Welcome, Guest. Please login or register.

Login with username, password and session length

 
Advanced search

1411281 Posts in 69324 Topics- by 58380 Members - Latest Member: bob1029

March 28, 2024, 11:08:13 PM

Need hosting? Check out Digital Ocean
(more details in this thread)
TIGSource ForumsDeveloperDesignSolving artificial creativity?
Pages: [1]
Print
Author Topic: Solving artificial creativity?  (Read 1731 times)
gimymblert
Level 10
*****


The archivest master, leader of all documents


View Profile
« on: July 05, 2015, 08:09:21 AM »

Quote
Quote
You seem to be asking, essentially, how new ideas are formed, and whether computers can have independent creativity.


I'm a bit more specific, It's not as general as "how idea are formed", I'm past the point  of asking what creativity is, generally accepting that creativity lies in a spectrum inside a triangle define by order, chaos and purpose (or fitness) and have different degree of complexity (the question being less about can computer have independent creativity but at which complexity or what kind).

I'm specifically asking for a brand of "creativity", called "assumptions". Creativity in computer is attacked from two angles:

1. Let the computer form "intent"
2. Manipulate structure

"Assumption" would be in the middle.

For example Angelina is brilliant (if we let the complexity scale outside), it let the system have an "experience" (scouring the web), form ideas and turn them into intent, then this intent is fed to a structure as constrains to create a structured composition.

However while Angelia is expressive and have some form of creativity, it just feed a system of game creation without reflecting on it like it does for web data, it does not experience its own creation, but for human the creativity is often very iterative. As a procedural generation system, it's "exploitative"  (vs "exploitative") as the exploration is emulate by the "randomness" of the experience of web site. The metaphor module being an additional "filter" and constrain to match human expectation of creativity.

Which also beg the question of independent creativity not just for machine but also human, are we that independent when our output is dependent on our experience too? But I see no evidence of the computer system (angelina) learning from previous experience to produce new experience, I don't think it's a requirement, just a problem of complexity scale, it does not invalidate the computer's experience.

To fight this exploitative assumption Angelina has a module of level design and movement evolved through GA, it's exploratory (as in trial and error), the fitness function being conceptually a simple way to experience the design (complexity left out but conceptually the same). However the fitness function itself is fixed and made of assumptions in the domain of the problem (it will only generate space traversal design which is the main design assumption).

A lot of experience designer handwave procedural generation because, when they look at them formally, they see the fixed assumption first, it's what matter to them not the simple permutation or fitness. For the same reason they see value in procgen as saving time when presenting as "tools".

They are quick to point that the output is samey and get bored because they recognize the same assumption applied all along. Generally the best generator implement system that vary assumption based on random input from a higher hierarchy (biomes as example) within a lower frequency but even that has limits, the same don't create new assumptions. They say that it cannot match human creativity pointing at work of great artist.

To contrast is that people interested in procedural generation as creativity are just fine with the infinite minute permutation. They tend to draw attention to meaningless big number (18 quitillions planet) and amazed by unexpected collision of property, effectively treating the simulation as a complex Rorschach test. They don't get why other people handwave it so easily, pointing that if it doesn't match human creativity, most human don't match that ideal too, showing the mass of clone and soulless rip off that always inundate the market, sure an algorithm can match that with consistent quality.

Most of the process of creativity for veteran, aside from applying what is learned, is to know where the assumption need to be broken to achieve the desired expression (How do I do that, aka exploitation) or challenge them to see if they lead to valid new experiences (Why do I do that, aka exploration). Ie they class problem as such:

1. The problem is known the solution is known (applying)
2. The problem is known the solution is unknown (solving)
3. The problem is unknown the solution is unknown (creating)

It's just moving the creativity problem up the complexity chain, 1 is typical generator, 2 is exploitative generator such as the level design of Angelina. 3 would be closer to my concept of "assumption" and closer to "why do I do that".

For example what if Angelina could not only experience web page, but also its own output and modify his own code to match better the combined experience of the exterior world and internal world? How could this be even possible?

The question being is there works that explore this particular question set of creativity? And is there a possibility for a machine to understand and create assumption based on its experience of the creative process and challenge it? Is there a conceptual framework (even on the low complexity scale) that mimic those concept? Is there a set of all assumption we can translate into a universal language of procedural generation? Will Angelina win Brad over Jennifer?

What do you think, do you have ideas about the problem open by this discussion?
Logged

ProgramGamer
Administrator
Level 10
******


aka Mireille


View Profile
« Reply #1 on: July 10, 2015, 06:26:43 PM »

You can't program a thing without understanding it's inner functioning. How does creativity work? By acquiring new memories and experiences and applying memories of them to each other in novel ways, themselves probably acquired through experience, forming new memories. Describe an algorithm that can replicate that behavior, and you've got programmable creativity.

Simple really.
Logged

baconman
Level 10
*****


Design Guru


View Profile WWW
« Reply #2 on: July 15, 2015, 01:40:07 AM »

Reflect on player input as experimental feedback?

Take note of what weapons/items/abilities the player uses, and how the player exploits level design or AI with them, then utilize that to create AI that replicates it - perhaps not perfectly, but to a reasonable enough manner that it presents the player with a new challenge to overcome, with which to gain more feedback.

*shrug*


PS: What IS up with the Angelina project, lately? I've been fiddling around with map generation algo concepts, and... after "yet another impossibly rough season of Real Life (TM)"; I'm back to reduxville.
Logged

s0
o
Level 10
*****


eurovision winner 2014


View Profile
« Reply #3 on: July 15, 2015, 10:37:33 AM »

sorry if this is a dumb question, but isn't this the same thing as the problem of artificial general intelligence? as much as i like this community, i don't think we're going to solve that in a tig thread.
Logged
gimymblert
Level 10
*****


The archivest master, leader of all documents


View Profile
« Reply #4 on: July 15, 2015, 07:49:50 PM »

@baconman

by framing it as weapon/item/etc you are alredy making fixed assumption about the games, that they must have them, and not creating a system that can move from them.

@silber

you never know, i never thought i would solve that crazy idea about generating infinite procedural character on a schedule without pre computingnpath finding but correlating their temporal displacement to the generation of a single procedural spatial tile. ie generating a spatial tile with all the currently traversing npc without query of past, present or nearby tile, and that extend to all the temporal npc states in that space. the solution is indexing using space and a time variable the range of a virtual list of all npc. that was a hard problem who only needed a leap of logic to be solved, ie replacing cause with correlation.

the simple fact of sharing thought on a forum is enough to help in my case, it can create a watson effect were people give the bad answer and that lead by narrowing the answer space to a good solution. i also just work better that way that spending thought alone. it's a documented fact /look for ted about superchicken effect/ that a group of scrappy but communicating average agent outperform group with star agents or star agent agent alone.



however reading baconman i thought i would need a starting description /the assumption/ nonetheless, the system would then challenge the element of the description to see which one lead to satisfaction. a bit like what happen to euclidian five axiom who where challenged and lead to non euclidian geometry.

the main problem here is the definition of satisfaction, it need to be described too. can a system challenge its own satisfaction /and when/. the problemn is kind of a circle like chicken and egg problem. but looking at human behavior can clue us, the fact we can change value other time and also how we evaluate that value itself. we can move to believing that A lead to happiness, then deciding that B was more important and then that decidely that responsability is more important /we changed the evaluation rather than the action/. so there must be a way to describe that too.

on creativity we can pose the problem as such: let say there is a set A of all creations, B is a set in A that is all created work, then creativity is all work in A that increase set B that was not in B prior /obviously as the set B is increased/. let's call d the distance between two different work, originality is such that d is as big as possible for single work toward all the other work in B. then there a fitness score f /the purpose/ such as the work please to an audience. creativity is maximizing a work such such as f and d is big when increasing B. low d is clone, low f is random. we can define two strategy to increase B creatively, one is exploitation by having a big f and the second is exploration by having a big d. the problem therefore is in defining f and d. in reality f and d vary accross culture and individual, so there is a sensibility modifier, however we can first solve them assuming they are constant first.
Logged

gimymblert
Level 10
*****


The archivest master, leader of all documents


View Profile
« Reply #5 on: July 15, 2015, 08:26:39 PM »

okay let's call 'description' the series of parametrized steps to create a work. the current description is the current 'assumption'.

let's call frustration the creation of works with low scores, it increase with the number of low score works, once the frustration reach a certain score the system challenge the description.

let's call boredom the creation of score with consistent high f low d, increased by each succesful works with low d, past enough boredom the system challenge the description.

the description is challenged by testing the validity of any of its proposition by switching, adding or removing elements of a proposition. then scoring the description to evaluate the contribution of the elements. similarly the system can try to add or remove element to test its contributions. so the system form belief about the importance of each element of a description.

how can we call low f high d ? ie original but purposeless random work



a more complex system evaluate a description per 'intent' and form multidimensional belief of the description with intent as context. intents being specific set of fitness born from experiential system such as angelina, pleasing a specific audience being just a subset of this, another one would be 'how to make the player cry', this last example prompt the question about how to make a crude system such as it can form such intent? how to give the system a set of standard that can evolve to create intent?
Logged

Zorg
Level 9
****



View Profile
« Reply #6 on: July 16, 2015, 12:48:18 AM »

I did NOT understand everythingmuch in this thread, but i think the main problem is how AI should 'score' or reflect on created pieces. An example with external feedback: AI creates images for children's books. The images are shown to kids and the reaction is filmed. Based on enough reaction videos the AI could be able to provoke certain reactions. Couldn't diverse solutions for same reactions be interpreted as a form of creativity?
Logged
ProgramGamer
Administrator
Level 10
******


aka Mireille


View Profile
« Reply #7 on: July 16, 2015, 03:33:40 AM »

This reminds me of that one game playing robot that paused Tetris in order to avoid losing. I think we should look there.
Logged

Artylo
Level 0
***


I'm just a voice, pal - a most talented failure, b


View Profile WWW
« Reply #8 on: July 16, 2015, 05:03:06 AM »

This reminds me of that one game playing robot that paused Tetris in order to avoid losing. I think we should look there.

That is so true, that it makes me feel stupid for not being able to think in that way. An AI has a goal or a mission, and the only thing that is important is to complete the mission. Just like a soldier would.
Saying that, this reminds me of the Metal Gear Solid 2 depiction of AIs. The intelligence to disconnect itself in order to complete the task, mostly due the lack of perception of mortality or death.
*shivers* Crazy
Logged
gimymblert
Level 10
*****


The archivest master, leader of all documents


View Profile
« Reply #9 on: July 16, 2015, 10:11:42 AM »

I did NOT understand everythingmuch in this thread, but i think the main problem is how AI should 'score' or reflect on created pieces. An example with external feedback: AI creates images for children's books. The images are shown to kids and the reaction is filmed. Based on enough reaction videos the AI could be able to provoke certain reactions. Couldn't diverse solutions for same reactions be interpreted as a form of creativity?

that's a good way to evaluate human compatible creativity, it would be a system that evaluate fitness f. In a more phylosipcal way it might also evaluate distance and fitness assuming we use the same sample of audience and evaluation would chase their interest, as the audience would automatically have boredom and frustration embeded this would lead to automatic adjustement of human compatible "creativity".
Logged

gimymblert
Level 10
*****


The archivest master, leader of all documents


View Profile
« Reply #10 on: July 17, 2015, 08:13:10 AM »

okay some more thoughts

let say the description is plain text, as we are responsible for elements in the description we know what semantic relation elements have and what structure they constitute as the description.

let assume the description is plain text so we can understand humanely (in the context of the example) ie "dan eat burger" we know this the structure of action that is "subject verb target" we know dan is a person, burger are food, eat an action.

One simple process is to punch up the structure to make it "generative" if we replace dan by person (variable constrain by a class) then we have a generative description, dan eat burger being one of the result which yield "cathy eat burger".

Why it's important? because we define the realm of the search space. We can move elements from class to relationship to structure, then score the resulted works from the new description and form belief as to which elements contribute to generation.

We can also form belief as which policy is best by evaluating iterations of the search and look at what type of mutation lead to better generated results.

as "the description" (generative statement) and the works (generated statements) is related, the system can totally start by looking at existing works (maybe with actual score attach to them) to try form belief about structure and elements contribution.

And that mean we can feed experiential simulation that generated output statement for the system to look for. Ie imagine a sims game that log the life of a sims and use that to generate works that match this life experience.

We can further use more ai learning technique to introduce new elements into the system's elements dictionary to increase the broadness of works.

that's my current hypothesis.
Logged

gimymblert
Level 10
*****


The archivest master, leader of all documents


View Profile
« Reply #11 on: August 16, 2015, 02:34:38 PM »


how can we call low f high d ? ie original but purposeless random work


Emptyness
Logged

Pages: [1]
Print
Jump to:  

Theme orange-lt created by panic