Talking our way out of reading and writing
We are building computers we can talk to right now. This has been a long time coming, of course, but it's finally getting good with full-duplex voice models like OpenAI's, which can listen and speak at the same time.
It's interesting to zoom out. For about 50 years, we could only read and write to computers. Now we can also speak to them and listen to them: we can talk. Zoom out even further: we started reading and writing roughly 5,000 years ago. Before that, for hundreds of thousands of years, we only communicated by speaking, making sounds, touch and what we could see.
So we've had reading and writing for 5,000 years, and now we have computers that might remove the need for it. They might let us go back to that older way of communicating without losing the advantages reading and writing gave us.
Not a new idea
While thinking this through, I found that Walter Ong described something like it in Orality and Literacy (1982). He argued that writing didn't just store our thoughts, it changed how we think. And he coined the term secondary orality for a return to spoken culture through technology. He was thinking of radio and television. I think AI that can actually listen is the real thing.
And it goes back much further. In Plato's Phaedrus, Socrates argues against writing. He says it will weaken memory, and that written words are like paintings: they look alive, but if you ask them a question, they stay silent. We've spent 2,400 years proving him wrong, or at least proving that the trade was worth it.
What reading and writing gave us
Writing brought a lot to the human race that animals never had:
- Remembering things long term
- Sharing big thoughts with big groups of people
- Writing down history
- Drawings and texts that can be interpreted, copied, shared around and moved somewhere else
But a computer that can talk like us can do all of that too. It can remember, it can share, it can carry knowledge across time and place. So why would we still need reading and writing?
The one advantage I could come up with: silence
The one property I could come up with is that reading and writing can be completely silent. Nobody in the room hears you.
That's cool, but we also have whispering, which is pretty quiet too. And if you really need other people not to hear you, you can just go somewhere else. So that problem is mostly solved already.
What's more interesting is what happens in a room full of people. Put 100 people in a room and they'll talk in groups of four, five, six, or one-on-one. Everybody can focus on the conversation in front of them just fine. But put one person in there behind a screen, reading and writing, and that person suddenly has a much harder time concentrating.
The two modes don't mix well. If 100 people are reading and writing on computers and five of them start talking loudly, everyone on a screen gets distracted. So that's another reason we still need quiet, and therefore reading and writing. Unless everybody starts talking. Then you don't need the quiet anymore, because people who are just talking to each other don't get distracted by other conversations.
There's something about talking. It's much more fluent, and much more what we're built to do as humans, I think.
A local optimum
So maybe we're in a local optimum right now, where we still need reading and writing.
There are a lot of rough edges with talking to computers. Voice doesn't work well at all yet, especially in noisy environments with multiple speakers. That's not a solved problem. But I think we're close.
When we really get there, we'll have a system that understands its entire surroundings: the people in the room, when it's supposed to speak, when it's supposed to listen, when it's supposed to stay quiet. At that point, maybe we can climb out of the local optimum of reading and writing into a new optimum where we generally don't read and write at all, because I don't really see the advantages anymore.
Counterarguments
I asked Claude for feedback on a first version of this post, and it came back with a few counterarguments. Here they are, together with my own, and why I mostly don't buy them.
"Writing and reading are different questions"
That's true. Writing is way less needed than reading already. Talking is much faster than typing, so writing is the easy one to get rid of. Arguing that we'll stop reading is much harder, but not impossible. I think reading might become obsolete too. The rest of the counterarguments are mostly about reading.
"Reading is faster than listening"
People read at maybe twice the speed of speech, and you can skim a page. But that comparison is between a static book and a static computer reading text out loud, or a television outputting a continuous signal you can't play with. Of course that's slower.
Talking isn't static. It's dynamic. When you talk to a person, you can interrupt them, ask them to skip ahead, or to go deeper on one thing. That's a way of skimming too, and it goes pretty fast. So reading might be twice as fast, but I doubt that's a big advantage. Twice as fast is not an order of magnitude.
And reading throws a lot of information away. When you read, you only get the knowledge. You don't get any visuals except the ones you imagine. When you talk to someone in front of you, you see their emotions, how they react, much more than just the words. So reading might be a bit faster for knowledge, but it's way slower for communicating everything else.
"People choose text, even though they have voice everywhere"
They do, still. Young people text instead of calling. But I'd argue that's because voice channels are still too dumb. When AI gets intelligent enough that talking to it is basically the same as talking to a human, text won't be needed anymore. We're not there yet, but I think we can get interfaces like that within this decade.
"Some things are hard to say out loud"
That's 100% true. But we're not animals that only communicate through audio. We're visual animals too, and we have touch. Those should be part of the interface as well, and computers can use visuals to make very good interfaces. It just doesn't need to be text.
So tables, diagrams, images and other visuals will definitely stay. But I don't think that's really reading. Addresses are hard to say out loud now, but better AIs will handle that easily. Tables are the interesting one: you can look things up in them much more easily than you could by listening. I'd like to think more about that.
Code and maths are a different story. I think they'll be used very rarely by humans in the future, because computers will do them behind the scenes to give us the right experience. The interface with humans can just be talking, without us needing to know how to do maths or code.
"Some people find it easier to write than to talk"
Some people find it really hard to put things into spoken words, while they find it easy to write them down. Many people find it easier to read about complicated stuff than to talk about it.
My counter to that: maybe that's just because we learned it that way our whole lives. People read a lot about complicated things but rarely talk about them, so of course talking about them feels harder. That might be exactly what keeps us in the local optimum. This is just a theory; there could be other reasons for it.
But if it's true, getting out of this local optimum is much harder than it looks, because it may take generations for people to adapt to only talking again.
A hundred years
Still, I can imagine it. Maybe it takes decades, maybe a hundred years. But once computers have a truly human-like interface (speaking, listening, making the noises we make), we might go back to that same interface, and stop reading and writing altogether. We'd keep the visuals, but lose the text.
Socrates' real complaint about writing was that it can't answer back. For the first time, the thing that holds our knowledge can.