AI Read 100,000 Pages of Ship Records. What Did It Reveal?

SPEAKER_1 (0:09)

I want you to picture the sheer physical weight of history for a second.

SPEAKER_2 (0:14)

Oh, that's a heavy thought.

SPEAKER_1 (0:16)

Right. Just imagine you walk into the National Archives at Kew, and sitting on a heavy wooden table right in front of you is a stack of paper.

But it's not a neat modern stack.

SPEAKER_2 (0:29)

Definitely not.

SPEAKER_1 (0:30)

You are looking at 100,000 pages of ship manifests.

The parchment is totally brittle. The iron gall ink they used in the 1600s, it's literally eating through the paper in places.

SPEAKER_2 (0:44)

Yeah, the acidity of that ink is notorious.

SPEAKER_1 (0:46)

And the spellings are completely bizarre, changing from like one sentence to the next. Plus the whole thing is covered in centuries of actual literal dust.

SPEAKER_2 (0:57)

Just waiting for someone to sneeze.

SPEAKER_1 (0:59)

Exactly. So your boss walks in, points at this mountain of decaying records and says, I need you to map out exactly what ships came in, who they traded with, what they were carrying, and on what dates for the next 200 years.

SPEAKER_2 (1:14)

Good luck with that.

SPEAKER_1 (1:15)

Right. Like where do you even begin with an assignment like that?

SPEAKER_2 (1:18)

Well, honestly, most people would probably just turn around and walk right back out of the building.

SPEAKER_1 (1:23)

I know I would.

SPEAKER_2 (1:24)

It is a completely paralyzing amount of information. And that paralysis is the great bottleneck of historical research.

We have the primary sources. They exist. They're sitting in boxes and archives around the world, but synthesizing them manually, line by line, it takes lifetimes.

SPEAKER_1 (1:41)

Yeah. I mean, you could spend your entire academic career just transcribing a few hundred of those pages.

SPEAKER_2 (1:47)

Easily.

SPEAKER_1 (1:48)

Which is the ultimate nightmare scenario of information overload for anyone trying to understand the past. But what if you didn't have to read it all yourself?

SPEAKER_2 (1:56)

No, that changes everything.

SPEAKER_1 (1:58)

It really does. And today, our Deep Dive is tackling exactly that. We're looking at a really fascinating Daily AI Chat news analysis, and this is based on an October 20, 2026 BBC News report by Ariana Botero.

SPEAKER_2 (2:12)

A great report, by the way.

SPEAKER_1 (2:13)

Yeah. Fantastic. And the mission for today's Deep Dive is to explore how a team of historians at the University of Southampton, led by Professor Craig Lambert, how they're using artificial intelligence to decipher these massive, impenetrable archives of centuries old port books.

SPEAKER_2 (2:29)

Right.

SPEAKER_1 (2:29)

They are taking this mountain of messy paperwork and turning it into a fully navigable interactive map of maritime history.

So, okay, let's unpack this because right out of the gate, I want to clarify a massive misconception. This is not a story about artificial intelligence coming in and replacing historians.

SPEAKER_2 (2:48)

Oh, far from it. If anything, this project highlights exactly why human expertise is more crucial than ever before.

SPEAKER_1 (2:54)

Yeah.

SPEAKER_2 (2:55)

You can think of the AI in this scenario, not as a replacement, but as a massive metal detector sweeping across a sprawling miles-long beach.

SPEAKER_1 (3:04)

Okay, I like that.

SPEAKER_2 (3:05)

Right. So, the metal detector can cover the ground incredibly fast, and it pings when it finds something buried. But, and this is key, it doesn't know the difference between a rare gold coin and a rusty soda can.

SPEAKER_1 (3:17)

It just knows there's metal.

SPEAKER_2 (3:18)

Exactly. You still need the human historian holding the tool to actually dig up the find, brush off the sand, and understand the historical value of what they're holding.

SPEAKER_1 (3:27)

I love that visualization. So, to really understand the findings of this project, I feel like we first have to look at the mechanics of the archival workflow itself.

SPEAKER_2 (3:35)

We do, yeah.

SPEAKER_1 (3:36)

Like, how does one even begin to digitize 100,000 handwritten pages? Because the source notes this is just a staggering amount of physical material.

SPEAKER_2 (3:45)

It's massive.

SPEAKER_1 (3:46)

You can't just throw these brittle 400-year-old pages onto a flat-bit scanner and expect, like, a neat spreadsheet to pop out.

SPEAKER_2 (3:52)

No, no, you definitely can't. The physical scanning is actually just step one, and honestly, it's the easiest part.

SPEAKER_1 (3:58)

Really?

SPEAKER_2 (3:59)

Yeah. Because scanning just gives you a digital photograph of an unreadable page.

To make that photograph actually searchable, the University of Southampton researchers had to train an AI system to read.

SPEAKER_1 (4:11)

And reading 17th century cursive is no joke.

SPEAKER_2 (4:14)

It's incredibly difficult. Teaching a machine to read early modern handwriting requires entirely different technology than the standard optical character recognition, or OCR, that you might use to scan a printed PDF today.