0:00
Until the tech worked, we weren't learning anything about the ancient world.
Search this video
0:02
Now we're actually reading the words and the authors are coming alive.
0:06
It's kind of a rescue mission, if not a sort of redemption, a redemptive story, kind of redeeming these lost works using AI as the tool to do it.
0:12
I'd like to find early gospel material.
0:15
Anything Christian, Herculanium would instantly become the oldest witness that we have.
0:26
I'm here today with professor of computer science of the University of Kentucky, Dr.
0:29
Brent Seals, to talk about his incredible work with the Hercuanium scrolls [music] at the Villa Papyrie um in Naples, Italy. Hey, Brent.
0:48
Good to have you here today.
0:51
Um, it's so good to see you in person again.
0:53
Um, you've been incredibly busy and productive this summer.
0:56
Uh, you are in the news literally all around the world, certainly here in the United States.
1:02
Um, but also in Italy, Germany, England, France for the work you're doing at Herculanium in Naples, Italy.
1:10
Um, we had a big June announcement.
1:10
Um, you know, I I I'm I'm privileged to know a lot about this because, you know, I honestly Adventures, we're literally doing the film on you guys.
1:22
It broke on deadline, you know, a few weeks ago.
1:26
But, um, I wanted to start uh my time with you now with an excerpt from an article in Scientific American. So, here it is.
1:38
So, how do you read a book you can't open?
1:40
That's precisely what Brent Seals, a professor at the University of Kentucky, has spent his career trying to figure out.
1:46
And on Thursday, his life's work reached a pinnacle.
1:50
Seals, alongside a huge group of volunteers and scientists working as part of the Vuvius challenge, has helped develop a technology to see inside books and scrolls that we can't open without destroying them.
2:02
I use the word um your life's work has reached a a pinnacle.
2:09
Um, so I don't think that's hyperbole.
2:09
I also don't think it's hyper hyperbolic to say that what you're doing here and what you've been able to do with AI and the X-ray tech that you've devoted your entire career for um has is probably a watershed moment for understanding our own civilization and understanding our own history um um our own civilizational history.
2:34
Um, so I guess just to start like how has the summer been?
2:37
Um, and how has your work changed even in the last couple of months just in terms of attention you've gotten around this?
2:43
And I want to talk about the actual um, discovery as well.
2:49
>> Well, thanks for that introduction. What a great piece.
2:50
Um, it is astounding and to hear it read back, you know, um, every time we think we've found a pinnacle, there seems to be a new place.
3:00
And this summer, I mean, has been fantastic.
3:03
The the June press event in Italy kind of brought it all back home because, you know, we we're standing there talking about reading these scrolls for the first time um using virtual unwrapping and we're looking across the bay at the actual volcano, you know, and the time scale is crazy, right? 2,000 years.
3:23
So, um it it was a very high high moment.
3:28
And um you know we've had a series of those over the last 5 years.
3:32
Um in fact in 2020 uh which was the anniversary of the Apollo moon landing right I uh I proposed to my research team you know this would be the year right and you know what happened in 2020 soon after those words came out of my mouth you know the entire world you know fought co upside down.
3:51
>> Yeah it was upside down.
3:51
Um, so you know, from there really to to now, um, has just been this constant climb and to get to the top, I mean, I think you're right that the word pinnacle, you know, kind of captures, uh, being pretty high on this mountaintop of having solved the problem and, uh, turning from the technology really more toward the lost voices that we're restoring, you know, by reading these manuscripts. It's amazing.
4:18
>> And so, um, talk to me about where these manuscripts actually live.
4:20
um in Herculanium and how many there are.
4:23
Give me a sense of scope and scale. >> Sure.
4:27
I I did not know about this collection uh as a student. >> Okay.
4:32
>> And it wasn't until we started developing this this particular technology that I started searching around for, you know, likely candidates and Herculanium came to the top immediately because it uh it's iconic.
4:42
I mean it was formed the library at Urculanium was formed by the volcanic eruption of Mount Vuvius in AD79 and most people are familiar with the story of Pompei.
4:57
>> Um Pompei uh is maybe the birthplace of archaeology of modern archaeology and it's also this fantastical you know ancient lost city that was uncovered by humanity right well the reason why it was lost is the explosion of Mount Vuvius.
5:15
What's what's unique about Pompei is the the architecture uh the scale and scope of the city.
5:20
What's unique about Herculanium, which is where the scrolls were found, is that it's the only library from antiquity that survived.
5:28
We have all the history in the medieval copies of written material that tells us, you know, there was a library at Alexandria. People were into books.
5:39
Books were traveling around the ancient world.
5:40
Um, but we don't have any of the actual libraries left and this is the only one.
5:45
So, you're talking about 1,800 scrolls, maybe a couple thousand books.
5:53
Uh, they were all discovered in the realm of 1750 to 1760 when the Herculanium site was first excavated.
6:02
And um many of them were badly damaged at excavation, but because uh they were being excavated by the royal family. >> Mhm.
6:13
>> Uh King Ferdinand and and others uh who were reigning in that part of Italy at the time, uh they were they were carefully kept.
6:20
Some of them were gifted to other kings as diplomatic gifts, but for the most part, the collection stayed in Naples.
6:27
And uh from 1750 until now, uh we have what we have.
6:30
And the the state of affairs is that the the scrolls were badly carbonized, which means that the volcano spared no expense in basically turning them in to what looks like charcoal.
6:43
So you have the only ancient library uh that is in existence in in the world from antiquity.
6:50
Um and then you have this profound physical problem that you can't really unroll or or very easily read any of the material. Right?
7:01
It's over the years, the easier scrolls were read and and have and have been read and researched, but there's just an entire portion that's unreadable and so that has basically stayed untouched until now. >> Yeah, absolutely.
7:17
The the ones that could be opened were some of them were were pretty readily able to be unrolled.
7:24
uh the the volcanic uh ash and mud that covered up the library uh varied in how it impacted the collection.
7:32
So early on you had some that were just immediately opened.
7:36
You have a lot of them that were destroyed because as they opened them they just scraped it away and they made sketches and now all we have are the sketches. >> Okay.
7:45
>> Um the worst of the worst are the ones that remain and for some reason I latched on to that as as the problem. >> Yeah.
7:51
And I don't know, looking back, you you say things like your life's work, but when your life is playing and you're doing that work, you you're never really sure what's going to become your life's work, right? >> Yeah. Yeah.
8:02
Well, what what were you doing before this?
8:04
Before Herculanium got onto your radar as a challenge.
8:06
What were you doing before that?
8:11
>> Well, I'm a computer vision and an imaging specialist and I did my PhD work at the University of Wisconsin in the field of computer vision.
8:18
Computer vision can be defined as sort of making computers and cameras work together to be able to either imitate or uh what humans can do visually or to solve problems in the world autonomously.
8:31
And so you see computer vision now every day succeeding >> with uh facial recognition, autonomous cars, uh all of the remote sensing uh that that happens.
8:46
Um the early days when I started that work led me into applications like manufacturing um autonomous navigation space exploration.
8:57
So I was I was doing work toward those applications.
9:00
We were doing welding inspection for example where you'd have a camera help you watch a weld being made and then uh the the control loop would help you improve the process of doing those welds. >> Cool.
9:16
it doesn't sound super exciting and I decided that it wasn't that exciting for me.
9:20
So, you know, I moved through some of those applications.
9:22
Um the thing I was doing immediately before Herculanium and a lot of the the heritage work that I started doing um was medicine. >> Okay.
9:33
>> Um you might be familiar with uh laparoscopic surgery. >> Yeah.
9:37
>> This is where surgeons will never actually open the operative field.
9:41
They'll do an entire procedure through scopes that have been inserted into the body cavity and it's incredibly hard to do that work.
9:49
And so the visualization part of doing that work is super important.
9:53
So I was working with surgeons to try to help them better visualize not only the operative field live as they were doing the surgery, but all of the pre-operative stuff that they were doing.
10:04
You might x-ray a patient in advance to see what you're dealing with and then the surgeon would what memorize that pre-operative stuff so that they know have an idea of what they're going to face when they go in.
10:16
Well, all of those things actually fed into what I eventually started to apply to virtual unwrapping because the X-ray, right?
10:22
I hadn't had a lot of experience with that. >> Okay.
10:28
>> But it actually became one of the key things that we've used to see inside of these books without opening them. And not just any X-ray.
10:33
It's uh what's it called? Computed tomography.
10:37
>> And um where are these machines at in the world?
10:42
>> The medical version of commuted medical version of computed tomography is everywhere. Every hospital has one.
10:46
In fact, tomography won the Nobel Prize in medicine in 1979. >> Oh, very cool. >> Yeah. The the two inventors.
10:54
Um >> and what about the X-ray machines you're using for for this project? Perculanium. >> Yeah. to travel.
11:02
You've had to travel to use these particle accelerators, haven't you? >> Yeah. Yeah.
11:05
The X-ray for Herculanium, we've settled on a synretron and that's >> that's a very very unique X-ray source uh that comes from a particle acceler accelerator and um you know there are 40 or 50 of these worldwide.
11:19
We've used a few of them in Europe and in America.
11:21
Uh there's one at Oxford. >> Okay.
11:28
>> It's called the Diamond Light Source. >> Okay.
11:29
We also used one at uh in Grenoval, France.
11:34
>> Uh it's the European Synretron light source.
11:36
And um what's interesting about the synretron is that it it produces a very carefully controlled extremely bright X-ray uh beam. >> Okay.
11:47
>> And brightness with X-ray means that you don't have to wait as long to collect the images. >> Okay.
11:52
um because you have so many photons going through the sample that you can take a really really quick snapshot because it's so bright.
11:58
So we can do an entire scroll at very high resolution very quickly. Okay.
12:06
How what do you mean quickly?
12:09
Well, the kind of resolutions that we're talking about now um which have really unlocked the the problem. >> Yeah.
12:17
>> Uh are kind of astounding.
12:17
Uh, you know, the size of a red blood cell is bigger than the resolution that we're getting. Wow.
12:26
>> To see the interior of the scroll.
12:26
Every one of the little voxels that represents the scan of the scroll. >> A voxil.
12:34
>> Voxil, which would be the 3D version of a pixel. Got it. Okay.
12:36
You add that third dimension, you call it a voxil.
12:41
Uh, the size of that little sample is um astoundingly small.
12:45
And that's what lets us see everything that's going on inside the scroll, including the ink. >> Including the ink. And this is the key.
12:53
And so once you can lift the ink, >> the ink signature, I'll say, off of the burnt scroll, then you've got something to work with, but it's still tightly wrapped in this sort of cylindrical form.
13:05
So now you need to This is where the virtual unwrapping unwrapping happens. >> That's right. >> Okay. Yeah.
13:09
The X-ray gives you a model of everything that's going on inside without having to open it, >> right?
13:15
just like it does on on the human body.
13:17
Uh but still all wrapped up and compressed and you know not in a form where you can actually do anything.
13:21
So all those software steps that take you from that scan to the final readable thing are that's the thing we invented to be able to do it.
13:31
>> And this is the vivious challenge.
13:33
>> Well before the vivious challenge that was the research problem which we essentially had solved to the point where we believed we had the formula. >> Got it.
13:41
The Vidius Challenge took us over the finish line because we we threw the contestants in there, broke through on the first ever scroll and then with the with the attention that we gathered and more funding.
13:52
I mean, we just ran at the finish line after that.
13:56
>> And so this is co-founded between you and Matt Freriedman.
13:58
Um Vibius Challenge is an open-sourced AI engineering challenge.
14:03
Um so open source like you know presented and set to the public.
14:09
anyone could contribute to this project.
14:11
Um, and you weren't really sure what you would be able to find when you when you opened it up like this and like and how did your like your partners in academia or other researchers in other lines of in other lines of work, maybe the paperologists, like how did they all were they all involved in this at the very beginning?
14:28
How do they react to this?
14:28
Is this just helpful to them?
14:32
Yeah, you know, the Vuvius challenge was not a no-brainer uh for for me to jump into and it it kind of evolved from an invitation that Nat Freeman sent me to just attend a conference >> um into what became the contest.
14:47
Uh the conference was the chance where I could meet Nat and his and his colleagues.
14:51
And in meeting him, I found out that uh it probably wasn't going to work to just ask for investment. >> Mhm.
15:03
>> Um because there wasn't that level of interest in the colleagues that he had gathered together for the meeting.
15:07
But Nat himself was incredibly interested and of course he's a really smart guy and very connected.
15:14
So he sat down after the conference before I went back to the to the lab and he pitched me this idea of a contest and that's an open source guy and what I knew that would require is for us to open up all of our research including our data.
15:29
So yeah I I was sent pack and back to the lab with this idea from that um hey why don't you just open up everything you've worked on for the last 10 years and we'll do a big contest.
15:41
[laughter] And you could see that maybe that isn't a no-brainer because, you know, you have to think, who's been supporting me?
15:48
>> What are they going to think about this?
15:50
Um, >> can I just do this on my own? Yeah. >> Right.
15:53
>> And maybe I should just do that.
15:53
Like >> maybe just put everything on hold and do it on on my own.
15:57
What what we soon decided as I talked it over with the folks in my lab, my my trusted team, >> Yeah.
16:04
is that um the acceleration that we were going to get from just going open source was going to far outseed uh it was going to far outpace our own ability to to just get there.
16:16
>> And what everybody wanted was to just solve the problem.
16:18
It wasn't so much about, you know, owning it or getting sole credit, right?
16:22
It was really about delivering these things back to humanity.
16:27
And we thought that we were right on the edge of being able to do that, right?
16:31
And so I think the excitement of being able to get over that that that sort of cliff that marker of of being able to say we're we're going to do this.
16:39
Uh that that put us o over the the line we needed to be over to be able to say, "Okay, it's worth opening this up."
16:48
But, you know, I'll be honest.
16:50
I mean, once in a while we'll we'll see even now in the press uh an article that says um you know, a teenage college kid solved this problem. Yeah.
17:01
>> And um while that might be true in a sense, right, it's also monumentally not true because we had this whole research program going and then we had this entire Vubius challenge happening, right?
17:12
So all of that is the mechanism that has helped us be able to have that kid um do that work.
17:20
>> Well, we were also talking about this before is like u you set up the Vibious Challenge and you know people that contribute to it in a meaningful way do get prize money for it.
17:30
But what ended up really happening was it became a selection.
17:33
It became an ident a signaling um like almost talent recruitment um function for you and your team because the people that ended up contributing to it ended up becoming a part of the team at least for a period of time. Correct. >> That's that's right. Yeah.
17:47
And uh I I think I I love that.
17:50
I loved that we got the interest that we got and I loved that it became this huge recruiting tool.
17:57
Um, a much bigger funnel than I could have ever cast on my own, right?
18:02
Um, combination of, you know, Nat's network plus the notoriety of what happened with the Vivius Challenge and just the buzz, you know, we had >> two, three, 5,000 people on a Discord channel, you know, talking every day.
18:14
Can't even, you know, so many messages you can't keep up, right? >> Yeah.
18:18
>> I mean, >> and this is from around the world.
18:19
>> From around the world. It's global. Yeah. Wow. >> Global.
18:23
>> So, >> and so it's still going today, the Pacific Challenge. >> That's right.
18:28
>> And you run the team still.
18:28
Um, what's the next challenge?
18:32
Um, well, first let's let's maybe talk about what happened in Naples this past summer because I think then the question becomes what's next?
18:41
So, tell me tell me about the the the scrolls that were presented in Naples this past summer. Yeah.
18:49
The moment that we signaled this summer in Naples was that we can we can do a complete scroll.
18:54
So you give me a scroll, it's never been opened, I can make a scan of it and then the software process is now good enough that whatever's inside we can unroll and unwrap and present um in in a kind of complete way. Mhm.
19:11
>> So we did that in Naples for three scrolls >> and we were able to show also what we found in those scrolls.
19:19
One one is is u an extensive work uh of at least 140 maybe 150 columns of text and I mean that's going to keep the paparologist busy for a long time.
19:32
This is all ancient Greek. It's very dense. Very dense.
19:36
Um another scroll uh itself was partial But everything that's there, we were able to unwrap and read, including the title and the author.
19:44
Um, which is kind of amazing.
19:48
And uh, a third scroll also gave us another title and an author uh, Filademus on God's book eight, which means that there were at least eight books on God by this guy and no one knew that Philis had written those books.
20:07
So again, you know, that can start more speculation about um you know, what's in the books, what's in book eight we're going to read.
20:15
>> Uh so th those things uh you know, we unpacked that in June and um what I think it did is it shifted the conversation now from the technical >> because we presented the complete technology that can do the whole thing um to the the lost voices um that are now going to be restored.
20:32
So there there may be 400 of these scrolls left that haven't ever been read.
20:35
They're the hardest ones and we're reading them.
20:37
Um, and that is astounding.
20:42
>> You gone from pursuing history to now actually reading history. >> That's right. >> Yeah.
20:46
That's So there's a sense of completion now that you have around this project.
20:51
>> Well, I've never been able to read any of the ancient languages because I haven't had that training. >> Okay.
20:56
>> So I feel like, you know, my part, not that it's completely done, but it's substantially done.
21:01
substantially done. like we achieved to this moment >> and now the people who can read the language are not talking about the technology anymore they're talking about reading language right understand the way >> so I think there's been a kind of change where I've completed uh the technical
21:16
work >> um substantially complete let's say where I don't have to make the case anymore that this is possible >> and of course we're going to refine the methods so that everything can be faster everything can be better but the conversation now is really about the works themselves Yeah. And the scholars
21:30
And the scholars who can read those works and what we're going to learn from the ancient world.
21:35
You know, until the tech worked, we weren't learning anything about the ancient world.
21:39
We're learning about synretrons and virtual unwrapping. Yeah.
21:43
Now we're actually reading the words and the authors are coming alive and that's an amazing and a remarkable moment.
21:49
>> How do you refine the process that you basically pioneered here um with your team? How do you refine that?
21:53
How long does it take today to read one one scroll and how long do you want it to take?
22:00
I'd love for it to be, you know, an instantaneous thing, completely automated.
22:05
Uh, that doesn't take much longer than the scan time. >> Okay.
22:08
Uh, that's, we're not even close to that right now, but, >> you know, imagine that you go to the, you know, scanner for your receipts from travel and um, as long as it takes for it to scan is how long it takes for you to get the OCR and the thing to be uploaded, right?
22:28
that happens right now for receipts and things like that for OCR.
22:30
Optical character recognition was one of those first finally solved computer vision programs and we would love for the scroll to be the same.
22:40
>> Okay, >> I don't think that we're that close to that right now.
22:43
I think we're still several years away from that kind of performance.
22:48
>> Okay, >> because the technologies that we're using are um not that robust yet. Okay.
22:54
>> But the computer vision and the machine learning that we're using constantly improves and you know we're in this era now that that is remarkable in that computer vision and uh machine learning with more data uh can get better. >> Yeah.
23:11
And so we actually see uh the ability to read the scrolls with the technology we've created as something that is going to create versions that get better over time as we read more scrolls.
23:23
So that that's never really been true with any other technology.
23:28
Like the photography doesn't get better the more photographs that you take, right? Yeah.
23:32
But because of the way the machine learning works the the more things that we capture we learn from that and then we get better at actually revealing the text.
23:40
So that's really rather than you know t time to text which is one thing that we talk about it's uh quality of text >> quality of text >> read readability um we want to make the job of reading this ancient language as easy as possible and overcoming the damage is really hard.
23:57
I mean this stuff is peeling it's fragile.
23:59
So, it's kind of a miracle that it works at all.
24:03
And then to say we we want it to be pristine, right?
24:07
You know, it's like >> and instantaneous. >> Yeah. And instantaneous. So American of us. I love it. >> Yeah. Well, it is.
24:12
It's always got to be better, right?
24:13
better, right? So >> well no I actually the you know for for the machine learning piece and and the AI of it all um that you know you and your team have been using to accelerate this process um in order to get it digitally unwrapped and and and handing
24:28
it over to the paperologist like it it comes across in the film in spades but it also is something that you know you know other other um and tech and tech and VC investors have have noticed and and even starting to make a little bit a little bit of waves in Hollywood which
24:42
is uh traditionally you know um you know lukewarm towards AI but in the film it comes across very very cleanly I think um and very authentically that this is a great use case for AI this is a positive use of AI um to further understand you know humanity and and and I think it's
25:05
currently presented in the national conversation as either or um you're either for AI and against humanity or for humanity and therefore you have to be against AI and and this is really a nice kind of marriage of the two uh in terms of a use case. >> Yeah, I see that it I think there's
25:21
>> Yeah, I see that it I think there's irony that AI is so forward thinking and premised on such you know modern and and current technology and yet we're using it in this case to look backward. >> Yeah.
25:35
>> Um but it's kind of a rescue mission if not a a a sort of redemption a redemptive story, right?
25:39
We're >> we're kind of redeeming these lost works uh using AI as the as the tool to do it. >> Yeah.
25:47
>> Um and in fact, I think, you know, I like that message, right, that every new tool, including AI, that comes into human purview can be used as a way to to, you know, redeem what's around us, right?
25:59
And um and so I think that that actually is a really good story and it's one we should focus on, you know. >> Yeah.
26:06
Um, so, so now that you've digitally unwrapped these three scrolls, is there a backlog of more scrolls specifically from Herculanium?
26:15
Are other labs or other researchers from around the world calling you up saying, "Hey, you know what?
26:19
I've actually got this I've got this this parchment here that I can't read. Um, what do you think?"
26:26
>> Well, it it's true that that is happening. Okay.
26:29
>> And I'm glad that's happening because congratulations.
26:31
>> We have the technology to to be able to look at some of that material.
26:33
For example, there are scrolls in Aman, Jordan from Petra. >> Mhm.
26:39
>> And they haven't been opened. They're carbonized.
26:41
Uh so there's more material in the world that that this could work on.
26:46
Um but there is a backlog of just the herculanium stuff.
26:49
I mean, we haven't plowed through that.
26:50
Uh there are 40 scrolls that we've already scanned and uh that's 10% of what we think actually remains. >> Unbelievable.
26:59
>> So, you know, that's a decade's worth of work right there.
27:01
And the entire the full villa hasn't even been excavated.
27:06
>> And that's the next thing, right?
27:06
The villa hasn't been excavated.
27:07
And I think people are a little bit naive about that.
27:11
They think we have what we have and that's it. >> Yeah.
27:14
>> But if you go down to the villa, which is not even open to the public.
27:18
>> The villa is a part of Herculanium that's separate from the the publicly available part of the town of Herculanium. >> Okay.
27:23
um you go in there and you realize they they really just sort of tunnneled along the uh the foundations of the villa and that's that's how they infer uh what what the actual architecture of the of the villa was and they just happened onto the artifacts that they have including the scrolls by doing that tunneling.
27:42
It's not fully excavated.
27:42
And you know, I'm not an archaeologist.
27:46
Um, but I would love to see now that we know we can read the material, um, full archaeology to happen there, uh, not only to find the scrolls, but really to just better understand the villa and the scrolls we already have. Right. >> Yeah.
28:01
Well, it's also, like you said, like a a part, um, it's the only ancient library that we know of.
28:05
Um, and and like what does that say about the authenticity of all these scrolls compared to maybe other artifacts that exist out in the world that that like the academia or re the research community currently has and uses?
28:19
>> Yeah, I love that you bring that up because the provenence of scrolls from merculanium is unassailable.
28:22
I mean, you can't fake the carbonization.
28:28
You can't fake uh something being in this collection for 200 years.
28:31
So the carbonization actually creates like an authentic fingerprint on this thing that cannot be >> exactly replicated.
28:40
>> Like think about the Dead Sea Scrolls and how that all went down right in the 40s.
28:43
They were parted out by the Bedawin.
28:45
So there were no archaeologists on site when they were found and then you know you end up with you know advertisements in the New York Times for like a scroll from from Israel. Yeah. Right.
28:57
Um, and what has happened since then is that we've we've seen a lot of fakes >> where somebody will say, "I have a a Dead Sea Scroll, a scroll from, you know, cave number nine, Kuman, whatever."
29:11
And it's really hard to prove sometimes that that's true or not true.
29:16
Not true with Herculinium at all.
29:16
The other thing that's amazing about a library from antiquity is that it's not been copied by medieval scribes.
29:25
So you have the original book that was actually on the shelf. The book. >> Yeah.
29:30
So fully authentic, verifiable and the original book not filtered through medieval scribes.
29:36
And you know that a lot of the philosophy and the religion right was tinkered with you know when the scribes of the medieval you know period right uh recopied and and um and preserved uh those works.
29:51
And uh the tinkering is not always obvious.
29:54
Like sometimes scholars have to spend a lot of time to like figure out how did this change.
30:01
They don't have the original work. They just have the copy. >> Right?
30:05
>> Every book that we've read from Urculanium has been a unique work.
30:07
It doesn't exist anywhere else in the canon >> and it's uniquely from antiquity.
30:12
So it's it's really a a very very compelling kind of collection. >> Yeah.
30:21
Um, so how long will it take you to do the next 40 45 scrolls that you have in backlog?
30:28
>> Right now we have a new competition that's been announced for million-doll prize through Vubis challenge. >> Okay.
30:34
>> And that is for um >> for someone else, a competitor. >> Mhm.
30:39
>> Using our software, using their own to do what we just did, read a full scroll >> from one of the 40. >> Okay.
30:47
>> And we're hoping that that'll push our techniques forward.
30:49
It'll engage the community.
30:52
Somebody will win a million dollars for doing that.
30:53
Um I think those 40 are going to be uh processed this year. That's my hope. >> Okay.
31:00
>> And uh and then we'll go after the remaining, you know, 360.
31:05
You know, the other 90% of what of what remains.
31:07
A lot of those are fragmentaryary.
31:08
So we know that we're not going to have a whole book.
31:10
We're going to have a piece.
31:12
Uh but the the game then becomes, you know, if we can read all the writing, uh how do we put these works back together?
31:17
Why not grab a Frontier Lab here to just say, "Hey, we'll we'll just do this." >> I'd love that. >> Yeah. H how do we do that?
31:28
>> We should have that conversation. >> We should do that. Yeah.
31:30
>> Um and then and then let's say these these scrolls get analyzed and scanned um or sorry, scanned and then how long does it take the paper to analyze them?
31:42
>> Paperology is not fast. >> Okay.
31:44
Um because what what a pepperologist brings to bear when they look at the text is uh all of the other knowledge about the ancient world because these texts are dense.
31:53
Um sometimes the texts aren't fully uh >> you keep you keep saying dense but what does that mean?
32:01
Like a lot of text um in a column or the the the the messaging is is dense.
32:07
>> I Yeah, I mean sort of intellectually dense. >> Got it. Okay.
32:10
So, you're talking about parachuting into the middle of a book that is probably in the middle of a phil philosophical argument, >> okay, >> that a a scholar is making and they're fully embedded in their time.
32:21
So, they're talking about a person you don't know whose name is maybe missing from the text, >> right?
32:29
And about an idea that uh is only partial, right?
32:32
And and the game is to try to figure out from the limited words uh all of that stuff, all the context.
32:40
And it can be really really tricky and take a long time.
32:42
Sometimes it it ends up just being a mystery.
32:43
It's like we don't actually know what's being talked about right here.
32:46
You know, we we we have all the words, right?
32:49
But it's really hard to know, you know, what the story is. >> Wow.
32:54
>> Um we're hoping that, you know, most of the virtually unwrapped texts that we can provide are are good enough in terms of readability and quality that uh that the scholars will be able to, you know, unpack that density and figure it out.
33:09
But the the extra knowledge that the paparologist has has be in their head besides just the uh the vocabulary you know and the structure of the language is is immense. >> Got it.
33:20
>> And that's what they're applying sometimes as a group and not just individually like sometimes they have to work on it together.
33:24
So I want to talk more about the tech um because and and the tech throughout the entire workflow.
33:32
So to include what um the part what participants were doing through at the Vubius challenge and the challenges that they that they faced either individually or as a group um in in mapping out um aspects of the of the of the scan, but then also the actual scan technology, the the synchretron, you mentioned it a few times.
33:54
like give me a sense of like scope um for like what a synretron is.
34:02
Maybe start there because that has to happen first before you even engage people um your lab and and other third party participants through the Cubius challenge to even look at this stuff. >> Yeah, that's right.
34:14
We we started by using a bench source and a bench source is one that you can move around.
34:18
a bench source for X-ray would be one that you'd find in the doctor's office or the dentist's office. >> Okay.
34:27
>> And so the first scans that we made way back in, you know, 2009. Okay.
34:31
>> First time I ever scanned a scroll, >> uh it was done with a a source that we took to the place where the scrolls were. >> Okay.
34:40
>> Um it was just so much easier to have access if we were within the perimeter of where the scroll lived. >> Okay.
34:46
>> At at the library in uh the Institute of France in Paris. Okay.
34:50
>> But what we discovered was that that that source wasn't going to give us um data that was good enough to do all the rest of the stuff. >> Okay.
34:58
>> So we quickly moved to the synretron which is a source that isn't movable.
35:01
A synretron is a an X-ray beam that comes from electrons that are running around in a ring speed of light. >> How big is the ring?
35:13
>> And the ring is like a kilometer across. >> Okay.
35:17
So just for this one scroll, >> well, so the the ring uh creates the the electrons that run around in a in a circle. >> Yeah.
35:29
>> And then when you manipulate those electrons, it it it creates all of the different beams that you can use from this instrument, right?
35:35
So it's actually a huge campus, right, around this ring.
35:41
And one of those beam lines, yes, would be uh used to do the work on the scroll at a certain point. >> Got it.
35:49
>> And um you can tune what that beam is going to be in terms of the intensity of the beam um and uh the the energy that it has, all of the things that you think are are needed in the characteristics to be able to elicit what you want from the scan.
36:04
Way more controllability than something uh that you can take to to the library, right? a lot of control.
36:12
>> So what it created though was this problem of now I have to take the material to this facility and I have to guarantee the custodial holder the people who own the scroll the institution that that's going to be okay >> that the scroll will be cared for it
36:26
won't be damaged and um so there was a time period where you know we had to shift to be able to use the synretron um but we hadn't convinced anyone who actually had a herculanium scroll that we could do that you know so uh you can imagine you know kind of going a little
36:44
bit nuts knowing that we could get better data but you know we have to be able to convince the institution to let us do this thing once once you get the scroll into the beam at the synretron at that point it's really all about you know how do you capture the best possible scan how do you optimize over a
37:00
bunch of different parameters >> which was a learning process >> which was a learning process and the first time that we did it we got what we got and um eventually we converged on a way of capturing the data which is a school better. >> Uh the synretron cost if if you're not
37:12
>> Uh the synretron cost if if you're not being gifted the time, >> okay, >> and typically we weren't being gifted the time. Okay.
37:20
>> Would be >> $10,000 per shift where a shift is eight hours. >> Wow.
37:24
>> So, you know, $1,000 $1,500 an hour and add add inflation to that over the last two decades, right?
37:31
And you imagine how much.
37:33
>> And it's the first time you're doing it, so you're you're definitely going to have to learn off that. >> Yeah. Sure. >> Yeah. Yeah. >> Sure. >> Okay.
37:38
And you know, you're not able to do the scan all in one go.
37:40
So, um, typically we do what's called, uh, grid scans.
37:45
You do a small section of the scroll and then another adjacent small section and then another adjacent small section.
37:52
And you do these sections individually and then have to assemble the whole thing.
37:56
So, it it's its own thing, you know, to just capture the data and then manage the sizes >> and uh, >> and not ruin the artifact.
38:05
>> And not ruin the artifact.
38:05
Um, you know, in a way we were lucky because these artifacts are completely inanimate.
38:11
They're completely desiccated and uh they're basically carbon and you you almost can't do anything to carbon with X-ray.
38:17
It just just goes through. >> Yeah.
38:20
>> Um but it's also the worst case for um solving the problem that we want to solve because the X-ray doesn't show the ink very well or at all.
38:27
There was a period of time, I don't know if you know this, Nick, that people told me we would never be able to solve this problem because it would be impossible to see. >> Who told you that?
38:38
>> Uh, colleagues, >> you don't have to say their name.
38:41
[laughter] >> Highle other professors. >> Got it.
38:45
>> Who felt like I was the guy tilting at the windmill >> and they tried to sit me down and say, you know, that you're never going to be able to see the ink using tmography. >> Carbon is invisible.
38:53
You won't be able to see the ink on top of, you know, this other surface. Yeah.
38:59
>> And um we refused to believe that because we felt like there would be evidence.
39:02
We just didn't know how to tease it out.
39:04
And um uh the the two miraculous things that occurred that let us tease that out in an non-obvious way was this the synretron first. >> Okay.
39:16
>> Super high resolution.
39:16
So you can see the effects at a microscopic level.
39:18
How >> how microscopic down to what?
39:22
>> Well, a couple microns.
39:22
A micron is uh 1 1,000th of a millimeter.
39:26
So you have a thousand slices at one micron.
39:30
You have a thousand slices in the space of 1 millimeter. It's tiny. >> Yeah, >> it's tiny.
39:38
So uh >> so so the two miracles that happened, >> the second miracle was uh AI machine learning. >> Got it.
39:46
uh because the evidence that is there is uh subtle >> and so you need a tool that lets you amplify that evidence in in embedded with so many other signals because there's a lot going on inside the scroll at that scale.
40:01
there fibers going everywhere and there's um damage from the carbonization itself and there rips and tears that folds um and then somewhere you know the ink is actually on the surface of this ancient paper which is not like normal paper.
40:19
It's not a composite that's really uniform you know it's fibrous everywhere. >> Okay.
40:26
And so, you know, you have this really weak signal and you have a pretty diminished modality, which is this X-ray modality.
40:34
You would love to have visible light, but you can't.
40:37
>> And then along comes this incredibly sophisticated and powerful tool of machine learning that lets you find needles in haststacks, you know, and boom.
40:45
And this is where actually the human comes in too because now you're taking um now you're now you're focusing AI engineering um AI engineers onto this problem via the challenge.
40:58
Um what are some of the obstacles that they had to overcome?
41:03
Well, we had this huge labeling problem. Um labeling what? >> Yeah.
41:09
To break through you you couldn't see the ink in the scans that we were getting with the naked eye. >> Okay. easily or at all. Okay.
41:17
>> But when we trained the AI based on ink we knew was there um and we did that by using open fragments that had writing visible on the open layer. Right. Cool.
41:30
Um and then we would scan those with the same tomography and then we knew where the ink was.
41:34
We would label it ink, no ink.
41:36
And then we could see that the machine learning would learn the evidence of the ink even though we couldn't see the ink with the naked eye in the scan. >> Right? >> Okay.
41:45
>> So all of that was scientific control to be able to prove to ourselves there is evidence of the ink. The AI is learning it.
41:52
The machine learning is learning it.
41:54
But we have to label it for it to learn. >> Yeah.
41:58
>> Well, how do you label the ink in a completely closed scroll?
42:00
This was where the competitors gave us a huge boost >> because now we have 2,000 people working on this problem and they're trying to create labels.
42:10
And what they would do is they would run a fairly diminished machine learning effort on an open section of the scroll virtually open, right?
42:19
>> And they'd see a response, some letters coming out of the mist, and then they would releabel and iteratively as more text gets exposed, they get more labels. >> Got it?
42:30
Now amplify that across the 2,00 competitors >> having 24/7 >> with people sharing stuff because we did inspire sharing through progress prizes. >> Okay.
42:40
>> And so now you have the acceleration that we we hoped we would get.
42:42
Uh the techniques weren't substantially different than what we had already been doing, right?
42:48
But the scale of what was happening and you know it iterations in terms of innovations were were happening too for sure.
42:54
people were building tools that were helping and um and so then eventually you're solving this bootstrap problem you know by do doing this kind of you know one-way ratchet everything sort of getting better slowly. >> Yeah.
43:09
>> And you know by the end of that year I think it was 2023 maybe uh we were headed toward the final submissions of the grand prize.
43:18
Things were just accelerating like crazy.
43:20
I mean, people sharing and text obviously, you know, being detected from inside the scroll. It was very exciting. >> Yeah.
43:29
And that's when we got wind of it.
43:31
Um, and and and so that at that point you have something uh digitally unwrapped then pass off to the paper and that's and that's and that's exa that's a whole flow of work. >> Yeah.
43:43
And we're trying to improve that actually because it it it's not really a one-way street at that point.
43:49
Paperologists have a lot of insight into what looks legitimate and what might be an artifact.
43:54
An artifact, for example, might be or or even a a mistake might be what we call a sheet switch where uh the virtual unwrapping actually gets off of the layer that it's on. >> Okay?
44:10
>> And it descends to a lower layer.
44:10
And the text, if you can read it, is going to tell you that that happened.
44:16
>> There's a like a break in the narrative.
44:17
>> There's a break in the narrative.
44:17
if you're reading the words or you can see sort of an obvious artifact. >> Okay.
44:23
>> That's not going to be in in a normal open scroll. >> Okay.
44:26
>> And so we can actually go back and fix those things if they've happened.
44:28
Um and usually the paparologists are the ones who pick it up. >> Oh wow.
44:32
So they're actually involved pretty early on in the process.
44:34
They're not just waiting for you to hand them something. >> That's right. >> Cool.
44:38
>> Um and then uh could you tell me a little bit about the scroll and geti?
44:47
Yeah, before we were making such good progress with Herculanium, uh, this girl from Engeti came to us through the Israel Antiquities Authority.
44:57
Um, and in particular the director at the IAA, who at the time was Pineina Shore.
45:02
She'd heard of our work and we didn't even know that this was happening, but she'd heard of our work and had found in their collection a scroll from Engeti that had been looked at by a number of professors, but um no progress could be made because it was too badly damaged and no one would wanted to take on the the formidable task of trying to open this little scroll uh physically because it just looked like it was in pretty bad shape. >> Okay.
45:33
Um, so she'd gotten that scroll back into the collection and then made a scan of it and she used a bench source, >> okay, >> not a synretron.
45:41
And then she gave us the data and said, "Give it a try.
45:43
You know, we can't do anything with this and uh we can't see any obvious writing, but I've read your work and maybe something can happen here."
45:53
So, it didn't take us long because this scroll was actually the perfect candidate for the moment in the development of our technology. Okay.
46:03
>> Uh it was animal skin and not papyrus.
46:06
So in that way it was less challenging.
46:08
[snorts] >> And then as we did the virtual unwrapping we found out that the ink was actually it must have been some kind of metallic ink. >> Okay.
46:17
>> Because the density of the ink in the x-ray was different from the density of the animal skin. >> Okay.
46:23
>> So we weren't looking at the same kind of really really subtle effects of herculanium where there's no density difference at all.
46:29
I mean, it it it showed up the ink showed up in the Enghetti scroll uh as a very bright signal sitting on top of this damaged animal skin and boom, it just like the text just appeared.
46:44
>> And and what did it say?
46:48
>> Well, that was text that had had never been previously seen before, even by the owners of it.
46:54
>> The scroll still remains uh wrapped up.
46:56
It's never been opened, >> right?
46:58
But we we could read everything inside because of these effects using virtual unwrapping. Okay?
47:02
And what we found is that it was uh a Torah scroll.
47:08
It was the first remnants of the first two chapters of the book of Leviticus from the Old Testament. >> Wow.
47:15
Um and what was that's old then.
47:15
So what was the scroll expected?
47:19
How old was the scroll that you were given?
47:21
How old was that expected to be versus what you found uh versus the text you found on the scroll?
47:27
Is that the right question to ask?
47:29
Like um we think it we don't know what this is.
47:32
We think it's this old and then you find this.
47:34
What what does that change about the the text definitely helped to date the scroll?
47:41
They had carbon dating evidence um that put the scroll second third century. Okay.
47:48
But it had been found at an archaeological site that was Byzantine, so a couple hundred years later.
47:52
So those results didn't make perfect sense.
47:57
There was disparity between the carbon dating and the archaeological evidence, which they knew this site is Byzantine, so it's probably fourth or fifth century. Okay.
48:05
Once we read the text, it made more sense because they could see the writing.
48:10
So p paleography, which is the analysis of the handwriting, dated it to an earlier era.
48:15
So it was probably a scroll that they'd had in their collection at this synagogue for a hundred or a couple hundred years. Wow.
48:25
And and as such, it was it was probably sacred.
48:28
Something likely catastrophic occurred at that archaeological site because normally uh none of the remnants would be in the holy ark.
48:37
I think that um it was tradition that uh the Genea you would uh you would dispose of of these artifacts because they're holy, right? Okay.
48:49
>> And you wouldn't just bury them or leave them partially burned, you know, for for others to have.
48:53
So this was the first time in a synagogue that they'd had ever uh discovered Torah.
48:59
And it also became known as one of the oldest copies of the first two chapters of Leviticus. Wow.
49:08
And uh from the text, the scholars were able to tell that it was the uh the canonical version, the Maseretic text. Wow.
49:18
And uh it it's a really early example of that settled text.
49:21
So hadn't changed, right?
49:24
So pushed it pushed the date for the settled messeretic text, you know, way back.
49:30
Um they do have examples in the Dead Sea Scroll collection itself of um variations from the settled text, but this one was dead on.
49:41
The uh the settled text of Leviticus was pretty exciting.
49:43
Well, you know, as we talk about like, you know, this era um and we think about the the Villa Papyrie at Herculanium, you know, we know that the the eruption happened in 79 AD, but the scrolls in this villa could be much much older.
49:59
Um, yeah, because it was I don't know, you tell me.
50:03
Is it like the sort of melting pot of a library that was colle that was collecting wisdom and scrolls from just around um antiquity at the time?
50:10
Um, like give me a sense for what you would hope to find or what what an historian would hope to find um and hope to see in the remainder of the library.
50:25
I think that the library probably was a place where they had an archive.
50:28
So, they had older texts.
50:30
Typically the way I understand it is that um resident philosophers would write new works but they would premise those new works on an archive of of other works often quoting them um extending the arguments in the earlier works debating with each other and then writing.
50:52
Uh, so I think it's very likely that there was an archive of much older works.
50:57
And if that's true and we discover those works, um, you you could you could have literally anything.
51:05
It was a Roman villa, so it's a little mysterious that there's so much Greek and not as much Latin. >> Mhm.
51:14
>> So, a lot of people have postulated that that's just an archaeological feature that that it hasn't been fully excavated.
51:21
If we were to excavate it because it's a Roman villa, we would find much more of the the Latin work.
51:28
And I mean, that opens up all kinds of possibilities.
51:30
We we know of so many things that have been lost from the ancient worlds. >> Let's speculate. Let's start speculating.
51:37
>> Well, I'm not going to hold you to anything, but I want I want to I want >> Libbyy's history of Rome, for example, >> okay, >> would be uh you know, this would be for Alex Pekkus, right?
51:48
because he knows he knows everything that's lost and found.
51:51
>> Cost of Glory, great podcast. >> Yeah, exactly. Incredible.
51:53
Um and and then you have all of the uh the literary and philosophical traditions uh of the Greek schools. >> Okay?
52:05
>> You know, so there could be much more of that.
52:07
that. uh it's thought that this particular philosopher since we have Filademus um was primarily an Epicurion and uh you know there were also the Stoics they argued with each other >> um and and other schools uh if you ask me what I what I wish we could find
52:26
because the first century was uh such an amazing point in human history um you know I'd like to find early gospel material this anything Christian that Herculanium would instantly become the oldest witness that we have to really New Testament Christianity. Yeah. I Yeah.
52:42
I mean, even the fragmentaryary stuff from the New Testament is second century.
52:46
For the most part, we have almost nothing from the first century symbols only.
52:52
>> But we know that this is >> Rome's backyard. >> It's Rome's backyard.
52:55
And we know the Apostle Paul uh was flitting around the the world writing letters to the churches which became part of the New Testament. Right?
53:03
So you have more than just the synoptic gospels which of course would be amazing. I mean AD. 79, right? Yeah.
53:12
>> I mean this is just shortly after the the life of Christ, right?
53:14
So uh if if there were writings that documented anything about his life um that didn't make it into the synoptic gospels, right, those writings could potentially be in a library like this >> in you know the recent results corner of the of the library. recent results.
53:32
I mean, most most people uh who who are academics and scholars would would argue that it it's far-flung to think about anything that's early gospel. Why?
53:43
Um well, because these these libraries facilitated the ancient philosophies >> uh and Christianity was new and and hadn't really taken hold. >> Okay?
53:58
And I mean I get that you know nent things at that time hadn't made their way into you know the the cannon the library the archive. >> Yeah.
54:07
>> But on the other hand uh you know it's fun to think about what could be there.
54:11
Uh the lost books of the history of Rome you know by Libby would give us a window into you know so many other events that would be better than just a philosophical tome. >> Yeah.
54:22
So rather than somebody talking about what what it means to live a a pleasurable life like an Epicurion, you you could have some telling you about all of the events that happened as Rome, you know, rose and fell.
54:36
And you know, imagine how that would be.
54:40
I think the historians would want history books. >> Yeah.
54:43
>> The philosophers want philosophy books and the Christians want Christian books, you know. >> That's great.
54:50
Um, well, here's to discovering new human knowledge. Thank you, Brent.