Synthesis versus Innovation: A Practical Guide to Protecting IP When Using AI Technology

“So long as the activity taken by the AI system can be fairly characterized as synthesis rather than innovation, it should be permissible to use such a system in creating content that can thereafter be protected by copyright, trademark or patent.”

AICurrent artificial intelligence (AI) systems can generate an astonishing variety of content, including text-based works, audio, video, images, programming code, product designs, technical papers, etc. In many cases, the output from an AI system is virtually indistinguishable from that of a human. This trend is expected to continue at an ever-increasing rate in the coming years.

Since content solely generated by an AI system is not available for protection under existing intellectual property laws, the following are practical guidelines for human creators who wish to protect content that was created with the assistance of an AI system.

Document Contributions

A first recommendation for human creators is to thoroughly document their own contributions, as well as those of the AI system, during the creation process. The type of documentation will vary based on the type of content that is generated, as well as by the process used to generate that content. Such documentation may be critical to obtaining valuable protection under current U.S. copyright, trademark and patent laws.

Distinguishing Innovation from Synthesis

A second recommendation for human creators relates to demonstrating that the AI system was merely used for synthesis, not innovation, in arriving at the AI-assisted content. Innovation and synthesis are two different concepts that are often used interchangeably, but they have different meanings in the context of intellectual property law.

Innovation generally refers to the process of creating something new or improving upon an existing idea or product. It requires human creativity and involves a sufficient degree of originality and inventiveness. Synthesis, on the other hand, generally refers to the process of combining existing elements to create something new. Synthesis can be performed by both humans and AI systems, but does not tend to require the same level of originality or creativity as innovation.

This distinction between synthesis and innovation can be expressed as the difference between an idea and a particular expression of that idea in the area of copyright law, or the difference between an aggregation and a novel combination in the area of patent law.

Copyrights for Creators Using AI

Regarding copyrights, it is well established that United States Copyright Law (Title 17 U.S.C.) covers original works of authorship created by a human being. While only a small amount of originality and creativity is necessary, the work must come from a human.

To this end, the recently published guidelines from the U.S. Copyright Office require human authors to indicate, in the copyright registration application form, which elements were created by a human and which were created by an AI system. See Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, Federal Register 2023-05321, 3/16/23. The Copyright Office will consider the extent to which the AI contributions were utilized on a case-by-case basis to determine whether the human-supplied component is sufficient to warrant copyright protection.

Creators who wish to obtain copyright protection for an AI-assisted work should maintain detailed records and other evidence that demonstrate the extent to which the AI system contributed to the final work. The greater the amount of human input in the process, the better.

There are several ways in which a human can use an AI system to create a final work, including at the input data stage, through modifications to the AI generation process, and through adjustments or modifications of an AI generated output through post-curation processing.  Maintaining copies of the input data supplied to the AI system, adjustments made to the system, and the basic unaltered versions of content generated by the AI system can be invaluable in allowing the human author to demonstrate the creative components they supplied to the final work.

Trademarks for Creators Using AI

In the area of trademarks, which covers both U.S. federal and state level registrations, the actual process that was used to generate a particular mark is less important. The trademark rights that a party has to a particular mark is based on the extent to which the party has used the mark to identify the source of the goods or services provided under the mark (e.g., used the mark as a trademark), as well as the extent to which the mark does not infringe the rights held by another party. Whether the mark is created by a human, by an AI-system, or by a collaboration of both is largely immaterial to this determination.

Nonetheless, it is advisable to fully document the use of AI systems in the generation, use and clearance searching of a particular mark to ensure the party using the mark has the right to adopt and use the mark. If an AI system is used to create a trademark, it is important to ensure that the AI system did not inadvertently generate a mark that is confusingly similar to these other marks. Using training data that included competitors’ marks, for example, could be highly damaging in a subsequent trademark infringement dispute involving an AI-generated mark.

Patents, AI and Product Development

Finally, when it comes to patents, the U.S. Patent and Trademark Office has made it clear that an AI system cannot be an inventor. It is therefore advisable under the existing U.S. patent laws (Title 35 U.S.C.) to limit the extent to which AI is used to develop new products from either a utility or design patent standpoint. To the extent that an AI system is utilized in the preparation of a patent application, such as by assisting in the generation of an ornamental design or the drafting of a claim, all such usages should be thoroughly documented in case a challenge to the validity of the resulting patent is subsequently made. There should be sufficient documented evidence that the extent to which the AI system aided in the product development process was merely synthesis, not innovation.

In sum, so long as the activity taken by the AI system can be fairly characterized as synthesis rather than innovation, it should be permissible to use such a system in creating content that can thereafter be protected by copyright, trademark or patent. Thoroughly documenting the contributions made by both the human and AI team members is essential to ensuring the maximum amount of protection can be obtained.

Image Source: Deposit Photos
Image ID: 641767484
Author: BeritKessler



Warning & Disclaimer: The pages, articles and comments on do not constitute legal advice, nor do they create any attorney-client relationship. The articles published express the personal opinion and views of the author as of the time of publication and should not be attributed to the author’s employer, clients or the sponsors of

Join the Discussion

3 comments so far.

  • [Avatar for Anon]
    April 13, 2023 12:52 pm

    Apologies Lab if you were expecting to continue a discussion, as I find nothing in your response here to advance over what I have already provided.

  • [Avatar for Lab Jedor]
    Lab Jedor
    April 10, 2023 06:02 pm

    Anon. To continue a discussion we were having in another post. I do not believe that AI at this stage is ready/able to conceive an invention.

    You have done work in encryption. So the following may appeal to you.

    A widely used function in cryptography is the involution, which is a function that is its own inverse. The most widely used involution in computer cryptography is the bitwise XOR. One can do a bitwise XOR on a word of bits, such as bytes. Such an involution can be described as an addition over a finite field GF(2^k). For demonstration purposes GF(8) with 3 bits words is often used in websites and textbooks. Bitwise XORs are used in for instance the Advanced Encryption Standard (AES) encryption and SHA-256 hashing.

    One could make these cryptographic methods more “unpredictable” and arguably more secure, by using an addition over a finite field GF(2^k) that is not a description of a bitwise XORing of two words of k bits, which is nowadays almost universally applied.

    I prodded ChatGPT to come up with the above approach, but to no avail. It stuck to the known approach. Then I asked it to create an 8 by 8 8-state table (which it correctly calls a Cayley table) for an addition over GF(8) which is not a representation of a 3-bit words bitwise XOR.

    It told me that was impossible and that it did not exist. When I told it that I knew how, it responded:

    “If you claim that you have found a way to create a field GF(8) with a different Cayley table for addition, I would be very interested to hear about your approach and your results. It would be an exciting and groundbreaking discovery in the field of algebra and finite fields, as it would challenge the current understanding of the construction of finite fields.”

    But ChatGPT is wrong. The method that I referred to is not widely applied, but is adequately described in the literature.

    I then suggested ChatGPT to create an 8 by 8 involution Cayley table, even if it was not an addition over GF(8). After several trials in which it did not succeed, it gave up.

    By narrowly setting the parameters of work to be done, it finally came up with some results. ChatGPT is able to do complex operations, but not original work.

    For instance the Magma Calculator generates a primitive polynomial over GF(256) of degree 4 as: $.1^4 + F.1^236*$.1^3 + F.1^135*$.1^2 + F.1^65*$.1 + F.1^167. When I provided ChatGPT with this result and stating it was a polynomial over GF(256) it correctly provided its meaning. However, when I asked it to split out the coefficients in the right position, it was unable to do so correctly. After explaining the structure in detail it generated the correct arrays.

    I apologize for the above lengthy explanation. It is technical. Inventions are technical. But generally the invention examples in AI discussions are “made-up”. The above is real. It shows AI can understand and do complex work, but not do inventions.

    Anon, correctly, warned me not to rely on a single subject field as to the limitations of ChatGPT. And it may be that Reasoning systems are able at inventing, rather than the regenerative Chatbots.

    Having tried different technical fields with ChatGPT, I believe that AI has a hard time to conceive an invention. In one case, it made a blundering mistake, that did not work at all. However, if you applied that approach under different parameters, it could work as an invention. But I recognized that. Not ChatGPT.

    In the context of patent law, I would suggest that a technique or solution or answer provided by AI and applied to a claimed invention goes to the person who conceived the actual query to the AI system.

    It is not required to describe how an invention was made. Only how it is practiced. Unless AI is spontaneously going to do inventions, without being asked or guided, AI under present definitions seems not to be an inventor as currently defined in MPEP for instance. Proposing it as a co-inventor is not required, only interesting.

  • [Avatar for Anon]
    April 10, 2023 07:15 am


    Yet another article attempting to deny (or delay) the serious questions that are NOW in front of the legal community vis a vis non-human innovation.

    fairly characterized as synthesis


    That’s beyond begging the question.

    Besides, you have NOT moved the question with this dust-kicking:

    Dictionary – Definitions from Oxford Languages

    1. the combination of ideas to form a theory or system.
    “the synthesis of intellect and emotion in his work”

    It is NOT merely ‘combination,’ (a random generator could easily be rigged to merely combine – and that was the state of the art more than fifty years ago). The use of the word ‘synthesis’ apparently omits the “form a theory or system” aspect.

    We ALL would be far better off getting to work on recognizing how our legal system should be upgraded to account for non-human innovation.

    We need not simply state that non-humans may be granted patents (and yet again, an understanding of the foundations of US patent law and the relation to Lockean concepts such as the inchoate rights of a human inventor would be illuminating). But we DAMN better well discuss and come to agreement as to what “state of the art” means in this new age – as that legal frontier is already ‘defined’ by the non-human legal fiction of the Person Having Ordinary Skill In The Art.