Who owns copyright to Deep Dream images? | Hacker News Reader

Who owns copyright to Deep Dream images?

plus.google.com

This is just a neural network fed with some dog pictures, it's not really "thinking", it doesn't "create" anything. In other words, it's not (yet) artificial general intelligence. As I see it, it's just some software which creates an output given an input. In essence, a tool.

So I guess, unless otherwise stated in the license of said software, the ownership lays on the operator.

Maybe, one can argue that the engineers who "fed" the NN do have the rights of the output.

But the software is not clever enough for me to see here a similar case to that of wikipedia, the photographer and the monkey.

To a sufficiently advanced alien species, _we_ are neural networks fed with some pictures (and sounds, smells, touches, and tastes) :)

You don't have to have general intelligence to be creative. Different networks can create different results with the same inputs. It's not as cut-and-dried as you make it out to be. https://www.youtube.com/watch?v=32GnAI3nrsQ

It is affected by that case though because of the policy and guidelines it spawned.

> “Because copyright law is limited to ‘original intellectual conceptions of the author,’ the Office will refuse to register a claim if it determines that a human being did not create the work,” said the US Copyright Office in its latest compendium of practices published Tuesday. “The Office will not register works produced by nature, animals, or plants.”

and

> In new guidance the USCO has ruled that only works created by a human can be copyrighted under US law, which excludes photographs and artwork created by animals or by machines without human intervention.

Source [0]

This feels like it falls into the 'by machines without human intervention.' There's no meaningful contributions from a human.

All that said this will eventually be an issue (hopefully) when machine intelligence actually takes off or human uploads become a viable operation. Though maybe by then we'll be less obsessed with every idea having to have an owner, but that's neither here nor there.

[0] http://www.theguardian.com/technology/2014/aug/22/monkey-bus...

Monetary concerns aside, I like seeing "by spiritplumber" on my work, even when I'm giving it out.

True and I'm a big proponent of attribution. I just think that in a lot of cases the person who's assigned ownership of things like a lyric, melody, etc. is really just the lucky one who got to have their name put on a long string of remixes. This is especially bad in music and specifically hip hop where building on earlier music is one of the core tenants.

It's easier when the thing in question is still whole and unchanged but ownership of bits and pieces of a copyright work is troublesome to me.

>There's no meaningful contributions from a human.

Humans provided the training data. Humans wrote the source. Humans compiled and ran the program.

Humans provided the input photograph

humans caused the photograph to be created...

authority is becoming confused. The notion of Intellectual Property is becoming obviously self-contradictory. Who owns the copyright to your soul? Is it Facebook? Is it Google? Did you accidentally assign all of your Intellectual Property to WR Hambrecht? Can you be required to perform on command like a dancing bear? Are the excrescences of your soul merely the feedstock of an industrial process that requires human fleshbrains for supervised learning processes the output of which will be the property of the state since they are formed without human agency?

Bullshit.

Intellectual Property is Ideal Theft. There are no original ideas. Nothing is true, everything is permitted.

There is nothing to be afraid of; resume shopping and avoid panic buying.

Overstuffed wording aside, copyright is just a set of laws that can be interpreted with more or less ambiguity; it's not some sort of ineffable concept. And laws aren't mathematical theorems; they don't become invalid just because they're self-contradictory.

Also, "original" in the law doesn't mean the same as in regular conversation; it merely means it's not an outright copy. That's why a work can be deemed to be derived from other(s), yet still have copyright protection for its own contribution.

>And laws aren't mathematical theorems; they don't become invalid just because they're self-contradictory.

No, they just become meaningless (and potentially dangerous).

The problem is that people are trying to apply the law to actual mathematical concepts (like bits). Since the law does not model mathematical reality, the law necessarily fails.

Many appear to be stuck in the fallacious belief that the law reflects reality, or at least that reality should reflect laws. Of course, this belief is incorrect, doomed to failure, and we see that failure play out all the time, including in this case.

A bit is no different than a letter/grapheme of a text, which was what copyright originally was designed to protect. They are both digital (discrete) encodings of information - a computer file is just a text in a language that only has two letters. The law doesn't concern itself with the specifics of the components of the work, it concerns itself with higher-level concepts like "copy" and "derivative", which still make perfect sense for information encoded as bits.

As for the failure, I disagree. This case merely represents ambiguity, which is inherent to law (it's why we have courts and judges and juries).

So the copyright belongs to google, even if you run it on your own photographs on your own machine? That doesn't seem right.

Consider if you took two copyrighted pictures and combined them in some way in photoshop. We can lay claim to the combined work, but we may not have consent of the original authors works to distribute thier work.

Now consider if you trained the NNet with the same two images, such that it was highly overtrained and basically produced a combined replica of the inputs. This is essentially the same as doing it manually in photoshop. That a computer done it does not take ownership away from the creators of the two images.

A NNet isn't trained with two images though, but millions. Do we abandon copyright because of scale? Should the NNet operator not be required to keep the entire training set such that copyright can be traced? Do we invent an entire new industry on determining the probability that a particular image was used to train a NNet (and by how much it affected it), such that it's owner can claim royalties on anything the NNet produces?

The question isn't about google versus operator, it's about whether or not we're going to continue investing in the madness of copyright for machines designed to mimic human brains, and if so when will it apply to ourselves - for we can't archive our own training set.

>The NNet was trained with millions of images not on your machine, whose ownership we cannot trace (unless we store the entire training set).

It doesn't matter whose images they were trained on. As long as you are not reproducing the images exactly it's not violating copyright.

But if it does, then I believe it was Flickr that provided the images.

> Do we abandon copyright because of scale?

This is an interesting question. I wonder if, due the number of items in the training set and the minimal impact of each individual creative work, it would be considered fair use.

If your training set significantly consists of images from someone else's training set in the same domain, you might have a conflict. But for arbitrary images, it may be analogous to search engines indexing (and learning from) copyrighted material, which is generally protected.

I agree with you: open source linear regression software trained on open source data creating amusing output when misused doesn't constitute an "individual without rights yet" in my book

> This is just a neural network fed with some dog pictures, it's not really "thinking", it doesn't "create" anything.

That output image, was indeed created by the program.

We tend to think that only we, humans, have the ability to create, but some animals create, like birds do nests for example.

Also, the fact that you can understand how a deterministic program produces a pseudo-random digital work (like an image in this example) does not make the program "dumb" nor intelligent-less --if we could only understand better what means being intelligent, maybe we not be having this conversation :)

derivative works

https://en.wikipedia.org/wiki/Derivative_work

The article (or the parts I've read so far; it's long and dense) assumes two things:

1) "An AI" should be considered a person. (This one is more subtle; I don't believe the author says it outright.) 2) "An AI", being a person, should be considered the author of its output.

Both of these suppositions are quite debatable, but the article doesn't bother to debate them:

  The law as it is currently configured cannot vest
  ownership of the copyright in a procedurally generated
  work in the work’s author-in-fact, because the work’s
  author-in-fact—a generative software program—has no legal
  personhood. Intuition and the principle of transitivity
  both suggest that the programmer of generative software
  is the logical owner of the copyright in the works
  generated by his or her software.

Whose intuition? Not mine. My intuition says that Deep Dream is a tool, and that by running the tool, I am the author-in-fact of the work. Or maybe that doesn't require enough human creativity for a court, and if not, fine -- the creators of the images that my instance of Deep Dream sourced to create its nightmare-scape are the authors-in-law. But Deep Dream is not the author, and the programmers of Deep Dream are not the authors.

That's _my_ intuition (possibly quite wrong!). The article would do better to distinguish intuition from fact.

Some time ago I had to address this kind of issue with fonts. Summary: the program generating text on the screen was copyrighted, but the procedural generated shapes (typeface text) was not.

Yes, but who plugged the computer into the socket and paid for its electricity? Does that count? It's just a minor contribution, but essential. Also, who configured the software - shouldn't the work performed by humans be considered a contribution to the end result?

Alternatively, why not attribute the Pixar movies to their rendering engines? Aren't they using enough neurons or procedural generation? I am sure there are lots of smart agents there in their procedural generation of trees, grass and minor characters.

> "An AI", being a person, should be considered the author of its output.

I guess "copyright holder" is a better term.

I can think of Music records, or Hollywood companies, which are not human but are definitely "persons" that hold tons of copyrighted material, and even some of those created that material.

I can't see why not a program that creates digital works couldn't hold the copyright of its output

Don't forget to consider the fact that humans select the best outputs from their NNs, thus, contributing with their esthetic sense to the final result. I am sure not all images generated from NN are worthy of Smithsonian yet. Also, it's humans who took the input pictures and fed them to the NN.

FTA:

    Q: So, Andy, who owns the copyright to DeepDream images?
    
    [...]

    A: Basically, "no one does," because AIs don't have legal
    personhood.

Meanwhile, William Gibson 31 years ago:

    "That's a good one," the construct said. "Like, I own your
    brain and what you know, but your thoughts have Swiss
    citizenship. Sure. Lotsa luck, AI."

Gibson was very, very right when he said "The future is already here — it's just not very evenly distributed." He just neglected to mention that the future is also really friggin' weird.

Where is that Gibson quote from? I'm intrigued. :)

it's from Neuromancer

Google: Neuromancer

That's from Neuromancer, when the hologram (like a chat bot) is having a conversation with the main character.

>Q: And [who owns the copyright to DeepDream images], according to [my law school mentor]?

>A: Basically, "no one does," because AIs don't have legal personhood.

Wait, what? So if you build a tool that helps you build the final product, the tool counts as the creator, since the tool doesn't have legal personhood? Do we argue that easels and spirographs don't have legal personhood so if you use that stuff in your work then no one has the copyright?

Reminds me of the quip: "Dismissing a graphic artist's work because it was computer-generated is like dismissing Michaelangelo's work because it was brush-generated."

If you used a spirograph to create art, you would own the copyright. If you created a spirograph that was robot controlled and could automatically create art then nobody would own the copyright. I don't see how an easel count create art on it's own.

The distinction is arbitrary. The only difference between the spirograph and the robot is you are touching the spirograph while it's operating.

If I write a computer program (say a Processing sketch) that generates an image, who owns the copyright to the image?

Unless your directly influencing the output you own the copyright to the program not the output. Basically, if your code generates a random image you don't get copyright, which is useful as generating say every English haiku is actually possible. It's under ~(15,831 syllable candidates) ^ 17.

Here's a real case: The Policeman's Beard is Half Constructed, a book released in 1984. The prose and poetry was written by Racter, a computer program written by William Chamberlain. The illustrations were done by Joan Hall, and the introduction to the book was written by William Chamberlain.

So it's clear that Joan Hall owns the copyright to the illustrations, and William Chamberlain the introduction. But what of the rest of the book?

So when you procedurally generate a world in mine craft does the guy who manually inputs the seed count as the creator, or Notch, or both, or none?

I think it's a much more interesting question to ask who owns the output if it is trained on copywritten data. It works by iteratively tweaking the image to look more like whatever it looks most like. It's not hard to imagine that a simpler model that has only seen one picture of a dog might iterate a mildly dog-like image into exactly that dog training image. Clearly that should belong to whomever the original photo belongs to. So then what about the grey areas beyond that simple thought experiment?

In the eyes of the law, wouldn't the images be considered derivative works of all the data the Neural Net was trained with?

And the input image.

Obviously presuming Deep Dream images to be the output of a strong AI or an otherwise sentient being is absurd, but it does raise a very good point.

That being: where do you draw the line in terms of what constitutes electronic sentience? The short answer is nobody knows. We only have vague ideas, and for all we know those may be completely wrong.

It becomes difficult in that the cognitive architecture of artificial intelligence is not bound by biological limitations, and thus may be quite alien in nature relative to mammalian cognition. As such, our existing ethical frameworks may be completely unsuitable if applied to artificial beings.

What I personally find unsettling about the Deep Dream images is their striking visual similarity to that of what a human sees on hallucinogenic drugs. The underlying algorithms utilized are themselves are heavily rooted in biological intelligence. It's not a huge leap of logic to say that if you were somehow able to excise part of a living human's visual cortex and hook it up such that it could receive input and generate output, you might see similar results.

It's also worth noting that attempting to grow human brain tissue to any large degree of scale or complexity within a laboratory environment would be considered a highly unethical horror show, despite the fact that we almost certainly have the technology to do such a thing today.

By the same token, one could do the exact same thing in digital form and the perception would be entirely different. This brings up a lot of issues, including the possibility that artificial agents of the future could potentially suffer in silence on unimaginable scales, double standards with regards to artificial versus biological intelligence, as well as fundamental questions about what constitutes artificial suffering and sentience in the first place.

Do blur filters have legal personhood?

Exactly. Calling this NN AI (as in general / strong AI) is ludicrous.

A neural net is just software, like Photoshop. Who owns the images created with Photoshop? The one who is paying for the electricity.

I tried reading that article, but for a topic that I find interesting (especially as someone who has studied AI and machine learning) I hoped it would be more approachable and easier to read. Can anyone explain the last few sentences for me?

>The increasing sophistication of generative software and the reality that all creativity is algorithmic compel recognition that AI-authored works are less heterogeneous to both their human counterparts and existing copyright doctrine than appearances may at first suggest. AI authorship is readily assimilable to the current copyright framework through the work made for hire doctrine, which is a mechanism for vesting copyright directly in a legal person who is acknowledged not to be the author-in-fact of the work in question. Through this legal fiction, the machinic creativity of generative code can be recognized for what it really is—something other than (but owing to) the human creativity of its coder

AI should be considered legal "persons". https://en.wikipedia.org/wiki/Legal_personality This is not treating them like humans, but for example businesses can be legal persons as well. It's an entity that can own things and enter into contracts. Instead of making some new, complex law about AI owning copyrights, we can recycle the "work-for-hire" doctrine. https://en.wikipedia.org/wiki/Work_for_hire That's where a person makes something (e.g. a book) as their job, and the person who hired them ends up owning the copyright.

That way, the AI is "hired" to produce images for the operator, and the operator owns the images.

This analysis is insane. It's just a very advanced photoshop filter.

And this comment is copyrighted by my "Left and Right hands", because it was them that typed, letter by letter. You know, my hands are so smart, they have lots of sensors and neural networks in them.

Reply on news.ycombinator.com