Wednesday, September 18, 2024

Automatic Against the People: Reading, Writing, and AI




Over the summer I posted a rant online (below after the jump), which was circulated enough that I was invited by my university to take the con position in the debate should students be encouraged to use AI in the classroom. This is what I wrote in response to that question. It is an an attempt to think about what is lost when we automate the acts of reading and writing. I am not really sure if what I wrote works, or if anyone will read it, I decided to share it here as well. 


My position is that so-called AI or Large Language Model (LLM) technologies such as ChatGPT should not be used for preparing writing assignments in college classes. There are multiple arguments that one could make against using such technologies. I am not going to address the ecological impact of AI, except to say in passing that it is substantial enough to lead companies like Google to completely reassess or scrap their objectives for lowering carbon emissions. I am also not going to address the ethical and legal issues brought up by the fact that all of these LLMs (and image generating software) are trained on published and copyrighted works. Those issues are best dealt by people who have expertise in that area. What I am going to address is what I know, and what I worry about, and that is what we lose when we automate or outsource reading and writing to technology. I am also not going to address the products of these technologies, the texts, images, and conversations that they can produce. I freely admit that they can be impressive as final products. My concern is not with the product, but with the process—with the process of reading and writing as part of education.

I will start with a basic axiom of teaching: we read in order to learn how to write. This is a fairly simple and often repeated formulation. It has many different variations, it is invoked as a building block of basic literacy and it is often uttered in graduate creative writing seminars which stress that a good writer is by definition someone who is well read. It bears repeating for two reasons. First, this kind of symmetry between reception and creation, consuming and producing, is not necessarily found in other forms of communication and other types of media. While watching films is integral to any directors apprenticeship, and directors will tell you that you have to watch a lot of films in order to make a film—filmmakers watch movies just like writers read, the act of watching does not directly tell me how to recreate what I see. I could also listen to music and never be able to sort out the sounds of what I hear, never distinguish between the part played by which instrument. Watching movies and listening to music might develop my taste, and make me a better and more informed consumer, but they do not by themselves teach me how to produce, how to create. It is in reading that the raw materials of writing are immediately apparent and given. The written text lays its construction bare, making possible a lesson in writing through that no other media reproduces.

Of course this only stands to raise the question, “why learn how to write?” Isn’t writing just one technology among others when it comes to the retention and communication of thoughts. I could just as easily make an audio or visual recording of my ideas as something to share with others or even as notes to myself. The difference between these different ways of recording is how each relates to time. Videos and audios have their own time span, a film is ninety or a hundred and twenty minutes, a podcast an hour or more, and so on. That cannot change without distorting it. (I have heard stories of people playing podcasts at one and a half speed to take in more information, but who wants to get their news from what sounds like Alvin and the Chipmunks?) When I read, however, the timing of the reading is more undetermined and less hardwired into the technology, if it does not sound too weird to call writing technology. I can read quickly, jumping from paragraph to paragraph, skimming for the most important ideas or facts, or I can read slowly, contemplating each word. This timing differs from individual to individual, what some find to be a page turner is for others an interminable slog. This is in part determined by our histories and literacies; what we read teaches us how to read. As Spinoza argued, when a farmer sees hoof prints in the mud he will think of planting but a soldier will think of war. The same sign, even a simple one like the mark of a horse, will mean different things to different people. Our reading is shaped by our experiences on and off the printed page, by what we have done and what we have read. This is true of genres, reading science fiction teaches you how to read science fiction and so on; it is also true of disciplines, sociology, philosophy, history, biology all have their vocabularies and ways of arguing. One of the most fascinating things about teaching, even teaching the same text year in and year out, is that in a classroom we are never on the same page even when we are on the same page. Discussing a text in a classroom is always the interplay of the common text and the singular histories that frame and shape how we make sense of it. It is fascinating to see the different things that people find and respond to in a text based on their lives, their education, and their experiences. How we read is not left entirely to us as individuals, some texts almost invite me to skim, and others demand to be read slowly to be understood at all. I can also skip ahead or reread and this happens seamlessly without fumbling with the controls that would make possible fast forwarding or rewinding, and I do this as I think. Reading has a unique relationship to the time of thinking. In some sense reading is thinking.

If we look at writing we find a different relationship to time and thinking. Every text from a grocery list to a philosophical treatise is written in a particular moment, and meets the exigency of that moment. At the same time every text from a grocery list to a philosophical treatise because it is written can exceed its moment. Even an old grocery list becomes a strange item of curiosity when found in a coat packet years later, “why did I need so much corn starch?” and “why did I underline garlic?” We all know how the passage of time has changed or altered the reception of philosophical texts, what seemed progressive even revolutionary during the time it was written has become commonplace or even reactionary with the passage of time. Texts from past ages are constantly proclaimed to be “dated” or to “hold up,” or, more often than not, to be a combination of the two. On a more individual level, when I write my own thoughts I am constantly caught in an interplay of identity and difference. I think I know what I think, but as soon as I put those thoughts on paper (or a screen) they seem to have a life of their own, they seem to say something else. Writing is a constant act of reading and rewriting in part because what I think and what I write are never quite identical. This becomes even more the case when I show my writing to someone else, someone who does not have the same thoughts, history, and context to make sense of what I wrote. I might be shocked or surprised to find that they read what I wrote differently than how I meant it. If reading can be understood as a process of the common becoming singular, as a shared text opens up different interpretations, then writing can be understood as an attempt to construct the common from the individual or the singular, to make what I think, see, or understand, something that can be shared with others. Communication is inseparable from the clarification and conception of what it means to think. As Eric Hayot argues, writing is not just a mode of expression of ideas that exist prior to their articulation, writing is thinking.

We are facing what the philosopher Bernard Stiegler called a “proletarianization” of thought. Stiegler borrows this term proletarianization from Marx; whereas Marx stressed that proletarianization meant that workers faced increased insecurity and poverty, Stiegler focused on the way that proletarianization was a change in the relation to the knowledge integral to the production process. Workers are proletarianized when the skill and knowledge, the know-how, that was part of the working process becomes externalized in the machine. One could think of the example of the difference between driving nails with a hammer and with a nail gun, the former, as anyone who has hung pictures can attest, requires skill to hammer the nail at the right angle, while the latter matches the angle through the machine. The history of technology in work can be understood as a technology of proletarianization, of deskilling. The dream of the fast food industry is to have zero training time for new employees, to fill a kitchen with so many devices, automated burger flippers and timed fryers, so that anyone off the street could be put to work. Or to take another example, Uber has replaced cab drivers with their intimate knowledge of city streets and traffic patterns, with a GPS map that can make anyone who can drive a cab driver. Of course this does not just relate to working technology. The skills of daily life have also been proletarianized, microwave meals replace learning to cook and the same google maps that an Uber driver relies on makes it so I never really have to know how to get around the city I have lived in for years. I am old enough to remember when driving around Maine required a copy of The Maine Atlas and Gazatteer and the knowledge of how to read a map. The skills of daily life have become increasingly automated and mechanized. Not all of these seem like a loss, anyone who has banged their hand with a hammer enough might welcome a nail gun or some other device for hanging pictures, google maps are in some ways an easier way to navigate than reading a map while driving, and it is hard to overlook the convenience of the automatically generated suggestions when sending a text. There are not that many different ways to say "Running late: on my way' anyway. What concerns Stiegler is the way in which proletarianization has come to affect the skills that are integral to what it means to be a human being, to be able to think and make sense of the world. Reading and writing are such skills. They are integral to how we acquire knowledge about the world around us, of ourselves (think of the journal or diary), and how we situate ourselves in the world. Learning to read a map of Maine makes it possible to read other maps, relying on google maps only teaches you how to rely on google maps. 






What is called AI or LLM is the latest example of this proletarianization of thought. It is advertised to us as a new example of a nail gun, as something which will do the onerous work of summarizing, freeing us from the terror of the blank page. This is a new development in the proletarianization of thought because it does not divide between two different people, the engineer who designs the automated fryer and the workers who just have to respond to the bells and buttons, but between two different sides of the process of writing or reading. Samsung recently ran an ad in which a phone could not just record the brainstorming suggestions from a meeting, but summarize them as well. We have been told that the technology can take care of the boring work of coming up with a first draft or an initial summary and all we have to do is provide the pure inspiration and creativity that will make it an interesting final product. This overlooks the simple fact that the act of typing up the suggestions from a meeting might create new ideas itself. That inspiration and creativity sometimes emerges from repetition. Google recently ran an advertisement for the Gemini AI system in which a father used the software to write a letter that his young daughter could send to an Olympic athlete who inspired her. As Ted Chiang has argued, it is unclear what exactly is being sold here, the point of such a letter is not to be eloquent or well written, the point is that the child has written it. It is a singular expression. I would go further than Chiang here, however, and argue that what is lost when we automate something, even writing a fan letter, is learning how to even begin to articulate and know what is that we know or think in the first place. I have heard people say that LLMs like ChatGPT can do the work of jotting down some ideas, or summarizing the secondary readings, and then they can take it from there in order to create something interesting. This creates a division between a part of thinking that is rote, repetitive, and mechanical and a part of writing that is creative and intentional. This seems to me to be an utterly specious and false division. For me writing is much more akin to playing an instrument, or sport, or learning an art or martial art in which the most mechanical basics and drills are foundational and must be returned to again and again in order to get inspiration to do the interesting stuff. Personally, I get my best ideas when I am doing something like transferring my notes for class from handwritten pages to something typed up, or copying down passages into a slide for presentation, in doing exactly the kind of work that could be automated. Reading and writing are thinking. Can technology such as ChatGPT give us a better product? Perhaps. What it cannot do, however, is replace the process of reading and writing, and that process is education.

No comments: