It's not a box, it's a bucket

I forgot to press record, and we were lucky we'd only just begun. But it was a fitting start, because we ended up spending the better part of an hour and a half on exactly that idea: that we have no guarantee of anything. Not even that the person on the other end is really there.

On the newest episode of Taming Technology I talked with Pietro Segreto, a PhD student at Tor Vergata University in Rome who studies how AI is reshaping the publishing industry. We'd spoken two weeks earlier and felt immediately that the conversation should have been an episode. So this time we held nothing back.

What does a publisher actually do?

I opened with the naive question: will there be a publishing industry in ten years?

Pietro turned it around. The question isn't whether books survive, it's what a publisher actually is. Tie the publisher to the object — the book — and you've misread the job. A publisher is someone who makes a choice. Who finds content, decides what should be published, works on it, and gives it a way to spread.

What changes, he said, isn't the choosing but the method. The big academic publishers — Elsevier, Springer, Wiley — already stopped selling you just books. They sell services on top of their data. They license that data to the large AI companies. The publisher discovers that the business is the data, and the real question becomes how you work with it.

That's where the first thread caught for me. A while back I sent Pietro a document: my pass over a few thousand years of human history, read through its written records. The reason is simple. What we know about the past, we know because someone wrote it down and someone kept it. The scribe, the monk, the printer. They were the gatekeepers of knowledge between generations. They chose what carried forward.

I think publishing is going to revolve far more around something I keep calling S-tier verification. Quality auditing. Real-world testing of information. Being interdisciplinary isn't optional anymore; it's the cost of entry.

The glass box from 1423

I'd been turning my master's thesis into a short story, and I told Pietro about it.

A tribal community is given a glass box. Inside it is all of human knowledge up to the year 1423. You can ask the box and it answers — but it knows nothing after 1423. And the society freezes. It stops inventing. It stops asking. It stands still, because the box already knows everything worth knowing.

Then, of course, a girl is born who's a bit of a rebel. She doesn't ask the box. She goes off and tinkers and discovers something new. She brings it to the box, and the box says: I don't know what this is. Can you teach me? Teach you what? I don't understand.

When the machine met something it had never seen — something that wasn't in its training data — it asked to be taught. And that's exactly what we're doing. With this podcast, with all of it, we're generating new training data as the first generation of humans living alongside large language models. I think we underestimate the training data behind these models and overestimate what they can tell us about a society they've never lived in. They have no data on a human society that lives with AI. We're writing that data right now, for the first time.

We're walking with our shadows

Somewhere in the conversation Pietro asked me to define AI. Not a scientific definition — my own. How I feel it.

I said: it's me interacting with data. It's a shadow, an extension of me, interacting with data. And once you figure that out, you can amplify yourself.

It isn't AI in the way the word suggests. We're just walking around with our shadows all day, because it's a statistical shadow — and the shadow is made of the past. Data always comes from what's already happened. It's a backwards-looking machine.

Pietro put it better than I did: the data is your shadow, but it's still you who has to define the road ahead. The machine rearranges, rephrases and reiterates what we already have. What it can't do is the forward look — that rare, difficult act of seeing toward a future no data covers. That, I think, is the human job.

I've gone through stretches where it felt like all of this would go badly. Not depression, but a real "oh god, this is going to be horrifying." I went through what I've called professional grief. But I'm in the acceptance phase now, and acceptance arrived with the understanding that the shadow isn't the one steering. I am.

Mathildur's birthday

My youngest turns six in the middle of June. We'll be travelling that weekend, so I made her a birthday app.

I sat down with her and we had a little conversation — used speech-to-text. What do you want for your birthday? Unicorns. Songs. Then I told my system: build her a birthday app, I want music, photos and videos. That was the whole of my input. By evening it was up on the television screen at the party, with ten custom songs and flying unicorns.

Pietro asked — and he was right to — whether that was really creativity, since in the end it's the generation of something that's been seen before.

If we take the narrowest definition of creativity, that's a hard road for all of us. Music is recycled. Stories are recycled. We're shaped by everything we've heard. But the creativity in that birthday app wasn't in the execution. It was in the conversation. The spark happened between me and Mathildur, in a question from a father to his child. That was all the creativity it needed. The rest was labor.

And my system knows it isn't part of that conversation. I just sat there with my phone, recording a raw exchange between me and her. Because what AI needs is imagination, curiosity, dreams, inspiration. It needs our soul. That's where it gets something to work with.

It's the same thought that runs through how I see school. A school is an institution, but the most important teachers in my daughters' lives are my wife and me. They clone what we do. They're our shadow, the same way the model is mine. So it matters less which device they hold and more how we live in front of them.

It's not a box, it's a bucket

Here's the correction that gave the episode its name.

Near the end of my story, the little community realizes one thing. This isn't a box. It's a bucket. A box is closed, bound to 1423. A bucket is something you keep filling. The job of each generation is to add good things to the bucket for the next.

And that's exactly what publishers have always done. From the very first thing ever published — the first drop in the bucket of information. It's the same information as always, the same knowledge passing from hand to hand. There's nothing new about AI in that respect. It just runs faster.

I find that calming. Not because nothing has changed, but because the work is the same work it has always been. Choose well. Verify. Add to the bucket. The shadow speeds up the execution; it doesn't decide what belongs in the bucket. We do.

It won't be a smooth ride. A human life is short, and we're dealing with one of the deepest technological shifts in our history, with no end in sight. Opus 4.8 landed last week, and it's just another layer. For a lot of families and a lot of businesses this will be a rough road, and it will take generations. But the machine doesn't learn for us. It only remembers what was. It's on us to decide what comes next — and to put the good things in the bucket before we hand it on.

Near the end Pietro said: if someone watched us from the outside, we'd just be making noises. Two men talking through fiber-optic cable, Italy to Iceland. He called the conversation a soul stone — something I could click into my system and let diffuse into everything I do.

Maybe he's right. But then at least it's a drop I chose myself.

This is a draft, written up from the cleaned working transcript with AI assistance and edited in my own voice. Quotes should be checked against the audio before this goes final. The episode is still in production.