Hacker News Reader

Comment by ekianjo | Hacker News Reader

> 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 think a few more.

Early games will establish things like "can ghosts go through each other", "if I hit the wall do I die" (like in Snake; though on Acorn Electron you could eat the wall sometimes!), "can ghosts go through the shortcut", "how long do pills last", do all pills last the same amount of time".

Then later efforts would be needed to know things like "what algorithm do the ghosts follow", "does the same colour ghost always follow the same algo", "do ghost algos change when you eat a fruit", "what order do the fruit go in" ...

Of course you could guess a lot of these, but that's not what "fully understand how the game works" means.

Something being glossed over is the difference between being able to play the game and being able to implement the game.

Also, Nvidia's thing doesn't do the fruits and possibly not even eating ghosts or points and probably not the ghost algos.

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.

Humans intelligence basically evolved (as far as we know) by retarding childhood development so we had neural plasticity longer to learn more. Crazy that we're giving our already-dumb AIs the extra challenge of learning things from nothing and doing so quickly.

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?

No, in humans the cycles involved in walking and 3d orienting carry over, so several thousand iterations of pre-game and a few of game.

It might. My understanding is, it is common to take trained models from one problem and use parts of them on other problems in a similar domain to speed up training.

This called transfer learning and is an active area of research. It doesn't work nearly as well as it "should" if "AI" was really Artificial "Human Intelligence".

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.

Indeed. Some work has actually been done to try to quantify how important various types of human prior knowledge are for learning video games: https://arxiv.org/pdf/1802.10217.pdf

This is a tired argument, experiments on humans with random textures show only a 10X slowdown, not 10,000x

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.

Not a very convincing argument. Even when Pac man was new and video games barely existed it took no time to get how it works.

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.

and how many games does a human need to transform into a pacman game playable by another human? ...lol

The same. Once you understand how it works its just about coding. Coding Tetris never required anyone to play millions of times either.

Reply on news.ycombinator.com