Do androids dream of electric copyright? Ownership of Deep Dream images


You may have come across Deep Dream, a visualization tool by Google that uses artificial intelligence to create unique, bizarre, and sometimes unsettling images. Deep Dream is an open source program that uses an artificial neural network to transform an image using  mathematical methods used in machine learning that resemble biological neural networks. In other words, the machine mimics thinking.

The researchers explain that they designed a neural network that takes an image and applies some sort of filter to it, but the machine decides what to amplify, so the result is unpredictable, but also it is a direct result of a decision made by the algorithm. They explain:

“Instead of exactly prescribing which feature we want the network to amplify, we can also let the network make that decision. In this case we simply feed the network an arbitrary image or photo and let the network analyze the picture. We then pick a layer and ask the network to enhance whatever it detected. Each layer of the network deals with features at a different level of abstraction, so the complexity of features we generate depends on which layer we choose to enhance. For example, lower layers tend to produce strokes or simple ornament-like patterns, because those layers are sensitive to basic features such as edges and their orientations.”

The result of a low level of abstraction produces some beautiful images, like these Seurat adaptations:


More layers produce much more amazing results, such as the one pictured at the heading of the post, leading some people to postulate that Deep Dream is actually producing art. When asked, art critic Ben Davies said: “Of course it’s art! There’s no limit to what you can classify as “art.” The question is only ever whether it’s good art. And people seem to be very amused by it.”

So, if it is art, who owns the copyright? I have to admit that I had not considered this to be an interesting legal question as I initially thought that it was evident that the copyright belonged to whoever took the original photograph, or painted the picture. But I have been surprised by the number of people who say that the resulting image has no copyright because machines are not human, and therefore are not subject to copyright protection. My opinion is that they are looking at it the wrong way, but I seem to be in the minority. It’s the monkey selfie all over again!

I think that the problem might be that people hear the words neural network, dreaming, and artificial intelligence, and get caught up in the technical process itself, believing that because the machine is making independent decisions, it is actually thinking. Roughly speaking, the law does not care how the image came about from a technical perspective; for the purposes of copyright law the question is one of agency and originality.

Copyright law actually can account for machine-made works. In UK copyright law, the definition of authorship allows for ownership of a work that has been created by a machine. Section 9(3) of the Copyright, Designs and Patents Act clearly states:

“In the case of a literary, dramatic, musical or artistic work which is computer-generated, the author shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken.”

So I took a picture I own and passed it through Dream Deeply, a web interface of the Deep Dream program, and selected Filter 12.


The system thought about it, and produced this nightmare.


So who owns the copyright on the resulting image? In the UK, I would say unequivocally that I still own the copyright based on section 9. I clearly made the arrangements necessary for the creation of the work. It doesn’t matter that the computer thought and independently produced a new image, the law is clear that the technical process is irrelevant, what matters is who made it happen.

To my mind, the interesting question is not one of ownership, but one of derivative work and transformative use. To what extent is the new image subject to its own copyright? Here we are presented with an issue of originality. In the UK I would tend to look at the Art 6 of the Copyright Term Directive (2006/116/EC), which says that “Photographs which are original in the sense that they are the author’s own intellectual creation shall be protected […]”. The Directive’s preamble defines original as a work that is the “author’s own intellectual creation reflecting his personality”, and the Court of Justice of the European Union interpreted the meaning of originality in Infopaq by stating that “choice, sequence and combination” of the elements of a work might be enough to warrant originality.

Personally, I think that Deep Dream images fulfil the requirement of originality described above, but I am willing to concede that one could make a strong argument to the contrary, and it all depends on the interpretation given to the law, the Directive, and Infopaq. I think that when selecting an image to place into the program, this is a choice that reflects my personality, but I am also aware that most of the operation is performed by the program without further input from me. I’ll be interested to read what others think, perhaps this is not an issue of originality at all!

There is one last question, and it is whether one would be infringing copyright by applying Deep Dream to an image one does not own. In general, the right to make an adaptation of a work (or a derivative in other jurisdictions) is an exclusive right of the author. So anyone applying through Deep Dream a picture they do not own, without the author’s permission, would be infringing copyright. But again, other jurisdictions where exceptions and defences are more lenient towards derivative and transformative use may have a different result.

Finally, the issue might be different in other jurisdictions. The originality threshold is much higher in the US as a result of Feist, so I find it unlikely that a computer-generated image from Deep Dream would meet the existing standards. It also seems that the question of computational authorship may be more open than it is on this side of the Atlantic, at least according to this very interesting article by Anemarie Bridy.

Comments 7

  1. In what possibly was the first piece on the issue of computer authorship, Pam Samuelson wrote that computers should not be treated as authors because they do not need incentives to create! (Allocating Ownership Rights in Computer-Generated Works, University of Pittsburgh Law Review)

    1. I have read some commentary online that has been concentrating on the authorship aspect, check the link to the Google+ discussion. I agree that this is not an issue about authorship itself, at least until true AI comes into play.

  2. If someone makes a painting of your photograph without your permission is that painting their intellectual property? (In this analogy deepDream is the painter. Talk amongst yourselves.)

    1. This is an area where the practice in the UK and the US is very different, and not only in name (we call derivatives “adaptations”). The definition of derivative work is narrower in the UK, and interestingly, it doesn’t include artistic works, mostly because these usually involve substantial copying.

      Roughly speaking, an unauthorised painting from a picture would be an infringement of copyright because it would be copying substantial part of the work. The question is one of whether the copying has crossed a threshold that makes the copying non-substantial.

      In the US this would be a derivative, and these are controlled more strictly by the author. However, under some circumstances it could come down as fair use.

Leave a Reply