How artificial intelligence works in relation to copyright

Kat friend Jakub Wyczik provides an enlightening discussion of how AI and the law, especially copyright, intersect, with particular attention to the technical operation of AI.

Images generated using text-to-image technology, including DALL-E 2, Midjourney, and Stable Diffussion, have become a main topic in AI. Not only do many express wonder about how these systems work, but there is even doubt whether such users even have rights to such outputs. Let us briefly examine how these systems work and then consider how intellectual property law relates to them.

How it's made

Knowing how something works is necessary when discussing law and technology. To create a system that can generate new paintings in the style of famous artists, one would have to learn to paint in their style. However, what is a style, if not a set of specific common characteristics. We can use computers and data science for this purpose. To do so, we need to create a large set of images labeled with descriptions of their content.

Then we perform the learning process. Most of the image-generating solutions employ a technique called "latent diffusion." It is a way of creating algorithms (models) that use noise. First, the software turns the original image into noise, namely a random pattern of pixels (such as white noise or snow). Then, by trial and error, the computer gradually tries to reproduce the original image until it learns to do it almost perfectly.

During the above learning process, the program tries to find statistical relations between certain words and images, primarily where pixels of specific colors are usually located relative to each other for a given object. With current computers, this process can be performed automatically on many billions of images. One of the latest systems, Stable Diffusion, uses the LAION-5B database of five billion graphics scraped from the Internet.

If we already have an algorithm that can reproduce familiar shapes, we can also use it to create new images. However, the starting point for creation must be some description of the desired result, the so-called "prompt." So, a user types in the prompt and runs the program. The computer then creates noise based on a "seed," which is some value generated randomly by default.

The starting point can sometimes also be a particular graphic to which the noise will be added. Then, similar to the learning process described above, one begins with noise and description and then tries to guess what the image looked like at the beginning.

The point is that this image never existed. Although the model absorbs artists' styles, it should never duplicate any image from the training set (see more, here, in my guide to Disco Diffusion).

Element of randomness eliminates the work

Some might say that the same prompt means the same picture. Generated images are different, however, because of a specific randomness factor. Two users who use the same system and prompt will get different results. First, it is all about the above-mentioned "seed." The illustration below shows the results using the same prompt with a random seed value (noise).

    "Highly detailed painting of wilderness by Florent Lebrun, trending on Artstation" (Disco DIffusion)

Even choosing the same seed mostly will not bring identical results. The images will be similar because the system performed the same generation process. However, they will be slightly different, as if two people painted the same landscape.

        "Beautiful painting of the Polish landscape by Thomas Kinkade," seed: 21 (Disco Diffusion)

Nevertheless, since there is some degree of predictability, can one speak of a work? According to the United States Copyright Office, there is no author-human in such a case (see Thaler's case). However, this is nothing new. Several cases have weighed in on the issue of whether there is a work when external randomness occurs.

Thus, the case concerning a selfie allegedly taken by a monkey concluded that there could be no work because copyright does not apply to a monkey. It is not enough to decide to travel to Indonesia, befriend the animals, and set up the camera (check IPKat post here).

In the same way, the garden in Chicago's Grant Park is not a work. Authorship must result from an entirely human undertaking. Plant colors, shapes, or textures come from nature, not the gardener's mind.

Even if a human creates an initial image or prompt, this does not extend to the generated elements. Thus, one could only obtain protection for the structure of reworked images. The same applies to the common additional functions of such systems: inpainting (changing elements already existing in the image) and outpainting (painting an image beyond its original borders).

The situation would be similar to photobashing (using multiple digital assets such as pictures, models, and textures to transform them into new forms). Nevertheless, one should remember that even human contribution can be excluded by the "merger doctrine" or certain statutory exceptions (e.g., as "words and short phrases" - 37 CFR § 202.1(a)).

      Initial image--"Highly detailed painting of  beautiful caste  by Canaletto" (Stable Diffusion)

Still, one may ask how to distinguish a human creation from a machine? That is the key, devilishly difficult issue. Sometimes it is possible to quickly determine whether there is a watermark on the work (DALL-E 2) or that the file contains relevant metadata. Nevertheless, both the former and the latter can be removed quite easily.

If not the user, then perhaps the manufacturer of the model?

Sometimes one might argue that the creator of the model is the author of its outputs. The author believes that the answer to is "no." The model does not contain any copies of works. It is only a set of parameters derived from training data.

It is impossible to predict all results, just as Microsoft cannot predict what an Office user will create. Of course, this does not affect some restrictions on using someone's programs, but it is just a matter of choosing the proper license (e.g., Disco Diffusion under MIT and Apache 2.0).

How to live with the issue?

Let's start with the hope that copyright will survive and include protection for the outputs of AI. In practice, however, unless there is international harmonization on the issue, it will lead to balkanization on what is protected in the context of AI. The upshot is that a revision or a new legal regime may be necessary, especially if the courts are reluctant to confer such creations.

In the interim, at least, related rights, trade secrets, or protection against unfair competition, might provide some helpful alternatives. [A similar view was expressed in the 1980's when the legal community began to confront protection of computer software.] Related rights will only apply to video or audio, and, in any event, they are not universally recognized around the world.

Unfair competition and trade secrets protection were never designed to address ownership issues. More generally, why copyright law cannot provide the solution is the topic for a separate post.

The post was written as part of the results of a research project funded by the National Science Centre, Poland, grant no. 2021/41/N/HS5/02726.

All graphics were created using generative systems (AI).

How artificial intelligence works in relation to copyright   How artificial intelligence works in relation to copyright Reviewed by Neil Wilkof on Friday, December 02, 2022 Rating: 5


  1. Despite the misleading headline, I think this is a very useful contribution to the debate about AI and copyright, not least because of the helpful way in which the author describes the process of machine learning with respect to artistic works. It is worth remembering that other AI projects, such as those intended to produce 'literary' or musical works, will use different approaches/methodologies.

    However I disagree with the author's concluding paragraphs, especially the sentence "Let's start with the hope that copyright will survive and include protection for the outputs of AI." I think that going into the problem with this preconception will inevitably lead to the wrong (that is to say, unworkable) outcome. Far better that in addition to looking at other IP rights and how their subject matter is defined and protected (trade secrets, really?), we should consider entirely new approaches, as was done within the EU with regard to databases.

    If the maker of the algorithms and the selector of prompts are removed as being ineligible as creators (rightly in my view), that really only leaves the AI process itself as the creator, which not only runs up against the strictures of US copyright law, it also makes a mockery of the copyright term based on the lifetime of the author. Furthermore if AI works are viewed through the lens of the European concept of droit d'auteur then the protection of AI works faces a very uncertain future, perhaps analogous to the early days of photography where a mere photograph was treated as an utterly inferior medium barely worth mentioning in the early iterations of the Berne Convention.

    The question I rarely see being posed in this context is, why is protection for AI works desirable? We all know that under the US utilitarian view, the purpose is to stimulate and incentivize authors to produce new works for the benefit of society. Under the droit d'auteur approach the purpose is to accord the author both the respect and the just renumeration due to his efforts. Both of these are profoundly human motivations. What would motivate an AI machine produce a picture or piece of music? Clearly it is not because (at present) the machine has any sentience or desires of its own. I think the motivation still lies outside the AI machine, with humans who still want the same things as traditional copyright protection offers them: recognition and renumeration. However the human who is thus incentivized is excluded unless he or she somehow gains a sort of vicarious ownership (for instance, per Section 9(3) of the Copyright Designs and Patents Act 1988). This approach is somewhat similar to the provisions of Database Right. And if a similar sui generis approach were to be taken with AI this helps to resolve the problem of how to determine the duration of protection: a fixed term based on when the work is made or first published. I would go further and say that we would then have the opportunity, assuming the sui generis approach is adopted, of breaking free from the restrictions of Berne with its no formalities requirement, and making registration a prerequisite for AI right protection. This also has the advantage of distinguishing between human creations and AI creations, the former remaining outside the registration requirement.

    This approach would bring AI right in line with other rights such as design right, trade marks and patents where the primary purpose of the right is to protect a commercial investment, which I suspect will also be the main purpose of much AI generated work.

    1. Thank you for your comments. I would be glad to answer them because the problem is essential, and the debate should have started at least a few years ago. By now, we should have seen amendments to the regulations.

      Of course, there are many methodologies for artificial intelligence. I chose the one currently gaining the most popularity as an example. Unfortunately, the character limit for the post makes it impossible to address all the issues. However, the point is similar. In any case, a human does not create such a result. Just as ordering a portrait from a painter does not yet make the client the author of such a painting.

      I agree that we should change our way of thinking in this case. I hope that the conclusions were clear that I am against expanding the concept of work in this regard. However, due to this character limit, I have planned to discuss the solutions in detail in a separate post. The partial conclusion applies more broadly to generative systems using AI, not just those that generate images. After all, we won't be creating the law of images, but the intellectual property right regardless of the subject. Trade secrets can answer some problems of systems commonly used as part of business intelligence solutions, such as demand or sales forecasting. As for the creator, I agree. That is why put the references in the text regarding randomness and its impact on the non-existence of the work.

      Motivation is a fascinating phenomenon, but the two concepts above are related. Ownership (broadly defined) allows people to get rich and encourages them to multiply their wealth and thus develop new ways to create value in products and services that someone will be willing to pay for. So it just provides the development of society. Hence, the American and European concepts are the same, only showing the essence from a different perspective. However, I'm not sure that Sec. 9(3) of the CDPA adds anything (the concept of work in the UK is still related to a human being and carries the findings of past CJEU case law as retained law).

      I suggest that better than sui generis is a related right similar to phonograms or videograms in the EU. But by no means the same, and I can't imagine such a solution without an open catalog of fair use. I also think that not only AI-generated works serve commercial purposes. All creativity today is inherently tied to the essence of ownership described above, which is about the reward for the one who makes an effort to acquire certain resources (similar to John Locke's view).

    2. Of course, by open catalog of fair use (Polish wording), I mean a non-exhaustive list of exceptions.

  2. Apologies for managing to misspell 'remuneration' twice!

  3. How about a radical solution. 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.

    1. Truly radical. Can't believe it's not already part of the CDPA 1988.

    2. That is a "solution" known, e.g., in the UK. The question, however, is whether it is the right one. The phrase "the person by whom the arrangements necessary for the creation of the work are undertaken" brings, in my honest opinion, an unnecessary discussion of what "arrangements" means. Should the author be the creator of the software/machine, the person who chooses the settings/parameters, or perhaps all of them together, including the user, are co-authors of such works? The better wording is in India, Sec. 2(d)(vi) Copyright Act, 1957, states that the author of CGW is "the person who causes the work to be created." However, more is needed to solve the problem of the general requirement that the work comes from the author's mind rather than from external factors. Therefore, a more practical solution would be a regulation based on related rights.

  4. Why should "the requirement that the work comes from the author's mind" exclude work produced with the assistance of AI, as long as the process leading to the work is driven by human beings (prompt, training data selection and processing, software config etc) and the selection/identification of an AI output as "work" (i.e. for which copyright protection is desired) is performed by human beings ? This implies that rather than focusing the analysis on the AI tool and the AI output, the human inputs in a broad sense must be brought to the fore.

  5. @francis hagel,
    As far as my own comment above is concerned, I was speaking about the future use of AI since obviously today section 9(3) CDPA covers exactly the circumstances you describe, in the UK at least.
    The point about true AI (as opposed to what is the state of the art at present, a sort of computer aided creation process) is that it will have the quality of being autonomous: literally just switch it on and it will create things without the need for human prompting. I maintain that when that occurs the section 9(3) approach will no longer be valid.
    If you think I am being too fanciful in my predictions, just take a look at some of the ultrafast trading programmes employed in the financial markets today to see how hands-off they have become.
    It is the future state of AI creativity which I think requires a new paradigm. Sadly I suspect we may see its fruition before the legislators get around to legislating for it. Even though the DABUS litigation has been in the realm of patents, clearly it will not be long before the courts have to deal with copyright cases with many of the same ingredients. Thus we could get a series of, possibly conflicting, judgments which incrementally become the de facto commonlaw precedent. This worries me, when you see two experienced Lord Justices of Appeal (Arnold and Birss LLJ) disagreeing about the correct way forward in the UK DABUS appeal.

    1. For you, it sounds like with AI, the name of the game is : make human beings redundant. To me, this is no dream, rather nightmare.
      Apart from my personal view, I think it essential when looking at IP protection for AI creations (both artistic work and inventions) to carefully identify the human inputs involved in the creation and distinguish the work or invention purported for protection by resp copyright or patent filing from the output of the AI. For example, in the case of inventions, an AI output is not patentable in its own right, it takes human inputs for the critical filtering regarding novelty and value assessment, leave alone the inputs needed to set the objective, specify the software, prepare training data and train the AI system. As to artistic work, if I assume that the AI will typically provide not one but a great number of outputs, it will take human input to select the output having artistic value and to meet the requirement of copyright law for a human being to qualify as an author entitled to copyright.

  6. As far as this criterion of deriving a work from the mind of the creator is concerned, the principle is that there must be an intellectual link between what the creator comes up with and what is created. So if I type a prompt e.g. "Write me a short text about what benefits the development of artificial intelligence poses to intellectual property" and I get a reply from ChatGPT:
    "The development of artificial intelligence (AI) poses several potential benefits to the field of intellectual property. One of the main advantages is the ability of AI to quickly and accurately analyze large amounts of data, including information related to intellectual property. This can help to identify potential infringements and facilitate the enforcement of intellectual property rights.

    Additionally, AI can be used to automate many of the tasks involved in managing intellectual property, such as searching for prior art and assessing the novelty and uniqueness of a particular invention. This can help to make the process of obtaining intellectual property protection more efficient and cost-effective.

    Furthermore, the use of AI in the field of intellectual property can help to improve the accuracy and reliability of decisions related to the granting of intellectual property rights. By using AI to analyze large amounts of data and identify patterns and trends, decision-makers can make more informed decisions about the validity and enforceability of intellectual property rights.

    Overall, the development of AI holds great potential for improving the effectiveness and efficiency of the intellectual property system, and can help to protect and promote the interests of creators and innovators."

    It is difficult to say that this text is my product of intellect and I am the author. To check out the ChatGPT yourself, just visit the website:

  7. You are right to point out the benefits of using AI in the field of IP. This is already today's reality in the operation of IP offices and private IP service providers as well.
    As to text creation, AI tools are becoming impressively efficient. But thwhy should copyright

  8. Dear Mr Wyczik,

    Sorry for the mishap.

    As to text creation, AI tools are becoming impressively efficient. The sample you quote is neat, but also perfectly neutral. Why should this text be eligible for copyright protection ? At first glance, it does not show a personal touch. I am not a copyright specialist, and I know there are variations across jurisdictions. But it seems to me the text would become eligible for protection if a human being rewrites the text output of the AI and makes changes reflecting his/her personality.

    1. The point is that, just keeping with copyright rules, this text is not protected (assuming it was autonomously generated on my prompt). The theory is that copyright only protects the creator's contribution as shown in the work, so if I correct something in the above text from ChatGPT then I should really only be entitled to the changes I made as a human being. Everything else is unprotected. In practice, however, making the distinction is difficult and inefficient, and, as a result, should be unnecessary for trading. If the result of AI can have some value then why should it not deserve protection? Especially, since a market for such images already exists, and one even won an art competition with images created by humans:

      If, in the case of AI and generative systems, the classical rationale and principles of copyright law fail, why maintain a fiction in the name of the aspiration that copyright must be able to handle anything? Law is supposed to be efficient by design, just as business is supposed to generate profits. That is why, as I mentioned in the comments, I see potential in related rights. These do not require a human creator and protect potentially even CCTV footage. This does not mean, however, that copyright should disappear. Instead gain a little sister that will deprive the argument of lawyers who postpone ruling the cases by arguing that the subject matter in dispute is not protected by copyright, even when identical elements have been used in both works.

  9. AI systems that generate images, such as DALL-E 2 and Stable Diffusion, use a technique called "latent diffusion" to learn how to create images in the style of famous artists. This involves using a large set of labeled images to train the system to identify statistical patterns and relationships between certain words and images. Once the system has learned to reproduce familiar shapes, it can be used to create new images based on user-provided prompts. The resulting images are unique due to the randomness of the generation process. Because of this, it is unclear whether the resulting images can be considered a "work" under US copyright law.


All comments must be moderated by a member of the IPKat team before they appear on the blog. Comments will not be allowed if the contravene the IPKat policy that readers' comments should not be obscene or defamatory; they should not consist of ad hominem attacks on members of the blog team or other comment-posters and they should make a constructive contribution to the discussion of the post on which they purport to comment.

It is also the IPKat policy that comments should not be made completely anonymously, and users should use a consistent name or pseudonym (which should not itself be defamatory or obscene, or that of another real person), either in the "identity" field, or at the beginning of the comment. Current practice is to, however, allow a limited number of comments that contravene this policy, provided that the comment has a high degree of relevance and the comment chain does not become too difficult to follow.

Learn more here:

Powered by Blogger.