Machine/Deep Learning and the Direction of the Gaming Industry

[omzdog]

You certainly can mathematically define fun.
But here we are assuming that fun is what makes video games 'good'.
Do submarines swim?
( ͡° ͜ʖ ͡°)

I am saying in plain terms that 'create a game' has no clearly defined goal and as such cannot be solved by a neural algorithm.

( ͡° ͜ʖ ͡°)
You are misinterpreting the idea and using the wrong words.
To create a game is not to solve a problem.
Just like the van Gogh example, the output was not a solution to an existing problem,
it was simply a pixellated result.
Video game code can replace pixels.
Hopefully this clears things for you.
( ͡° ͜ʖ ͡°)
Also the part about AI for the AI was a joke.

 

[Splintert]

Let's see it then, your mathematical definition of fun, or someone else who has.

You are stating exactly my point, creating a game is not a problem thus a computer cannot solve it. I'm not going to keep stating this over and over again if you are not going to read it.

 

[ColonicAcid]

can i just stop you right there and ask you why are you putting lenny faces everywhere.


is this 2012. did i roll back to the past.

 

[Desideratus]

Let's see it then, your mathematical definition of fun, or someone else who has.

You are stating exactly my point, creating a game is not a problem thus a computer cannot solve it. I'm not going to keep stating this over and over again if you are not going to read it.

I'm in no way agreeing with Omzdog but I think you are misinterpreting the idea of a "problem". This is Computer Science's fault and not yours because as far as i'm aware on the subject of algorithms, the way we define "Problem" is a task to be done. Or more plainly "Whatever we want it to be".

The problem with the problem is that the problem "A game must be made which is considered fun" needs a definition for fun (And possibly 'game'). And... Until we get one the problem is unsolvable.

 

[Antoine de Saint-Exupéry]

I keep thinking you started a thread called 'Amontadillo' and am perpetually disappointed.

I saw a thread labelled "Amontadillo" and thought, "This could be cool."  Imagine my disappointment.

Imagine my further disappointment when I find myself in agreement with Docm. 

I have no idea how that happened I blame wap2

 

[Splintert]

I'm in no way agreeing with Omzdog but I think you are misinterpreting the idea of a "problem". This is Computer Science's fault and not yours because as far as i'm aware on the subject of algorithms, the way we define "Problem" is a task to be done. Or more plainly "Whatever we want it to be".

The problem with the problem is that the problem "A game must be made which is considered fun" needs a definition for fun (And possibly 'game'). And... Until we get one the problem is unsolvable.

I'm not sure what you're trying to say here. From what I can tell it's a restatement of what I have been saying. It is well defined what computers in their current state are capable of computing, but it seems like Omzdog thinks that you can just solve anything by throwing more computational power at a problem in a fancy way which is simply not the case.

 

[omzdog]

you can just solve anything by throwing more computational power at a problem in a fancy way which is simply not the case.

You and I are agreeing and I refuse to argue further on the matter if you believe my point is the inverse. ( ͡° ͜ʖ ͡°)

can i just stop you right there and ask you why are you putting lenny faces everywhere.

I couldn't resist the gratuitous lenny face stamp of approval.
It also helps me keep cool in the face of non-believers.
( ͡° ͜ʖ ͡°)

 

[Jace]

I couldn't resist the gratuitous lenny face stamp of approval.
It also helps me keep cool in the face of non-believers.
( ͡° ͜ʖ ͡°)

Please do your best to resist. They're getting highly spammy in nature.

 

[Arcadius112]

What is it with admins on TW and abusing power...

 

[Austupaio]

 

[Arcadius112]

Oh I was reminiscing about Archonsod.

 

[Desideratus]

I'm not sure what you're trying to say here. From what I can tell it's a restatement of what I have been saying. It is well defined what computers in their current state are capable of computing, but it seems like Omzdog thinks that you can just solve anything by throwing more computational power at a problem in a fancy way which is simply not the case.

I'll put it more plain terms.

creating a game is not a problem thus a computer cannot solve it.

That's wrong. It is a problem. Not because it's well defined or makes sense but because it's a task to be done.

 

[AreJohn]

It'll only be a task that needs to be done if there is an end aim/goal. How would you explain that to a computer?
EDIT:
Just realised you'll probably end up needing to make a game that you're wanting the computer to make to give it an idea of what to make. 

 

[omzdog]

Just realised you'll probably end up needing to make a game that you're wanting the computer to make to give it an idea of what to make. 

Yes [instert lenny face].
There is a stage of data retrieval that allows the machine to learn from video games that already exist.
You're beginning to get it.

 

[AreJohn]

Better break the machines before they learn too much. Who needs machines when you have a mass of serfs?

 

[Splintert]

Yes [instert lenny face].
There is a stage of data retrieval that allows the machine to learn from video games that already exist.
You're beginning to get it.

And how is a random game mashup going to make anything worthwhile? This is what I am saying, it does not know what to optimize towards. Even if you had someone tell it yes or no, to the computer those decisions are arbitrary and it will continue to just randomly throw crap together.

 

[omzdog]

No. Back-progagation is a technique that allows the network to actually learn.
It's modeled after how the neurons of a brain work.
You likely won't read this but it explains it all.
https://en.wikipedia.org/wiki/Backpropagation
Basically you retrain weights inside the network and its an emulation of how 'learning' is done.
This is the 'optimization' you are looking for and it allows the forward propagation to output better results.
The stylistic conjunction of pictures was only an example of what neural networks are capable of doing.
It was never programmed to do that, it learned it through taking other examples of pictures.
[insert lenny face]

 

[Do not look here]

I could understand using it on the smaller scale. Have a septic level and throw a few 'bones' around, then let the players in and let the machine populate it depending on monitored activity of players.

I think that's the biggest problems, if you want to achieve something with system like that you have to repeatedly let players swarm in. It definitely could work as a one-shot gimmick, but I can hardly see it becoming a trend. A more fail-proof method than allowing game to 'decide what it wants to be' on its own, though.

 

[Splintert]

No. Back-progagation is a technique that allows the network to actually learn.
It's modeled after how the neurons of a brain work.
You likely won't read this but it explains it all.
https://en.wikipedia.org/wiki/Backpropagation
Basically you retrain weights inside the network and its an emulation of how 'learning' is done.
This is the 'optimization' you are looking for and it allows the forward propagation to output better results.
The stylistic conjunction of pictures was only an example of what neural networks are capable of doing.
It was never programmed to do that, it learned it through taking other examples of pictures.
[insert lenny face]

Backpropagation requires a known, desired output for each input value in order to calculate the loss function gradient

The goal of any supervised learning algorithm is to find a function that best maps a set of inputs to its correct output.

The motivation for developing the backpropagation algorithm was to find a way to train a multi-layered neural network such that it can learn the appropriate internal representations to allow it to learn any arbitrary mapping of input to output.

the loss function is a function that maps values of one or more variables onto a real number intuitively representing some "cost" associated with the event.

There is no known, desired output for a video game, there is no correct output, there is no correct mapping of input to output, there is no way to assign cost to features of a video game. It's not a magical sentient-AI algorithm that magically makes computers capable of doing anything, they are still bound by the fact that the only thing they can do is crunch numbers.

 

[omzdog]

You read. I'm happy. [you know what time it is]
The intro paragraphs are enough.

The motivation for developing the backpropagation algorithm was to find a way to train a multi-layered neural network such that it can learn the appropriate internal representations to allow it to learn any arbitrary mapping of input to output.

This is what you're looking for.
And you quoted it correctly.
Like we mentioned on the first page,
we might use ludometry to judge the outputs.
This might include concrete statistics such as sales figures, replayability, etc.
There doesn't need to be a 'correct' output (unless the game doesn't function which would be unlikely if the network were exposed to enough real-world examples).

I hope this is enough to satisfy you.