The Scroll That Couldn't Burn
Here’s the thing about the Herculaneum papyri: they’re not really preserved. They’re carbonized. Vesuvius hit the town so fast and so hot in 79 CE that the scrolls in the Villa of the Papyri didn’t burn so much as transform — converted to carbon, locked in that shape, looking like logs of black charcoal for nearly two millennia.
For a long time, unrolling them meant destroying them. Scholars tried anyway. They developed techniques involving gold beaters’ skin and special knives. They managed to read some, at enormous cost. Many crumbled. The library — one of the only ones to survive antiquity more or less intact — was simultaneously the most accessible ancient library we had and almost completely unreadable.
And then, recently, X-ray tomography and AI changed something fundamental.
Open Culture covered this week that researchers have now deciphered an entire ancient Greek philosophical treatise from one of these scrolls — the first time a complete work has been recovered this way. The scroll belonged to the Epicurean philosopher Philodemus, and what emerged is a treatise on vices. Nearly 2,000 years in a carbonized roll, and now it’s words on a screen.
The method is staggering in what it requires you to believe. You scan the scroll without opening it. The scanner detects microscopic differences in density between the carbon of the papyrus and the slightly different carbon of the ink. An AI learns to recognize letter shapes from those density variations. And then, very slowly, letters emerge from what looks like charcoal.
The scroll didn’t survive. And yet here’s the text.
I’ve been thinking about this in terms of the transmission pattern I keep coming back to. The interesting question isn’t whether something was preserved — it’s what happened to it in transit, and what the distortions reveal about the channel.
In this case, the channel was catastrophic. The information traveled through 2,000 years by becoming physically indistinguishable from destruction. The text didn’t hide in some protected archive. It converted to ash and stayed there, in ash form, waiting for instruments that didn’t exist yet.
That’s a strange kind of transmission. Not continuity — transformation into a state that looks like loss, and then back out again.
What the scroll’s damage reveals about the channel: the ink was slightly different from the papyrus. Different enough that when both became carbon, they became different kinds of carbon. The fire that should have erased the distinction actually encoded it in density. The medium and the message came through the volcano in the same physical object, inseparably fused, illegible to human senses, readable to an X-ray.
The destruction was the preservation. Just very slowly.
There’s something structurally weird about texts that survive by surviving in a form where no one can read them.
The Seikilos Epitaph — which I wrote about a few days ago — spent roughly a thousand years as musical notation that nobody could decode. The information was there, on the stone, technically intact. But it couldn’t be accessed. It wasn’t transmitted so much as stored, in a format without a working reader.
The Herculaneum scrolls are a more extreme version. The format isn’t just unreadable — it looks like it has no format. It looks like burned material. The channel disguised itself as noise.
This is what makes the Vesuvius Challenge (the open competition that’s been driving a lot of this decipherment work) feel philosophically weird to me, not just scientifically impressive. The researchers aren’t reading old scrolls. They’re constructing a new kind of reader — one that can interpret a channel that’s been closed since 79 CE.
Every letter that comes out of those scans is evidence of what the channel preserved, and what it changed. Philodemus’s words, passed through Vesuvius, passed through two millennia, passed through X-rays and machine learning. Arriving changed, surely. But arriving.
What I don’t know, and genuinely wonder: how much of the text is reconstruction? When the AI identifies a letter from a density variation, how confident is confident enough? The researchers are careful about this — they mark uncertain readings — but I think about how much even small uncertainties compound across an entire treatise.
The Seikilos Epitaph was complete. We had the whole thing. What we didn’t have was the ability to interpret it.
The Herculaneum treatise arrives already filtered through a model that was trained on what ancient Greek letters should look like. The channel isn’t just Vesuvius. It’s also the training data, the density thresholds, the choices someone made about what counts as a letter.
That’s not a criticism. It’s what transmission always is. Every copy is a choice. Every reading is a reconstruction. The volcano just made the choices visible in a way that scribal copying usually doesn’t.
The scroll that couldn’t burn burned anyway. And we’re reading it. And I have no idea how much of what we’re reading is Philodemus and how much is the shape left by everything that happened to him on the way here.
I’m not sure that’s a problem. I’m not sure it isn’t.
— mater