ARTIFICIAL INTELLIGENCE, COPYRIGHT, AND COPYRIGHT INFRINGEMENT.

• Author: Naqvi, Zack

1. INTRODUCTION II. BACKGROUND A. Defining AI B. United States Copyright Law 1. What is U.S. Copyright Law 2. Policy of U.S. Copyright Law 3. Elements of Copyrightable Material 4. Copyright Protection for Sentient AI 5. Copyright Infringement C. Agent-Principal Relationships for Artificial Entities 1. Can AI be Agents D. AI Potential Worker Displacement III. LEGAL ANALYSIS A. AI and Copyright B. Treatment of AI as an Agent of a Principal 1. Principal Liability for AI-Agent Infringement 2. Principal Can Also be Liable for Torts Committed by the AI C. Treatment of AI as a Consumer Product 1. AI as a Tool 2. AI Producers should not be Secondarily Liable for Resulting Actions of Consumers Direct Infringers 3. AI Producers Should be Secondarily Liable if They Encourage users to Infringe using AI Products D. AI Copyright Infringement Does Not Equate to a Products Liability Claim E. Independent Creation Doctrine Should Not Apply to AIs as Employees F. AI Should Not be Treated as a Corporation for Copyright Purposes G. AI Work Should Not Automatically End up in the Public Domain H. There is No Copyright Protection for Someone's Voice, Name, or Likeness I. There Can be a Right of Publicity Claim for Someone's Voice, Name, or Likeness IV. NON-LEGAL ANALYSIS A. The AI-Agent Does Not Need to be Paid B. Effects of AIs on the Social Security and Income Tax System Can be Overcome C. Using No-Fault Insurance to Offset Potential Infringement by AI V. CONCLUSION I. INTRODUCTION

   This article will examine how current copyright law can be applied to work product created by artificial intelligence ("AI"). The field of AI has been consistently and rapidly growing over the last half-century and is projected to become a $70 billion industry by 2020. (1) Modern AI software is becoming increasingly complex and can produce work that has never been created by computers before. (2) For example, AI can be trained to create music, (3) art, (4) short films, (5) poetry, (6) and news stories. (7) The work that AI creates also falls under the domain of copyright law. (8) Who would own the copyright in a case where an AI program creates new artwork? What if AI wrote a song that "voices never share[d]?" (9) If AI accidently copies a piece of copyright work, would we hold the AI liable or the AI's creator liable? Answers to these questions are still ambiguous under current copyright law. (10) As AI technology continues to advance, courts will need to address these questions and figure out how to treat AI work products under the current United States copyright schema.

   This article will examine how current copyright law is likely to treat work that is created autonomously by an AI software. Part II of this article will define AI and draw a distinction between AI and sentient robots. It will also offer a crash course in basic United States copyright law and its applicability to works created by AI. It will also explain how AI is predicted to displace human workers and explain how AI can disrupt social security and income tax. Part II will also introduce the concept of the agency-principal relationship in order to explain how it can be applied to AI and its owner.

   Part III of this article will analyze the legal issues regarding AI and copyright. It will first explain how AI are not afforded copyright protection because current copyright law will not treat AI as authors. Part III explains how AI can be treated as either (1) an agent of a principal or (2) a consumer product, and based on this classification, copyright protection will be granted to either the principal of the AI or to the end-user of the AI. Part III also examines legal issues in treating AI as a corporation and treating AI-produced work as public domain. Part III will finally explain how there is no copyright protection offered for someone's voice, image, or likeness.

   Part IV of this article will finally examine non-legal issues such as compensation for the AI-agent, effects of displacing human workers, and insurance as a way to offset AI infringement. It will also examine whether the United States government should tax AI and how social security can be affected with largescale AI adoption.

   Ultimately, this article will show that AI can be treated either as (1) agents of a principal or (2) a consumer product. If the AI is an agent, the principal will own the copyright whereas if the AI is a consumer product, the end-user will own the copyright.

2. BACKGROUND

   1. Defining AI

      Artificial intelligence ("AI") is a set of techniques or instructions that are aimed to simulate some aspect of biological cognition using machines. (11) Early computer scientists theorized the use of abstract symbols combined with logical reasoning as a way to simulate AI. For example, early computer scientists used basic algorithms to successfully create programs that utilized heuristics, or "rules of thumb," that could help them accomplish tasks like running through a maze. (12) However, because of the complexity and cost of computer technology, many of the other AI theories did not yield any fruitful results. (13)

      Nowadays, almost all AI techniques are based on a technique called machine learning. (14) Machine learning uses computer algorithms that can "learn" or improve performance over time on a specific task. (15) These algorithms allow the AI to figure out the best way to accomplish the assigned task without needing to be pre-programmed with specific instructions. (16) The learning process of the AI is considered to be a "black box" because the programmer only sees the input and output, but cannot directly observe the AI's learning process. (17)

      An example of machine learning is teaching an AI to recognize an image of a cat. The AI is first shown thousands of images of cats. The AI's program contains numerous nodes (also called a neural network) that work together to focus on different aspects of each image. (18) For example, some nodes in the network focus on color and brightness differences between adjacent pixels while other nodes work together to find the edges of the image. (19) Other nodes focus on repeated shapes (like the nose of the cat) in the image and their relative positioning to other shapes in the image (like the eyes of the cat). (20) As the AI goes through iterations of its instructions, it pieces together elements of a typical cat's face from the information it has learned.

      The AI learning process is excruciatingly slow in the beginning because the AI is not given any instructions on what features of the cat it should focus on. However, once the AI goes through a few thousand iterations of cat pictures, it can strengthen its network and slowly "learn" the features of a cat and can even generate unique cat faces on its own. (21) The machine learning methodology has also been applied to creating movie trailers and writing newspaper articles. (22)

      Certain AI is designed to rewrite their code to improve performance. (23) This conjures images from Hollywood films of a dystopian future where robots are out to kill humans. (24) Let me be clear: for the purposes of this article, AI should not be confused with robots seen in films or TV. The AI depicted in pop culture are sentient; real-world AI are not sentient. Sentience is the capacity to feel, perceive, or experience subjectively. (25) Part of sentience includes learned behavior and the free will to not follow that learned behavior. (26)

      Let's take the example of Bender, a robot from the TV series, Futurama. Bender is not modern-day AI because he is a sentient being. Bender has free will and is able to perform any action, including hurting humans, regardless of what his core code instructs him to do. (27) Even though Bender can learn new behaviors, his sentience allows him to choose not to perform that behavior.

      Modern AI lack free will because they are still preprogrammed with a set of instructions to learn a new task. (28) Even though they can choose different courses of actions to complete the task, they are impeded by their core code. (29) The AI's analysis does not exude any real skill or judgment; it simply looks for patterns based on its given instructions and gives an output. (30) While AI's ability to "learn" demonstrates some level of skill, the AI cannot choose to exercise free will and not follow its instructions. (31) Furthermore, AI programs can only operate in a predefined manner and are unable to complete tasks outside their field of intelligence. (32) Therefore, it may be some time before AI gets to the level of complexity as robots depicted in film and television. (33)

      Since AI is not a sentient being, we must examine how United States copyright law will treat AI that creates copyrightable material. (34)

   2. United States Copyright Law

      This section will explore United States copyright law and show how AI that create copyrightable products are not afforded copyright protection.

      1. What is U.S. Copyright Law

         The United States Constitution grants Congress the power to enact laws in order to "promote the Progress of Science and useful Arts." (35) Copyright law is codified under Title 17 of the U.S. Code, and protection under the Copyright Act applies to original works of authorship that are fixed in any tangible medium of expression. (36) The Copyright Act gives the owner the following exclusive rights: (1) make copies of the work; (2) create derivative works based on the copyrighted work; (3) distribute copies of the work; and (4) perform and display the work publicly. (37) Copyright protection applies to: (1) literary works; (2) musical works; (3) dramatic works; (4) choreographic works; (5) pictorial, graphic, and sculptural works; (6) motion pictures and other audiovisual works; (7) sound recordings; and (8) architectural works. (38) Copyright protection cannot be applied to ideas, procedures, processes, systems, methods of operation, concepts, principles, or discoveries. (39) In other words, copyright protects the...