Burying the Black Box: AI Image Generation Platforms as Artists' Tools in the Age of Google v. Oracle. (artificial intelligence)

• Author: Silverman, David
• Date: 01 January 2024

TABLE OF CONTENTS
                I. INTRODUCTION 116
                II. BACKGROUND 119
                A. Copyright, Fair Use, and its Implications in Google 119
                1. The Key Factors of the Fair Use Doctrine 121
                2. Google v. Oracle Expands Application of Fair Use
                Doctrine on Technology 122
                3. Transformative Use and its Role in the Fair Use Analysis 124
                B. Artificial Intelligence Image Generation Platforms 125
                1. User Experience on Image-Generation Platforms 126
                2. Inner Workings of Image-Generation Platforms 128
                3. The Black Box Problem in AI 129
                4. Conceptualization of AI as a Tool 130
                III. ANALYSIS 132
                A. The Court's Analysis of Purpose and Character of the Use in
                Google as it Relates to AI 133
                B. The Court's Analysis of Proportion and Substantiality in Google
                as it Relates to AI 137
                C. The Court's Analysis of Effect on the Market in Google as it
                Relates to AI 141
                IV. CONCLUSION 142
                

I. INTRODUCTION

When the Colorado State Fair's fine arts competition opened for entries, fantasy games designer Jason Allen submitted into the "digitally manipulated photography" category a piece he called "Theatre D'opera Spatial." (1) It was a rousing piece, to say the least, with a bold command of light, contrast, color, and detail, depicting a lavish futuristic pseudo-Victorian ballroom wherein figures dressed in finery observed a gaping portal to another fantastical world as if at a show. (2) It handily won first prize. (3) The crowd, however, was startled to learn that the art was largely the product of an artificial intelligence (AI) art generation platform called Midjourney, which Allen used to create the piece. (4) Allen attests that he spent many hours inputting the textual prompts that ultimately rendered the award-winning piece, claiming that his work was no less valid than anyone else's. (5) In response to his critics, Allen defied, "You said AI would never be as good as you, that AI would never do the work you do, and I said 'Oh really? How about this? I won.'" (6) Allen continued forebodingly, "[Artificial Intelligence Art Generating Tools are] here now. Recognize it. Stop denying the reality. AI isn't going away." (7)

As AI art generation platforms become increasingly prolific, the truth of Allen's defiant warning rings louder and louder. (8) But what exactly are AI art generation platforms? Generally speaking, AI art generation platforms are machines that receive user input in the form of written text and produce images that match the user input. (9) For example, you could input text prompts such as "an oil painting of a corgi wearing a party hat," and the platform would generate one or several images to those specifications. (10) The internal mechanisms represent a black box, where the AI's method of processing data is so dynamic and complicated that it is presently impossible to model manually. (11) With the Internet offering millions of points of data as the bot's database from which to pull information, trouble necessarily arises when the AI creates an image that clearly violates existing copyright, doing so either at the behest of the user or by maligned happenstance. (12)

Recently, the Supreme Court held in Google v. Oracle that the creator of a software was protected from a copyright infringement claim by the doctrine of fair use (fair use) where the allegedly infringing code was copied directly from copyrighted code. (13) The Court relied on a qualitative and quantitative analysis, holding that Google did not infringe Oracle's copyright because the final product used only a small portion of the copyrighted code, the portion it used was small in proportion to the total volume of original code, and the purpose that the final product served was different from the purpose of the copyrighted code. (14) This case heralds important consequences in how to navigate the murky waters of AI-generated imagery.

This Note argues that because AI image-generation platforms treat the copyrighted works from which they gather information as data in a larger string of information, Google v. Oracle presents the best model for assessing whether the generated image is protected by fair use because the case provides a framework for analyzing code as a tool in the production of a creative work and is instructive on each of the fair use factors.

With the burgeoning technology and all its commercial implications just on the horizon, legal discussion, or more aptly litigation, seems unavoidable to determine the usage of such technology, particularly about how image generation platforms learn from copyrighted material or produce a final product that is otherwise very similar to copyrighted material. For example, a recent class action lawsuit was filed against Microsoft, GitHub, and OpenAI on behalf of programmers who submitted lines of code to the open-source database GitHub. (15) The complaint alleges that GitHub Copilot, an AI platform that uses the GitHub database to generate code in response to plain text user inputs, violated the copyrights of those who contributed code to the database by failing to attribute the code it produced to the code from which it learned, even when the two were substantially similar. (16) In a separate case, a complainant who works in developing and applying AI sued the Register of Copyrights and the Director of the United States Copyright Office after they denied a copyright application. (17) The complainant listed himself as the owner and listed the author of the work as an AI image-generating program of his own making. (18) The complainant echoes Allen's warning in his insistence that AI is "going to be profoundly economically and socially disruptive, as [AI programs] evolve from essentially academic pursuits to those having significant commercial value, including in the context of personalized music, journalism, and digital art." (19) In fact, the United States Copyright Office has recently shifted course to focus more heavily on issues arising from AI. (20)

This Note will first discuss the current status of copyright law as it pertains to AI, with a specific focus on fair use and how courts interpret whether an allegedly infringing work is "transformative" by way of "altering the original with new expression, meaning, or message." (21) Next, this Note will offer a brief introduction on the methodology employed by AI image-generating platforms when creating an image in response to a user prompt. This section will then discuss the difficulties in mapping exactly what data points, and in what proportions, the platforms use when creating an image, also known as the "black box" problem. This section will also discuss different suggestions for viewing AI image-generation platforms as tools of the artists. Finally, this Note applies each of the fair use factors discussed in Google v. Oracle to AI image-generating platforms to demonstrate why the Court's analysis in the aforementioned case provides a pathway forward for cases surrounding fair use and AI-generated art.

With the rising popularity of various AI programs across different industries and markets, (22) the focus of this Note must necessarily be narrow. As such, this Note will not discuss fair use as it pertains to anything other than AI image-generation platforms. Moreover, this Note will not argue whether the user, programmer, or platform is fit for consideration as the primary author of a work fit for copyrightability, given the Copyright Office's stance on machine authorship. (23)

II. BACKGROUND

A. Copyright, Fair Use, and its Implications in Google

In general, Section 102 of the Copyright Act of 1976, codified as 17 U.S.C. [section] 102, sets the stage for applicable copyright law by establishing the broad categories of "original works of authorship fixed in any tangible medium... either directly or with the aid of a machine or device." (24) The phrase "original works of authorship" is vague by intent so as to incorporate the standard of originality "established by the courts under the present copyright statute." (25) The broad aspect of this language is specifically meant to mimic "the empowering language of the Constitution." (26) Critically, in the case of AI, the Copyright Office does not consider work generated by a non-human to be copyrightable. (27) Other countries hold similar schema, with Spain and Germany holding that works created solely by machines are ineligible for copyright protections. (28) Spain goes further, eschewing fair use altogether, instead preferring a payment obligation scheme. (29)

A key aspect of authorship within the confines of Section 102 is the topic of creative or original authorship. Title 37 of the Code of Federal Regulations requires that "[i]n order to be acceptable as a pictorial, graphic, or sculptural work, the work must embody some creative authorship in its delineation or form." (30) Section (b) of the same rule qualifies that "[a] claim to copyright cannot be registered in a print or label consisting solely of trademark subject matter and lacking copyrightable matter." (31) The Supreme Court referenced this requirement for creative authorship in determining that a telephone company and telephone book producer lacked a copyright over the content of its white pages as it was merely a compilation of names, towns, and telephone numbers, which lacked an essential aspect of creative authorship that turns otherwise publicly accessible information into a "copyrightable expression." (32) The manual for Copyright Office practices expands on the concept, adding that the work must be "the author's tangible expression of his [or her] ideas," paired with the conveyance of that expression in a tangible medium. (33) Expressions that convey a sense of message or meaning either in a definite sense, as might be the case with a sculpture of the human form, or more abstractly, through color or "modernistic form," are copyrightable. (34) Such original authorship may manifest in content but also in form, such as "the linear contours of drawing, the design and brush...