Introduction[1]
New York Times journalist Cade Metz has given readers the opportunity to “book snoop” at OpenAI’s beautiful two-story Victorian library within its high-tech San Francisco headquarters. Metz’s article, “The Old-Fashioned Library at the Heart of the A.I. Boom,”[2] is interactive, allowing the reader to zoom in on titles sitting on the shelves, thanks to Christie Hemm Klok’s brilliant photography skills. The library is remarkably low tech, inspired by the Rose Reading Room at the New York Public Library, Stanford University’s Green Library, and a “library-like bar inside the now defunct Nomad Hotel, 15 blocks south of the Rose,” according to Metz.
Some of the titles include the Oppenheimer biography American Prometheus by Baird and Sherwin; The Precipice by Tony Ord, which is about the existential risks facing humanity; Endurance by Lansing, about the doomed Antarctic journey of Ernest Shackleton; 2001: A Space Odyssey, of course; Cosmos by Carl Sagan; Old Man’s War by John Scalzi; the 2020 climate fiction book The Ministry for the Future by Kim Stanely Robinson; and Homer’s The Iliad. Thomas Jefferson’s Architecture is there, as is a book about birds and a parody of the same book written by ChatGPT. There are several titles by Neal Stephenson, a favorite author among tech elites; and naturally, Philip K. Dick’s dystopian novel Do Androids Dream of Electric Sheep?—which is set in San Francisco and the inspiration for the movie Blade Runner—was among the books.
Klok’s title picture captures the disquieting paradox of having an old-fashioned library inside the walls of a company whose technology has led to several lawsuits brought on by authors, periodical editors, and other artists for copyright infringement and plagiarism. The title picture was taken from within the tranquil wood-paneled library. A ladder leans against the bookshelf, and the lighting within the library is yellow-hued and non-LED, emanating from lamps scattered around the room.
This is contrasted with the window that looks out to some indiscernible part of OpenAI’s facilities, like a window into the modern world. The light from the window is blindingly bright compared to the library. The viewer can make out some kind of tubes and wires leading from somewhere and going somewhere else. It reminded me of a giant-sized version of the entangled cables behind my smart TV. The paneling outside the library is barely visible at the top of the window and starkly contrasts the wood paneling within the library with an airport terminal aesthetic complete with dingy off-white accents.
Whatever the library is trying to recapture, it is merely a simulacrum of the real. The library nestled within OpenAI’s headquarters is recreated, undoubtedly with pristine accuracy, but devoid of its time and place, and therefore, devoid of context. The library is a metaphor for the output ChatGPT produces, a simulacrum of reading and writing.
Three Ethics Questions
OpenAI CEO Sam Altman is a bibliophile who enjoys being surrounded by physical books. One of his favorites is The Beginning of Infinity by David Deutsch. Deutsch’s book is a celebration of the Enlightenment and human progress through better and better explanations for everything from how the world works to beauty to morality. OpenAI’s library, as seen through this lens, makes sense. The titles highlighted by Metz celebrate human ingenuity and caution against its powers. So far as I could tell, the shelves were devoid of any kind of homage to God, religion, or a higher being. Like the Enlightenment, the shelves enjoy the fruits of the Western tradition and the Scientific Revolution without acknowledging its religious foundations.
In considering OpenAI’s library and AI’s relationship to the human activities of reading, writing, and creativity, let’s consider three ethics questions:
- Are ChatGPT and other LLM chatbots engaging in a kind of sophisticated plagiarism? Or to put it another way, are these technologies engaging in “deepfake” reading and writing?
- What are we gaining and what are we giving up with this technology?
- What is the telos of this technology? Who does it serve?
Deepfake writing
The first question is whether what ChatGPT does is the same as what a human being does when he or she reads, acquires knowledge, and then writes a paper about it. To get to the ethics question at the heart of this, consider how I, as a human being, am constructing this article.
Right now, I am writing this article after reading the New York Times article that prompted the idea. It led me to ask certain questions about the details of the lawsuits, which also led me to ask about the nature of reading because I seem to recall Mortimer Adler saying something about reading being an active endeavor, and ChatGPT is passively consuming. I conducted some research and in so doing, happened to encounter an article in The New Atlantis that I thought would inform this discussion. I looked back at a couple of books and articles that I’ve read on the topic, and then synthesized these ideas into an essay.
I bring into my writing my experiences as a scientist, bioethicist, writer, and Christian who lives and works in a large city in the United States in the first half of the twenty-first century. Whether I am aware of it or not, I also bring my presuppositions and my experiences.
Importantly, I reject some ideas as untrue, speculative, in poor taste, or unnecessary, something that humans are capable of doing but ChatGPT cannot, at least not without human input. That is why companies like OpenAI hire actual humans to sift the data that is used to train ChatGPT and test its answers. (As a note, the Victorian-esque library, with all its romantic charm, is for the San Francisco employees. No word on whether the Kenyan, Ugandan, and Indian workers who sift data for a few dollars per day have such luxuries.[3] Or whether they had a say on which books to include in the library.)
When I write, I am not just sifting through the trove of digital knowledge available online. I am sifting through the very small number of materials that I have encountered in my life. I am also sifting this information through the lens of my particular life experiences. That is something that cannot be replicated by an algorithm, or another person, for that matter. The digital world wants to homogenize users into a flattened world, but the reality is that we each have a unique mental bookshelf and set of experiences that inform how we encounter and interpret the world. Even among Christians, where we ostensibly have a common worldview, a Christian living in London during World War II is going to have a very different perspective than one living in Dallas during Covid.
ChatGPT cannot produce writing from a lived experience. It can only mimic something that has already been written in the past. ChatGPT cannot digest, take in, or provide a sense of what it is like to be in this time and place, or even what it is like to be decades removed from a time and place and looking back. That’s why companies have to hire millions of gig workers to train their algorithms.[4] Writing is about a common human experience, and ChatGPT, for all its sophistication, is little more than an automated content mill.
Novelty at the cost of connection
Mortimer Alder’s How to Read a Book[5] helped me understand what we lose when ChatGPT “reads” books for us and then tells us about them. Adler likens reading to baseball. The pitcher and the catcher are both actively participating in the initiation and reception of the ball. Similarly, the writer is like the pitcher, the reader is like the catcher, and the book is like the ball. ChatGPT is upsetting because it intercepts the pitch. It takes what the writer wrote and utilizes it in a different way than intended. Rather than an asynchronous conversation between writer and reader, the writing is treated as mere feedstock. The conversation, interaction, and shared human experience between writer and reader has been co-opted for a billion-dollar efficiency tool.
Megan Hauser writes in her essay “AI Is a Hall of Mirrors” in The New Atlantis that AI cannibalizes as it serves.[6] According to Hauser, AI misunderstands something that is fundamentally human: the power of encounter. When it comes to reading, the human encounter is still there; as Adler points out, it is like pitching and receiving. When we feed books to an algorithm and allow it to distill down the ideas, it removes the two things that Hauser says we want from our digital institutions: to encounter other people and to discover new things. ChatGPT does away with both. When ChatGPT digests books and regurgitates a mediocre paper of aggregated “information,” it is an intrusion of the machine on a human activity that is different from the machine directing one human to another. It is the difference between a dating app and a robotic date. One facilitates connection while the other simulates it. Similarly, searches, websites, and online library portals facilitate connecting readers with writers’ ideas, while ChatGPT simulates reading and writing.
Despite the hype, ChatGPT does not serve anything new. It serves up amalgamations of the old, but not in ways that lead to greater understanding. Hauser colorfully describes AI as a “Cubist portrait on the things we have already said and made, that by combining our facets becomes either passing or passing strange.”[7] She describes it as giving us knowledge without knowing. It leads to the fundamental question: Does it matter that there is another person at the other end of this sentence or this essay? It depends on what you want out of your technology.
Telos of technology
Walker Percy, in his book Lost in the Cosmos: The Last Self-Help Book, summarizes the paradox of our modern predicament through a rhetorical question: Why is it possible to learn more in ten minutes about the Crab Nebula in Taurus, which is 6,000 light-years away, than you presently know about yourself, even though you've been stuck with yourself all your life?[8] Percy’s tongue-in-cheek self-help book is a commentary on the modern (i.e., twentieth-century) self. We know so much about so many things, but very little about ourselves, or anyone else, for that matter. AI is similar to searching within to find answers, only the “within” is all the content we have collectively created on the internet.
Rather than a twenty-first-century novelty, ChatGPT, and the AI hype surrounding it, is like the idols of old:
What profit is an idol
When its maker has shaped it,
A metal image, a teacher of lies?
For its maker trusts in his own creation
When he makes speechless idols!
Woe to him who says to a wooden thing, Awake;
To a silent stone, Arise!
Can this teach?
Behold, it is overlaid with gold and silver,
And there is no breath at all in it. (Hab 2:18–19, ESV)
Hauser contends that the true aim, the telos, of artificial intelligence is to make something that can transcend anything that has been made by man; as she puts it, “They want emergence. They want to create something for us that, finally, would be only for and in and of itself.”[9] Perhaps the Silicon Valley mortals want emergence, but I think a better word might be “deliverance.” In Isaiah 44, the culmination of the absurdity of man-made idols is when the carpenter burns half of his wood for warmth and food and then worships the other half saying, “Deliver me, for you are my god!”
Perusing OpenAI’s library virtually through the New York Times article is not the same as walking through a real library, running your finger across the book spines until you come to the book you’re looking for, and sitting down in a comfortable chair for an hour to read. You cannot smell the musty odor of old books or the aroma of coffee brewing. You cannot let your eyes wander over the titles until something serendipitously catches your attention. The experience of visiting OpenAI’s library via an interactive New York Times article is analogous to reading something written by ChatGPT—No matter how advanced the technology is, you are not experiencing it in a fully human way unless you are there.
References
[1] AI was not used to write any portion of this article for the same reasons why someone bothers to have a vegetable garden in the middle of a city of grocery stores, cooks a meal for guests when we have Door Dash, paints a landscape when everyone has smartphone cameras, or plays the piano when we can stream any music we want.
[2] Cade Metz, “The Old Fashioned Library at the Heart of the AI Boom,” New York Times, May 15, 2024, https://www.nytimes.com/2024/05/15/technology/openai-library-office.html?unlocked_article_code=1.sE0.QarB.-42PJYf2N94t&smid=url-share.
[3] Billy Perrigo, “Exclusive: OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less Toxic,” Time, January 18, 2023, https://time.com/6247678/openai-chatgpt-kenya-workers/.
[4] Max Zahn, “‘Overlooked’ Workers Who Train AI Can Face Harsh Conditions, Advocates Say,” ABC News, May 17, 2024, https://abcnews.go.com/Business/overlooked-workers-train-ai-face-harsh-conditions-advocates/story?id=110303586.
[5] Mortimer J. Alder, How to Read a Book: The Art of Getting a Liberal Education (Simon & Schuster, 1940).
[6] Megan Hauser, “AI Is a Hall of Mirrors,” The New Atlantis (Spring 2024): https://www.thenewatlantis.com/publications/ai-is-a-hall-of-mirrors.
[7] Hauser, “AI Is a Hall of Mirrors.”
[8] Walker Percy, Lost in the Cosmos: The Last Self-Help Book (Picador, 1983), 1.
[9] Hauser, “AI Is a Hall of Mirrors.”