Creative computers

[DrZoidberg]

I've always wondered why so many people are so cocksure about computers never being able to be creative. As if that's unique for humans. I can't even imagine it being particularly hard to program.

1) Randomly change or switch up sections of the program
2) Test run it toward some metric (can also be randomised)
3) If it's an improvement, keep it. Otherwise revert to old programming.
4) repeat.

Isn't that all creativity is?

Thoughts on this?

 

[rousseau]

The primary problem with giving computers true A-I, is that almost all 'programming' is based on if-else logic conditionals.

if(thisThingIsTrue)
{
    doThis()
}
else
{
   doThis()
}

Even when solving some of the simplest of problems you can get a good number of 'cases', like this.

if(this && that)
{
    doThis()
}
else if (this || that)
{
   doThis()
}
else
{
    doThis()
}

The problem with giving computers 'intelligence' then, is that intelligence requires an almost unimaginably large number of test cases, that for a computer to parse would take extremely sophisticated algorithms. That doesn't even take into account the problem of ontology, or 'how the computer should be'.

For the most part, we don't program computers to do things for themselves, we program computers to do what we tell them to do. So even if we give them 'intelligence-like' capabilities, they are not really self directed.

I don't really know how else to say: it ain't that easy

 

[DrZoidberg]

The primary problem with giving computers true A-I, is that almost all 'programming' is based on if-else logic conditionals.

if(thisThingIsTrue)
{
    doThis()
}
else
{
   doThis()
}

Even when solving some of the simplest of problems you can get a good number of 'cases', like this.

if(this && that)
{
    doThis()
}
else if (this || that)
{
   doThis()
}
else
{
    doThis()
}

The problem with giving computers 'intelligence' then, is that intelligence requires an almost unimaginably large number of test cases, that for a computer to parse would take extremely sophisticated algorithms. That doesn't even take into account the problem of ontology, or 'how the computer should be'.

For the most part, we don't program computers to do things for themselves, we program computers to do what we tell them to do. So even if we give them 'intelligence-like' capabilities, they are not really self directed.

I don't really know how else to say: it ain't that easy

I wasn't thinking about making a AI that can think like we do. That is uninteresting. We already have a brain that thinks like us and we know it's got faults.

What we want is computers that can think creatively about certain very narrow and limited scopes. We can have a computer that paints a certain type of paintings... creatively. Or does interior design. Composes music. That shouldn't be impossible. I'm not saying it's easy. But it should be real creativity. As real and genuine as our creativity. What I'm saying is that there shouldn't be any difference in quality necessarily. And it should already be within our grasp to programme. If we want.

I actually know of people who have been working on a computer program composing hit music, simply by analysing already existing hits, and creatively making new one's. They're good. But not good enough. None of these songs have made it anywhere in the charts.

 

[Bronzeage]

I've always wondered why so many people are so cocksure about computers never being able to be creative. As if that's unique for humans. I can't even imagine it being particularly hard to program.

1) Randomly change or switch up sections of the program
2) Test run it toward some metric (can also be randomised)
3) If it's an improvement, keep it. Otherwise revert to old programming.
4) repeat.

Isn't that all creativity is?

Thoughts on this?

We can worry about the creative computer problem when we have solved the lack of creativity in humans.

 

[rousseau]

The primary problem with giving computers true A-I, is that almost all 'programming' is based on if-else logic conditionals.

if(thisThingIsTrue)
{
    doThis()
}
else
{
   doThis()
}

Even when solving some of the simplest of problems you can get a good number of 'cases', like this.

if(this && that)
{
    doThis()
}
else if (this || that)
{
   doThis()
}
else
{
    doThis()
}

The problem with giving computers 'intelligence' then, is that intelligence requires an almost unimaginably large number of test cases, that for a computer to parse would take extremely sophisticated algorithms. That doesn't even take into account the problem of ontology, or 'how the computer should be'.

For the most part, we don't program computers to do things for themselves, we program computers to do what we tell them to do. So even if we give them 'intelligence-like' capabilities, they are not really self directed.

I don't really know how else to say: it ain't that easy

I wasn't thinking about making a AI that can think like we do. That is uninteresting. We already have a brain that thinks like us and we know it's got faults.

What we want is computers that can think creatively about certain very narrow and limited scopes. We can have a computer that paints a certain type of paintings... creatively. Or does interior design. Composes music. That shouldn't be impossible. I'm not saying it's easy. But it should be real creativity. As real and genuine as our creativity. What I'm saying is that there shouldn't be any difference in quality necessarily. And it should already be within our grasp to programme. If we want.

I actually know people of people who have been working on a computer program composing hit music, simply by analysing already existing hits, and creatively making new one's. They're good. But not good enough. None of these songs have made it anywhere in the charts.

Creativity and AI are the same thing. For something to be creative it requires intelligence, you're just limiting the scope a bit, so the same challenges are still there, but things get slightly easier.

We could easily program a computer to write 'good music', but not necessarily 'good and original music' which would take intelligence.

 

[Keith&Co.]

2) Test run it toward some metric (can also be randomised)
3) If it's an improvement, keep it. Otherwise revert to old programming.

Hand waiving on 2 doesn't help bring 2 to come to be.
And the criteria for 3? How would you go about establishing this?

 

[Keith&Co.]

What we want is computers that can think creatively about certain very narrow and limited scopes.

Oh. You didn't specify what 'scopes' these are, or that they're limited.
When i think of 'creativity' i think of people writing political cartoons or comedy sketches or comedy movies. If it was just 'try random shit until you come up with something better' we'd have a lot more successful comedy sequels. Most are either doing the same thing over again, or trying to copy what someone else did.

We can have a computer that paints a certain type of paintings... creatively.

Then you really need to define 'creatively.' Elephants and chimps can make paintings. But so can people on an assembly line.
Does someone who figures out how to crank out two dozen paintings a day for hotel rooms display creativity?
Do elephants?

 

[rousseau]

Basically billions of years of evolution vs carefully crafted inorganic material for 50 years.

 

[Juma]

We could easily program a computer to write 'good music', but not necessarily 'good and original music' which would take intelligence.

And the problem here lies in how to measure that. You do not talk about objective "good and original music" cause that does not exist. You mean "good and original music" to a human.
To create that the AI must be have almost exact the same inner workings as a human. It must "tick" the same way. Or at least have a modrl of how humans tick.

 

[Bronzeage]

We could easily program a computer to write 'good music', but not necessarily 'good and original music' which would take intelligence.

And the problem here lies in how to measure that. You do not talk about objective "good and original music" cause that does not exist. You mean "good and original music" to a human.
To create that the AI must be have almost exact the same inner workings as a human. It must "tick" the same way. Or at least have a modrl of how humans tick.

Years ago, I was a member of a poetry forum. It was a small group, but there were some very talented writers among us. One day, some guy named Jasper, from Australia showed up. He believed poetry should be spontaneous, subject to no editing or revisions. Even so, he had lots of suggestions for other people's poems and was a general pain in the ass. Of all the different trolls on the internet, poetry trolls are the most pathetic.

One of my poetchick girlfriends and I hatched a plan. We went into our email's spam filter and found emails that were created by a random sentence generator. The nouns and verbs are all in the right place, but read something like, "Toxic penguins read the paint very well." We took the emails and arranged them into poemy line length and posted them on the forum.

I reviewed and critiqued her work and she did the same for mine. I explored the complexity of her imagery and dissected every metaphor, showing how well she wove it into the overall theme and motif. It was total bullshit, but great fun.

Jasper loved them. He said they were the best work on the forum.

If it's possible for a mindless random word generating program to create something pleasing to a human, a real AI program would have to have some small advantage. Human's can't do much better. Much of what we create is little better than random chance. On some occasions we string a few of these random sequences together and find something really nice. The one thing that humans and computers will never be able to do is accurately predict the reactions to the products of our creativity. That is just too random.

 

[dystopian]

I've always wondered why so many people are so cocksure about computers never being able to be creative. As if that's unique for humans. I can't even imagine it being particularly hard to program.

1) Randomly change or switch up sections of the program
2) Test run it toward some metric (can also be randomised)
3) If it's an improvement, keep it. Otherwise revert to old programming.
4) repeat.

Isn't that all creativity is?

Thoughts on this?

Computer programs have done creative shit on their own for a while now. Many years back, I read this article about a computer programmed to autonomously build stuff out of transistors and what not; simply tell it in general terms what you want, and it'd build it. So the researchers told it to build them an oscillator. Sure enough, it builds a device that shows an oscillating wave; except to their surprise it cheated in a way it wasn't 'supposed to': it had actually built a radio transceiver instead of an oscillator, picked up an oscillating signal from a nearby PC, and displayed it. If that isn't creativity (unforeseen at that) then I don't know what is.

Then there's more recent stuff like ANGELINA; an A.I program that designs its own videogames when given a theme. It even describes its own games afterwards, explaining some of its design choices. Here's something on it: http://motherboard.vice.com/blog/this-eerie-game-was-made-by-artificial-intelligence

 

[Jayjay]

I've always wondered why so many people are so cocksure about computers never being able to be creative. As if that's unique for humans. I can't even imagine it being particularly hard to program.

1) Randomly change or switch up sections of the program
2) Test run it toward some metric (can also be randomised)
3) If it's an improvement, keep it. Otherwise revert to old programming.
4) repeat.

Isn't that all creativity is?

Thoughts on this?

The creativity in this algorithm is contained in "some metric". Not in the algorithm itself. In other words, whoever defines that metric is being creative, not the computer that produces the result.

 

[Sajara]

http://www.ted.com/talks/guy_hoffman_robots_with_soul

Robots that groove to the beat...does this count?
At ~9 minutes they have a robot playing music with a human too.
At ~15:30 they show the robot grooving to the beat.

 

[DrZoidberg]

2) Test run it toward some metric (can also be randomised)
3) If it's an improvement, keep it. Otherwise revert to old programming.

Hand waiving on 2 doesn't help bring 2 to come to be.
And the criteria for 3? How would you go about establishing this?

For music we can programme it with harmonies. We have it randomly compose music and then check how well the randomly generated music fits the types of harmonies we accept.

Same goes with colour combination. Or light composition for photography.

We do have aestethic quality metrics we can use.

 

[DrZoidberg]

Oh. You didn't specify what 'scopes' these are, or that they're limited.
When i think of 'creativity' i think of people writing political cartoons or comedy sketches or comedy movies. If it was just 'try random shit until you come up with something better' we'd have a lot more successful comedy sequels. Most are either doing the same thing over again, or trying to copy what someone else did.

Comedy is a fascinating field to study if one is interested in creativity. We have this enormous market for comedy. We just can't get enough of it. Still, comedic talents are few and far between. Even those fairly far down on the success ladder rake in grotesque amounts of money. Why is it so hard?

All comedy follows the same pattern. 1. Set up to establish the norm. 2. An end the breaks that norm in a surprising way.

Yet, it's incredibly difficult to pull in the target and to make them accept the context enough to be fooled. As we're all aware of considering the low quality of most of what is produced.

We can have a computer that paints a certain type of paintings... creatively.

Then you really need to define 'creatively.' Elephants and chimps can make paintings. But so can people on an assembly line.
Does someone who figures out how to crank out two dozen paintings a day for hotel rooms display creativity?
Do elephants?

A chimp did manage to fool the Gothemburg art community for a while.

http://en.wikipedia.org/wiki/Pierre_Brassau

My definition of creativity is adapting something or combining things to add value. Make it somehow better than what was before. Yes, it's quite vague