Make Mistakes · 3:52am May 27th, 2019
… and keep the best ones
When I worked for Zoesis, I once did a test animation of one block chasing another block around a… third block. (Drawing's hard, okay?) I animated by writing computer code to control each character. About 15 times a second, at least one behavior for each character would execute that updated its speed and heading according to where it wanted to go and what was in its way. (It used a behavior-based robot architecture, if you know what that is.)
The first character was supposed to run in a circle. I programmed it to try to stay a constant radius away from the center of that circle, by steering towards the center if it was too far out, and steering out if it was too far in. The second character chased the first character.
Body turns weren't instantaneous, though, so each character always ran in the direction that would have been the right direction to go about 30 milliseconds ago. The result was that the first character ran almost in a circle, but with a little oversteering, so it wobbled back and forth as it ran, like it was panicked. The second character had much smaller angular changes, so it pursued the wobbling figure cooly and precisely, like a great cat on the hunt.
I was about to fix it, when I realized I liked it that way. I showed it in our theater when we did the dailies (that's when you show each other your day's animations), and our artist loved it. "That's perfect," he said. "But it was totally an accident", I said. "I just decided to keep it." He said, "Half of art is knowing which mistakes to keep."
(Which later became the theme of "Keepers", tho now I feel that story suffers from having been written with a message in mind.)
I can't tell you how many typos I've made in the writing process that turned out to be better than what I was trying to write.
My old art teacher called this the square centimetre of chance.
As with life, so with art. It can be planned, but mostly it involves seizing any nearby opportunities and thinking, "Hey, that could work."
I guess that makes humanity an artform, given that most of us started off as accidents.
5065651 Humanity is a more typical example of the more general pattern: evolution. So think of it in the other direction: It makes artistic creation a type of evolution.
That's in the essay this came from, which got too long to post as a single piece.
Philosophical question: Is creativity possible without randomness?
Probably a question which would dissolve upon defining both creativity and randomness.
5065783
There's a sense in which the answer is no and another sense in which the answer is yes. It depends on the definition of creativity.
If creativity only requires looking at things in ways that previously have not been considered, then abstraction and analogies are sufficient for generating creative ideas, and both can be done systematically.
If creativity requires looking at things in ways that previously could not be considered, then creativity requires some mental flexibility outside the range of what one can consider. We can say that randomness (in ideas one generates) is defined by cross-entropy of the idea-generation distribution expecting the idea-anticipation (i.e., what one can consider) distribution. This definition of creativity requires infinite entropy, which is about as random as you can get.
5065783
Re: Artistic creation is a form of evolution.
Organisms evolve within an environment. Is the output of artistic creation an organism or an environment? If it's an organism, what's the environment? If it's an environment, what are the organisms?
I get the sense that it could be either. Artistic creations are composed of artistic ideas, which spread through humans. In that sense, artistic ideas are the organisms, and humans are the environment. On the other hand, artistic creations are open spaces for artistic interpretation, and humans project their own thoughts onto those creations. In that sense, the creations form the environment, and our interpretations are the organisms.
It's interesting in this case that it works both ways. Artistic ideas spread through humans to yield new creations that contain artistic ideas. Artistic creations enable artistic interpretations, which spread through humans to yield further artistic creations. Thus spin the interlocked circles of life.
I expect there's an answer somewhere here on whether the creation conveys an idea or encourages people to form their own. I think it depends on how "open" a space the creation is. If there's only one interpretation that fits, that creation is certainly an organism and not much of an environment. If there are many mutually-inconsistent interpretations possible, then it's certainly an environment and not an organism. Somewhere down the middle, I'd guess a creation has both a coherent theme and a multitude of interpretations.
5065853
This is so for every question asked in human language.
Can they be done systematically? I take it that means ordering all possible abstraction and analogies, then iterating thru them. But the set you'd need to iterate through would have a cardinality at least that of aleph-one, so good luck with that. Also, I don't think you can enumerate "all possible abstraction and analogies", because most rely on prior abstraction and analogies which you haven't conceived of yet.
If the model is to begin with the concepts you have, and iteratively consider all concepts that can possibly be derived from them, then it seems likely that all people would iterate through the same set of concepts--like beginning with the set-theoretic definition of zero, and iterating thru all numbers that can be constructed from it.
Also I think you'd need the axiom of choice to do this enumeration, and doesn't the axiom of choice require randomness for sets that aren't ordered?
5065853
I think this is a tautology, not a real problem. If creativity is defined as looking at things in ways that previously couldn't be considered, then it requires looking at things in ways that previously couldn't be considered. And the tautology is built on a definition, not on a real question: Is creativity defined as arriving at a conclusion which needs an inference chain of more than a single step?
Okay, I think that makes sense, if the anticipation and generation distributions are different things. But this requires a Platonic conception of the idea-anticipation distro as a thing in the world, rather than as a distribution you build empirically by observing ideas generated. The empirically-built distro would be the same as the idea-generation distribution, so randomness is zero by construction.
You mean, because the set of ideas is infinite? Infinite sets don't have infinite entropy. The continuum from 0 to 1 is infinite. Entropy is normalized by the probabilities of individual events, which sum to 1.
5065858
Organisms evolve within an environment, which includes themselves and all the other organisms in the environment. Not all environment is organism, but all organism is environment. Exactly the same with art: all art is environment, tho not all envir. is art.
Exactly the same with biological evolution.
If the latter, then random patterns are creations, and idea based on those creations required randomness. If the former, we go back to the beginning, asking whether a human can create a "new" idea without randomness. Where the question depends entirely on how we define "new".
I don't think there's going to be anything but definitional questions, and it will all get hopelessly muddled and metaphysical when someone inevitably injects the metaphysical words (note I don't say "concepts") "cause" and "free will". (E.g., If free will is not the same as randomness, then creativity doesn't require randomness. Spiritually-minded people will feel that "creativity" must be a product of "free will", so they'll then spin the question around and say, "We know creativity requires free will; therefore, methods of producing new art that rely on randomness are not creative." And thus possibly define creativity out of existence.)
5065879
You don't need to enumerate over all abstractions and analogies. Just enough of them to find something people haven't thought of. Also, I don't doubt that randomness is required somewhere along the way, but "new" randomness isn't necessary for each instance of creativity per the first definition.
I think for most cases where people think I'm being creative, I'm actually just following what seems to me to be a fairly rote process. Abstract the apples out of a problem, fit it to one of a few basic patterns, translate the solution back into the concrete problem at hand. I don't think you can really say that randomness is required for this since it feels like a fairly robust process (i.e., I'd get the same result if I did it ten times, discounting the effects of sleep deprivation). Maybe there's some low-level necessity for randomness in cognition in general, but I wouldn't know of that.
5065988
The infinite cross-entropy comes from the fact that something with 0 likelihood of showing up per the idea-anticipation distribution has a non-0 likelihood of showing up per the idea-generation distribution.
My intention with that second definition was to assume that a creative solution can't even be anticipated by the person that comes up with it. That happens often enough to me that I think it's reasonable to assume such a thing for a practical definition. You can ignore that though and say that the idea-anticipation distribution can't be associated with the same person as the idea-generation distribution (i.e., the person saying that a solution is creative can't be the same as the person that comes up with the creative solution).
As for the "actual" definition of creativity, I think that's more of an empirical question. I can guess properties of creativity and show them to require or not require randomness, but I don't know what the universally-accepted properties of creativity are, or if they even exist. Without those, it's hard to state that something is the definition of creativity.
5065992
We can just skip the definitions and go straight to the properties then. That way if people agree that word W should have property P for enough words and properties, then they can agree with the conclusion regardless of any additional nuance in the definitions.
I didn't intend to say that all interpretations are artistic interpretations. Just the artistic ones
.
5066191
I think part of the question at hand is whether that counts as "creativity". I suspect every concept with a spiritual patina, like "creative", "conscious", "free will", of referring to some algorithmic process with a complexity so great that it appears mysterious. If we replaced all your neurons with error-free components and locked you in a white room, you could still appear to be creative.
Ah. That's not a problem with the cross-entropy; it's a problem with how you computed the idea-anticipation distribution. There shouldn't be any zero probabilities. Given a non-zero prior, it would take an infinite amount of evidence for Bayesian reasoning to change that prior to 0 or 1.
That's like saying "the outcome of this simulation can't even be computed by the computer that will compute it". The process of "anticipating" seems to be the same as the process of being creative, just carried out to fewer ply or something.
If people agree that word W should have property P for enough words and properties, that's a definition.
5066321
Why do you call it a problem? I did that deliberately to make it clear what cases I was considering and what the relationship was to randomness.
I don't understand this statement. I don't equate "anticipation" with "able to compute" or "able to come up with," so the analogy is lost on me.
Some things that I'd guess are getting in the way of the analogy:
How about this: people are never surprised by the things they anticipate. Sometimes people surprise themselves, therefore sometimes people do/create things that they don't anticipate.
Sometimes I'll do things out of habit and come up with results that surprise me. It seems like in these cases, there's neither randomness in the habit nor anticipation of the surprising results.
Sometimes it's a definition. Otherwise, you could fill a book with a hundred thousand words next to the phrase "something, or not" and call it a dictionary.
Socrates was a man, but "a man" is not much of a definition of "Socrates." It is enough though to conclude that he was mortal, given some assumptions about the relationship between "man" and "mortal." We can agree on the relationship between "man" and "mortal" without fully defining Socrates beyond the fact that he was a man.
5066647
A probability distribution can't have zeroes or ones in it (unless it's estimating probabilities within a formal system rather than within the real world). If it does, you goofed.
I don't think I can address the rest of this comment until June 4.
5066663
Sure thing. When you get back, can you explain this as well:
5066668 A probability distro is part of a formal system, but if you're using that formal system to model reality, zeroes or ones are not epistemologically rational. Priors of zero and one can't be allowed because they're epistemologically unsound, and because they'll break your system. Posteriors of zero & 1 can't be computed except from priors of zero or 1.
(Analytical priors model things that shouldn't be part of a model of the real world, because they exist only for formal systems. They should be part of the code.)
If people come up with hypotheticals of 0 or 1, they're doing it wrong and their model will probably crash. In most cases they've forgotten to use some smoothing of a sample distribution, like Laplace smoothing.
Sometimes it's computationally a lot easier to have 0 or 1s, but then you can't just turn an inference engine on and let it go; you typically have to write error traps or conditionals in, or restrict the inference. My larger objection is that ordinary people assign things 0 or 1 probability ALL THE TIME in their mental models, and this is a large part of what makes them stupid.
5066676
Sure, that works. The model is certainly either incorrect, incomplete, or inefficient when modeling probability distributions over observable natural phenomena if it can hypothesize a zero or one probability output for any measurement that doesn't encompass the entire space of possibilities.
Because you didn't convey the message well, or because it had a message at all?
We don't make mistakes, just happy little accidents. ~ Bob Ross
5068112 Because I started with the message. I think that writing a story while knowing the message you want it to tell comes from the tradition of writing fiction as propaganda. That's a tradition in which the purpose of writing fiction, rather than nonfiction, is to convince people by emotion rather than reason. I think it's fundamentally dishonest.
I believe the honest use of fiction is to entertain, and to explore topics that you aren't sure what to think about. To put an abstract idea into a concrete situation & see if it works.
5069329
Hm, I certainly agree stories can be dishonest, and some subjects are more prone to this than others. Perhaps, even, most stories are dishonest.
Personally, I don't think they can't begin with/know their message else they're dishonest. I don't believe writing with a conviction is what causes dishonesty. In fact I myself think real conviction in the author, born of experience and honest reflection, is what produces honest fiction (though I do not think this is the only way).
Also, it's very hard, I think--likely impossible--to write a story that asserts no theme without asserting anything else along the way, sometimes subconsciously.
I think using fiction to explore is one of its important functions. But I also think this function too is prone to dishonesty. Because no matter what if you have a dishonest person doing the writing--or the exploring--the result will be dishonest. You might also just have a bad explorer as the author.
I could be wrong though :P
5072690
You're right. I didn't, tho, say that writing with conviction causes dishonesty. I said I think it's a dishonest use of fiction. Fiction written with conviction communicates the honest conviction of the writer, but the choice to write fiction rather than an essay is dishonest, because when you have a message to communicate, the difference between choosing fiction vs. an essay is that fiction persuades by emotion rather than via reason. So to make that choice is to say, "I want to convince people of something, but I want to circumvent their reason."
5073821 5072690
What idiot wrote that? Oh, me, yesterday.
I've changed my mind. Better to say, "fiction persuades by gestalt rather than via rational dialectic." Gestalt perception isn't distinct from reason except in the rationalist model of reason. So it's legit to use stories In a realistic mode, where you pile on realistic details to produce a gestalt. But I still think it's illegitimate to use a story to persuade if the story Is so highly artificial that any gestalt it produces can only be one designed to convey a particular meaning. For instance, allegory seems inherently dishonest to me. But I have used allegories, so I should disagree with this also in some way that I haven't yet figured out.
5074187
I'd argue that anyone using rational dialectic to persuade is also being dishonest. I see reason as a method of linking two otherwise-disjoint concepts so that the invalidity of one reveals the invalidity of the other, and so that clashes between two ways of framing a thing become obvious when stated. I don't see reason as a way to demonstrate that one thing is correct just because another thing is correct, at least not outside of the logical framework within which the statement is framed.
I think the distinction is openness to being wrong. Sometimes stories are the only good way to show people what you're thinking. For example, when there are objects/properties involved that you don't know how to define, the only good way to describe the hypothesized relationship between them is by analogy, and stories are great for setting up complex analogies.