Episode 12

full
Published on:

15th May 2025

Unlocking Ancient Secrets with AI

In the final episode of season 2 of the Confluence podcast, Brent Seales of University of Kentucky joins the show to talk about his groundbreaking work using AI and CT scanning to digitally unwrap ancient Herculaneum scrolls buried by Mount Vesuvius, making previously unreadable texts accessible. The conversation includes insights on collaboration with Nat Friedman, the Vesuvius Challenge, and the intersection of technology and humanities scholarship, along with Seales' new role directing an AI and Humanities Institute funded by Eric Schmidt's foundation.

Episode Links:

Watch this episode on YouTube or Spotify.

About Brent Seales:

Dr. W. Brent Seales is the Stanley and Karen Pigman Chair of Heritage Science and Professor of Computer Science at the University of Kentucky. He earned a Ph.D. in Computer Science at the University of Wisconsin-Madison and has held research positions at INRIA Sophia-Antipolis, UNC Chapel Hill, Google (Paris), and the Getty Conservation Institute. The Heritage Science research lab (EduceLab) founded by Seales at the University of Kentucky applies techniques in machine learning and data science to the digital restoration of damaged materials. The research program is funded by the National Science Foundation, the National Endowment for the Humanities, the Arts and Humanities Research Council of Great Britain, the Andrew W. Mellon Foundation, and Google. Seales is a co-founder of the Vesuvius Challenge, an international contest formed around the goal of the virtual unwrapping of Herculaneum scrolls. He continues to work with challenging, damaged material (Herculaneum Scrolls, Dead Sea Scrolls), with notable successes in the scroll from En-Gedi (Leviticus), the Morgan MS M.910 (The Acts of the Apostles), and PHerc.Paris.3 and 4 (Philodemus / Epicureanism). The recovery of readable text from still-unopened material has been hailed worldwide as an astonishing achievement fueled by open scholarship, interdisciplinary collaboration, and extraordinary leadership generosity.

-----

The Confluence podcast is a collaboration between TRXL and AVAIL, and is produced by TRXL Media.

Transcript
Randall Stevens:

Welcome to a new Confluence podcast.

2

:

I'm joined as usual with Evan Troxel.

3

:

Uh, but I've got a friend of mine

that's been a friend for a long

4

:

time, Brent Seals, that's joining us.

5

:

Brent is a, uh, a professor at the

University of Kentucky, a research

6

:

professor, and we've probably known

each other, Brent, 20, 25, 26,

7

:

27 years, probably late nineties.

8

:

Uh, that we got to know each other.

9

:

the kind of background story

that I always tell is that, uh,

10

:

you know, I was doing computer

graphics work in the nineties and.

11

:

You know, when you're in a place

like Lexington, Kentucky, there's not

12

:

that many people that even understand

what you're talking about when you're

13

:

working on those kinds of things.

14

:

So we kind of became, uh, fast early

friends, uh, through the university

15

:

and connections over there about what

we were doing on the graphics side.

16

:

but, um, you know, most recently, you

know, I've been watching Brent, you know,

17

:

like I said all this time, and really he's

been working on a project that we're gonna

18

:

dig into today and talk about for the

last 25 years that I, I now claim is like

19

:

one of the sexiest, coolest projects that

you could imagine on the planet, right?

20

:

So it's a lot of fun.

21

:

And we had, uh, one of Brent's

colleagues, uh, Christie, participating

22

:

in our Confluence event, uh, this fall.

23

:

I am, you're, you're not

gonna get away this fall.

24

:

I'm gonna have you, uh, come and visit

the crew when they're, when everybody's

25

:

in town, but, uh, welcome Brent Seals.

26

:

Brent Seales: Thanks so much.

27

:

I can't wait.

28

:

Randall Stevens: So maybe, uh,

maybe we can kick this off.

29

:

Just give us a little bit of your.

30

:

Background, Brent.

31

:

And then we'll, we'll dig into, um,

into this cool, sexy project that

32

:

I've, uh, that I'm alluding to.

33

:

Brent Seales: Sure.

34

:

Yeah.

35

:

I have been at the University of Kentucky,

uh, for 30 plus years now, and it's my

36

:

first and only academic appointment.

37

:

Actually.

38

:

I came from the University of Wisconsin

Madison with a PhD in computer

39

:

science, specialty in computer vision.

40

:

Uh, and I arrived in 1992

in Lexington, Kentucky.

41

:

Um, so I've held various roles at the

university, um, including department chair

42

:

for two terms, that would be eight years.

43

:

Um, and I also ran a research

center here and, uh, I've been

44

:

continuously doing research, uh,

since I arrived in the nineties.

45

:

Randall Stevens: Great.

46

:

So I will attempt to describe this

grandiose project and, uh, you can correct

47

:

everything that I get wrong, but, uh,

48

:

Brent Seales: Well, let

me hit you with this,

49

:

Randall.

50

:

See if you can do it.

51

:

One thing I love about this project is

that you can say it in one sentence.

52

:

Evan Troxel: Good luck Randall.

53

:

Brent Seales: So just do it.

54

:

Randall Stevens: I don't know.

55

:

That's a big challenge.

56

:

Uh, the, uh, so, so what Brent

has been working on in general,

57

:

I'll, I'll just kind of start

from what I've saw, what I've seen

58

:

you doing over the last 25 years.

59

:

The idea was that, that you could take

artifacts, historical artifacts that

60

:

are, I'll just say roll rolled artifacts.

61

:

So lots of times, paper parchment

artifacts that end up, uh,

62

:

you know, antiquities that,

that are become very brittle.

63

:

You can't open these things.

64

:

The, the theory that Brent had and has

really worked on and, and, you know.

65

:

I, I don't wanna say perfected, but

you've, you've actually cracked,

66

:

you've cracked the code on, uh, is

to be able to scan these things with

67

:

using, uh, you know, CT scanners or

x-ray type technologies and digitally

68

:

unfold them based on the data that

you recover from these CT scans.

69

:

So everybody's probably seen, you know,

CT scans when they think about the human

70

:

body, imagine being able to take rolled

up, uh, scrolls of paper that are, uh,

71

:

and then running a CT scan on those and

then being able to digitally unfold those.

72

:

So I kind of, you know, got to watch

this play out from afar just, you

73

:

know, every year or two, you know,

kind of catching up with Brent about

74

:

what they were doing over in the lab

and how they was thinking about this.

75

:

But.

76

:

And then, so he kind of perfected this

technique of being able to do this.

77

:

And then the real breakthrough, so

the, the, the project or the, uh, the

78

:

kind of main project was to take these,

uh, the veian, the Vesuvius scrolls

79

:

from the Pompeii where the volcano

covered up what, hundreds sprint,

80

:

hundreds or even thousands of these

scrolls that were buried in a library.

81

:

That was Julius Caesar's father-in-law,

I think, if I've got that correct.

82

:

And so they've got these artifacts and

they're, you know, a few of 'em are

83

:

in museums around the world, but they

basically are, looked like charcoal

84

:

is the best way to describe 'em.

85

:

I, I have other ways to describe

it, but we'll say it's charcoal.

86

:

Uh,

87

:

but, uh, yeah, but, you know,

and they're so delicate.

88

:

You, you can't, you can't unfold

'em, you can't unwrap 'em.

89

:

So they've been able to now,

uh, you know, scan these things.

90

:

And now with the help of

AI actually taking the.

91

:

Output of that as they get this data

and using these techniques to, uh, to

92

:

basically read what's inside of them.

93

:

So I don't that

94

:

Evan Troxel: That was a

long sentence, Randall.

95

:

Randall Stevens: It is,

96

:

Evan Troxel: But you did it.

97

:

I'm gonna say

98

:

Randall Stevens: it is,

uh, it, and it works.

99

:

It is like amazing.

100

:

It is literally one of the most

amazing things once you start going.

101

:

And now Brent's kind of world famous.

102

:

He's, he's, uh, on 60 Minutes,

he's being interviewed and all the

103

:

famous things around the world.

104

:

You can literally go out now and

just, just go YouTube, search

105

:

for Brent Seal's, uh, Vesuvius

Project and a Scrolls project, and

106

:

you'll see all kinds of videos.

107

:

Uh, they're making a

documentary that he's filming.

108

:

He's, he's, he's the

coolest cat in town now.

109

:

So,

110

:

Evan Troxel: Nice.

111

:

Randall Stevens: uh, but anyway, let,

112

:

let's, uh, let,

113

:

Evan Troxel: I wanna

hear Brent's sentence.

114

:

Randall Stevens: yeah, let's go.

115

:

Brent, tell, tell us how this, like,

when did the, when did the first

116

:

idea for this pop into your head?

117

:

Brent Seales: Well, we, we invented

the technology before we really knew

118

:

what the materials should be because,

you know, I'm a tech geek, right?

119

:

And we were starting on just doing,

um, digital restorations of open pages.

120

:

You know, things that aren't

flat, but you have to photograph

121

:

them without making them flat.

122

:

So can you, can you fix it in post?

123

:

Right.

124

:

You know, like, like a filmer would do.

125

:

And um, it turns out there's

a lot you can fix in post.

126

:

Uh, we were worried that there were

gonna be things, you know, in the

127

:

internet age that wouldn't make it

into the digital library because

128

:

they are too hard to digitize.

129

:

So, you know, we were doing

flattening and then that evolved

130

:

into complete unwrapping.

131

:

And, you know, I, I thought about

it at the time, more like a party

132

:

trick, you know, when we were kids,

you know, you crumple up pieces of

133

:

paper from your notebook and you,

you create a little ball, right?

134

:

And then you open it and you do

that again, and it creates a kind of

135

:

fabric right out of the, the paper.

136

:

So that was a fun thing to do as kids.

137

:

Um, in the end we imagined, okay, you

have a crumpled up piece of newspaper.

138

:

Would it be possible to put that in a

scanner and build a representation of

139

:

it so that you could still read what was

there, even though it's all crumpled up?

140

:

and we built a prototype.

141

:

So the technology actually

evolved really quickly,

142

:

Evan Troxel: I,

143

:

I'm just.

144

:

Brent Seales: on in the project.

145

:

Evan Troxel: I'm laughing because it's

like, yeah, the, the paper's crumpled

146

:

up, but it's also been set on fire

and like, like where you, where you

147

:

ultimately ended up with this, with this

148

:

project and with this technology, but,

149

:

but

150

:

Randall Stevens: like,

151

:

let's just make it more complicated.

152

:

Let's set it on fire and see

153

:

Evan Troxel: just like, like if when you

actually wrap your head around where this

154

:

eventually got to, it's absolutely insane.

155

:

Brent Seales: Well, when we finally

discovered what some of the iconic

156

:

artifacts are, you know, that that is what

the Herculaneum scrolls, you know, are

157

:

that they are pieces of papyrus wrapped

up and crumpled and then set on fire and

158

:

then, you know, baked for 1800 years.

159

:

You know, I mean, crazy.

160

:

Right?

161

:

Uh, didn't even know they

existed when we started the work.

162

:

Um, but my one sentence would, would

be, you know, when I finally met that

163

:

material and I walked into the library

and I saw that collection, I mean,

164

:

the magic happened because I looked at

it and I'm like, this is the easiest

165

:

project in the world to describe

to a lay person, which isn't always

166

:

true when you do technology right.

167

:

And the description is, I wanna find a way

to read what's inside that scroll right.

168

:

Without opening it.

169

:

And that's it.

170

:

Uh, you know, it gets more involved

when you say how you're gonna do

171

:

that, but you know, the, the ultimate

goal is just really easy to say,

172

:

right.

173

:

Randall Stevens: Well, it's gotten a

lot of, it's gotten a lot of prominence,

174

:

uh, the last couple of years, just

because I'll let you describe,

175

:

you know, how that all came about.

176

:

But there was a official challenge.

177

:

There was, uh, some money that was put up.

178

:

Uh, and then with the, uh, with

the new advances in ai, using

179

:

those techniques now to kind of

decipher, uh, the data is actually

180

:

now proving out that it's working.

181

:

But, you know, to me, the, because

I was able to kind of see this, uh,

182

:

from a far, uh, Brent, you know,

it's, it's the classic, you know,

183

:

25 year overnight success story.

184

:

It's like all these little incremental

things that were built up, you

185

:

know, and, and you know, I think

when we had Christie on from your

186

:

team talking about this, just the

testament to, you know, not being told.

187

:

Uh, that something's not possible, right?

188

:

It's that, that, that fortitude really,

because if you knew how long it was it

189

:

gonna take, you probably would never,

it's one of those falls into those camps

190

:

that I would just not bang my head against

the wall for, from so many, um, so many

191

:

challenges along the way, but, uh, yeah.

192

:

Kuda, kudos to you to keeping this

moving, and I'm sure a lot of students

193

:

over the years that were in and outta

your lab, uh, you know, con contributed

194

:

to this, uh, ability to get here.

195

:

So, um, yeah, maybe you can, uh, I

don't know if you want to, do you want

196

:

to describe maybe what happened around

the pandemic time and the way that

197

:

the, uh, meeting that and, uh, and,

and how all that challenge came about?

198

:

Brent Seales: Sure.

199

:

Yeah.

200

:

I mean, we've, we've made systematic

progress, you know, for two decades,

201

:

and a lot of the things that were

roadblocks were not technical.

202

:

They were things that had to do with

rivals or access or lack of funding,

203

:

you know, momentary lack of funding.

204

:

Um, there were some

roadblocks that involved, uh.

205

:

Inability to meet a technical

challenge where we had to go back

206

:

to the drawing board and figure

out, okay, what do we do now?

207

:

So for example, the, the scans we were

using weren't high enough resolution, and

208

:

because it wasn't a typical camera, right?

209

:

We were doing tomography at

a high energy physics place.

210

:

You know, it wasn't really

directly obvious how to get the

211

:

resolution we needed, um, because

there were some barriers there.

212

:

But we, we figured that out During,

during the covid period, we, we had

213

:

just finished acquiring what proved

to be the golden scans of two scrolls,

214

:

and we had them in 2019 actually.

215

:

And so you might be thinking, well, if

you had 'em in:

216

:

you until 2023, you know, to crack

that open and, and, and break through.

217

:

Uh, COVID was one of the reasons, and

another reason was that, um, we needed ai.

218

:

To be able to improve the very

weak signal inside that scroll that

219

:

represented the evidence of the ink.

220

:

And we were systematically working

in that direction and had a

221

:

bead on exactly how it worked.

222

:

We, in fact, we published a paper in 2019.

223

:

That was the recipe.

224

:

Right?

225

:

But, you know, it's hard, it's hard work.

226

:

And it was slow going and,

and Covid slowed us down.

227

:

And also, um, you know, we had, we had

a small team and uh, you know, when you

228

:

start wading through the data inside

a scroll that's been scanned at, at

229

:

an eight micron resolution, which is

the size of a red blood cell, okay?

230

:

That's how small that is.

231

:

Um, that's a lot of work.

232

:

That's just a lot of grunt work to,

you know, that's spade work that,

233

:

uh, we knew we could accelerate.

234

:

And that's where Net Nat came in.

235

:

Yeah,

236

:

Randall Stevens: So.

237

:

Evan Troxel: And can, can you just

describe real quick, Brent, like this,

238

:

you, you're, you're scanning this really,

really, really, really, really tiny

239

:

thing, but I think Christie mentioned

it's like this three dimensional,

240

:

it's a 3D, 3D pixel, a voxel kind of

a thing where it's, it's locating, I.

241

:

in material or an application of ink

or you know, whatever it, whatever that

242

:

ink is made out of on top of the papyrus

in 3D space, which you then unfold.

243

:

And maybe you can just get a little

nerdy here for a second and describe

244

:

exactly what's going on there.

245

:

Brent Seales: Yeah.

246

:

I mean, in fact, we named our toolkit, um,

the, uh, volume cartographer because what

247

:

it allows you to do is, is map every three

dimensional point through the thickness

248

:

of something, not just on the exterior.

249

:

That's what tomography gives you is, is

an estimate at every point all the way

250

:

through something of what its density is.

251

:

Um, and that comes from the X-ray, and

most people are familiar with that from

252

:

having medical, you know, attention, uh,

because you can see your, you know, your

253

:

insides, uh, non-invasively when you

do these, uh, CT scans at the doctor.

254

:

And we're basically leveraging the same

technology, but what we're seeing is

255

:

the internal structure of the scroll,

every layer, all the way through

256

:

the thickness of the scroll, right?

257

:

Without having to open up those layers.

258

:

Evan Troxel: Nice.

259

:

Randall Stevens: So maybe, uh, let's pick

up with, uh, you mentioned Nat Friedman.

260

:

I think he, if I've got this right,

he was the founder of GitHub.

261

:

So if people are used to the code

repository, GitHub, uh, he eventually,

262

:

that company was sold to Microsoft.

263

:

He's on the Microsoft team.

264

:

He's become independently

wealthy, you know, beyond belief.

265

:

Uh, hits one of those

kinds of home runs and.

266

:

Uh, maybe pick up, uh, you know, I

think he reached out and contacted you.

267

:

Maybe you can kind of pick

up there on the story.

268

:

Brent Seales: Yeah, so I, I don't know

NA's resume perfectly, but I, I think

269

:

actually they, uh, Microsoft acquired a

different company that he had founded,

270

:

and then he became a Microsoft executive,

and then he advocated for acquiring

271

:

GitHub, which they did,

and then he became the

272

:

Randall Stevens: Got it.

273

:

Brent Seales: um, you know,

which is just in incredible.

274

:

So when he sent me an email, I, I didn't

think it was probably real, because I

275

:

get a lot of email from people who claim

to be this and that, and, and, you know,

276

:

it's not that, uh, it really was him.

277

:

He, he cold called me.

278

:

He reached out to me and said.

279

:

I'm having a conference and I think

it would be great for you to come out.

280

:

Maybe we can help you raise some money,

maybe we can find interested partners.

281

:

And so I did.

282

:

I went out and, um, joined him

for one of his Silicon Valley,

283

:

uh, uh, weekend conferences.

284

:

It was an unconference, which meant that

when we got there, we all kinda self,

285

:

self-organized and I got to give a talk.

286

:

It was fantastic.

287

:

And you know, a number of times he said

to me, you know, maybe we can kick a

288

:

few rocks and start an avalanche, right?

289

:

Because he knew that we were pretty

far advanced in what we were doing

290

:

and had seen some of our work online.

291

:

So, you know, Nat was very

well prepared and true.

292

:

CEO style knew everything

about what we'd done.

293

:

Um, what I didn't expect though, was

for him to pitch me because I was

294

:

going out there to pitch him, right?

295

:

Um, you know, he was one, the one

with all the connections and the

296

:

money and everything else, and,

uh, he turned it around, which I

297

:

love, you know, it's a great story.

298

:

And he pitched me on the idea of doing

a contest, which I hadn't thought of.

299

:

And I, I wouldn't have known

how to do it by myself anyway.

300

:

But in partnership with na, we launched

this crazy idea of forming the Vesuvius

301

:

challenge and seeing if we could

accelerate to the, to the punch, um,

302

:

the place that we were already standing.

303

:

You know, basically kick the

rocks, start the avalanche, and

304

:

that is exactly what happened.

305

:

Randall Stevens: So the, the idea was

to raise some money, put up, uh, put up.

306

:

Rewards for being able to

advance the technology.

307

:

And, you know, I think, I think, you

know, we, we've talked about this, but

308

:

to your credit, the, the idea of this,

you know, your baby that you had worked

309

:

on for all these years, the idea that you

were gonna like now kind of expose that

310

:

you know, the data and let other people

kinda open, you know, it's an open source.

311

:

Maybe, maybe because he was running

GitHub and, and involved in all that,

312

:

that he's always thinking that way.

313

:

But, you know, this idea that there is

this open source of all this data, uh, I

314

:

think it's, you know, a testament to you

and we've talked about this a little bit,

315

:

about, you know, that's the fear point

is like, all this stuff that I've worked

316

:

on, am I gonna like now put it out there?

317

:

But, you know, magic happens.

318

:

It sounds like, you know, when, when

you, when you took that next step.

319

:

So maybe you can give us more.

320

:

Brent Seales: Yeah.

321

:

Uh, I, I think CEOs maybe, and you know,

you, Randall, you're an entrepreneur.

322

:

I don't know about you, Evan, you

know, what your various incarnations

323

:

in life have, have been like, but,

um, I think CEO types and founder

324

:

types are, are pretty used to the idea

that you can't hold onto everything.

325

:

You have to build a company, which

is a group of people, not just you.

326

:

Right.

327

:

Professors, not so much.

328

:

I mean, we are not ingrained with

this idea that we, uh, we, we gift

329

:

the world, you know, with, with

our thing or, or build a company.

330

:

Everything is pretty tightly held.

331

:

So he pitched me and I had to come

back to my team and, uh, kick, kick the

332

:

idea around quite a bit because yes,

we had invested a ton in collecting

333

:

data that no one in the world had.

334

:

And we firmly believed it was the golden

data that inside that data we would be

335

:

reading a book from the ancient world.

336

:

And if I had held onto that and said,

um, this is gonna be just for me, for

337

:

us, I think I probably would've gotten

there, but it would've taken a lot longer.

338

:

And you know what?

339

:

The impact in being able to

walk alongside all these other

340

:

people would never have happened.

341

:

And I mean, we can talk about that, but I.

342

:

But Nat and Daniel Gross alone, the impact

of being able to meet them and being able

343

:

to get to know them and being able to be

astounded by the fact that they didn't

344

:

just raise money, Randall, they actually

put up their own money and a lot of it,

345

:

a lot more of it than I make in a year.

346

:

They brought to the project of their own

money and said, we believe in this so

347

:

much that we're gonna be the first donors.

348

:

Right.

349

:

I mean, I just loved that

about them, that they did that.

350

:

Um, it, it, it secured for me,

you know, a real confidence

351

:

that we were gonna be partners.

352

:

Right.

353

:

That instead of me just being

sort of acquired, you know?

354

:

Right?

355

:

And I needed that confidence because

it was a big step and it wasn't part

356

:

of my thinking until I met them.

357

:

Randall Stevens: Yep.

358

:

So how much money describe the

challenge and how that was structured.

359

:

How much money did you raise and, uh.

360

:

Brent Seales: I think we launched with

a quarter of a million dollars, and that

361

:

was mainly money from Nat and Daniel and.

362

:

Nat had said, we're gonna launch

and we're gonna see if we can keep

363

:

raising money because we can run

the contest and pay for the prizes.

364

:

Um, but by raising money we can

also go ahead and and hire people.

365

:

Um, and what happened is that within

a week that doubled and then within

366

:

another week it doubled again.

367

:

So we were over a million dollars

in prize pool and that launched the

368

:

competition because, um, a million dollar

prize right, is much more fancy right.

369

:

Than a quarter, quarter of a

370

:

million.

371

:

Evan Troxel: Right.

372

:

Brent Seales: And also it just started to

build this, this community of other people

373

:

besides just Nat and Daniel who started

to believe and root for us to succeed.

374

:

And you know, some people, um, there's

a website somewhere where people put

375

:

odds on various things and it's, they

trade fake currency by betting, you

376

:

know, um, I don't remember the name

of it, but we were tracking that

377

:

and there were a lot of people who

were like, oh, this is 30% they're

378

:

gonna make it otherwise No way.

379

:

And.

380

:

I was always a hundred percent

believer, but, uh, it, it generated

381

:

a lot of, of attention once

we got over a million dollars.

382

:

Right.

383

:

And then what it was really about

is, is structuring the competition

384

:

so that the funnel Right.

385

:

Didn't lose all the people and in the

end, we actually had viable competitors.

386

:

And, uh, you know, Nat was

really brilliant at doing that.

387

:

And, and I'm sure you probably talked

to Christie about some of that because

388

:

she's actually doing thesis work right.

389

:

In, in understanding the community

that formed in the, in the Discord

390

:

channel around the contest.

391

:

And, um, the goals that we had,

392

:

Randall Stevens: So how many, how many

people you know would you estimate that

393

:

got, you know, actively started working

on it or testing theories and code?

394

:

Brent Seales: well, we, we started

the competition with a first phase

395

:

that was, uh, Kaggle competition.

396

:

So we partnered with Kele and it was

sort of phase one and we had over a

397

:

thousand teams in the Kele competition.

398

:

So that was a pretty good start.

399

:

And then we started to monitor the

Discord channel and saw, you know,

400

:

regularly two or 300 people active on

the Discord channel, you know, monitoring

401

:

what's going on or contributing to a

conversation In the end, I am not sure

402

:

exactly how many people worldwide, you

know, dipped their toe in the water.

403

:

Um, we had, we had 20 or so, uh,

submissions for the final grand prize,

404

:

uh, from the first year's competition.

405

:

And, um, many of those were viable.

406

:

So, you know, that was pretty good.

407

:

And that was a $700,000 prize.

408

:

Randall Stevens: Yeah.

409

:

Nothing to sneeze at.

410

:

And the, uh, the young man who

first I guess was able to read the

411

:

first words out of it, he was a

Nebraska boy, if I remember right.

412

:

Uh,

413

:

Brent Seales: That's right.

414

:

Yeah.

415

:

We had a competitor who was

basically an undergraduate at

416

:

the University of Nebraska.

417

:

His name's Luke Ferrato.

418

:

And he submitted, for

the first letters prize.

419

:

First words prize, and he won

that back in October,:

420

:

And the word that his AI identified from a

section of the scroll from deep within the

421

:

scroll was, uh, the word purple in Greek.

422

:

Uh, a really interesting kind of rich

piece of vocabulary that, that, you

423

:

would hope would be the first word.

424

:

Right.

425

:

You know, I mean, we got made fun of

a little bit because why is everybody

426

:

jumping up and down that you found

one word inside a burnt scroll?

427

:

I.

428

:

But what we knew was that,

uh, it was opening up, right?

429

:

The whole thing technically was

working soup to nuts all the way

430

:

through, and that the first word

was just that it was the first word.

431

:

But what we were gonna get were

many, many columns of a book from

432

:

antiquity that no one's ever read.

433

:

Right?

434

:

Randall Stevens: And then

he was part of the team.

435

:

Then a couple of others that that decoded

paragraphs or several pages is that.

436

:

Brent Seales: Well, Luke won the first

letters prize, but he stayed with the

437

:

competition, which a few months later,

you know, included the grand prize.

438

:

And what happened is that Luke teamed up

with a couple of the other competitors

439

:

strategically because I think originally

Luke was working by himself and he

440

:

made it to first letters by himself.

441

:

But I think strategically they realized

collaboration would probably get them

442

:

closer to winning the grant prize

than fighting against each other.

443

:

And so they made an agreement about

how they would spl split the prize.

444

:

And then, um, Yusef, uh, and Luke

and, uh, Julian were the three who

445

:

teamed up and ultimately won the grand

446

:

prize.

447

:

Randall Stevens: and it was

like, what,:

448

:

Like multiple or several thousand

449

:

Brent Seales: Yeah.

450

:

We had asked as the grand prize

threshold to find four separate passages.

451

:

Was it four of, of about 128 letters with

no mistakes, basically confirmable by

452

:

someone who knows the ancient language.

453

:

And they did it.

454

:

I mean, they met the threshold.

455

:

It was, and we thought that was way out

there, you know, um, based on all of my

456

:

work, you know, uh, we were working with

fragments that were open because the open

457

:

fragments gave us kind of the treasure

map for how to detect the ink with ai.

458

:

Right.

459

:

You know, you, you, you know,

from photographs where the ink is.

460

:

And so you can align that with

the tomography over here and um,

461

:

then you can learn the ink and

that's an easier problem, right?

462

:

Than diving in to the center of the

scroll where you have no reference

463

:

and just trying to find the ink.

464

:

Right.

465

:

Um, but they met the threshold

and we were able to award the

466

:

grand prize, which was fantastic.

467

:

Randall Stevens: Hm.

468

:

So maybe you can describe

Brent, the, you know.

469

:

Where does AI come in on this?

470

:

And you know, from what you've told me

and I've read, um, you know, the ink

471

:

is, is not perceivable by the human eye.

472

:

You can't even tell.

473

:

So, and then in the scans even, it was

very difficult to see with the na, with

474

:

the human eye, any perceivable difference.

475

:

So how did AI come to play on this?

476

:

Brent Seales: Yeah, a

couple things about that.

477

:

Um, the examples we had that were

fragments with visible ink on the

478

:

top layer, we were using those as

control experiments in the lab.

479

:

We had real trouble seeing

where the ink was in the x-ray.

480

:

Even though we knew it was there, um,

and yet we could train the AI to, to

481

:

learn to enhance where the ink was

correctly, even though the evidence of

482

:

it to our naked eye was wasn't there.

483

:

I always had this theory that the

ink on the exposed layers, because

484

:

they've been hundreds of years, you

know, open, was probably weaker than

485

:

the ink that we would find inside

the scroll, but I had no evidence to

486

:

really back That up.

487

:

Um, but what happened with the competitors

was that we had, you know, a thousand

488

:

people hammering away on this problem.

489

:

And we had virtually unwrapped

brute force with, uh, you know,

490

:

our volume cartographer tool.

491

:

We had, we had brute force unwrapped,

a bunch of surface area from inside.

492

:

Um, but again, most of that surface

area, it didn't have ink that

493

:

was visible to the naked eye.

494

:

One of the competitors though,

very, very carefully observed.

495

:

Uh, hundreds of square inches of

opened surface area and was able

496

:

to detect evidence of the ink.

497

:

It looked a lot like a crackle

in a mud puddle that's dried up,

498

:

you know, so, you know, the, the

puddle, let's say, is on concrete.

499

:

So you have concrete, uh, as one

texture, and then the mud creates another

500

:

texture with some cracks in it, right?

501

:

So when you look at that, you

say, okay, that's mud, and it's

502

:

different from the pavement.

503

:

That's what they were seeing

on the surface of the papyrus.

504

:

The papyrus has fibers that run in,

in two directions, cross hatched,

505

:

and then on that cross hatch pattern,

they were seeing this, this pattern

506

:

that looked like crackled mud.

507

:

And it turned out that

was evidence of the ink

508

:

and the part that they were able to

detect with the naked eye gave them

509

:

a foothold to successively train.

510

:

The AI detector to get better

and better at seeing the ink.

511

:

And so then it was just a race.

512

:

It was just a race.

513

:

And so Luke was on the front end

of that race with the word purple.

514

:

Um, and then there was an

acceleration to the finish line.

515

:

Who could get the best AI to

amplify that evidence and see,

516

:

see the ink board clearly?

517

:

Randall Stevens: so the, the work

that they were doing on the AI side,

518

:

ultimately you're trained, you know.

519

:

My novice understanding is you, you need

inputs, you need a lot, normally you need

520

:

a lot of input in order to train a model,

521

:

Brent Seales: Mm-hmm.

522

:

Randall Stevens: uh, to

have enough data points.

523

:

It sounds like in this example,

there's very few, there's very

524

:

little data or very limited data.

525

:

How, how can you describe, how do they.

526

:

Brent Seales: Well, yeah.

527

:

So our initial approach, uh, before

we started the contest that we, we

528

:

believed would bear a fruit because

it was working in the lab, was to, um.

529

:

To use, uh, inputs that were just

very, very small sub regions.

530

:

Uh, let's say eight by eight

region of voxels from the scan.

531

:

Um, so you take the center nine by nine,

five by five, you take the center voxel

532

:

and then some sub region around it.

533

:

And it's very, very

small, just a tiny point.

534

:

And then you map that to uh,

just a yes or no label that says

535

:

this is ink or this is not ink.

536

:

And we did that mapping because we

had a photograph of the top layer

537

:

where we knew where the ink was.

538

:

And having done that mapping, we were able

to train up a system that reliably with,

539

:

you know, an enfold kind of experiment,

would detect the ink and would generalize.

540

:

Um, but you know, it wasn't very good.

541

:

It was hard to see that it was readable.

542

:

So we made that our first

Kaggle contest actually.

543

:

Um, but to answer your question.

544

:

Randall, the reason why we had

lots of labels is because those

545

:

regions were really small.

546

:

So if I have an image that's got 16

million pixels in it, right, and I have

547

:

only an eight by eight window, right?

548

:

Then you're gonna have a bunch

of, you know, a million labels

549

:

in this single image, right?

550

:

Just by taking that region and, and

mapping it all, all, all over the surface,

551

:

um, still, you know, not at scale

like we are now, but for the time

552

:

we still had enough examples of

the ink and knowing in that we were

553

:

able to get this system to work.

554

:

Randall Stevens: So are Brent, uh,

with the challenge and the prize money,

555

:

uh, is the goal that as people make

these discoveries, they have to teach

556

:

and or expose the tools that they've

written, like is how, how is, how is

557

:

this gonna progress and, and, you know,

um, make this more widely available?

558

:

Brent Seales: Yeah.

559

:

You know, Nat and the folks that he

hired to run the challenge were really

560

:

masterful in the way, uh, that they,

they created an open en environment,

561

:

even though it was competitive.

562

:

Uh, one of the things that they

wanted to do was level everyone

563

:

up every time a prize was given.

564

:

So the terms of receiving the prize

terms required, uh, open sourcing tools.

565

:

So did a lot of progress prizes, we,

where we rewarded people for, you

566

:

know, building sort of the next tool

that just helped you do something

567

:

a little bit more efficiently.

568

:

So then all the competitors get that,

and so then it kind of creates this

569

:

community of people who are still

competing with each other, right?

570

:

But they're also being rewarded along

the way, um, in these sub races, right?

571

:

With, you know, not

trivial amounts of money.

572

:

10 K, 20 k.

573

:

Right.

574

:

For something that might have taken a few

weeks, but really removes a pain point.

575

:

And again, this is something that's

so hard to replicate with a regular

576

:

research group in the university.

577

:

Like, you know, how do you

really, how do you really inspire

578

:

that kind of, you know, focus?

579

:

I can't give people here

a cup of coffee, right?

580

:

Like from the university, I have

to pay it outta my own pocket

581

:

because, you know, republican institution,

there's no such thing as a spot

582

:

bonus or a, uh, you know, a prize.

583

:

Um, and yet, you know, with the

competition we were able to do

584

:

exactly that and we could get sort

of pinpoint, um, results that also

585

:

created a, an incredible community.

586

:

Um, and you know, Nat, Nat thought

that up with the help of JP Psma and,

587

:

and, uh, Daniel heavier who, uh, were

guys that, that nat through his huge

588

:

network of folks he was able to hire.

589

:

Uh, to help him basically launch.

590

:

And then ultimately my PhD student, Steven

Parsons, became a part of Nats team and

591

:

he works to this day, uh, with that group.

592

:

I think it's been a really fruitful, sort

of postdoctoral experience for Steven.

593

:

Randall Stevens: so, so maybe Brent

talk about, you know, this is.

594

:

This is literally probably two worlds

colliding technologists with people

595

:

in the humanities and historians.

596

:

And, um, one, how has the work been

perceived, you know, on that side?

597

:

And then you kind of hinted at earlier,

one of the challenges is just getting

598

:

access to these antiquities are

probably closely held for good reason.

599

:

A lot of times, I think Christie

talked about that when we had her

600

:

on too, is it's, uh, but maybe you

can talk about, you know, are these

601

:

two, two worlds are colliding?

602

:

Uh, is it good, bad,

moving things along faster?

603

:

Uh.

604

:

Brent Seales: Well, it's

definitely accelerating.

605

:

Um, it, it isn't, uh, trouble free

because there, there are collisions.

606

:

Uh, I think generally though, um,

the colliding worlds are moving

607

:

in the, in the right direction.

608

:

Ologists and humanities scholars

are, are becoming more collaborative

609

:

and also more accepting of the value

of things like ai, uh, to really

610

:

tremendously open up their fields.

611

:

We're talking about books that are

new from the ancient world, which is

612

:

something that hasn't happened for 50 or

a hundred years in the field of pathology.

613

:

You just don't have whole

books being discovered anymore.

614

:

Um, but typically humanities scholars and

Ologists, they don't work together, right?

615

:

The way they work is at a

pretty slow pace individually.

616

:

Um, and then they, they reveal

or publish their, their stuff,

617

:

and then they ar argue about it.

618

:

Um, and then they publish something else

and, and keep, keep the arguments going.

619

:

And I, I don't mean argument

in, in a nasty way, but you

620

:

know, it's, it's academic

621

:

argument, right?

622

:

To try to ultimately converge on the

truth of whatever is being debated, right?

623

:

Um, technology folks are different.

624

:

They want things to be, uh,

pretty quickly happening.

625

:

Um, open source, uh, revisions.

626

:

AI is fundamentally iterative.

627

:

So you don't create an image that

is definitive, but you create an

628

:

image that's on a, an optimization

pathway toward a better image, right?

629

:

And that drives.

630

:

Humanities scholars nuts because they,

they want to have something that's not

631

:

moving, it's fixed, and then debate it

and, you know, understand it and edit it.

632

:

Those are some of the pain points.

633

:

Um, but ultimately, you know, I think

the biggest thing that's helped me

634

:

is that I've, I've approached the

problem as one of, not technical

635

:

virtuosity, although that's there,

but scholarship, like the goal is to

636

:

enable scholars who know the languages

to understand and do scholarship,

637

:

not to be gaga about the technology.

638

:

And I think that more than anything

has just helped those scholars realize,

639

:

hey, they're, they're trying to help

us here, you know, to unlock a library

640

:

that's never been unlocked and let

us do the scholarly analysis of that.

641

:

Um, other, other of my

competitors didn't take so.

642

:

Generous of you, I would say in, in that

they just wanted to focus on their own

643

:

invention of the technology and have that

sort of be the end of the conversation.

644

:

And I think that you, you miss,

you miss something, right?

645

:

If you focus only on what the tech

can do for the tech's sake, right?

646

:

Randall Stevens: So like with the,

uh, with the Vesuvius Scrolls, one,

647

:

how many are there and where are they?

648

:

And is there gonna be a foreseeable

pipeline of starting to try to

649

:

get these things scanned and.

650

:

Brent Seales: Yeah.

651

:

There are a couple things there,

uh, that we're trying to do.

652

:

The first, to answer your question,

how many, um, there were probably

653

:

about 1200 books in the original

cash, uh, excavated in 17 50, 52.

654

:

And the, the reason why the

numbers uncertain is because

655

:

over time some were destroyed.

656

:

Many were experimented on, bunch were

opened and they created fragments.

657

:

And so now it becomes much

more difficult to, to count.

658

:

You know, it's like giving a

toddler a bag of cookies and then

659

:

at the end trying to count the

660

:

pieces.

661

:

How many cookies were there?

662

:

It's like, who knows?

663

:

Yeah.

664

:

You know, and, um, we can count large

chunks now in the, uh, on the order

665

:

of three or 400 fairly sizable chunks.

666

:

And, um, experts are saying that

maybe that represents a couple

667

:

hundred books, maybe 150 books,

certainly more than a hundred books.

668

:

Each one unknown and unique

from the ancient world,

669

:

probably not a copy of anything.

670

:

This, this archive has proved

for the most part to be unique

671

:

books, so that's really cool.

672

:

Um, as far as our, our ability to

access them, we still have, um, a

673

:

great relationship with the library

and they are allowing us to scan

674

:

a batch of 20 coming up in March.

675

:

So, um, that will be huge news because

this, you know, we're gonna be escalating

676

:

in sort of orders of magnitude.

677

:

So from two to 20, right?

678

:

And then we hope next year to 200.

679

:

And by doing that, of

course the AI is going to

680

:

improve because we will see more papyrus,

we'll see more geometric configurations

681

:

of what these things look like.

682

:

Um, and, you know, we'll also just

generate, it's gonna change the field.

683

:

I mean, Randall, we're really on the edge

of bringing from the classical period

684

:

more new material, uh, to the scholarly

world than has happened in 500 years.

685

:

I mean, the last time was the

Renaissance when people like, you

686

:

know, brachii, who is the subject

of this book, the Swerve Right?

687

:

Found, um, books by Lucious and, you

know, and other pla other, um, authors

688

:

in libraries around medieval Europe

in the early Italian Renaissance.

689

:

It, it's been since then, right?

690

:

That we've had such a large

cache and trove of potential

691

:

material from the ancient world.

692

:

So, you know, how often do we get

to stand on this threshold of,

693

:

you know, something like that.

694

:

I think it's pretty exciting.

695

:

Randall Stevens: It's very cool.

696

:

I, uh, I think I probably, uh.

697

:

Um, I told a little bit of this

story when Christie was on, but the,

698

:

the whole Confluence event that we

put on was, was literally me having

699

:

coffee or lunch with Brent one

day in:

700

:

And I was, I had sent him a note and

was like, you know, I need to get up.

701

:

I'm looking for somebody who can

help us think about what's going

702

:

on with AI and how we might apply

it, you know, in our business.

703

:

So I go up, Brent was kind enough to,

uh, join me for coffee and, you know,

704

:

we just had one of the normal kind of

conversations, interesting conversations,

705

:

and I remember walking back to the office,

walking back downtown from the university

706

:

and I'm like, I'm gonna have an event

where I'm gonna bring all my friends from

707

:

the architecture and design world and I'm

gonna mix 'em up with Brent's friends.

708

:

'cause he knows all these great people.

709

:

You know, Brent's a Google fellow.

710

:

He's like connected with all of these

famous scientists doing all this.

711

:

I'm like.

712

:

And I'm gonna call it Confluence.

713

:

We're gonna bring these people

together and once a year, you know,

714

:

talk about some of this kinda stuff.

715

:

So I'll have to give, uh, uh, Brent

credit for, uh, kind of the impetus to

716

:

get this, uh, you know, really it is, it

was these worlds, you know, thus the name.

717

:

That's why we ended up

calling it Confluence.

718

:

But it's like, can we bring together

people from even outside of protection

719

:

engineering world that are working on

things and be inspired about how this

720

:

either, uh, you know, could apply or

to bring these technologies to bear on

721

:

the kinds of things that we're trying

to work on inside of our industry.

722

:

So, um, just lots of, and

actually the first year, um.

723

:

You know, I am gonna get you to,

to, to, to join us this year, Brent,

724

:

because you've put me off twice.

725

:

I've had the first year, uh, James from

your team, uh, joined us at the first

726

:

Confluence and gave a little overview.

727

:

This was back in 2018 about what you all

were doing and had a great experience.

728

:

James, I think, enjoyed it too,

and I think he came back the next

729

:

year and joined us even, even

though he wasn't given a talk.

730

:

And then this past year, uh, Christie

from your team gave, gave a great

731

:

talk that everybody really enjoyed.

732

:

So at some point we're gonna get you to

733

:

Brent Seales: There are no probably

no talks left for me to give though.

734

:

Everybody's given

735

:

Randall Stevens: No, no, no, no.

736

:

It's just, uh, well, it's such a, it's

like I said, it's uh, it's like one of

737

:

those perfect projects where I think it's

just that combination of SI technology

738

:

and humanities and, you know, kind

of bringing all that stuff together.

739

:

I.

740

:

It has a lot of romanticism to it

and, uh, you know of, and, and also

741

:

now opens up, I think, kind of all

these thoughts about, well, what else?

742

:

Like with the, is there any sh do they

think that there are more of these

743

:

scrolls that just haven't even been

discovered, that were buried, you

744

:

know, by, by the volcanic ash That that

745

:

Brent Seales: If you look at the, uh,

the archeological site of the villa

746

:

of the Papyri, which is actually

in the town of Herculaneum, it's

747

:

around the, around the coast from

where Pompeii is, a different town,

748

:

but you know,

749

:

nearby, if, if you look at that

site, you'll see the original

750

:

tunnels that were dug in in 1752.

751

:

Very little extra excavation.

752

:

With some exceptions, but

not that much has been done.

753

:

So yeah, it's entirely possible

that to the left or the right of

754

:

any of those tunnels by one foot

is a case full of scrolls laying

755

:

there waiting to be discovered.

756

:

We feel like we've taken one of

the big, big reasons why they're

757

:

not gonna do that off the table.

758

:

And so it might be possible in the

next iteration after we've read these

759

:

400 scrolls to start excavation.

760

:

And you know, I don't, I don't know why

we would not wanna fully excavate the

761

:

villages, the only library from antiquity

that's ever been discovered in situ.

762

:

And you know, I mentioned the 500

year high of reading what we have.

763

:

If we discovered more material in the

library through archeology, it would

764

:

be the biggest discovery in human

history from the classical period.

765

:

So.

766

:

Randall Stevens: Incredible.

767

:

Evan Troxel: One of the things you

mentioned, Brent, was like when you set

768

:

out that the, those challenges early on,

you said you thought you set the goal

769

:

pretty far out and, and, then obviously

surpassed kind of your, maybe your

770

:

wildest dreams about what was possible

within such a short amount of time.

771

:

Part of that is, you know,

bringing in people and them teaming

772

:

and this open source kind of

incentivization of how this all works.

773

:

And then there's the tech and how fast the

tech is moving and, and I'm just curious,

774

:

maybe you could paint a picture for us

because I think it starts to speak to.

775

:

The way we look at tools in different

industries, but how we should maybe

776

:

look at them similarly, which is

like, whoa, this is moving way faster

777

:

than anybody thought was possible.

778

:

And, and I'm just curious, maybe it

sounds to me like it's the backbone of

779

:

your, your research and your process now.

780

:

And it maybe it always was, I know

you started in computer vision,

781

:

but this AI thing has really

made computer vision on steroids.

782

:

Right?

783

:

It's just done that to it and it's

kind of doing that to everything.

784

:

And so, you know, there's a lot of

people in a lot of industries, not just

785

:

architecture and not just academia who

are kind of sitting on the sidelines

786

:

watching this happen, saying, I'm

just gonna sit over here and watch.

787

:

And then there's teams like you

who have definitely benefited from

788

:

this in incredible acceleration.

789

:

Could you just talk about that and, and

like how your eyes have opened or how

790

:

your perspective has shifted in this, you

know, last six years of, of doing this?

791

:

Brent Seales: Well, yeah.

792

:

Uh, it, it really has accelerated and

also, um, AI in the way it works has

793

:

turned engineering a little bit on

its head where you can solve problems

794

:

without necessarily understanding

all of the reasons why it's being

795

:

solved, which is kind of weird to say,

796

:

but people ask me, you know, what,

what is the AI actually seeing in the

797

:

data that it's using to detect the

difference between ink and not ink?

798

:

And the, the correct answer has to

be, you know, we're not exactly sure

799

:

here are the things that are possible.

800

:

You know, maybe the fibers get

coated, uh, with ink and that

801

:

makes a, a, a different pattern.

802

:

Maybe it's a little thicker there.

803

:

Maybe there's a texture that's different.

804

:

But at the end of the

day, who cares, right?

805

:

We're solving the problem

of enhancing the ink.

806

:

We're doing it reliably and industry

wide, you know, sort of, we're,

807

:

we're seeing a lot of who cares?

808

:

This works, so just use it.

809

:

We don't necessarily need to

tease out all the, the things that

810

:

are happening that make it work.

811

:

And that was never how computer

vision worked in the past.

812

:

We, we had to decompose everything into

the pieces parts, and then put it together

813

:

and sort of understand every step, right?

814

:

And then kick the tires on the weaknesses

of each step and try to ablate what

815

:

was the weakest part, make that better.

816

:

You know, that was basically how it

worked, and it's not really true anymore.

817

:

And that's a huge shift in problem

solving and is part of the acceleration.

818

:

Stop under trying to

understand everything.

819

:

Just, it's like, don't try to

figure out how the piano works.

820

:

Just play the piano.

821

:

It's beautiful.

822

:

right.

823

:

So.

824

:

Evan Troxel: Um, that, that opens up

another question as far as like tools.

825

:

I mean, you, you talked about

like who cares what, what tools.

826

:

You're, like, you're talking about

the anti, the antiquities side, the

827

:

people who are in the humanities,

and it's like, well, this is a tool.

828

:

And it's kind of like that too.

829

:

Who cares?

830

:

Like it's amazing that what it does right

is magic and it's helping us do what we

831

:

do better and, and push our field forward.

832

:

and and I, I just want to emphasize

that like, I mean, in architecture

833

:

it's kind of the same with clients.

834

:

Like they don't care what tools we

use, but there's a lot of excitement

835

:

in, especially in Randall's, in my

circle about the tools themselves.

836

:

But it's not why we do what we do.

837

:

And I think that that's kind

of true for architects, right?

838

:

It's like we don't do it so that

we can use these tools or so that

839

:

we can evangelize these tools

or so that it's not about that.

840

:

It's about solving these big, bigger

problems in culture and society and humans

841

:

and the way they interact with space and

what it can do for people and how it can

842

:

make them potentially better citizens

because they have better spaces to live

843

:

in and work in and things like that.

844

:

So like not just pursuing text for tech.

845

:

Tech for tech's sake, right?

846

:

It's like, what, what is the big picture?

847

:

Let's step back.

848

:

And I just think this is a kind

of a brilliant reminder of that.

849

:

It's like it's, it's, these are the

things that are coming alongside and,

850

:

and as humans we're using for leverage,

which is what we do with tools.

851

:

It could be a digital tool, it

could be an analog tool, right?

852

:

But the the why is the big

picture, and I think that's

853

:

what's such an important message.

854

:

Even when, when Randall asked the question

about how is it benefiting these other,

855

:

and how, how is that relationship going?

856

:

That that's just kind of a, a, a good

thing to step back and, and really take

857

:

stock of because that's, that is super

important just to remember why this is

858

:

happening in the first place and these

tools come along and help make it happen.

859

:

Brent Seales: I agree with that

and, and, um, I think one of the

860

:

things that seems so magical about

this moment is the humanities side.

861

:

It's that these tools are, are so human

862

:

and they're, um, whereas in the past,

you know, they weren't so mysterious

863

:

because it seemed, uh, kind of obvious.

864

:

You can build a backhoe and

it can move earth, right?

865

:

That's not really a human thing.

866

:

But you know, when we can build a keyhole

back in history:

867

:

Uh, summarize a corpus of enormous

mountains of material and in, with

868

:

pinpoint accuracy, ask a tool to give

us a summary of that five different

869

:

ways and in five different styles.

870

:

That always seemed like that

would be beyond the grasp of

871

:

anything that we could invent

872

:

that would require a human right.

873

:

And so I, I actually think that I'm in

this space, which is gonna be the future

874

:

of, of this, this whole enterprise, which

is the confluence of humanities and ai.

875

:

I mean, I think that's where it's at.

876

:

Randall Stevens: So

maybe, maybe it's a good,

877

:

Evan Troxel: yeah, go for

878

:

Randall Stevens: ahead.

879

:

No, go ahead.

880

:

Evan Troxel: Okay.

881

:

So one, one last question is,

is you've gone through this

882

:

process over the last few years.

883

:

You met Nat, you've had these incredible

people come along because of these

884

:

challenges and prize money, but also just

kind of the challenge of figuring it out.

885

:

Right?

886

:

And, and you mentioned earlier

that like, this isn't traditionally

887

:

how academia has worked, right.

888

:

And you talk about kind of the silo

protectionism that happens over your

889

:

research, your findings, your tools,

your processes, all those things.

890

:

And, and so like you mentioned, like

you didn't, you didn't know there was

891

:

another way necessarily, but I think

entrepreneurs and technologists and,

892

:

and specifically are, are pretty much

like, this is how they work, right?

893

:

And so they, they took their,

their thing and applied it to this.

894

:

I'm curious if it has actually changed

anything about how you go about.

895

:

Problem solving or maybe how you

envision how it could be and, and

896

:

what you would change if you could.

897

:

I know like that we have these,

these organizations and these,

898

:

you know, institutions that

are really not easy to change.

899

:

Right.

900

:

So, I don't know even know if it's

possible, but I'm curious, like, has

901

:

this inspired you to kind of rethink

the way you could or wish you could

902

:

kind of move, move around these

kinds of subjects in the future?

903

:

Or, and, and obviously

enable others to do also?

904

:

Brent Seales: it has, it's been

eyeopening for me to, to know there

905

:

is this other way that can work.

906

:

I, I don't think it's a, you know,

it's panacea for, you know, I, I think

907

:

that we kicked the rocks and started an

avalanche because we had carefully set

908

:

up the scaffolding through conventional

research that was systematic and

909

:

convinced us that we were ready.

910

:

Um, but, but I do think it, it

has changed my mind about, um,

911

:

those points when you are ready.

912

:

There are these other options

besides just, you know, digging in

913

:

the same old way, uh, to try to do

the same old thing on a task that,

914

:

you know, maybe is oversized or,

or for which we're ill-equipped.

915

:

Um, and, and I think that's, that's

the part of collaboration, right?

916

:

It was, it was Nat kind of flipping

the script, pitching me, and then

917

:

me being, you know, brave enough

and, and not, not, too stupid to

918

:

recognize the upside was probably

gonna be way bigger than the downside.

919

:

I mean, there has been downside, you

know, I mean, I, uh, I'm human and so,

920

:

you know, sometimes I think I'm missing

out on a piece of glory, but, you know,

921

:

the minute that I start thinking that

way, I, I look at what's happening and

922

:

I'm like, yeah, that's silly thinking.

923

:

I mean, look at the

upside to what's going on,

924

:

right?

925

:

Evan Troxel: How hard was it for him

to convince you in the beginning?

926

:

I know, and it wasn't just you, right?

927

:

Like you had to basically present

this to your team, and I'm sure there

928

:

were a lot of objections, but like how,

how big of a job was that for him to come

929

:

along and convince you and your team?

930

:

Brent Seales: he didn't

have to do a lot of arguing.

931

:

I mean, we laid out

what, what was possible.

932

:

I think that the convincing that had

to take place was, um, me making sure

933

:

I was gonna do right by my team, uh,

because it wasn't just a decision for me.

934

:

Um, and so I had to, had to make sure

that was true and then me, um, deciding

935

:

whether I could really trust Nat.

936

:

Um, because the foundation

of collaboration is trust.

937

:

And I've been in some collaborations

where the, the collaborations broke

938

:

down because people did things that

were bad or that you didn't expect

939

:

or that didn't consider other people.

940

:

And, um, I had to get to the

point where I could trust Nat

941

:

Randall Stevens: Was there any, I

don't think I've asked you, Brent.

942

:

Was there any challenges with the

university as far as intellectual property

943

:

or, you know, what had been done under

the university, uh, umbrella versus

944

:

opening that up then for the world?

945

:

Brent Seales: Well, you know,

I, I got pretty good advice.

946

:

That was to just keep everything outside

the university because that would've

947

:

added an extra layer of difficulty

for me to have to cut through.

948

:

And I said before that the long

timeline on some of this work had to

949

:

do with more than just the technical.

950

:

That's an example where I bring this

back to the university, it goes to

951

:

legal, then they decide, do we want to

experience any kind of risk, uh, for this?

952

:

And then they go back and forth

and it's out of my control and

953

:

I don't know what's gonna happen

and it takes time and so forth.

954

:

Right.

955

:

So rather than doing that, and rather

than having to sort of start up my own

956

:

company and sever my relationship, we

just went ahead and kicked it to Nat.

957

:

He ran with it.

958

:

And, um.

959

:

He gave me an affiliation as

co-founder, and, and, and that's worked

960

:

great.

961

:

And the fact that we, we have a,

a strong trust in each other that,

962

:

you know, we're not gonna go off

and, and do things that are silly

963

:

or unexpected, you know, it works.

964

:

Randall Stevens: Yeah, so maybe, uh,

maybe we can turn to, earlier this

965

:

year it was announced that you are

now going to lead an AI and Humanities

966

:

Institute, I think it's officially

called, that is, uh, funded by Eric

967

:

Schmidt, who was the former CEO of

Google and his wife have a foundation.

968

:

Maybe you can tell us more about that.

969

:

Brent Seales: yeah.

970

:

Eric and Wendy Schmidt have,

um, a philanthropic foundation

971

:

called, uh, Schmidt Sciences.

972

:

And through that mechanism, they, um,

they very, very generously have set up an

973

:

endowment that allows, uh, funding across

a lot of different science endeavors.

974

:

Uh.

975

:

And one big institute that

they've set up is the AI and

976

:

Advanced Computation Institute.

977

:

In, in, um, around which, uh, the

program I'm going to run is, is fit.

978

:

So, uh, they, they are very interested,

both Eric and Wendy in how, you know, AI

979

:

is going to be shaping, uh, our future.

980

:

And so one of their signature programs

is called AI:

981

:

envision what AI will do to change the

orld between now and the year:

982

:

And, um, you can think about it as,

uh, a similar program to the, um,

983

:

MacArthur Fellow program where they,

uh, they fund young scientists who

984

:

are making a huge impact so that those

folks with very few strings attached

985

:

can use those resources to, um.

986

:

Move the needle to change the world.

987

:

Uh, I think it's an incredible program.

988

:

The humanities and AI Virtual Institute,

we call it Javi, um, Javi, which,

989

:

uh, is going to fund projects at the

confluence of, um, I keep using that

990

:

word, Randall, for your benefit

991

:

Randall Stevens: It's a good word.

992

:

Brent Seales: at the intersection right.

993

:

Of, um, humanities and ai.

994

:

Um, is I think the future.

995

:

I think moving, uh, humanities

scholars, um, more toward the

996

:

center of the magic of ai, right?

997

:

So that we can really understand how

deeply human these things are, are

998

:

becoming, and, and how we can leverage

the way they're working, understand

999

:

better how to make them better.

:

00:56:46,773 --> 00:56:48,603

Uh, I think that's, that's where it's at.

:

00:56:49,203 --> 00:56:54,693

And so there's gonna be a lot of work,

probably a lot of, um, interest in my

:

00:56:54,693 --> 00:57:01,323

portfolio and, you know, things like large

language models, multimodal ai, um, and

:

00:57:01,383 --> 00:57:04,053

new AI techniques that try to, to grasp.

:

00:57:04,548 --> 00:57:10,938

Significant humanities collections

and um, uh, collections that, that are

:

00:57:10,938 --> 00:57:16,098

unusual that you're not gonna see from

a sensor, for example, but, uh, that are

:

00:57:16,098 --> 00:57:19,848

still really interesting and, um, can,

can sort of push the technology forward?

:

00:57:20,850 --> 00:57:22,830

Randall Stevens: It's pretty

incredible and you're gonna be

:

00:57:22,830 --> 00:57:26,940

able to keep your position at the

university while you're simultaneously

:

00:57:27,120 --> 00:57:27,925

helping with this institute.

:

00:57:28,460 --> 00:57:33,440

Brent Seales: yes, uh, Schmidt has,

Schmidt Sciences has, um, created a role

:

00:57:33,440 --> 00:57:38,420

for me through the University of Kentucky

where, uh, in, in a way that's similar

:

00:57:38,420 --> 00:57:42,890

to working for the National Science

Foundation, um, I can act as a program

:

00:57:42,890 --> 00:57:49,160

director for, um, a percentage of my time

and, um, still maintain my affiliation

:

00:57:49,160 --> 00:57:50,150

with the University of Kentucky.

:

00:57:50,150 --> 00:57:53,570

So I'm actually really grateful to

Schmidt Sciences for arranging it

:

00:57:53,570 --> 00:57:57,680

that way, um, rather than me having to

leave my post because I, I love being a

:

00:57:57,680 --> 00:57:59,180

professor at the University of Kentucky.

:

00:57:59,480 --> 00:58:02,300

And I'm also grateful to, you know,

my dean and the leadership here

:

00:58:02,300 --> 00:58:04,100

for allowing me to accept that.

:

00:58:04,610 --> 00:58:07,310

Um, because they, you know,

they, they don't have to, right.

:

00:58:07,310 --> 00:58:09,620

But, um, I think it's good for all of us.

:

00:58:10,490 --> 00:58:13,340

Randall Stevens: No, it's definitely,

it's great for the university, uh, you

:

00:58:13,340 --> 00:58:18,110

know, uh, on behalf of the university and

Lexington and Kentucky in general, we're

:

00:58:18,110 --> 00:58:22,400

all very proud that, uh, of the work that

you've done over all these years and that

:

00:58:22,400 --> 00:58:24,320

you've been able to, uh, you know, I.

:

00:58:25,475 --> 00:58:28,235

Stay, you know, stay here at the

university and do this kind of

:

00:58:28,235 --> 00:58:31,295

work, which is obviously now,

you know, it's world renowned.

:

00:58:31,415 --> 00:58:34,505

Um, so I'm, I'm proud of you.

:

00:58:34,565 --> 00:58:35,945

I'm, I'm, I'm glad.

:

00:58:35,945 --> 00:58:39,875

And, and thanks for coming on the

podcast and sharing some of the story.

:

00:58:40,145 --> 00:58:43,715

Uh, Evan, we're gonna have to

figure out how interspersed in here.

:

00:58:43,715 --> 00:58:47,345

We're gonna have to at least show maybe

a few little clips or something so people

:

00:58:47,345 --> 00:58:48,935

can understand what we're talking about.

:

00:58:49,145 --> 00:58:52,415

They've got a lot of media that, uh,

that they produced and is out there.

:

00:58:52,415 --> 00:58:55,445

And then in the, in the show

notes and stuff, we'll put links

:

00:58:55,445 --> 00:58:58,625

out to some of these videos

and, uh, yeah, it's pretty cool.

:

00:58:58,745 --> 00:59:02,195

And aren't they, didn't you tell me

that there, there's like a documentary

:

00:59:02,200 --> 00:59:05,285

crew or something kind of following

you around or have, have been

:

00:59:05,690 --> 00:59:07,460

Brent Seales: There is, there's

gonna be a feature film.

:

00:59:08,495 --> 00:59:10,775

Um, they're almost wrapping up production.

:

00:59:10,775 --> 00:59:11,885

They'll go to post-production.

:

00:59:11,915 --> 00:59:17,255

Um, the target, I believe is going

to be film festivals in early:

:

00:59:18,005 --> 00:59:21,395

and it'll be a full feature

around, uh, this work.

:

00:59:21,995 --> 00:59:27,485

My, the members of my team and the related

members who have all, you know, helped in

:

00:59:27,485 --> 00:59:29,345

a collaborative way to, to break through.

:

00:59:30,185 --> 00:59:32,015

I think it'll be really, really fun.

:

00:59:32,015 --> 00:59:36,905

Executive produced by Scott Free, which

is Ridley Scott's documentary film

:

00:59:37,100 --> 00:59:37,550

Randall Stevens: Okay.

:

00:59:38,480 --> 00:59:39,080

Very cool.

:

00:59:39,530 --> 00:59:43,130

Yeah, I, I started out by saying it's

one of the, I think it's like, it's

:

00:59:43,130 --> 00:59:47,600

hard to think of something that's like

a sexier, uh, kind of cool combination

:

00:59:47,600 --> 00:59:49,610

of, and I, and I think you nailed it.

:

00:59:49,610 --> 00:59:53,390

It is that it is the human part of it

that really makes it so interesting.

:

00:59:53,390 --> 00:59:57,590

And it's this, now all of a sudden this

technology's come to bear, uh, you know,

:

00:59:58,370 --> 01:00:03,500

in this way to, uh, reveal and kind

of open up all this new possibility.

:

01:00:03,500 --> 01:00:05,450

So, incredibly cool.

:

01:00:05,450 --> 01:00:10,160

I know you've been, uh, sharing with

me some other projects and stuff, uh,

:

01:00:10,160 --> 01:00:13,340

of, of, there's just gonna be a lot of

applications for this, and I think it's

:

01:00:13,340 --> 01:00:15,290

gonna open up a whole new world of, uh.

:

01:00:15,650 --> 01:00:16,520

Of possibility.

:

01:00:16,520 --> 01:00:17,960

So congrats on that front,

:

01:00:18,780 --> 01:00:19,210

Brent Seales: Thank you.

:

01:00:20,600 --> 01:00:20,930

Randall Stevens: Evan.

:

01:00:20,930 --> 01:00:22,100

Any, uh, any last

:

01:00:22,375 --> 01:00:23,155

Evan Troxel: No, I'm good.

:

01:00:23,155 --> 01:00:26,665

I, this was a fascinating conversation

and it, it builds on what Christie

:

01:00:26,665 --> 01:00:30,475

shared the last time, and I feel

like, uh, it's just an incredible

:

01:00:30,475 --> 01:00:32,455

story and it's still unfolding, right?

:

01:00:32,455 --> 01:00:34,196

Like this isn't over and,

:

01:00:34,285 --> 01:00:34,615

and,

:

01:00:34,775 --> 01:00:35,785

Randall Stevens: literally unfold?

:

01:00:35,785 --> 01:00:38,395

Evan Troxel: cool moment in time where

it's like, we can share this with a

:

01:00:38,395 --> 01:00:44,395

wire audience and they understand what's

going on because the technology has

:

01:00:44,395 --> 01:00:48,445

been democratized to a certain extent

where, okay, like, oh, I've heard of

:

01:00:48,445 --> 01:00:51,865

this, I've used this, and, and maybe not

the exact tools you're talking about.

:

01:00:51,865 --> 01:00:55,495

Obviously you're talking about

super high tech eight micron, you

:

01:00:55,495 --> 01:00:57,115

know, level stuff that you're doing.

:

01:00:57,115 --> 01:01:00,235

But at the same time, it's like,

oh, the, the Bri the, the bridge has

:

01:01:00,235 --> 01:01:03,685

already been built of understanding of

what's going on here, and I feel like,

:

01:01:03,685 --> 01:01:08,155

uh, what a great time to be able to

present it and get the story out there

:

01:01:08,485 --> 01:01:12,115

because there are so many people who,

who would love to hear about this and.

:

01:01:12,635 --> 01:01:16,505

Um, be way beyond this podcast,

obviously, 60 Minutes and all, all

:

01:01:16,505 --> 01:01:18,680

those other second rate channels

:

01:01:18,695 --> 01:01:20,645

Randall Stevens: This is, this

is gonna get more views in

:

01:01:20,645 --> 01:01:21,965

your 60 minute story, Brent.

:

01:01:22,230 --> 01:01:22,800

Brent Seales: I hope so.

:

01:01:23,530 --> 01:01:23,970

hope so.

:

01:01:24,330 --> 01:01:26,970

You know, somebody asked me how long

will it be before one of those books

:

01:01:26,970 --> 01:01:30,960

is actually on the bookshelf, you know,

edited and, and I, I don't know, but I,

:

01:01:31,260 --> 01:01:35,010

I know it'll be a lot shorter than if we

hadn't done the VASIs challenge, right?

:

01:01:35,700 --> 01:01:38,880

Because if I had just hacked away

on my own, and then, you know,

:

01:01:38,910 --> 01:01:43,470

one ologist takes what we did

and they hack away on their own.

:

01:01:43,500 --> 01:01:43,920

Right?

:

01:01:44,490 --> 01:01:48,540

But instead, we've built a, a community

and an expectation that we can go faster.

:

01:01:48,540 --> 01:01:49,350

And I, and I love that,

:

01:01:49,440 --> 01:01:49,680

you know.

:

01:01:50,555 --> 01:01:53,045

Randall Stevens: I think, uh, I was,

uh, kinda giving you a hard time

:

01:01:53,045 --> 01:01:55,805

about, um, you know, but I am serious.

:

01:01:55,805 --> 01:02:00,815

I'd love to get you, uh, this next

October for our, uh, confluence of here

:

01:02:00,815 --> 01:02:05,075

in Lexington, uh, to get, to spend a

little bit of time with this group.

:

01:02:05,075 --> 01:02:07,895

I think when you get a kick out

of some of the people that come in

:

01:02:07,895 --> 01:02:11,135

for this, I know they'd love, uh,

having this conversation with you.

:

01:02:11,555 --> 01:02:16,895

Um, and, um, yeah, I mean that,

that's to, to me, that's what

:

01:02:16,955 --> 01:02:18,335

doing these events is all about.

:

01:02:18,335 --> 01:02:22,655

You kind of get this cross pollinization

of ideas and fresh, fresh thoughts.

:

01:02:22,685 --> 01:02:23,375

Like you said, you.

:

01:02:23,900 --> 01:02:26,870

This wouldn't have come about if you

didn't take the opportunity to go

:

01:02:26,870 --> 01:02:30,920

out to Net's, uh, weekend and, you

know, engage in some new way that,

:

01:02:30,920 --> 01:02:32,330

uh, that you hadn't in the past.

:

01:02:32,330 --> 01:02:40,160

But, um, we'll, uh, we'll keep everybody

up to date and, uh, and, and I, oh,

:

01:02:40,160 --> 01:02:41,300

I know what I was gonna say, Brent.

:

01:02:41,360 --> 01:02:45,680

Uh, I think one of the reasons

that you weren't able to, uh, like

:

01:02:45,680 --> 01:02:50,810

participate last year 'cause you bring

people to, to Lexington too, from,

:

01:02:50,870 --> 01:02:52,670

from that world and that community.

:

01:02:52,670 --> 01:02:56,300

And sometimes we end up kind of

overlapping and, uh, just kind of,

:

01:02:56,480 --> 01:02:57,920

uh, not being able to work it out.

:

01:02:57,920 --> 01:03:00,290

But hopefully this year

we'll be able to have you.

:

01:03:00,325 --> 01:03:01,615

Brent Seales: Well, I

heard that last year.

:

01:03:01,615 --> 01:03:04,795

You know, the, the party

was, the party was amazing.

:

01:03:05,785 --> 01:03:07,945

You know, I don't know what had me

outta town, but the, you know, the

:

01:03:07,945 --> 01:03:10,375

year before that, Randall, it was first

:

01:03:10,375 --> 01:03:10,945

letters.

:

01:03:11,180 --> 01:03:11,390

Randall Stevens: Yeah.

:

01:03:11,965 --> 01:03:16,165

Brent Seales: We had all of the Italian

ologists in town, and the Friday of your

:

01:03:16,165 --> 01:03:20,155

Confluence event was the Friday we had

a press conference to announce purple.

:

01:03:20,450 --> 01:03:20,660

Randall Stevens: Yep.

:

01:03:21,040 --> 01:03:21,460

Evan Troxel: Thanks.

:

01:03:21,505 --> 01:03:22,555

Brent Seales: so I had a good reason for

:

01:03:22,555 --> 01:03:23,185

missing, you know,

:

01:03:23,255 --> 01:03:24,065

Randall Stevens: yeah, it was very cool.

:

01:03:24,395 --> 01:03:24,785

No, no.

:

01:03:25,685 --> 01:03:26,585

Uh, well, cool.

:

01:03:26,645 --> 01:03:30,455

Well, thanks for, for giving us

this time coming on the podcast.

:

01:03:30,485 --> 01:03:32,015

I've been looking forward to this.

:

01:03:32,105 --> 01:03:35,135

I think this is gonna end up being

the final episode of our season.

:

01:03:35,495 --> 01:03:39,005

Uh, we kind of break these up into

seasons, but, uh, really appreciate

:

01:03:39,005 --> 01:03:42,935

you coming on and sharing this and,

uh, we'll get plenty of, uh, links and

:

01:03:42,935 --> 01:03:46,445

fodder for people to go dig in as deep

as they want to go on this project.

:

01:03:46,475 --> 01:03:46,715

So,

:

01:03:47,465 --> 01:03:48,065

Brent Seales: Well, that's cool.

:

01:03:48,065 --> 01:03:51,695

I mean, maybe you can have me back next

season because we're gonna have a big

:

01:03:52,085 --> 01:03:54,425

scanning session in a, in a few weeks

:

01:03:54,905 --> 01:03:58,415

Randall Stevens: yeah, I, you know,

I, I, I had sent you when I was out

:

01:03:58,415 --> 01:04:02,105

at Autodesk University, you know,

there's some friends of mine, Paul Bin.

:

01:04:03,140 --> 01:04:07,820

And, uh, Rob Sinclair, who we've had on

the episode, I sent you some pics because

:

01:04:07,820 --> 01:04:11,720

they've been involved in some scanning

projects that they volunteer, they go over

:

01:04:11,720 --> 01:04:16,940

to Italy and have been scanning, you know,

using some, uh, scanning technologies to

:

01:04:16,940 --> 01:04:18,440

scan some of the antiquities over there.

:

01:04:18,440 --> 01:04:21,890

So I for sure, there's some people

that come in, uh, for our Confluence

:

01:04:21,890 --> 01:04:25,160

event that you'd love to probably have

some conversations with about this.

:

01:04:25,985 --> 01:04:26,225

Brent Seales: Absolutely.

:

01:04:26,255 --> 01:04:26,545

Yeah.

:

01:04:27,230 --> 01:04:27,620

Randall Stevens: Great.

:

01:04:27,710 --> 01:04:29,180

Alright, thanks for joining us, Brent.

:

01:04:29,425 --> 01:04:29,725

Brent Seales: All right.

:

01:04:29,725 --> 01:04:30,245

Thanks so much

All Episodes Previous Episode

Listen for free

Show artwork for Confluence

About the Podcast

Confluence
The director's commentary track for AEC industry software development.
The Confluence podcast is the director's commentary track for AEC industry software. Go behind the scenes with us to learn how and why decisions were made in the creation of your favorite software for the architecture, engineering, and construction industries.

It's a collaboration between Randall Stevens of AVAIL and Evan Troxel of TRXL.

About your hosts

Evan Troxel

Profile picture for Evan Troxel
An industry-leading design and technology expert with a passion for connecting people, Evan is a licensed architect in California and is most well known for his podcasts that focus on the AEC industry.

He has over 25 years of experience in the practice and technology in the architectural profession working with large teams to deliver large public projects for clients. He now brings his experiences together on the Archispeak and TRXL podcasts, and now on the Confluence podcast.

Randall Stevens

Profile picture for Randall Stevens
An AEC industry veteran with 25 years of software development, and sales and management experience, Randall offers a unique combination of expertise in software and graphics technology — coupled with a background and degree in architecture.

In 1991 he founded ArchVision, a software firm specializing in 3D graphics, specifically Rich Photorealistic Content (RPC). Through ArchVision, Randall has built an extensive network with the industry’s leading experts, architectural firms, and visualization software companies, which led him to product development of the AVAIL platform.