Machine/Deep Learning and the Direction of the Gaming Industry

[Calradianın bilgesi]

I think there's still much way to go, artificial intelligence is still not good enough in well defined tasks like recognising patterns in images, and you expect them to recognise a pattern of 'being fun'

隐藏

 

[Vieira151]

This isn't much of a point, but I've already seen AI compose music based on a chosen style. It was quite interesting, but i suppose if you go listen to it with predetermined notions of what makes a 'good' song, then you'll probably think it's ****. It was interesting, nothing less, nothing more. http://www.flow-machines.com/

I don't see how it would be such a huge, monumental leap for it to apply those same principles to designing a game. After all, designing a game is essentially using tech that other people have developed and implementing ideas that have already been implemented. Besides the very few, very rare games that one can consider 'innovative'.

Sure, it won't be a human free process, but what would? Besides the Dawn of Robots, that is.

 

[Do not look here]

The thing is, it's not as easy as requesting a song in certain style. What you see is a result of many cycles of teaching.

And then, no matter whether it was supervised or unsupervised learning, you have the basic problem Calradianın Bilgesi's graphic shows. Whether you teach it to focus on certain patterns or it assigns weights to them on its own, it will search for them and replicate the results (and yes, I'll insist that making one pic in style of other is replicating patterns once you deconstruct both into basic shapes). And now you have this complex network of patterns that create a game, world it is set in, the soundtrack, consistent art direction and everything else.

I'm not arguing that you can't use neural networks to create games, as I said, I can imagine it as a tool to populate a game under direction of human. It just doesn't seem plausible to create one that can just churn out games on its own can just churn out games recognizable as such on its own.

Though this thing was made and played, so who the **** knows.

 

[AreJohn]

Just a thought, but if we could teach computers how to make art. Then they'll be capable of making video games.

 

[Vieira151]

That relies on you being able to define art. And let's not get into that debate...

Let's face it - when it comes to art, some people will like one thing and others another. A robot could draw anything and people could find meaning in it. A game made by an AI would be something of a mixture of the song and the art - it would need to take some predefined things it's learned about styles of games (FPS, RTS etc) and add it's own splash of AI 'art', which some people may or may not find attractive.

The thing is, it's not as easy as requesting a song in certain style. What you see is a result of many cycles of teaching.

And then, no matter whether it was supervised or unsupervised learning, you have the basic problem Calradianın Bilgesi's graphic shows. Whether you teach it to focus on certain patterns or it assigns weights to them on its own, it will search for them and replicate the results (and yes, I'll insist that making one pic in style of other is replicating patterns once you deconstruct both into basic shapes). And now you have this complex network of patterns that create a game, world it is set in, the soundtrack, consistent art direction and everything else.

I'm not arguing that you can't use neural networks to create games, as I said, I can imagine it as a tool to populate a game under direction of human. It just doesn't seem plausible to create one that can just churn out games on its own can just churn out games recognizable as such on its own.

Though this thing was made and played, so who the **** knows.

Oh, I don't think it would be possible for robots to completely make a game. But I think that it'd be possible to get to a point where human influence would be minimal - a basic input of what kind of game they want. If the neural network and knowledgebase server is big enough they'll eventually start to learn from their own games too. I mean, I doubt the "learning" in that case would be anything other than scouring the data mined from behind the scenes analytics in the code to find patterns, but still.

 

[Untitled.]

You could argue that there are things in aesthetics that are objectively appealing. Art in general doesn't have to be appealing (in fact, the point of a piece of art could be that it is utterly disgusting), but it might be possible to isolate and define the patterns we do find appealing.

 

[Vieira151]

You make a good point, yes. What aesthetic traits would you consider objectively appealing and unappealing?

 

[omzdog]

I remember going to the National Gallery of Art with a friend of mine and we talked about the kind of art where a woman splurged her period blood all over the canvas and mounted it.

 

[AreJohn]

You make a good point, yes. What aesthetic traits would you consider objectively appealing and unappealing?

Maybe we could somehow plug the computers into donated brains with methods of seeing and hearing things to find out?

 

[Vieira151]

Genius. Who's going to donate first? David Dire is banned.

 

[Untitled.]

You make a good point, yes. What aesthetic traits would you consider objectively appealing and unappealing?

For instance, symmetrical patterns tend to be appealing. Suddenly breaking the pattern in one spot while still keeping the overarching pattern also appears to be appealing to most people.

I honestly don't know anything about aesthetics. I'm just quoting some ancient Greek architect or whatever.

I remember going to the National Gallery of Art with a friend of mine and we talked about the kind of art where a woman splurged her period blood all over the canvas and mounted it.

I don't think most modern/abstract art is supposed to be highly appealing visually. Part of the point of doing stuff like that is that the artist assumes the viewer knows about the creation process. And thus shock factor became a thing.

 

[omzdog]

Yes, I spoke to him saying that the simplicity of technique does not diminish the value of a piece,
and that the times when artists are demanded to reflect reality in image are gone.
The period blood example is an extremity.

 

[ColonicAcid]

So the last post I wrote here I described human minds similar to computers.

I've been reading Homo Dues A brief history of Tomorrow by Harari and there was an interesting tidbit that he said: "In the nineteenth century, scientists described brains and minds as if they were steam engines. Why steam engines? Because they were the leading technology of the time." It goes to talk about how the Freudian argument that Armies harness the sex drive by limiting it in young men and then channeling that fustration onto the enemy is synonymous to how a steam engine works.

This is a direct argument versus the whole "brains are like computers".  He argues that because computers are our current leading technology that is the reason why our current understanding of the brain is so much like computers. Perhaps in the future we will look back at our understanding of the brain alike to a computer just like we look back at the theories of Freud. Perhaps our brains and computers are absolutely nothing a like and it's merely found to be "similar" purely through our biases.

When I read this it completely changed my mind about the entire concept of AI. I actually don't think we will be able to find proper AI for a long, long time until after we truly discover how the mind works. We can create simple workarounds ala deep learning but never a truly self aware machine.

Though I must preface that whilst Harari as an author is very thought provoking some of his points are super far fetched. He does make a lot of connections that make you think he's trying far too hard to connect stuff that doesn't need to be connected.

 

[omzdog]

Indeed, I've always been suspicious of anyone who likens biological components to mechanical ones.
In linguistics, and generally in most all fields relating to philosophy, during the 50's and 60's there was this wave computationalists that glorified the likening of logical gates and some internal logical coordinator in the mind.

You should understand though, I'm not asking a machine to replicate the workings of the brain or compensate for the faults of the mind.
I just want it to be able to make a video game or at least that's the question being brought up in this thread.
They already make art so I'd imagine simple puzzles given in various forms would only be a step up.
I don't think Harari is looking to uninspire his readers but instead to make them aware that connectionism, the theory that logical nodes can replicate neuronal nodes, has faults as a science.

Also, I'd like to point out that Machine Learning is not a technology akin to computers and steam engines.
It has no physical implications its just a logical replication of biological behavior.
But the idea that there is too much zeal over the capabilities of AI doesn't escape me.

 

[Splintert]

If we can make a computer physically into a biological brain then we've defeated the idea of AI and at that point it's just I, real intelligence, just artificially constructed. And it would probably run into all the downsides that biological intelligence runs into.