Play in the gears
I remember watching an interview of Edward Snowden a few years back where a journalist asked him what he would reply to someone telling him that they didn’t care much about mass surveillance because they didn’t have anything to hide. I quite liked Snowden’s answer. He said-and I paraphrase-that he would enjoin this person to think about it in terms of having a right to secrecy, of having the right to make a mistake in private and to grow from it without the world knowing about it. This interview stayed with me because over the course of my lifetime, I have seen this right to privacy erode in more and more insidious ways. When we think about privacy, we often refer to government surveillance, cctv, internet tracking, etc. Here I’d like to focus on how language and modern communication have participated in this great disappearance.
Harold Pinter gave a great speech on writing for the theatre at the National Student Drama Festival in Bristol in 1962. The entire speech is worth reading (and parts of it are quite funny). I was particularly captivated by Pinter’s depiction of his creative process and his relationship with words:
“Language, under these conditions, is a highly ambiguous business. So often, below the word spoken, is the thing known and unspoken. My characters tell me so much and no more, with reference to their experience, their aspirations, their motives, their history. Between my lack of biographical data about them and the ambiguity of what they say lies a territory which is not only worthy of exploration but which it is compulsory to explore. You and I, the characters which grow on a page, most of the time we’re inexpressive, giving little away, unreliable, elusive, evasive, obstructive, unwilling. But it’s out of these attributes that a language arises. A language, I repeat, where under what is said, another thing is being said. […] One way of looking at speech is to say that it is a constant stratagem to cover nakedness.”
My question is: have we lost our precious ability to cover nakedness with words? It feels like nowadays, speech is rarely inconsequential. In every part of our lives, our words are taken seriously, as if they’re perfect conveyors of facts and thoughts, as if they’re to be definitive. We ignore that, as Sartre once wrote, “there is play in the gears of language”. Every time I go back to France, I’m re-invigorated by how people converse there. We play with words, we have fun, we say things and argue and then forget all about it. The French have truly mastered the art of hiding behind words. But then again, there is this weird dichotomy between real life where we speak freely and without thinking too much about it, and politicians, journalists or people on social media who all seem to be making definitive statements.
I guess our ability to live hidden goes away when we start addressing more than a few people at once. We forget about our human body and its need for privacy. We try to share thoughts with thousands if not hundreds of thousands of people who lack any biographical data that would allow them to understand, or even better, fully dismiss what we said. Social media “conversations” are thus most often a surreal mix of misunderstandings, people clumsily trying to make a point, and others who use a lot of words to end up saying nothing. I remember listening to a podcast about young French women who had decided to join Isis. One of them simply explained “I wanted to be free of saying what I thought”. This is a good description of how everyone feels, we want to say what we think, and we want to say it to thousands of people just like we would say it to our friend at the coffee shop. It never works, of course, because in one case, we are words or a video on a screen and in the other, we are a body facing another body. So, we have slowly developed an alternative language to live online, a language that is partially remote from reality, but that works on social media and the internet in general. With this Bizarro language, everybody is free to say what they think and as a result, every day, a new concept pops up on the internet. In La Dame aux Camélias, Alexandre Dumas wrote about this feeling that we have when we discover a new fact and suddenly, everybody else seems to be thinking and talking about this same fact. This is the power of online communication, to make us forget about our nakedness, make up realities, build upon these new realities, and share, share, and share some more. Bizarro language is much more descriptive and prescriptive than Real language. It has to be, because it mainly speaks about made-up things, and for everyone to be able to join the conversation, we need to keep justifying, explaining, refining. Words that were meant to let us live privately are used for the opposite purpose, to expose as much of our thought process as possible.
I’ve seen this new way of communicating seep through and infiltrate real life. Bizarro language is used more and more to force unnatural communication, to formalize relationships that should happen organically. I’m the parent of a toddler and a lot of the advice I read or hear really is about how to shape every single aspect of my toddler’s life through Bizarro language.
This is the way to handle tantrums. You must teach your kid to recognize and name all her emotions and you must discuss big feelings. You must say this to keep that one random thing that happened yesterday from becoming a permanent habit. Never say this, instead, say that.
And so on. We are teaching kids from their infancy about not having any privacy, smothering them with our interventions, when really kids mostly need time and space to grow on their own. I don’t want my daughter to have to explain every single one of her emotions to me, or worse, for me to try to do it for her. We have to accept that language is not learnt through dictionaries and associating one experience with one word, language is highly intuitive and comes from years and years of experiencing and mapping things together. Kids are very smart, and they don’t need us to teach them everything, they learn a lot on their own. I’ve seen it in the workplace as well with some managers taking words spoken carelessly way too seriously. Instead of developing the emotional intelligence that is necessary to understand when people just need room and space to learn and grow on their own, they become obsessed with giving the timely feedback and seizing the teachable moment. Not that feedback is not a good thing, but there is a sweet spot between forcing people to hear something that they already know (think of the new form that needs to be filled in Office Space) or something that they learned by themselves (like toddlers, most adults are also smart and capable of introspective self-improvement), and actually saying something useful. The common self-improvement tips often start the same way: when you say X, this is what you do, and here is what you should say instead. If language used to be a cloak, Bizarro language is primarily a tool.
Relationships were easier before Bizarro language, because we were safer, safer to say whatever and forget about it, safer to have things to talk about that didn’t involve us fully as a person, safer in our privacy. We have taken our dehumanization of language one step further by training AI models on Bizarro language. LLMs mostly know internet speak. They know a version of us that doesn’t really exist. I worked for a long time in conversational AI and I often referred to Brice Parain’s philosophy to separate the language of conversational AI from human language. Parain first thought that orders were the only efficient way of communicating, that we could only be understood correctly if we specified a precise action for someone else to perform. He thought that words only made sense in a tightly ordered situation, with one giving the order and the other receiving, executing. Later on, Parain moved away from this view and ended up theorizing the existentialist nature of language: words shape us as much as we shape them because every time we use a new word, we have to invent its meaning, we link it to our own experience and perception. When we built the first conversational AI models, we focused on what we called task-oriented dialogue systems. These systems were designed to execute orders given from the user, e.g., find me a restaurant nearby. They had very basic general conversational skills. Most of the time, they could only tell a joke or two. With the rise of end-to-end systems and then large language models, we’ve blurred the lines between order-executing agents and general chatbots that try to use-a Bizarro version of-existentialist language. This has led to weird products that are at the same time very useful and potentially very harmful. It is no wonder to me that their so-called alignment is so brittle and can be relatively easily bypassed or broken. These models cannot be shaped by language in any other way than to predict the next Bizarro thing to say. Perhaps true alignment will be achieved when we find ways to teach these models that the internet is not an accurate encapsulation of human communication. Perhaps true alignment is about teaching these models to speak as if they had something to hide.
Layla El Asri