhaps, I was talking about games in general.
>>330213273
how do you determine the norm though?
>>330213225
I think he was referring to video games (and specifically fighting games), not Go, unless there's a version of Go where reaction speed is a factor.
>>330213417
Average human being of course.
>>330213275
You mean like the AI can be not sure if it heard something or not?
>>330213209
They should hear it perfectly. The point of these "AI vs human" competitions is to see who is better, a machine is inherently better certain things so if you deliberately hold it back all you get is "human vs simulated average joe" at best.
>>330213446
I would assume the first step would be to make such an AI for turn based games before you improve it for the use of real time games.
>>330213492
Average human being is bad at video games
>>330213949
average human being is actually average at video games
It'll probably be a while before we get quality AI again
Lets wait for the newest vidya fad to cycle back to "great AI" and then we'll get results
>>330214065
>>330212635
When the paying majority wants to be challenged in their videogames.
Which is never.
http://www.youtube.com/watch?v=xPrB8jJ4oGU
Even played against the motherfucker a few times.
He's certainly not as dangerous as really good players but holy shit he crushes scrubs.
>>330214065
It would depend on how you take the average. For example, imagine there was a test given to 20 people. 19 of them get 100%, and 1 of them gets 0%. That would mean the average person has an above average test score. I wouldn't be surprised if the average human was below average at video games.
>>330214307
>holy shit he crushes scrubs.
that's not hard at all really, even in assfaggots hard bots destroy baddies, when i try out builds and shit in dota i see plenty of people that just can't deal with the bots unless they cheese the ai
>>330214213
The last game that had good AI was FEAR. When will the cycle come back?
>>330213643
>The point of these "AI vs human" competitions is to see who is better
Not always the case though. Often enough the point of AI is not to do something as efficiently as possible, but to be as human-like as possible.
Remember that competitive players' skills often efectively deteriorate when they exclusively train against AI opponents. That's because the human players they will end up playing against on a competitive basis will not behave like the AI. To have a hyperefficient AI enemy doesn't really help in this case.
>>330213151
genetic algorithm
>>330214361
>imagine there was a test given to 20 people. 19 of them get 100%, and 1 of them gets 0%. That would mean the average person has an above average test score
didnt know you can challenge the AI i Pokemon Go
>>330214361
thats not how averages work
>>330214631
> Median
> Mode
>>330214361
I don't think so. Average is always average. Provided that the test is possible to be completed with a test score from anywhere between 0 and 100, the average in this case would be around 95%, right?
That means that when you introduce a new person into the test routine, judging from the average you can assume that he might score a 95% as well. Just because every person tested so far scored either a 100% or a 0% and nothing in between doesn't mean that
1) you can ONLY score either 0% or 100%
2) the average is not still the average in its original meaning.
Basically, the only way for the AVERAGE human to NOT have an average value would be by having the value average include more than just the human average. Since videogames are pretty much only played by humans, the average human will be average at videogames.
And, by extension, the average vidya enthusiast will most likely be above "human average". But he will be exactly on "vidya enthusiast average".
Final thought, "average" human beings, or anything "average", is always a thought-out concept. The average person can have like 1.2 kids in a certain country, but they won't have one fifth of a baby lying around in the closet.Well, some people might. That's not the average though.
>>330214835
>has no place in the previously stated argument
>>330214852
The average human also has 1 testicle, 1 breast and less than 4 limbs
>>330214737
>>330214631
Why not? That's exactly how they work. Its more of an example about why there is also a median and a mode, because sometimes the mean is a shitty metric to use.
>>330214852
That doesn't mean you can expect the average test score to be a 95%. What if it was a 1 point test? Averages aren't some magical thing that warp reality around them -- they are a mathematical tool to help analyze data. All I was trying to show is that its fully possible for the vast majority of a population to be above average at something, and vice versa.
>>330215089
And here are more examples about why just taking the average can be a shitty way of analyzing things. It takes a bit more thought than that.
>>330215132
There's a difference between median and average
>>330215217
What's your point? Are you saying the taking the mean of something is "right" way to get an average? The median is just as much the average as the mean is. There are multiple equally valid ways to take the average of something.
anyone else get the urge to rewatch hikaru no go?
>>330212635
The DeepMind developers want to develop an AI for Starcraft after it masters Go.
>>330215424
Starcraft:BW or Starcraft 2?
>>330215089
In what country, m8? Mexico or Colombia?
>>330213053
>>330212635
>>330215492
They didn't specify. Just "Starcraft".
>>330215492
https://twitter.com/deeplearning4j/status/706541229543071745
i assume this is where that anon saw that. the tweet is a quote from a video but the video is like 5 years old so idk if they're just tweeting this because they're going ahead with it or if they're just doing filler. or maybe there's something else entirely i havent seen about it.
looks to be just SC in general though which probably means SC:BW, since that already has a big AI scene.
>>330215424
I don't think it should be too difficult to make one that can beat top players if you let it use all of the inherent advantages of being an AI.
What I mean is perfect micro while also perfect macro and perfect map awareness - those alone have a huge impact on the game.
I'd much rather see them develop something for a more complex game that doesn't rely nearly as much on mechanical skill.
