Pac-Man recreated with a GAN trained on 50k game episodes • Hacker News Board

Re: Pac-Man recreated with a GAN trained on 50k game episodes

#31

Post by madenine » Fri, May 22, 2020, 2:37 PM UTC

cooljacob204 wrote: post #20

QuixoticQuibit wrote: post #16

This research reminds me of one of their April Fool’s videos [1] with a gaming USB stick that you plug into your computer to offer various gaming enhancements. One of which is a “Ghostplay” feature where an AI program can mimic you after training on your gameplay. I remember that video because it always struck me as not being far off from reality. It’ll make the current set of online game cheats/hacks seem like child…

Servers are already running AI to try and counter cheating in quite a few games.

2018 video on Valve's use of deep learning for anti-cheat in CSGO https://www.youtube.com/watch?v=ObhK8lUfIlc

Re: Pac-Man recreated with a GAN trained on 50k game episodes

#32

Post by lonelappde » Fri, May 22, 2020, 2:41 PM UTC

__s wrote: post #15

ekianjo wrote: post #13

> Trained on 50,000 episodes of the game Meanwhile an actual human will probably need only about 5 games to fully understand how the game works and reacts.

I wouldn't say "fully" https://www.gamasutra.com/view/feature/3938/the_pacman_dossi... & also that humans come with a huge amount of training beforehand. We've experienced eating & ghost stories & moving around in 3d spatial dimensions before playing pacman usually

OTOH if you told a current AI a ghost story it wouldn't be able to do anything useful with that info a pacman context.

Re: Pac-Man recreated with a GAN trained on 50k game episodes

#33

Post by hola123456789 » Fri, May 22, 2020, 2:42 PM UTC

hola

Re: Pac-Man recreated with a GAN trained on 50k game episodes

#34

Post by lonelappde » Fri, May 22, 2020, 2:43 PM UTC

gallerdude wrote: post #30

I wonder what the copyright implications of this are. Obviously it's Pacman, but none of the source is the same.

Game rules aren't copyrightable. The visuals are a copy but obviously fair use, and it's obviously trademark use, and Nvidia probably has a marketing partnership with NAMCO for this

Re: Pac-Man recreated with a GAN trained on 50k game episodes

#35

Post by lonelappde » Fri, May 22, 2020, 2:44 PM UTC

coldpie wrote: post #10

itchyjunk wrote: post #8

Thought experiment: Is my brain also generating reality without a physics engine and is filling in details that it thinks should be there? If true, what are the advantages/disadvantages of it?

Yes, absolutely. You can catch a baseball because your brain predicts where it will go before your eyes see its actual movement. This may be a good place to start learning more: https://en.wikipedia.org/wiki/Direct_and_indirect_realism

How is that "not a physics engine"?

Re: Pac-Man recreated with a GAN trained on 50k game episodes

#36

Post by andr » Fri, May 22, 2020, 2:46 PM UTC

It would be super interesting to see if this knowledge generalizes with transfer learning. For example, after seeing 50,000 episodes of PacMan, would the GAN be able to recreate Space Invaders with just 5,000 extra episodes?

Re: Pac-Man recreated with a GAN trained on 50k game episodes

#37

Post by zeroxfe » Fri, May 22, 2020, 2:47 PM UTC

CShorten wrote: post #4

I made a video explaining the details of this paper if you are interested! https://youtu.be/H8F6J7mYyz0

Thanks for doing this. I stumbled upon one of your previous videos a while ago and really liked it. I'm super indebted to YouTubers like yourself that take the time to distill and explain complex topics, papers, and new research.

Re: Pac-Man recreated with a GAN trained on 50k game episodes

#38

Post by jcims » Fri, May 22, 2020, 2:49 PM UTC

flixic wrote: post #9

Interesting to explore how this can be used. You still need to develop a game with an actual engine first, to generate all the samples for training the AI how to generate frames without an engine. But maybe if training sample size can be considerably reduced, game making can open up to far less technical people. Instead of developing with an engine, game makers could show the AI "when I press this button, change the…

Seems like it's basically simulating an environment which it has agency to affect. To the extent that games are simulating some universe, it seems like this could also be applicable for simulating at least one universe that has a game engine we can't quite reproduce yet.

Re: Pac-Man recreated with a GAN trained on 50k game episodes

#39

Post by jcims » Fri, May 22, 2020, 2:52 PM UTC

lonelappde wrote: post #35

coldpie wrote: post #10

Earlier quoted context omitted.

Yes, absolutely. You can catch a baseball because your brain predicts where it will go before your eyes see its actual movement. This may be a good place to start learning more: https://en.wikipedia.org/wiki/Direct_and_indirect_realism

How is that "not a physics engine"?

It's more general than a physics engine in that it's predicting an expected outcome...it just so happens that the outcome here is largely dictated by physics. (At some level they all are but some more obviously than others)

Re: Pac-Man recreated with a GAN trained on 50k game episodes

#40

Post by Ajedi32 » Fri, May 22, 2020, 2:55 PM UTC

QuixoticQuibit wrote: post #16

This research reminds me of one of their April Fool’s videos [1] with a gaming USB stick that you plug into your computer to offer various gaming enhancements. One of which is a “Ghostplay” feature where an AI program can mimic you after training on your gameplay. I remember that video because it always struck me as not being far off from reality. It’ll make the current set of online game cheats/hacks seem like child…

But if the AI is functionally indistinguishable from a human player (human-like and undetectable), does it even matter really? https://xkcd.com/810/

(Conversely, if they're _not_ functionally indistinguishable from a human player, then that means it's possible to write an AI or other algorithm to detect them.)