Bard: Even though you're training your LLM on out-of-copyright works, the resulting outputs could be considered derivative works if they share substantial similarities with the copyrighted versions

Question for AI services: What if I have a multimodal LLM ingest and train on 'out of copyright works' relevant to superheroes, and then I produce new versions of those superheroes and use them in comics and movies in 2024. Will I get into legal trouble because more modern versions of those images of these superheroes - which may still be copyrighted - actually resemble the ones the multimodal LLM created on the basis of the 'out of copyright' versions?

ChatGPT-4’s answer:Using a multimodal large language model (LLM) to ingest and train on "out of copyright" works related to superheroes and then producing new versions of these superheroes for comics and movies in 2024 could lead to legal complexities. While "out of copyright" works are in the public domain and free to use, the resemblance of newly created versions to more modern, copyrighted versions of these superheroes could potentially raise copyright infringement concerns. This is because copyright law protects specific expressions of ideas, not the ideas themselves. If the new creations are substantially similar to protected expressions of these characters in copyrighted works, it could result in legal issues. It's crucial to consult with a legal expert in copyright law to navigate these potential challenges and ensure that your use of AI-generated content complies with current copyright laws and respects existing intellectual property rights.

Google Bard’s answer:It's impossible to say definitively whether you'd get into legal trouble without a full understanding of the specific situation, but there are definitely risks involved in your proposed scenario. Here's why:

Copyright and "derivative works":

• Even though you're training your LLM on out-of-copyright works, the resulting outputs could be considered derivative works if they share substantial similarities with the original, copyrighted versions. This could be the case even if the new versions appear different at first glance.

• Copyright law protects against unauthorized derivative works, meaning you'd need permission from the copyright holders of the modern iterations (e.g., comic book publishers, movie studios) to use your AI-generated characters.

Fair use:

• You could potentially argue fair use, an exception to copyright law that allows for limited use of copyrighted material for transformative purposes like commentary or criticism. However, fair use hinges on several factors,and using AI-generated characters in commercial works like comics and movies might not meet those criteria.

Additional factors:

• The extent of protection for character designs varies depending on jurisdiction and how unique the character is. Generic superhero tropes likely wouldn't be protected, but specific costume elements, backstory details, or personality traits could lead to copyright infringement.

• Even if you believe your work doesn't infringe on copyright, defending yourself in court can be expensive and time-consuming.

Recommendations:

• To minimize risk, consider consulting with a lawyer specializing in intellectual property to assess your specific case and potential legal concerns.

• You could also explore creating entirely original characters or modifying the AI-generated ones significantly to ensure they don't resemble copyrighted versions.

• Obtaining licenses from the copyright holders of the modern versions would be the safest route, but it might not be feasible depending on the specific characters and rights holders.

Remember, this is just general information, and it's not a substitute for legal advice. Always consult with a qualified legal professional for specific guidance based on your unique situation.

MS Copilot:This is a complex question that touches on the intersection of copyright law, artificial intelligence, and creative works. Here are some general points to consider:

1. Copyright Law: Copyright law varies by jurisdiction, but generally, it protects original works of authorship. If a work is out of copyright, it’s typically free to use. However, creating derivative works based on copyrighted characters could potentially infringe on the rights of the copyright holder.

2. Artificial Intelligence: AI can generate content based on its training data. If the training data includes only out-of-copyright works, the AI-generated content should theoretically be free of copyright issues. However, if the generated content closely resembles copyrighted works, it could potentially lead to legal issues.

3. Resemblance to Modern Works: If the AI-generated superheroes closely resemble modern, copyrighted versions of those characters, there could be a risk of infringement. This would likely depend on how similar the characters are, and whether the similarities go beyond non-copyrightable elements (like basic superhero tropes).

4. Legal Precedents: As of now, there are few legal precedents regarding AI and copyright law. The law is still catching up to the technology, and different courts may interpret the law differently.

Disclaimer: This information is intended to provide a general understanding of the subject matter. It should not be used as a substitute for legal advice from a qualified professional. If you have specific legal questions, you should consult with a lawyer.

Claude’s answer: I would be very cautious about creating and commercially exploiting new versions of copyrighted superhero characters without permission, even if the AI model was trained only on public domain materials. Here are a few key considerations:

• While older, public domain depictions were used for training, the newly generated images could end up including elements or visual similarities to more recent copyrighted versions of those characters. If the new versions are deemed substantially similar, it could still be considered copyright infringement.

• The most iconic superhero characters (Superman, Batman, Spiderman, etc.) are owned by major media companies who actively protect their copyrights. Even if you believe you have a legal argument for your actions, they have the resources to tie you up legally if they choose.

• You'd be exploiting the commercial value and goodwill associated with these characters which were built up by the IP owners over decades. Even if not strictly illegal, courts may look unfavorably on these types of derivative activities that appear to "free ride" for profit.

• Simply the threat of legal action from a large corporation, even if you ultimately prevail, could disrupt any commercial plans and be financially draining. The risks may outweigh benefits.

The safest approach commercially would be creating entirely new and original characters and stories not derived from copyrighted works. But I'd be happy to discuss further if you have additional questions! Let me know if any part of this is unclear.