generative AI
Subscribe to generative AI's Posts

Senate Policy Roadmap Steers Generative AI Toward Transparency

In May 2024, the Bipartisan Senate AI Working Group released a roadmap to guide artificial intelligence (AI) policy in several sectors of the US economy, including intellectual property (IP). The group, which includes Senate Majority Leader Chuck Schumer (D-NY), Mike Rounds (R-SD), Martin Heinrich (D-NM) and Todd Young (R-IN), acknowledged the competing interests of positioning the United States as a global leader in AI inventions while also protecting against copyright infringement and deepfake replicas. According to the Working Group, a careful balance can be achieved by establishing two requirements for generative AI systems: transparency and explainability.

Under the current regime, AI inventors may hesitate to reveal datasets used to train their models or to explain the software behind their programs. Their reluctance stems from a desire to avoid potential liability for copyright infringement, which may arise when programmers train AI systems with copyrighted content (although courts have yet to determine whether doing so constitutes noninfringing fair use). Such secrecy leaves artists, musicians and authors without credit for their works and inventors without open-source models for improving future AI inventions. The Working Group proposed protecting AI inventors against copyright infringement while simultaneously requiring them to disclose the material on which their generative models are trained. Such transparency would provide much-needed acknowledgment and credit to holders of copyrights on content used to train the generative AI models, according to the Working Group. Although attributing credit does not absolve an alleged infringer of liability under the current legal framework, such a disclosure (even without a legislative safe harbor) may promote a judicial finding of fair use. The Working Group also identified the potential for a compulsory licensing scheme to compensate those whose work is used to improve generative AI models.

The roadmap also recommended a mechanism for protecting against AI-generated deepfakes. Under the Lanham Act, people receive protection against the use of their name, image and likeness for false endorsement or sponsorship of goods and services. But deepfakes often avoid liability through humorous or salacious misrepresentations of individuals without reference to goods or services. The Working Group advised Congress to consider legislation that protects against deepfakes in a manner consistent with the First Amendment. Deepfake categories of particular concern included “non-consensual distribution of intimate images,” fraud and other deepfakes with decidedly “negative” outcomes for the person being mimicked.

If Congress legislates in accordance with the roadmap, the transparency and explanation requirements for generative AI could impact IP law by creating a safe harbor for copyright infringement. Similarly, an individual’s name, image, likeness and voice could emerge as a new form of protectable IP against deepfakes.

Nick DiRoberto, a summer associate in the Washington, DC, office, also contributed to this blog post.

read more

Tragic Ending: Award-Winning AI Artwork Refused Copyright Registration

The US Copyright Office (CO) Review Board rejected a request to register artwork partially generated by artificial intelligence (AI) because the work contains more than a de minimis amount of content generated by AI and the applicant was unwilling to disclaim the AI-generated material. Second Request for Reconsideration for Refusal to Register Théâtre D’opéra Spatial (Copyright Review Board Sept. 5, 2023) (S. Wilson., Gen. Counsel; M. Strong, Associate Reg. of Copyrights; J. Rubel Asst. Gen. Counsel).

In 2022, Jason Allen filed an application to register a copyright for a work named “Théâtre D’opéra Spatial,” reproduced below.

The artwork garnered national attention in 2022 for being the first AI-generated image to win the Colorado State Fair’s annual fine art competition. The examiner assigned to the application requested information about Allen’s use of Midjourney, a text-to-picture AI service, in the creation of the work. Allen explained that he “input numerous revisions and text prompts at least 624 times to arrive at the initial version of the image.” He went on to state that after Midjourney created the initial version of the work, he used Adobe Photoshop to remove flaws and create new visual content and used Gigapixel AI to “upscale” the image, increasing its resolution and size. As a result of these disclosures, the examiner requested that the features of the work generated by Midjourney be excluded from the copyright claim. Allen declined to exclude the AI-generated portions. As a result, the CO refused to register the claim because the deposit for the work did not “fix only [Mr. Allen’s] alleged authorship” but instead included “inextricably merged, inseparable contributions” from both Allen and Midjourney. Allen asked the CO to reconsider the denial.

The CO upheld the denial of registration, finding that the work contained more than a de minimis amount of AI-generated content, which must be disclaimed in a registration application. The CO explained that when analyzing AI-generated material, it must determine when a human user can be considered the “creator” of AI-generated output. If all of a work’s “traditional elements of authorship” were produced by a machine, the work lacks human authorship and the CO will not register it. If, however, a work containing AI-generated material also contains sufficient human authorship to support a claim to copyright, then the CO will register the human’s contributions.

Applying these principles to the work, the CO analyzed the circumstances of its creation, including Allen’s use of an AI tool. Allen argued that his use of Midjourney allowed him to claim authorship of the image generated by the service because he provided “creative input” when he “entered a series of prompts, adjusted the scene, selected portions to focus on, and dictated the tone of the image.” The CO disagreed, finding that these actions do not make Allen the author of the Midjourney-created image because his sole contribution was inputting the text prompt that produced it.

The [...]

Continue Reading

read more

Copyright Office Seeks Comments on Artificial Intelligence

The US Copyright Office (CO) issued a notice, seeking comments on copyright law and policy issues raised by artificial intelligence (AI) systems. Artificial Intelligence and Copyright, 88 Fed. Reg. 59942 (Aug. 30, 2023).

The purpose of the notice is to collect factual information and views relevant to the copyright law and policy issues raised by recent advances in generative AI. The CO intends to use this information to advise Congress by providing analyses on the current state of the law, identifying unresolved issues and evaluating potential areas for congressional action. The CO will also use this information to inform its regulatory work and to offer resources to the public, courts and other government entities considering these issues. The questions presented in the notice are grouped into the following categories:

  • General high-level questions
  • AI training, including questions of transparency and accountability
  • Generative AI outputs, including questions of copyrightability, infringement and labeling or identification of such outputs
  • Other issues related to copyrights.

The specific questions can be found in the notice. Given the importance of using shared language when discussing AI, a glossary of terms is also provided, on which commentators can provide feedback. The CO indicated that it does not expect every party choosing to respond to the notice to address every question raised. Instead, the questions are designed to gather views from a broad range of stakeholders.

Written comments are due no later than 11:59 pm (EDT) on October 18, 2023. Written reply comments are due no later than 11:59 pm (EST) on November 15, 2023.

read more