Table of Contents for Who Wrote This?
Prologue: Human Writers Meet the AI Language Sausage Machine
The Prologue invites us into the burgeoning world of artificial intelligence as author. Since the dawn of literacy, writing has been a uniquely human preserve. But with the coming of AI, we now have competition. This chapter highlights both potentials and challenges in today's AI developments, especially as they relate to language issues. We're introduced to AI text-generation programs like GPT-3, along with examples of the kinds of bias such programs can unleash and some potential effects on people in writing-intensive professions. The chapter concludes by laying out eight key questions that Who Wrote This? explores and by previewing how the book is organized.
1.The Journey to Literacy
Chapter 1 describes the origins of writing. While this development happened multiple times in human history, the purposes it has served varied, as did the number of people who became literate. A central question is what mental effects literacy has on us, including on the way we think and on our brain structure. We look at hypothesized connections between emergence of the Greek alphabet—or literacy more generally—and the rise of Greek philosophy. The chapter closes with examples of intriguing neurological research on how reading and writing rewire our brains and affect their functioning.
2.Why Humans Write—and Rewrite
Chapter 2 delves into what it means to be an author or writer. We explore how, in the English-speaking world, people gained ownership over their written words. This discussion leads to issues of copyright: What do laws say, and where do works produced by computers figure in? We then turn to human motivations for writing, ranging from writing for a living to sharing with others, self-discovery, or personal release. These purposes are compared with AI programs that generate text but are incapable of motivation. The chapter concludes by looking at rewriting as part of the larger writing process.
3.English Comp and Its Aftermath
Chapter 3 recounts how English composition became a ubiquitous college requirement in the United States. The birth of English comp at Harvard University in the late nineteenth century reflected a broader transition from education based on rhetorical skills to pedagogy grounded in written assignments and examinations. From the earliest days, a major challenge was the amount of time needed to grade student compositions. With the development of computers, a logical question was whether machines might do some of the work. That prospect would lead to the Educational Testing Service creating automatic scoring programs for some of its standardized tests, using natural language processing. Similar technology would later surface in software programs for assisting everyday writers.
4.The Dream of Language Machines
Chapter 4 presents the rise of artificial intelligence as a research endeavor. From the start, language was a central puzzle to be tackled. The chapter introduces Alan Turing's work on cryptography during World War II, but then follows his pursuit of the question of whether computers might be capable of intelligence. The story of AI then shifts to the United States, where in the mid-1950s, the term "artificial intelligence" was coined. The remainder of the chapter is a layperson's roadmap to what the field of AI encompasses, dealing not just with language but with information, embodiment, vision, creative works, games, and science. In later chapters, we'll see how AI techniques developed for one branch of inquiry have been harnessed for addressing questions in other areas, including written language.
5.The Natural Language Processing Sausage Machine
Chapter 5 offers a tour of natural language processing. It opens by introducing concepts that lie at the heart of how AI is used to generate and "understand" human language. These include machine learning, neural networks, and transformers—the computer model underlying today's powerful tools like GPT-3. After probing what is meant by a "natural language," the chapter turns to elements of human natural language to be processed. The discussion looks at how AI handles production and perception for both speech and writing. The chapter closes by summarizing how AI takes on many of the tasks for which humans have traditionally used natural language.
6.Machine Translation Rises Again
Chapter 6 leads us from the earliest days of AI to the present by exploring the early failures and eventual successes of machine translation. As far back as the late 1940s, there was speculation that cryptography was a possible model for computers translating from one "code" to another. US efforts were driven by hopes of rapidly translating Russian scientific documents, though by the mid-1960s, funding stopped due to lack of progress. However, independent translation projects in the United States and Canada proved that basic machine translation was viable. The enterprise eventually blossomed into tools like Google Translate. Despite these successes, problems remain. AI translations are not always accurate, plus they commonly show bias, especially regarding gender.
7.Machines Emerge as Authors
Chapter 7 begins in England, with Christopher Strachey using 1950s computer technology to write love letters. But most of AI's early steps in authorship were American, and they involved prose that had a story line. These stories were initially embedded in computer games, which helped lay the groundwork for the flowering of interactive fiction. The chapter introduces Tale-Spin, which inspired subsequent AI storytelling projects, and hypertext fiction, which proved a short-lived genre. We then turn to AI as author in pragmatic contexts, beginning with the rise of scientific abstracts and then getting computers to generate them. The chapter closes with an example of how AI was harnessed to write customer service letters and more recent instances of AI generating text for advertising and marketing.
8.AI Comes for the Writing Professions
Chapter 8 weighs effects of AI on professions that heavily involve writing. We begin by looking at how earlier technologies affected the labor market, and then compare having a job versus having a "good" one (measured not just by salary but by providing mental stimulation and a sense of purpose). The chapter focuses on journalism, law, and translation. In the journalism world, AI has already been writing stories for over a decade. It's unclear how much AI or other factors account for the ongoing decline of humans in the newsroom. In the legal domain, AI research and writing tools have become increasingly sophisticated, though there's debate over how much they will replace lawyers. For translation, the handwriting is already on the wall. Opportunities for full human translation are increasingly being replaced by jobs doing post-editing of what machines have already translated.
9.The Creative Side of AI
Chapter 9 asks if it's meaningful to speak of AI as being creative. To help answer the question, we begin with the human side of creativity. There's a broad literature attempting to define what constitutes human creativity. We look at rubrics for measuring levels of creativity, along with whether it matters that multiple people come up with the same idea. How do people come to be creative? Among the potential factors that have been proposed are IQ, genes, the zeitgeist, and personality and frame of mind. The topic then shifts to the AI side of creativity, presenting examples from music, art, and writing. The chapter concludes by comparing how AI creative works stack up against human creativity. ChatGPT is invited into the conversation.
10.AI as Jeeves
Chapter 10 proposes Jeeves (the valet to a character in humorist P. G. Wodehouse's novels) as an image for AI software that can edit text that humans write or that has a hand in new writing. The chapter opens with AI tools that everyday writers use for correcting and completing what we write. These include spellcheck, autocomplete, and predictive texting, along with grammar and style programs like Microsoft Editor and Grammarly. AI as Jeeves also generates new text with options like Gmail Smart Reply and commercial products for drafting longer passages. While all these tools lend efficiency, they aren't always accurate, they reduce incentives to develop and retain human writing skills, and they diminish our individual writing voice. The chapter concludes by looking at plagiarism and at cheating more broadly, along with AI programs for detecting them.
11.Human–AI Symbiosis
Chapter 11 follows evolving notions of the relationship between people and computers. One perspective sees AI as automating cognitive tasks and replacing people, while another looks to AI augmenting human abilities. Since the 1980s, computer scientists have been talking about ways in which humans and computers might interact. The initial concept was "human–computer interaction." However, as the discipline of AI evolved, the language shifted to "humans in the loop" (indicating that computers weren't functioning independently) to most recently "AI in the loop" (stressing that humans remain in charge). The chapter presents examples of human–AI collaboration and co-creativity.
12.Do We Always Welcome AI?
Chapter 12 reports on survey research I undertook with about 200 young adults in the United States and Europe. My goal was to explore what everyday users thought about AI as a writing technology. Main survey topics included spelling, editing proofreading, and software that writes for you, such as predictive texting and programs that can generate essays. In addition, I asked questions about writing by hand (as opposed to using a digital device) and about the potential impact of AI on writing-intensive jobs. Highlights of survey results include rich examples of what individual participants had to say, in their own words. For handwriting, the chapter goes on to connect the dots between people's attitudes regarding advantages of handwriting (such as retaining your own writing voice) and concerns about AI as Jeeves constraining what we write.
Coda: Why Human Authorship Matters
The Coda takes stock of what we've learned over the course of the book, with the underlying focus on why it matters that humans retain their role and skills as writers. But it's up to individuals to make their own choices about what their personal approach looks like. To help frame those choices, the chapter proposes keeping an informal scorecard that centers around eight questions posed back in the Prologue: What's your motivation for writing? Is AI a threat to human written creativity? Which writing skills are worth keeping? Can you AI-proof your personal writing voice? Is AI redefining authorship? Does AI threaten professions built on writing skills? Where do you draw the line between collaboration and handing over the keys? Will disclosure rules help? The chapter closes with a personal example for my own scorecard.