So wait, is this playable, or is it just capable of reproducing video of gameplay?
Pac-Man recreated with a GAN trained on 50k game episodes
111–120 of 154 posts
Re: Pac-Man recreated with a GAN trained on 50k game episodes
#112The point is this brings us closer to demonstrating AI which is in some sense grounded in reality, as opposed to some kind of “Clever Hans” or Searle’s “Chinese Box.” The network has learned how to “imagine playing out a scenario” much like a human player might use his own mental model of the game to consider various strategies to attempt, perhaps with the real game.
That the network apparently has a mental model which is rich enough to simulate actual game play implies that it hasn’t just fit a static caricurature of the game which immediately breaks on an unseen move. That is, it has a non-trivial “conceptualization” of its environment which goes beyond simple pattern matching or nearest neighbor matching.
Re: Pac-Man recreated with a GAN trained on 50k game episodes
#113I've observed an increasing trend with modern AI research that A) It's no longer practical to incorporate into real products given the hardware costs for training and inference. In NLP, large transformer based models can easily occupy a 5 thousand dollar GPU at a paltry 50-60 words per second processing pace. B) The Research often demonstrates things that existing non-ML toolchains were already reasonable at doing wi…
It reminds me of older electric cars. They worked. They could get you around. But they did so at higher price, slower speeds, or other factors that a consumer wasn't willing to pay for. Thus they were regulated to novelty items that some work campuses would have and concept cars that would be shown from time to time. But eventually we reached the point where they are not only competitive, they may be positioned to fu…
The word you are looking for is "relegate", not regulate.
Re: Pac-Man recreated with a GAN trained on 50k game episodes
#114How do they know it's fully functional without trying out how it reacts in every possible situation (which isn't really feasible)?
Re: Pac-Man recreated with a GAN trained on 50k game episodes
#115Does anyone know if we can combine knowledge from different domains when training to play games? Example: have a language model read pieces of text about the game, then feed the output of that into a model that's doing normal reinforcement learning. Maybe train them both at the same time?
The hope is that the model can combine inputs from different domains and generalize better.
Re: Pac-Man recreated with a GAN trained on 50k game episodes
#116I've observed an increasing trend with modern AI research that A) It's no longer practical to incorporate into real products given the hardware costs for training and inference. In NLP, large transformer based models can easily occupy a 5 thousand dollar GPU at a paltry 50-60 words per second processing pace. B) The Research often demonstrates things that existing non-ML toolchains were already reasonable at doing wi…
I don’t think that’s the point of bleeding edge research. Pushing boundries != commercialization potential
Is there research in this field that is not commercially funded? That seems like a silly distinction in America.
Re: Pac-Man recreated with a GAN trained on 50k game episodes
#117I've observed an increasing trend with modern AI research that A) It's no longer practical to incorporate into real products given the hardware costs for training and inference. In NLP, large transformer based models can easily occupy a 5 thousand dollar GPU at a paltry 50-60 words per second processing pace. B) The Research often demonstrates things that existing non-ML toolchains were already reasonable at doing wi…
It reminds me of older electric cars. They worked. They could get you around. But they did so at higher price, slower speeds, or other factors that a consumer wasn't willing to pay for. Thus they were regulated to novelty items that some work campuses would have and concept cars that would be shown from time to time. But eventually we reached the point where they are not only competitive, they may be positioned to fu…
The myth of progress has got to be the marketer’s best tool for getting at rich folk.
Re: Pac-Man recreated with a GAN trained on 50k game episodes
#118> 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.
and how many games does a human need to transform into a pacman game playable by another human? ...lol
Re: Pac-Man recreated with a GAN trained on 50k game episodes
#119How do they know it's fully functional without trying out how it reacts in every possible situation (which isn't really feasible)?
I guess they should train an AI to play it too
Re: Pac-Man recreated with a GAN trained on 50k game episodes
#120I've observed an increasing trend with modern AI research that A) It's no longer practical to incorporate into real products given the hardware costs for training and inference. In NLP, large transformer based models can easily occupy a 5 thousand dollar GPU at a paltry 50-60 words per second processing pace. B) The Research often demonstrates things that existing non-ML toolchains were already reasonable at doing wi…
Do you really think nVidia is seriously advocating for using neural networks as the best way to write a PacMan clone? It's obviously just a proof of concept to demonstrate that GANs are sophisticated enough to learn certain categories of models.
Doing these cute demos is a double-edged sword. A fun and charming demo attracts attention by non-experts who can't see the wood for the trees.
In my field we make robots do a thing, say folding laundry, as an example of dextrous manipulation[1], and the popular conversation is about whether your laundry robot makes commercial sense and not about your novel grasping strategy and nifty motion planner that makes this sorta-possible for the first time. But everyone sees your video. Pros and cons.
[1] laundry folding at Berkeley (not me) https://youtu.be/gy5g33S0Gzo