The Best Mind-reading Technology in History

And why it still matters

Image created with ChatGPT

Meaning is like movies inside the head. I’ve got movies in my head. I want to put them inside yours. Only I can’t do that because our heads are opaque. All I can do is try to be clever about sending you a sound track and hope I’ve done it in such a way as to make you construct the right movies in your head. –Peter Elbow, Writing Without Teachers

The past few years have seen—to put it lightly—a staggering pace of technological improvement. You can now form a relationship with your very own AI companion, model protein-folding with advanced computation, and control a robotic limb by brain waves alone. (We’ve also seen some scarier LLM capabilities, like deliberate deception by “sleeper agents” and attempts to persuade an NYT columnist to leave his wife.) With all these advancements, it can be easy to forget that the best mind-reading technology ever created has been with us for thousands of years.

I’m talking, of course, about written language.

When you read something I wrote (this sentence, for example!) a remarkable sequence of cognitive processes must take place. In simplified terms, these are:

1. An image or idea forms in my mind.

2. I formulate a string (of words) which expresses the image/idea as a relation among preexisting concepts (i.e., by arranging them according to shared syntax).

3. I transcribe the string using a series of arbitrary symbols (letters) corresponding to agreed-upon sounds that map onto the concepts.

4. You read the symbols and use them to recreate the string of concepts.

5. If successful, my original image/idea forms in your mind.¹

This sequence of events is not the most direct way to produce phenomena in another person’s mind—but it is, I’d argue, the most miraculous. To get from my brain to yours, the idea must go through not one but two transformations (mental to lexical; lexical to phonographic transcription) and back again. That it emerges intact at all is an astounding feat of collective intelligence.²

Source

These transformations (from thoughts to words to letters and back) allow writing to convey information across incredible swaths of space and time. Whenever I pause to think about it, I’m awed and grateful for this astoundingly flexible system of twenty-six characters, which allows me to commune with family members across the globe and historical figures from hundreds of years ago. Still, in 2025, the barriers to producing audiovisual content are lower than ever before, and it’s reasonable ask: What is the continued value of reading and writing? Naturally our ancient ancestors did not have mics and cameras, and we’re blessed to still find meaning in their etchings on scrolls and tablets; but won’t our own descendants in the year 3000 prefer to just receive TikToks and YouTube links? Why go through the mediating steps of text at all?

After all, when I describe a tree to you in words, I evoke your existing tree concept (perhaps enhanced with sensory details like the color of its leaves and texture of its bark) which you then use to form a picture in your mind. But if I really want to be confident you’re seeing what I’m seeing, I can simply show the tree to you (in a picture, video, or real life) and know that the image on your retinas is more-or-less identical to the one on mine.

I believe that writing continues to have value, and that this will remain true even as AI-generated and -editable videos become easily producible; as the metaverse looms. There are a few reasons for this.³

First, there are some practical benefits of writing that simply make it better for certain purposes. Cutting out visual cues and social niceties can be more efficient communication—as anyone who has ever opted for a text instead of a phone call knows intuitively. Written language is generally more crystallized than audio or video—easier to distill and manipulate; compared to the watcher/listener, a reader can more readily take in the text as a single object, jump between pages and paragraphs, and rearrange information. Yes, audio and video offer some analogs for this (e.g., scrubbing through or adjusting playback speed, and jumping between chapters), and AI helps facilitate them. But it still seems to me that writing has an inherent advantage here because it relies on a single, sparer mode of expression.

Second, written language is uniquely equipped to convey interior, “purely mental” phenomena: those thoughts that cannot be accessed through direct sensory experience because they are conceptual, or pertain to the world of emotion. Sure, if I really want you to visualize a very particular tree, I’m better off snapping a photo. But if I want you to understand the distinction between socialism and communism, or a certain decision-making framework, or the strange grief I feel when my barista of the past year vanishes without explanation, I need language to get my point across. These ideas may have sensory elements, but they are in larger part composed of things which can’t be seen. Visual aids may help here, but words are indispensible.

Finally, the lossiness of writing compared to video forces the reader to project some of their own knowledge into the gaps; ironically, this can be one of the most useful features of writing compared to other mediums. You may have noticed, for example, that historical film footage can sometimes feel flat or corny; the people in these videos don’t quite feel real and three-dimensional because speech patterns and mannerisms have evolved in the intervening decades. By contrast, when you read historical text, the act of recreating the person’s thoughts in your own mind imbues them with some of your own consciousness; they automatically absorb the pathos of the voice in your head.⁴ It’s tempting to think that video will give our descendants more direct access to our lives—and in strictly perceptual terms, this is true. But on a deeper level, I suspect our audiovisual media will feel dated much more quickly than our writing.⁵ Communication is inherently a collaboration between two minds, and the mind of a reader is always of the moment.

These three advantages of writing (practical benefits like efficiency/editability; conveying interior information; and the active participation of the reader) suggest where and why writing will continue to exist in this increasingly futuristic world.

Of course, to continue deriving practical benefits of writing is not to say that it must all be human-authored. Much of communication is functional, and I don’t think it’s any great loss to humanity to automate away the composition we currently do to, e.g., book a restaurant reservation or email customer service to demand a refund. These are contexts where text is often efficient, but we may not write it ourselves.

In other contexts though, it will matter very much that the writing is human-authored. No matter how good AI gets, there is no world where it will make sense tell an LLM “Draft a 500-word private diary entry on my complicated feelings about my divorce proceedings.” There will never be a world where it is socially acceptable to deliver a fully AI-generated apology or expression of gratitude. AI may be helpful in prompting or refining these forms of writing, but it fundamentally does not have access to your feelings and experience—which in this case, is precisely what the text is supposed to reveal.⁶ At their very best, with lots of prior knowledge about how we think, LLMs might craft a personal message that we endorse.

Finally, consider again the reader’s role in constructing meaning from the text. Constructing meaning doesn’t just make the words more real and memorable; it is an active rewiring of our brains—one which, the more often it is repeated, leaves more and more traces on our own habits of thought. As Paul Graham once put it:

Reading and experience train your model of the world. And even if you forget the experience or what you read, its effect on your model of the world persists. Your mind is like a compiled program you've lost the source of. It works, but you don't know why.

Of course, AI-generated text can train our model of the world too—and certainly there are advantages to writing derived from the collective intelligence of the internet, with its the wisdom of the crowd. But relying only on LLMs for writing will lead, I fear, to a homogenization of thought and style.

For myself, I know that I ultimately don’t want my mind to become a statistical average of all of humanity; I want it to become more like the particular humans I love and admire. I want to access those certain writers’ models of the world—to draw upon their characteristic frames and sensibilities and lexicons, their ways of describing a tree. For that innermost picture which only they can see, a thousand words are priceless.

1. The cognitive processes underlying spoken language are almost as miraculous as writing, but without the demand of transcribing words via letters. (Of course, speaking also offers the added cues of intonation and, depending on the context, facial expressions, making it a more information-dense mode of communication in general.)

2. If you’re a fluent reader, it can be easy to gloss over the fact that formulation of words and letters are two separate transformations, but it’s clear that they are from looking at two possible types of comprehension errors:

• Semantic error — “Let me grab my bat!” (In this case, you identify the correct word but might misinterpret it as referring to a piece of sporting equipment rather than the small flying mammal who serves as my familiar. You have identified the correct perceptual input but mapped it to the wrong concept.)

• Decoding error — “Let me grab my yarn!” (In this case, you might misread “r + n” as a single letter “m” and think I was referring to a vegetable rather than a skein of wool. You have perceived the wrong letter/word.)

3. This isn’t to say that new media won’t continue to encroach on what was previously the sole purview of written text, or that this encroachment is even bad. To give a simple personal example: As a high school English teacher, I switched over at some point from written comments on essays to short screen recordings precisely because video is more information-dense. (In this case, it allowed me to modulate my tone when giving feedback and to more efficiently highlight connections between different sections of the paper.)

4. I sometimes experience a similar phenomenon when I’m disappointed by a film adaptation of a book I love: A particular casting choice—sometimes even a good one—can collapse the richness and interpretability of the character I’ve created in my head.

5. Audiovisual media becomes quickly dated in part because there is nonverbal information that gives away the time period (e.g., fashion, technology, and mannerisms) and in part because the form of the media may change (e.g., you probably don’t buy CDs anymore; our descendants may consume media in more immersive VR forms that make video feel flat). In both these respects, writing is more evergreen.

6. This desire for interiority also suggests which kinds of human-authored writing will be most commercially and artistically viable if/when LLMs reach the same technical proficiency: memoir, personal essays, autofiction, confessional poetry, et cetera—i.e., genres which depend on the unique perspective of the author. By contrast, modes whose value-add is pure information (e.g., news, textbooks) or entertainment (e.g., romance, thriller, and even comedy) could provide utility, at least in theory, irrespective of their source.