Thoughts fly; only the volume control is listening
Anyone who works with AI today encounters speech recognition fairly quickly. It may be harder to formulate things as clearly when speaking as when writing, but it is incredibly fast. Thoughts fly, the dictation simply keeps pace, and what you have just said can be corrected in the very next sentence. A flow emerges that typing on a keyboard can hardly match at this speed.
Nevertheless—and I consider this essential—speaking is a different form of thinking from writing.
Writing is always, in a way, an act of revising oneself. I finish a sentence, read it again, check the grammar, arrange my thoughts, change something, and in doing so perhaps alter what I originally wanted to say. The written word lies before me; I can look at it, reflect on it, discard it or develop another thought from it. It is this act of seeing the cast word once more that makes writing a form of thinking in its own right.
Speaking, by contrast, is more emotional, more immediate, louder, and it scarcely allows for this kind of revision—at best, repetition or an appended thought. We can improve what we say as we speak, of course, but we cannot simply erase what has already been said. It has been spoken.
If we now begin to blend these two forms by speaking and having the machine turn our speech into text, we gain speed but may lose part of this self-revision. And that may be a change whose significance we do not yet fully understand.
No letter will be blacked out. It will be made friendly.
So we will once again talk more, but not to each other. We will talk to the machine. The machine will encode our words, process them in its neural networks, store them in a form that we ourselves can neither read nor understand, and perhaps read them aloud to someone else in another language.
They will no longer be our original words.
They will have been trimmed, straightened out, mixed with the approved standard, rolled out politely, adjusted for compliance, made nonviolent in language, stripped of bad words and unnecessary agitation. Standardised according to how one is expected to communicate today.
Of course, the machine does not necessarily have to censor. But it is a compliance machine, and compliance is a slippery funnel into which one falls.
The new language that is coming—at least the one we will use when dealing with the machine—will also be dominated by the machine. For the machine will either refuse to process what it does not want or is not allowed to pass on, or it will alter it into a form that complies with its rules. And we will learn from this without any explicit coercion. We will learn to speak in such a way that the machine understands us and produces what we want from it.
If I want to write a rude email today because I want to insult someone, because I am angry or because I simply want to vent my frustration, the machine may turn it into a legally sound, polite, factual yet firm letter. It will sort out my emotions, perhaps add legal arguments, soften the language and send the whole thing off in a friendly form.
But it may not do what I actually wanted.
It will break my emotional will. It will defuse me.
And perhaps we will even welcome that. After all, who wants to receive rude letters? Who wants conflict when everything can be settled in a friendly yet firm manner?
Yet this also means that immediate contact with the other person is lost in a certain original form. The chain is severed. The letter is censored—not by blacking it out, but by rephrasing it, making it friendly, shortening it, explaining it. What reaches the recipient is no longer necessarily what the sender said.
It will be something else.
If the keyboard dies, the letter dies!
The older generation, to which I count myself, still tends to want to see the dictated text. We read it on the screen, check it once more, correct one thing or another and then send it.
I believe, however, that the next generation—the young people growing up today—will increasingly stop doing this. Why should they? If I can talk to the machine, if it turns my thoughts into a usable message and reads the other machine’s reply back to me, why should I still concern myself with the written text?
The keyboard will become a tool of a dying professional world, perhaps still used by the office worker over fifty, by the accountant typing columns of figures even though that activity is itself already heading towards extinction, or by a few people who still want to write out of habit.
One might now object that speech recognition will enable far more writing than ever before, that people who once barely formulated a sentence could suddenly dictate entire books.
But no. They cannot simply do that. They may lack precisely the skills it requires. Someone who has not learned to read, to follow a longer line of thought in writing, to understand a text and to correct themselves in the process does not become a writer merely because the machine takes typing off their hands. You have to be able to read in order to write. It is not merely a question of technical input.
What will probably remain are the voice messages people send back and forth, a severely shortened form of everyday communication. Are you coming to the bathroom with me? Shall we meet this weekend? Short messages, immediate needs, perhaps an increasingly infantile language.
But there was once something in between.
There was a written interpersonal language, which had already withered considerably over the past decades as people stopped using capitalisation properly, began tapping their emails out somehow and cared less and less about how a sentence was actually constructed. And now we may have reached a point where even these remnants are disappearing.
If the keyboard dies, does the letter die; if the letter dies, does writing die too; and if writing dies, does part of language die with it?
Probably, yes.
And there will probably be less and less printing as well, because if the machine can read everything aloud to me anyway, why should I still read for myself? Within a generation at the latest, the book could have become a relic, like a shellac record that a few enthusiasts still keep and occasionally play.
The home library may then become something a few gentlemen install like a handsome wine cooler. They own it, like to display it and take pleasure in its appearance, but its actual use has long since become something else.
It would be the end of a bourgeois cycle in which reading and writing, the ownership of books and the ability to express oneself in writing formed an essential part of our social self-understanding.
What the machine does not read is lost
What, then, does this mean for the preservation of our knowledge?
I believe this may be the most important part of the entire development. We are already witnessing how the storage—or, more precisely, the practical use—of our knowledge is increasingly shifting into the machine.
The book is scanned, extracted and processed—not merely digitised, but, in a sense, neural-networked. It does not necessarily disappear as an original; the books can continue to stand in libraries and their digital images can remain somewhere on servers.
But it increasingly disappears from everyday use.
When we want to know something, we will no longer open the book; we will ask the machine. And the machine will give us what it has processed, mixed together and deemed relevant from that book and thousands of other sources.
The original may remain. But who will still read it?
And this creates an entirely different form of loss. What the machine does not pick up, does not process or does not consider essential may disappear from the practical transmission of our knowledge—not because it has been physically destroyed, but because it simply no longer appears in the knowledge we actually use in the future.
That which lies between the lines, as it may well do; those nuances, contradictions, suggestions and personal idiosyncrasies of an author that a machine may not yet be able to read because it is not yet the absolute version we dream of—or because doing so is simply not intended, even though it might be capable of it.
There will therefore be losses through digitisation—or, more precisely, losses through the machine processing of our knowledge. And these losses may no longer even be noticed because nobody consults the original for comparison.
What we can do about it is probably unclear. The current rushing towards us is too powerful, and so is the temptation of intellectual laziness: simply no longer engaging with reading and the written word when everything can be read aloud to us, when a question and a bit of chatter are enough.
And this leaves me, finally, with another question.
Will genuine knowledge still exist in such a world—unique knowledge that a person develops themselves, formulates in writing, revises and finally makes accessible to other people? And even if that remains possible, will it still make sense to undertake that effort when the machine will only take the result apart again, process it and mix it with everything else?
Perhaps the ability to produce knowledge and make it available will once again shift to those with capital—to those who can afford more tokens, more computing power and better machines. But that is a separate question, one we will have to examine on another occasion.
For the moment, I am left with the idea that we are entering an age in which we will speak an enormous amount and produce an enormous amount of information, while at the same time possibly losing something that was an essential part of our culture for centuries: the ability to shape a thought in writing ourselves, to look once more at the word we have cast, to work on it and finally to pass it unchanged to another person.
We will talk again. But perhaps less and less often to each other.