I've never had a harder time explaining something this impressive to my gf.
Pac-Man recreated with a GAN trained on 50k game episodes
41–50 of 154 posts
Re: Pac-Man recreated with a GAN trained on 50k game episodes
#42Earlier quoted context omitted.
Fundamentally, this is because we have already been trained on years of living in and reacting to the real world, which, to some abstract degree, the game mimicks. The Nvidia model starts with nothing.
So you are saying that if you first trained it on a different game with similar real-world mechanics, then training it on this game would require far fewer iterations?
Pac-Man isn't a purely abstract game, it's loaded with human cultural references. Pac-Man himself is a creature who loves eating food: mostly white pellets but also various kinds of fruit and of course the iconic power pellets. Fruit of course is something humans immediately identify as something good to eat and in the game they're rewarded with points which are denoted by numbers on the screen that grow larger.
It almost needn't be said that human culture puts a great deal of emphasis on making numbers grow larger, particularly those associated with personal wealth and economic growth. To a computer without this context, why should there be any reason to prefer larger numbers over smaller ones? Heck, if you're going all the way and denying the AI direct access to the game's memory, then it's got to learn to recognize the digits first before it can even begin to decide whether it's good to increase them.
Even more abstractly, the game relies on the human concepts of good and bad. It's good to eat the fruit and pellets to increase your score counter. It's bad to let a ghost catch pac-man, causing him to die and lose a life. Without being taught, human children can identify all of these elements but the computer would seem to have a more difficult task.
Re: Pac-Man recreated with a GAN trained on 50k game episodes
#43This 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.)
AI-based detection systems have to be fairly conservative about banning accounts. so there will probably be a transition period where the clientside AI mimics a human well enough not to get banned, but makes enough "dumb computer" type mistakes to get noticed by other human players. this would be very annoying for the humans.
if the AI is truly indistinguishable from a human player, then it sort of depends on the human player we are comparing to. if the AI plays like the best human players in the world, this would be a big problem for games without a good ELO system.
Re: Pac-Man recreated with a GAN trained on 50k game episodes
#44This 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.)
>> does it even matter really?
Because, even if acting like a human, AI can play/fight 24/7. Many online games still have leader boards that can be climbed through steady play rather than short-term bursts of skill.
Re: Pac-Man recreated with a GAN trained on 50k game episodes
#45How long until this method evolves to solve business requirements? (Argument could be made that the game is a set of business requirements)
Imagine visual programming but instead of connecting systems and dealing with high level schemas, the stake holder just defines the data flow, and the model creates a (hopefully) more extendable system than current visual programming systems.
But the complexities/black box nature of some neural nets (and the CPU designed by AI that was enormously complex) may transcend to this use case. Meaning we can create systems that are more efficient, more easily, but we ourselves can’t interpret the entire system, let alone each piece sufficiently.
Re: Pac-Man recreated with a GAN trained on 50k game episodes
#46> 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.
Because it resembles a world you are very familiar with. Let’s say you have something we don’t have experience with, for example docking two satellites with orbital mechanics, the humans would be struggling because their intuition is wrong. AI may do better with this.
Re: Pac-Man recreated with a GAN trained on 50k game episodes
#47How long until this method evolves to solve business requirements? (Argument could be made that the game is a set of business requirements) Imagine visual programming but instead of connecting systems and dealing with high level schemas, the stake holder just defines the data flow, and the model creates a (hopefully) more extendable system than current visual programming systems. But the complexities/black box nature…
In most real world business situations we can't even define the requirements and constraints in a manner amenable to AI solutions. It only works for the most trivial issues.
Re: Pac-Man recreated with a GAN trained on 50k game episodes
#48I made a video explaining the details of this paper if you are interested! https://youtu.be/H8F6J7mYyz0
Very interesting video! How does the GameGAN image discriminator work? I don't know so much about GANs but it seem to me that classifying PacMan images as real or fake would be a very hard problem because of the low resolution.
Suppose PacMan is in a corridor and there is a dot to the left of him and three to the right: ".P..." That is plausible because PacMan hasn't eaten any dots yet. But if the second dot to the right is missing: ".P._." something is wrong because the first dot to the right is still there and PacMan can't jump over dots. It must be very hard to get the discriminator to understand that the second image is fake.
Re: Pac-Man recreated with a GAN trained on 50k game episodes
#49Thought 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
Another fun example of how your brain predicts physics stuff is when you misjudge what something weighs. For example, if you subconsciously judge a glass is full of water but it's actually empty, as you try to lift it your arm will fly up awkwardly. This all happens without conscious thought.
Don't tell me it never happened to you!
Re: Pac-Man recreated with a GAN trained on 50k game episodes
#50Earlier quoted context omitted.
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.)
>> does it even matter really? Because, even if acting like a human, AI can play/fight 24/7. Many online games still have leader boards that can be climbed through steady play rather than short-term bursts of skill.
One more reason to fix those games, I guess.