iSchools Book Talk
Plagiarism and Artificial Intelligence
presented by Michael Seadle
The Book
Questions about authorship and originality have become urgent in an era where artificial intelligence (AI) and Large Language Models can write essays, generate images, and mimic human creativity. This authoritative book explores the intersection between plagiarism and forms of artificial intelligence that can generate written content copied from human-written texts.
This work discusses the legal and ethical history of plagiarism and shows how the rise of AI is changing how scholars define and measure plagiarism. The book also explores how AI could be used to discover plagiarism and highlights the risks of using AI when working with data and images.
Written by an established thinker on the topic of AI and plagiarism, Plagiarism and Artificial Intelligence will help librarians and researchers to understand the relationship between AI and plagiarism as well as those who seek solutions to the problems caused by using artificial intelligence systems.
The Author
Michael Seadle has been a Professor at the Berlin School of Library and Information Science at Humboldt-Universität zu Berlin, was Dean of Humanities, and was Executive Director of the international iSchools organisation. His most recent publication is Information Science: History, Ideas and Applications. He has previously published The Measurement of Information Integrity, Quantifying Research Integrity and Educating the Profession: 40 Years of the IFLA Section on Education and Training.
What makes this book special?
Plagiarism is a widely discussed problem for artificial intelligence, and one of the goals of this book is to explain why large language models depend heavily on plagiarism in order to function. In the simplest terms, the AI/LLMs get their speech ability by taking other people's words and restructuring them into sentences that the original author may never have spoken and may never have wanted to speak. The goal of this book is not to explain the mathematics behind how the AI/LLM systems work, but to explain how these systems transform people's words into mathematical tokens that the AI/LLM systems can combine using statistical probability to create entirely new sentences that no one ever actually said.
Companies use proprietary restrictions to keep the algorithms behind these systems secret, and one consequence is that outsiders cannot predict what the system is going to say. The goal of this book is not to show people how to build their own AI/LLM systems, but to help them understand what the risks are when these systems have unrestricted access to their words. All of the AI/LLM systems rely to some degree on plagiarism. They could use older language sources where the words are already in the public domain. That would be a legal approach, but the developers prefer to avoid giving an archaic tone to the speech that their systems generate.
Why should people read this book?
One of the reasons to read this book is both to understand the ways in which plagiarism laws try to protect you from plagiarism and how the laws leave openings for systems to take your words. So far, most AI/LLM systems come out of the United States, and many developers believe that the US "fair use" exemption applies to how they take other people's words and use them without permission. There is no evidence that AI/LLM systems have sufficient legal awareness to know under which circumstances the US Fair Use law does not apply. There is no European equivalent to the U.S. fair use law. The British "fair dealing" laws are far more restrictive than U.S. Fair Use. The internet makes the problem even greater because it sets no physical boundaries on who can get access to a person's words. Equally troubling is that some established companies that offer word processing tools have AI tools built into them, which makes them particularly vulnerable to AI/LLM theft..
What will people learn from this book?
One of the simplest things that people reading this book should learn is that everything they put into computing systems, including internet sites and word processors, could increase the likelihood that an AI/LLM system will reuse their words without any attempt to get permission. The fundamental assumption behind laws preventing plagiarism is that writers and speakers have a right to control over the words that they use. Many of the AI/LLM systems simply ignore those rights and take any sets of words that they can access. Designers who create AI/LLM systems have made no effort to build in controls to prevent plagiarism. That does not mean that such controls are impossible to construct. What it means is that the corporations building these systems care less about the importance of laws on plagiarism than about the profits that they can generate using other people's words.
Until law enforcement systems begin to crack down on the plagiarism inherent in AI/LLM writing, this will continue. There is no compelling reason why the process of turning words into tokens could not include a link to where they were used originally, which would make discovering plagiarism in AI/LLM systems much easier. This will only happen once countries establish far more serious financial penalties for AI/LLM systems that plagiarise.
iSchools Book Talk is a new iSchools feature series spotlighting books that members of iSchools would recommend to the iSchools community. These books can be presented either by the authors themselves or by any other iSchools member.
Please contact admin@ischools-inc.org in case you would like to feature a book as well.
Comments