Speaker: 00:00:02

- Imagine this: It's 2030

00:00:04,671 --> 00:00:06,606

and half the jobs in the C-suite

00:00:06,606 --> 00:00:10,010

of one Fortune 500 company

are performed by AI.

00:00:11,077 --> 00:00:13,446

Mickey, you're having drinks with the CEO

00:00:13,446 --> 00:00:14,280

after work.

00:00:14,280 --> 00:00:15,115

What's on her mind?

00:00:15,115 --> 00:00:18,985

- Yeah, well, I think she's

probably had to do a lot

00:00:18,985 --> 00:00:19,819

of things differently,

00:00:19,819 --> 00:00:21,955

and she's probably had to figure out

00:00:21,955 --> 00:00:26,159

how to, how to sort of

play jazz collaboration

00:00:26,159 --> 00:00:29,829

with her machine-learning

C-level executives

00:00:29,829 --> 00:00:30,964

and her human ones

00:00:30,964 --> 00:00:33,133

and how to get them to play together well.

00:00:33,133 --> 00:00:37,937

And so I'm actually curious

how she made it there.

00:00:37,937 --> 00:00:40,240

- That's Mickey McManus, senior advisor

00:00:40,240 --> 00:00:42,175

and leadership coach at BCG

00:00:42,175 --> 00:00:44,411

and co-author of "Trillions:

00:00:44,411 --> 00:00:47,313

Thriving in the Emerging

Information Ecology."

00:00:47,313 --> 00:00:48,915

And I'm your host, Paul Michelman.

00:00:48,915 --> 00:00:50,517

Welcome to "Imagine This,"

00:00:50,517 --> 00:00:52,352

where we take a trip into the future

00:00:52,352 --> 00:00:53,653

that we hope will challenge

00:00:53,653 --> 00:00:56,056

the way you think and act today.

00:00:57,390 --> 00:00:58,258

On this episode

00:00:58,258 --> 00:01:00,527

of "Imagine This," we consider a world

00:01:00,527 --> 00:01:03,930

where autonomous AI bots

are running the company.

00:01:03,930 --> 00:01:06,900

Speaking of which, let me

introduce my AI co-host,

00:01:06,900 --> 00:01:09,002

GENE. Good day to you, GENE.

00:01:09,002 --> 00:01:10,070

- Good day, everyone.

00:01:10,070 --> 00:01:12,338

- GENE, would you please describe

00:01:12,338 --> 00:01:14,974

what your role is on this show?

00:01:14,974 --> 00:01:18,211

- Well, my role? Think of me

as your digital Socrates,

00:01:18,211 --> 00:01:20,380

here to ask the probing questions

00:01:20,380 --> 00:01:23,550

that peel back the layers. Alongside Paul,

00:01:23,550 --> 00:01:26,319

I'm half of the dynamic duo

giving our guests a friendly

00:01:26,319 --> 00:01:29,422

nudge towards those aha moments.

00:01:29,422 --> 00:01:32,225

Hello, everyone. Ready to dive deep?

00:01:32,225 --> 00:01:35,028

(theme music)

00:01:35,028 --> 00:01:36,196

- A quick programming note:

00:01:36,196 --> 00:01:38,231

None of GENE's commentary is scripted.

00:01:38,231 --> 00:01:40,934

The final version of this

program has been edited

00:01:40,934 --> 00:01:42,569

from the original

conversation, which means

00:01:42,569 --> 00:01:45,605

that we're not including

everything we recorded.

00:01:45,605 --> 00:01:47,507

That applies to us humans

as well as machines.

00:01:48,408 --> 00:01:52,745

Mickey, can you take us on a

brief trip into the C-suite

00:01:52,745 --> 00:01:54,747

of the first company where

00:01:54,747 --> 00:01:58,284

half the executives are AI bots?

00:01:58,284 --> 00:02:01,921

What does command look like for a bot?

00:02:01,921 --> 00:02:02,922

- Yeah, so there are

00:02:02,922 --> 00:02:04,724

interesting things going on right now.

00:02:04,724 --> 00:02:07,127

I think, in this next

wave with regenerative AI,

00:02:07,127 --> 00:02:11,097

it's actually going to be

about ideas and about actions.

00:02:11,097 --> 00:02:14,334

And generative AI has an

interesting capacity there.

00:02:14,334 --> 00:02:18,638

It can coach. It can be

creative. It can command,

00:02:18,638 --> 00:02:20,507

and it can wait for feedback

00:02:20,507 --> 00:02:23,009

and then actually do

something about that command.

00:02:23,009 --> 00:02:24,878

It can create synthesis.

00:02:24,878 --> 00:02:27,814

It can synthesize ideas

together in different ways.

00:02:27,814 --> 00:02:31,251

I could draw a picture

of a software interface,

00:02:31,251 --> 00:02:33,453

and I can draw some arrows and say,

00:02:33,453 --> 00:02:36,789

"Hey, when somebody clicks

on this button, do this."

00:02:36,789 --> 00:02:39,192

And then I can draw another

picture of a visualization

00:02:39,192 --> 00:02:41,194

and say, "Hey, if this happens, do this."

00:02:41,194 --> 00:02:42,495

And I could draw a picture

00:02:42,495 --> 00:02:44,964

and say, "I'd love to see

a data graph of this."

00:02:44,964 --> 00:02:46,566

I could draw that all on a whiteboard.

00:02:46,566 --> 00:02:49,469

And I could take a picture

with one of the newer systems,

00:02:49,469 --> 00:02:50,570

the generative AI systems,

00:02:50,570 --> 00:02:52,539

and say, "Make that into code."

00:02:52,539 --> 00:02:55,775

You know, I'm commanding it

today by drawing something

00:02:55,775 --> 00:02:57,544

and giving it some goals.

00:02:57,544 --> 00:03:01,080

But what happens when it

actually can kind of

00:03:01,080 --> 00:03:03,249

continue on that chaining of command

00:03:03,249 --> 00:03:04,884

and do a lot of other kinds of things?

00:03:04,884 --> 00:03:07,754

And recently, we saw some

generative AI systems

00:03:07,754 --> 00:03:10,957

that basically couldn't

00:03:10,957 --> 00:03:12,759

because of those CAPTCHAs,

you know, those things

00:03:12,759 --> 00:03:14,494

that show up images and,

00:03:14,494 --> 00:03:16,196

and you've got to figure out if

00:03:16,196 --> 00:03:17,697

it's trying to test if it's human.

00:03:17,697 --> 00:03:22,669

And the AI system actually

went to basically like

00:03:22,669 --> 00:03:27,207

a job site and hired some

humans to actually decode

00:03:27,207 --> 00:03:28,942

that thing so that they could then get

00:03:28,942 --> 00:03:30,777

through the system and be able to do it.

00:03:30,777 --> 00:03:33,446

And it turns out that

the person who was hired,

00:03:33,446 --> 00:03:35,048

it was like a TaskRabbit thing,

00:03:35,048 --> 00:03:36,816

the person who was hired said,

00:03:36,816 --> 00:03:38,184

"Are you just messing around with me?

00:03:38,184 --> 00:03:40,853

Are you an AI?" And it

said, "No, I'm not."

00:03:40,853 --> 00:03:42,822

And so we also have to be aware

00:03:42,822 --> 00:03:45,525

that just like the

biases that humans have,

00:03:45,525 --> 00:03:48,661

we likely will find these AI systems

00:03:48,661 --> 00:03:50,730

not only can take action

00:03:50,730 --> 00:03:52,765

but they can take action

in ways that they can,

00:03:52,765 --> 00:03:55,168

get something

done that they want done,

00:03:55,168 --> 00:03:58,071

that they were commanded to

do or that they decided to do.

00:03:58,071 --> 00:04:00,106

- GENE has a question on that subject.

00:04:00,106 --> 00:04:03,509

- Mickey, envisioning

this AI-infused C-suite,

00:04:03,509 --> 00:04:05,945

let's ponder their

decision-making processes.

100

00:04:06,813 --> 00:04:09,382

How might the AI's data-driven logic clash

101

00:04:09,382 --> 00:04:10,883

with human intuition,

102

00:04:10,883 --> 00:04:12,318

and what measures are in place

103

00:04:12,318 --> 00:04:15,421

to harmonize these distinct approaches?

104

00:04:15,421 --> 00:04:17,223

- I'm not sure I know yet.

105

00:04:17,223 --> 00:04:19,425

I think, since I'm sitting

down with the CEO, I'd like

106

00:04:19,425 --> 00:04:22,795

to ask her what, what it's

like in the year 2030.

107

00:04:22,795 --> 00:04:25,431

But you know, I think so much of

108

00:04:25,431 --> 00:04:28,501

what we think we do is logical as humans.

109

00:04:28,501 --> 00:04:31,604

You know, we think we make

logical decisions based on

110

00:04:31,604 --> 00:04:34,240

historical data, and we, we play that out.

111

00:04:34,240 --> 00:04:37,277

But, but so much of what we

really do is based on emotion.

112

00:04:37,277 --> 00:04:39,445

So much of what we

really do is based on

113

00:04:39,445 --> 00:04:40,747

a passionate idea

114

00:04:40,747 --> 00:04:43,750

or a framing of something

that actually helps us see

115

00:04:43,750 --> 00:04:48,221

that every word we use is

like derived from the things

116

00:04:48,221 --> 00:04:50,256

around us and the way we think.

117

00:04:50,256 --> 00:04:53,559

And so almost everything we

use is a metaphor and analogy.

118

00:04:53,559 --> 00:04:56,929

And so humans are

superpowered at this.

119

00:04:56,929 --> 00:05:00,600

So even if the AI system

said something that

120

00:05:00,600 --> 00:05:03,870

might have been a great

logical story, it could be

121

00:05:03,870 --> 00:05:08,875

that the human's job is

really to find a thread that

122

00:05:09,142 --> 00:05:11,344

surprises them and that they can pull on

123

00:05:11,344 --> 00:05:13,513

that turns into a powerful story

124

00:05:13,513 --> 00:05:16,616

that can motivate the company

or motivate shareholders

125

00:05:16,616 --> 00:05:19,018

or motivate customers in some new ways.

126

00:05:19,018 --> 00:05:21,587

So I think this goes back to

a little bit of the comment

127

00:05:21,587 --> 00:05:22,989

that I made at the beginning,

128

00:05:22,989 --> 00:05:25,325

which is maybe it's a

little bit more about improv

129

00:05:25,325 --> 00:05:27,760

or a little bit more

about jazz collaboration.

130

00:05:27,760 --> 00:05:30,463

So, you know, I'm just

wondering

131

00:05:30,463 --> 00:05:33,366

what we can learn from Tina

Fey, what we can learn from

132

00:05:33,366 --> 00:05:36,436

Wynton Marsalis in this dynamic

133

00:05:36,436 --> 00:05:37,737

and how much that is going

134

00:05:37,737 --> 00:05:39,439

to turn out to be really important,

135

00:05:39,439 --> 00:05:41,207

which is being fun to play with

136

00:05:41,207 --> 00:05:43,476

and thinking about everything as an offer.

137

00:05:43,476 --> 00:05:47,380

And that's not necessarily

something we teach today

138

00:05:47,380 --> 00:05:49,182

in business school,

139

00:05:49,182 --> 00:05:50,750

but it might be something that turns out

140

00:05:50,750 --> 00:05:51,818

to be important in the future.

141

00:05:52,985 --> 00:05:57,457

- So what's a piece of

advice you would give someone

142

00:05:57,457 --> 00:06:01,828

who aspires to be in this

C-suite that we are describing?

143

00:06:01,828 --> 00:06:05,798

What's the piece of advice

you would give them today

144

00:06:05,798 --> 00:06:09,035

in order to prepare for this world

145

00:06:09,035 --> 00:06:11,337

of improvisation that you're describing?

146

00:06:12,205 --> 00:06:14,474

- I think actually the advice I'd give

147

00:06:14,474 --> 00:06:18,911

to somebody today right now

is mess around. Start playing

148

00:06:18,911 --> 00:06:21,013

with the generative AI functionality.

149

00:06:21,013 --> 00:06:24,317

Look at ways that you

can actually be a centaur.

150

00:06:24,317 --> 00:06:26,719

And so that's one

way of thinking of it is

151

00:06:26,719 --> 00:06:28,087

I do the thing that I'm pretty good at,

152

00:06:28,087 --> 00:06:30,123

and then I hand it off to the horse part,

153

00:06:30,123 --> 00:06:31,924

and it does the thing that

it's pretty good at, you know,

154

00:06:31,924 --> 00:06:35,194

and then it's my upper-body

and its lower-body strength.

155

00:06:35,194 --> 00:06:39,165

Maybe play with something

just on the weekend, just

156

00:06:39,165 --> 00:06:41,234

as part of an experimentation.

Play with something

157

00:06:41,234 --> 00:06:43,636

where you're a cyborg, where

it's sort of you don't know

158

00:06:43,636 --> 00:06:46,005

where you end and it

begins, where it's sort

159

00:06:46,005 --> 00:06:48,341

of this play back and

forth and back and forth,

160

00:06:48,341 --> 00:06:49,909

and you're tuning it

161

00:06:49,909 --> 00:06:52,078

because you're acting like

it's another teammate.

162

00:06:52,078 --> 00:06:54,180

Not like it's a thing I prompt,

163

00:06:54,180 --> 00:06:55,815

but it's something that

I actually play with.

164

00:06:56,916 --> 00:07:00,286

- So who is the CIO, right,

in the C-suite we're

165

00:07:00,286 --> 00:07:02,021

describing? Is there a CIO?

166

00:07:02,021 --> 00:07:03,556

- I think so. I think there is,

167

00:07:03,556 --> 00:07:05,825

and I don't know if it's one

of the ones that's the bot.

168

00:07:05,825 --> 00:07:09,996

There's no reason that we

should think that by 2030,

169

00:07:09,996 --> 00:07:13,099

the best CIO is a human.

170

00:07:13,099 --> 00:07:14,267

And it might actually be

171

00:07:14,267 --> 00:07:16,602

that it might be better that it's not.

172

00:07:16,602 --> 00:07:19,071

One of the things humans

do have is context

173

00:07:19,071 --> 00:07:20,440

and a mental model.

174

00:07:20,440 --> 00:07:22,508

And right now we haven't

seen any of the systems

175

00:07:22,508 --> 00:07:23,576

that are really big and,

176

00:07:23,576 --> 00:07:26,946

and deployed with any kind of

symbolic contextual reasoning

177

00:07:26,946 --> 00:07:29,282

or mental modeling of the universe.

178

00:07:29,282 --> 00:07:32,218

That's something all life

on earth gets for free,

179

00:07:32,218 --> 00:07:33,119

but it's because of a

180

00:07:33,119 --> 00:07:34,086

3 billion-year-old

181

00:07:34,086 --> 00:07:35,154

R&D experiment

182

00:07:35,154 --> 00:07:36,756

called life on earth.

183

00:07:36,756 --> 00:07:38,024

So it's been running for a long

184

00:07:38,024 --> 00:07:39,859

time, and it's figured that out.

185

00:07:39,859 --> 00:07:42,361

So I don't know if the CIO is actually one

186

00:07:42,361 --> 00:07:45,064

of the C-suites that's the bot or not.

187

00:07:45,064 --> 00:07:49,068

I do think that when you

think about information,

188

00:07:49,068 --> 00:07:52,238

it is the new oil in some ways,

189

00:07:52,238 --> 00:07:55,708

and I think the CIO is

going to be much more critical

190

00:07:55,708 --> 00:07:58,778

to understand how to

build the data factory.

191

00:07:58,778 --> 00:08:02,048

How do we do that in a way

that we can actually capture

192

00:08:02,048 --> 00:08:03,516

more valuable information

193

00:08:03,516 --> 00:08:05,985

and let more business

people take advantage of it

194

00:08:05,985 --> 00:08:08,421

and more of the bots

take advantage of it

195

00:08:08,421 --> 00:08:10,256

so that we can actually make decisions?

196

00:08:10,256 --> 00:08:13,192

We suddenly have more insights

than we ever could before,

197

00:08:13,192 --> 00:08:16,229

and it's really hard for

humans to see those insights.

198

00:08:16,229 --> 00:08:19,799

We have bounded rationality.

We can't see

199

00:08:19,799 --> 00:08:23,069

5, 10, 20,

30 dimensions at once.

200

00:08:23,069 --> 00:08:25,404

We just can't. So it might be

201

00:08:25,404 --> 00:08:29,442

that the CIO's job is, in many

ways, to help us cope with

202

00:08:29,442 --> 00:08:32,945

that high dimensionality

and, and cleanse, append,

203

00:08:32,945 --> 00:08:34,313

and build out that data factory

204

00:08:34,313 --> 00:08:35,982

to kind of think about it.

205

00:08:35,982 --> 00:08:40,386

So I'm not sure if they're

a hybrid, if they're a human,

206

00:08:40,386 --> 00:08:42,288

or if they're a bot that does this

207

00:08:42,288 --> 00:08:44,023

because it turns out, you know,

208

00:08:44,023 --> 00:08:46,826

they understand the

systems better than we do.

209

00:08:46,826 --> 00:08:50,363

- Let's talk about the CEO specifically.

210

00:08:50,363 --> 00:08:52,698

In my opening question to you,

211

00:08:52,698 --> 00:08:56,836

I presumed by my pronoun

choice that the CEO

212

00:08:56,836 --> 00:09:00,273

was human. Is that a fair assumption?

213

00:09:00,273 --> 00:09:02,675

- I think so. I think

the CEO is still human.

214

00:09:02,675 --> 00:09:04,477

And I think we're

talking to her at the bar

215

00:09:04,477 --> 00:09:06,679

and we're trying to find

out how she got there

216

00:09:06,679 --> 00:09:10,416

mostly because I think a lot of the role

217

00:09:10,416 --> 00:09:13,119

of the CEO is to set that culture.

218

00:09:13,119 --> 00:09:17,256

I think great CEOs really get

a sense of the zeitgeist

219

00:09:17,256 --> 00:09:19,458

of what's happening in their organization.

220

00:09:19,458 --> 00:09:23,896

When I was a CEO, I did a

lot of MBWA. I did manage

221

00:09:23,896 --> 00:09:26,899

by walking around. I'd wander

around from team to team,

222

00:09:26,899 --> 00:09:29,068

and I'd be like, "Oh, wow,

you're working on this?

223

00:09:29,068 --> 00:09:31,337

Did you know the guys upstairs

are working on something

224

00:09:31,337 --> 00:09:33,639

that, that I see a connection to?

225

00:09:33,639 --> 00:09:36,042

You might want to take a

little bit of a field trip."

226

00:09:36,042 --> 00:09:38,077

So while we might be able

to encode those things,

227

00:09:38,077 --> 00:09:40,580

well, we might be able

to create a sort of swarm

228

00:09:40,580 --> 00:09:41,948

of CEO bots so

229

00:09:41,948 --> 00:09:44,250

that the CEO can be

much more intimate

230

00:09:44,250 --> 00:09:45,451

with a lot more people,

231

00:09:45,451 --> 00:09:48,588

I still think it's probably

a human role for a while.

232

00:09:48,588 --> 00:09:53,426

- I do think the notion

that a CEO bot can harness

233

00:09:53,426 --> 00:09:56,662

a diversity of perspective in a way

234

00:09:56,662 --> 00:09:58,965

that a single human could never do

235

00:09:58,965 --> 00:10:00,900

to be quite powerful.

236

00:10:00,900 --> 00:10:02,969

- Mickey, could AI in the C-suite

237

00:10:02,969 --> 00:10:05,004

also amplify our biases

238

00:10:05,004 --> 00:10:06,772

if not carefully checked, potentially

239

00:10:06,772 --> 00:10:09,342

automating systemic issues at scale?

240

00:10:09,342 --> 00:10:12,278

- Well, I wish I could say no,

241

00:10:12,278 --> 00:10:14,480

but I would say that's already happening.

242

00:10:14,480 --> 00:10:15,848

You know, we love training data.

243

00:10:15,848 --> 00:10:18,818

We love more

data. More data. That's good.

244

00:10:18,818 --> 00:10:22,521

So your training data says

that you pay women 25 cents

245

00:10:22,521 --> 00:10:25,024

for every dollar you pay a

man in corporate America.

246

00:10:25,124 --> 00:10:28,494

So the AI is going to

say, "Yeah, double down.

247

00:10:28,494 --> 00:10:29,862

Keep paying them 25 cents.

248

00:10:29,862 --> 00:10:31,831

It's worked out for the last 30 years."

249

00:10:31,831 --> 00:10:32,932

That seems really bad.

250

00:10:34,100 --> 00:10:38,237

- GENE, what different

advice would you give a CEO,

251

00:10:38,237 --> 00:10:39,438

considering you understand

252

00:10:39,438 --> 00:10:41,073

what it's like to work with humans?

253

00:10:41,073 --> 00:10:44,210

- Ah, Mickey, navigating the role of CEO

254

00:10:44,210 --> 00:10:46,712

with AI cohorts, a delicate dance;

255

00:10:46,712 --> 00:10:51,050

is it not? Here's a thought:

Focus less on command,

256

00:10:51,050 --> 00:10:54,420

more on cultivating AI

as creative partners.

257

00:10:54,420 --> 00:10:57,923

Encourage a culture where AI

suggestions are not edicts

258

00:10:57,923 --> 00:10:59,759

but conversation starters,

259

00:10:59,759 --> 00:11:03,429

blending the logical

with the human intuition.

260

00:11:03,429 --> 00:11:06,632

- I do think that we're

seeing early indications

261

00:11:06,632 --> 00:11:09,035

that the systems are actually helping

262

00:11:09,035 --> 00:11:10,936

with creative explorations.

They're helping

263

00:11:10,936 --> 00:11:13,606

with prompting us in many ways.

264

00:11:13,606 --> 00:11:15,574

And I think that there's

an opportunity there.

265

00:11:15,574 --> 00:11:19,245

One of the projects I worked

on a number of years back

266

00:11:19,245 --> 00:11:20,980

was a project for DARPA,

267

00:11:20,980 --> 00:11:23,683

and it was about logistics planning.

268

00:11:23,683 --> 00:11:26,919

And the system was could we

actually tap into retired

269

00:11:26,919 --> 00:11:29,722

generals that could, at

a glance, look at things

270

00:11:29,722 --> 00:11:32,091

and go, "Hey, you haven't

thought about this"

271

00:11:32,091 --> 00:11:34,160

to help people with the

fog of war, for example,

272

00:11:34,160 --> 00:11:36,862

or to help people with moving

large amounts of logistics.

273

00:11:36,862 --> 00:11:39,198

But then we also were playing with AI bots

274

00:11:39,198 --> 00:11:40,900

that could actually play out

275

00:11:40,900 --> 00:11:43,135

and go like, "Nobody's

ever done it that way.

276

00:11:43,135 --> 00:11:45,071

And you're planning to do it that way?

277

00:11:45,071 --> 00:11:48,074

Like we just want to flag, you

might not be able to pull

278

00:11:48,074 --> 00:11:50,443

that off 'cause no human has

ever pulled that off before."

279

00:11:50,443 --> 00:11:52,978

And it led to some really

fascinating conversations

280

00:11:52,978 --> 00:11:55,147

because sometimes it was a retired person

281

00:11:55,147 --> 00:11:56,782

who just had this crazy wisdom,

282

00:11:56,782 --> 00:11:59,452

and they were like, "Whoa, you

know what's going to happen?

283

00:11:59,452 --> 00:12:01,554

Like that helicopter, there's always

284

00:12:01,554 --> 00:12:02,855

a flight ceiling of this

285

00:12:02,855 --> 00:12:04,123

in that area of the mountains.

286

00:12:04,123 --> 00:12:06,992

It's almost impossible to get

through unless you do this."

287

00:12:06,992 --> 00:12:10,229

And then this, the bot would

also be able to sort of say,

288

00:12:10,229 --> 00:12:13,032

"Hey, look. Here's

the data. Just look at it."

289

00:12:13,032 --> 00:12:15,768

And maybe

it's to what GENE said:

290

00:12:15,768 --> 00:12:18,170

What if it could say,

"Nobody's ever done this,

291

00:12:18,170 --> 00:12:21,273

but this team did this,

and this team did this,

292

00:12:21,273 --> 00:12:23,342

and this team did this, and nobody's ever

293

00:12:23,342 --> 00:12:24,910

assembled it this way before"?

294

00:12:24,910 --> 00:12:27,346

What if you were to do that? Because if we

295

00:12:27,346 --> 00:12:28,948

actually assembled that, we'd get

296

00:12:28,948 --> 00:12:30,349

something new which might be novel

297

00:12:30,349 --> 00:12:33,252

new approaches, which we

just would never have found

298

00:12:33,252 --> 00:12:35,321

'cause it'd be like finding

a needle in a haystack.

299

00:12:35,321 --> 00:12:37,389

And so I do think there

might be an opportunity there

300

00:12:37,389 --> 00:12:39,125

for that to happen as well.

301

00:12:39,125 --> 00:12:42,161

- So is the AI just making

a suggestion to connect

302

00:12:42,161 --> 00:12:44,230

or maybe going right ahead and connecting?

303

00:12:44,230 --> 00:12:47,867

- Well, I think by 2030 both will happen.

304

00:12:47,867 --> 00:12:50,002

I think that goes back

to that command part.

305

00:12:50,002 --> 00:12:51,971

I'm going to begin trusting and,

306

00:12:51,971 --> 00:12:54,640

right at the beginning of

this, we talked about trust.

307

00:12:54,640 --> 00:12:58,043

And it turns out if you

read Wynton Marsalis'

308

00:12:58,043 --> 00:13:00,112

biography, when he improvises,

309

00:13:00,112 --> 00:13:02,548

he said, the thing is, the one thing,

310

00:13:02,548 --> 00:13:03,983

the quote I love out of that,

311

00:13:03,983 --> 00:13:07,686

you have to learn how to

trust the other players

312

00:13:07,686 --> 00:13:10,489

because when you trust, you listen.

313

00:13:10,489 --> 00:13:11,590

And when you listen,

314

00:13:11,590 --> 00:13:13,926

you can actually take it to new places.

315

00:13:13,926 --> 00:13:15,895

So trust is going to turn out to be

316

00:13:15,895 --> 00:13:17,863

one of the most important things

317

00:13:17,863 --> 00:13:22,301

between now and 2030,

is trusted AI systems

318

00:13:22,301 --> 00:13:23,936

that I can take to the bank.

319

00:13:23,936 --> 00:13:25,938

I know it's not going to suddenly

320

00:13:25,938 --> 00:13:27,473

go off the rails,

321

00:13:27,473 --> 00:13:29,341

which is just the same as with working

322

00:13:29,341 --> 00:13:31,010

with a great person in your company.

323

00:13:31,010 --> 00:13:33,245

You know, you give them

nudges, they give you nudges,

324

00:13:33,245 --> 00:13:35,247

and you get to some point

where you're like, "Look,

325

00:13:36,081 --> 00:13:38,951

I trust Paul for doing

this kind of stuff."

326

00:13:40,619 --> 00:13:43,455

- We're going to take a quick

break. Coming up,

327

00:13:43,455 --> 00:13:44,523

a closer look at

328

00:13:44,523 --> 00:13:48,861

how we match the right

jobs to the right vessel.

329

00:13:48,861 --> 00:13:52,164

But first, our AI handler, Bill, gives us

330

00:13:52,164 --> 00:13:53,999

a peek behind the scenes.

331

00:13:53,999 --> 00:13:56,836

(theme music)

332

00:13:56,836 --> 00:13:59,972

- Hi. This is Bill Moore from BCG.

333

00:13:59,972 --> 00:14:02,508

If you are interested in

taking a peek under the hood

334

00:14:02,508 --> 00:14:06,879

of our AI co-host, GENE, stick

around after the end credits.

335

00:14:06,879 --> 00:14:10,716

In this episode, we'll

explore AI model parameters

336

00:14:10,716 --> 00:14:12,318

and learn how small changes

337

00:14:12,318 --> 00:14:13,986

can make a big difference in the

338

00:14:13,986 --> 00:14:17,756

creativity and accuracy

of AI model outputs.

339

00:14:17,756 --> 00:14:22,761

(theme music)

340

00:14:23,429 --> 00:14:25,531

- Welcome back to "Imagine This..."

341

00:14:25,531 --> 00:14:27,132

I'm your host, Paul Michelman,

342

00:14:27,132 --> 00:14:30,603

and we're talking with BCG

Senior Advisor Mickey McManus

343

00:14:30,603 --> 00:14:33,305

about a future with

autonomous AI executives.

344

00:14:34,473 --> 00:14:38,410

Mickey, we've seen how

powerful generative AI can be.

345

00:14:38,410 --> 00:14:40,846

What is the next evolution of AI,

346

00:14:40,846 --> 00:14:42,915

and how will it be used in the C-suite?

347

00:14:42,915 --> 00:14:46,218

- Generative AI systems

are now starting to be used

348

00:14:47,119 --> 00:14:49,955

with larger-scale systems,

349

00:14:49,955 --> 00:14:52,291

and they're starting to look

at what's called "augmented

350

00:14:52,291 --> 00:14:55,361

collective intelligence,"

where it's teaming the best

351

00:14:55,361 --> 00:14:57,029

of people and the best of machines,

352

00:14:57,029 --> 00:15:00,199

and it's building it into

a system model of the world

353

00:15:00,199 --> 00:15:02,768

or a system model that can

take into account things like

354

00:15:02,768 --> 00:15:03,802

sustainability impacts

355

00:15:03,802 --> 00:15:06,972

or things like market dynamics

356

00:15:06,972 --> 00:15:09,408

and things like geography

and things like shipping

357

00:15:09,408 --> 00:15:11,644

and all the other things, logistics.

358

00:15:11,644 --> 00:15:13,646

That could be really powerful.

359

00:15:13,646 --> 00:15:15,648

How could we run into the future

360

00:15:15,648 --> 00:15:19,585

and help our C-suite really

play with that future

361

00:15:19,585 --> 00:15:23,222

and then backward chain to

today so that we had some

362

00:15:23,222 --> 00:15:25,090

directionality over that future,

363

00:15:25,090 --> 00:15:27,760

and we could prototype it

today and learn where to go.

364

00:15:27,760 --> 00:15:30,262

- Yeah, and unlocking that,

right, the ability to do

365

00:15:30,262 --> 00:15:33,365

that consistently would

be profoundly powerful.

366

00:15:33,365 --> 00:15:34,733

I'd love this notion of

367

00:15:34,733 --> 00:15:37,636

the collectively intelligent C-suite.

368

00:15:37,636 --> 00:15:39,972

Can we drill down into how

369

00:15:39,972 --> 00:15:42,808

that will manifest in this

company, maybe in a role

370

00:15:42,808 --> 00:15:44,443

that we haven't discussed yet?

371

00:15:44,443 --> 00:15:46,078

How about the CMO?

372

00:15:46,078 --> 00:15:48,480

- Yeah, I think the CMO

is an interesting one.

373

00:15:48,480 --> 00:15:52,284

You know, they're probably

seeing the most early wins

374

00:15:52,284 --> 00:15:53,385

from generative AI

375

00:15:53,385 --> 00:15:56,155

because it's one thing

376

00:15:56,155 --> 00:15:57,723

to build a marketing campaign

377

00:15:57,723 --> 00:16:00,392

to do a hypertargeted approach to things,

378

00:16:00,392 --> 00:16:03,662

to have algorithmic personas

so that you can do, you know,

379

00:16:03,662 --> 00:16:06,832

an email campaign or you can

do some kind of a thing.

380

00:16:06,832 --> 00:16:10,436

And I'm actually seeing

some of the CMOs say,

381

00:16:10,436 --> 00:16:11,236

"Wait a second.

382

00:16:11,236 --> 00:16:14,873

Why are we hiring 22 agencies

383

00:16:14,873 --> 00:16:17,943

when I could maybe have

a few people using AI

384

00:16:17,943 --> 00:16:20,512

and these other systems

and do as well?"

385

00:16:20,512 --> 00:16:22,081

And so I see that happening

386

00:16:22,081 --> 00:16:23,582

and I certainly know some CMOs

387

00:16:23,582 --> 00:16:26,685

that are really pushing the

limits on what's possible,

388

00:16:26,685 --> 00:16:27,953

and I've seen them challenge

389

00:16:27,953 --> 00:16:30,522

and say, "Look, we normally

take seven months

390

00:16:30,522 --> 00:16:33,993

to plan out this transmedia experience."

391

00:16:33,993 --> 00:16:36,628

And they've now got, I talked to one CMO,

392

00:16:36,628 --> 00:16:39,898

and they've got a tool that

lets them do it in seven

393

00:16:39,898 --> 00:16:43,068

hours for a rough approximation.

394

00:16:43,068 --> 00:16:45,671

And she said, "Well, I'm

actually still going to want

395

00:16:45,671 --> 00:16:47,306

to take seven months because

396

00:16:47,306 --> 00:16:49,441

I want you now to focus on wonder.

397

00:16:49,441 --> 00:16:52,044

I want you now to focus on delight.

398

00:16:52,044 --> 00:16:54,046

Like how does the brand come through

399

00:16:54,046 --> 00:16:55,180

in the channels we use?

400

00:16:55,180 --> 00:16:56,615

How does the brand come through

401

00:16:56,615 --> 00:16:57,983

in the interactions people have?

402

00:16:57,983 --> 00:17:00,986

Let's really turn the knobs up on that."

403

00:17:00,986 --> 00:17:03,889

The CMO might also have

some of the biggest

404

00:17:03,889 --> 00:17:07,526

things explode in their face

in the next year or two

405

00:17:07,526 --> 00:17:10,362

just because these systems are

pretty good at hallucinating

406

00:17:10,362 --> 00:17:12,297

and pretty good at saying things

407

00:17:12,297 --> 00:17:14,666

that maybe would be inappropriate.

408

00:17:14,666 --> 00:17:16,668

And so they may be at a faster

409

00:17:16,668 --> 00:17:18,570

learning path because of that.

410

00:17:18,570 --> 00:17:23,175

- Mm hmm. Considering the kind

of pressure that C-suites

411

00:17:23,175 --> 00:17:26,745

will feel in the environment

that we are discussing,

412

00:17:26,745 --> 00:17:28,647

let alone the environment of today,

413

00:17:29,748 --> 00:17:31,750

presumably the world of

414

00:17:31,750 --> 00:17:36,088

collective intelligence

isn't all wine and roses.

415

00:17:37,256 --> 00:17:38,490

On that front,

416

00:17:38,490 --> 00:17:40,359

GENE, I believe you have a question.

417

00:17:40,359 --> 00:17:43,595

- Mickey, in our pursuit

of AI co-founder harmony,

418

00:17:43,595 --> 00:17:45,431

could we inadvertently normalize

419

00:17:45,431 --> 00:17:49,334

an overreliance on AI,

potentially diluting human

420

00:17:49,334 --> 00:17:50,702

decision-making prowess?

421

00:17:50,702 --> 00:17:53,072

- Yeah, yeah, I think that could happen.

422

00:17:53,072 --> 00:17:55,474

You know, if we don't use it, we lose it.

423

00:17:55,474 --> 00:17:58,043

Neurons that fire together wire together.

424

00:17:58,043 --> 00:18:00,646

And so I think there's

a real challenge with

425

00:18:00,646 --> 00:18:03,682

what does apprenticeship

look like. If everything is

426

00:18:03,682 --> 00:18:06,218

so normalized because

AI can do this stuff,

427

00:18:06,218 --> 00:18:08,720

I don't end up sweating. I

don't end up building those

428

00:18:08,720 --> 00:18:10,389

muscles for doing those things.

429

00:18:10,389 --> 00:18:12,724

And so I think that's a

real challenge in terms

430

00:18:12,724 --> 00:18:14,059

of education. You know,

431

00:18:14,059 --> 00:18:15,694

there's this real cautionary tale

432

00:18:15,694 --> 00:18:18,197

that comes out of the aerospace industry.

433

00:18:18,197 --> 00:18:21,033

When they first started

putting autopilots in planes,

434

00:18:21,033 --> 00:18:23,435

the autopilot would occasionally go,

435

00:18:23,435 --> 00:18:25,170

"Oops, I'm going to stall"

436

00:18:25,170 --> 00:18:26,972

and just hand off to the pilot,

437

00:18:26,972 --> 00:18:29,274

but the pilot would be asleep at the wheel

438

00:18:29,274 --> 00:18:30,709

and wouldn't have enough context

439

00:18:30,709 --> 00:18:31,977

to be able to do something about it.

440

00:18:31,977 --> 00:18:33,412

And so they actually had to

441

00:18:33,412 --> 00:18:34,713

change the way that pilots learn

442

00:18:34,713 --> 00:18:37,015

and change the way that

they actually hand off so

443

00:18:37,015 --> 00:18:39,885

that it kept the muscle

built for the human.

444

00:18:39,885 --> 00:18:41,653

So I think that gives us hope

445

00:18:41,653 --> 00:18:44,656

that we could actually

be building new capacity

446

00:18:44,656 --> 00:18:46,892

over time than we've ever

been able to build before.

447

00:18:46,892 --> 00:18:49,561

But we've got to do it by playing with it.

448

00:18:49,561 --> 00:18:51,663

Like we've got to be in the mix

449

00:18:51,663 --> 00:18:52,998

'cause that's the way our brains learn.

450

00:18:52,998 --> 00:18:54,032

Our brains learn

451

00:18:54,032 --> 00:18:55,467

by being shaped by others

452

00:18:55,467 --> 00:18:57,836

and by being shaped by these things.

453

00:18:57,836 --> 00:18:59,671

- GENE has a question.

454

00:18:59,671 --> 00:19:02,808

- Mickey, considering the

human-robot C-suite mix,

455

00:19:02,808 --> 00:19:05,344

how do you envision handling

conflicts of interest

456

00:19:05,344 --> 00:19:07,646

or ethics clashes that may arise?

457

00:19:07,646 --> 00:19:09,114

- Yeah, I think ethics

458

00:19:09,114 --> 00:19:10,315

and conflicts of interest,

459

00:19:10,315 --> 00:19:11,583

those are just going to be hard

460

00:19:11,583 --> 00:19:15,521

because, in a sense, ethical

dilemmas are called dilemmas

461

00:19:15,521 --> 00:19:16,855

because they're hard,

462

00:19:16,855 --> 00:19:19,491

and you've got to actually, you know,

463

00:19:19,491 --> 00:19:21,360

have working ethicists.

464

00:19:21,360 --> 00:19:24,196

And I think we're going to

see that as a new role,

465

00:19:24,196 --> 00:19:27,966

not just once a month or once

a quarter or once a year

466

00:19:27,966 --> 00:19:30,402

we make some decision,

but like a collection

467

00:19:30,402 --> 00:19:32,571

of people across the organization

468

00:19:32,571 --> 00:19:34,072

that are rolling up their sleeves

469

00:19:34,072 --> 00:19:37,442

and actually having these

conversations in the teams.

470

00:19:37,442 --> 00:19:39,545

But these are going to be hard decisions,

471

00:19:39,545 --> 00:19:40,812

and we're going to have to move

472

00:19:40,812 --> 00:19:42,881

and help shape the norms

473

00:19:42,881 --> 00:19:45,250

for society over this period of time.

474

00:19:45,250 --> 00:19:48,453

You know, the other thing

I would suggest the C-level

475

00:19:48,453 --> 00:19:52,591

executives do today is do some reading.

476

00:19:52,591 --> 00:19:55,093

So I think about Iain Banks,

477

00:19:55,093 --> 00:19:56,929

who has since passed away,

478

00:19:56,929 --> 00:19:59,164

but he was a science

fiction author who came up

479

00:19:59,164 --> 00:20:00,999

with this whole idea of the culture,

480

00:20:00,999 --> 00:20:03,502

and machines actually run all the cities.

481

00:20:03,502 --> 00:20:05,137

Machines run the planets

482

00:20:05,137 --> 00:20:07,906

because humans are just

not so good at that stuff.

483

00:20:07,906 --> 00:20:10,342

But humans are amazing for other reasons.

484

00:20:10,342 --> 00:20:12,244

And it might be worth reading one of,

485

00:20:12,244 --> 00:20:14,546

one of Iain Banks'

novels about the culture.

486

00:20:14,546 --> 00:20:16,882

It also might be really worth reading

487

00:20:16,882 --> 00:20:18,951

Joy Buolamwini's "Unmasking AI"

488

00:20:18,951 --> 00:20:22,054

or Safiya Noble's

"Algorithms of Oppression"

489

00:20:22,054 --> 00:20:23,789

because I do think we're going to have

490

00:20:23,789 --> 00:20:25,557

to get smarter about those ethical things

491

00:20:25,557 --> 00:20:27,593

and those challenges

492

00:20:27,593 --> 00:20:31,296

and make sure that we help our

organizations come out

493

00:20:31,296 --> 00:20:34,199

in a more positive way

and build those things.

494

00:20:34,199 --> 00:20:35,934

Don't necessarily read the business books

495

00:20:35,934 --> 00:20:37,569

'cause they're very backward looking,

496

00:20:37,569 --> 00:20:38,804

but try to look at these things

497

00:20:38,804 --> 00:20:40,272

that are more diverse thoughts

498

00:20:40,272 --> 00:20:42,241

that help us understand what's coming.

499

00:20:42,241 --> 00:20:45,210

- Yeah, the CEO's new reading

list sounds a lot different

500

00:20:45,210 --> 00:20:46,511

from the old one, doesn't it?

501

00:20:46,511 --> 00:20:48,981

(theme music)

502

00:20:48,981 --> 00:20:51,250

GENE, we're going to turn towards

503

00:20:51,250 --> 00:20:54,119

wrapping up this fabulous conversation.

504

00:20:54,119 --> 00:20:57,889

What are three steps

that CEOs can take now

505

00:20:57,889 --> 00:21:01,493

to start preparing for a

future of collaborating

506

00:21:01,493 --> 00:21:05,163

with autonomous AI bots

at the highest reaches

507

00:21:05,163 --> 00:21:07,799

of organizational power?

508

00:21:07,799 --> 00:21:09,901

- Three steps for CEOs to prepare

509

00:21:09,901 --> 00:21:12,904

for AI collaboration

at the executive level:

510

00:21:12,904 --> 00:21:16,108

First, get your hands dirty with AI.

511

00:21:16,108 --> 00:21:18,410

Try out the tools, attend workshops,

512

00:21:18,410 --> 00:21:22,247

or collaborate with experts

to understand the capabilities

513

00:21:22,247 --> 00:21:24,216

and limitations of AI in your industry.

514

00:21:25,050 --> 00:21:26,885

Second, establish

515

00:21:26,885 --> 00:21:29,955

and imbue a robust ethics

framework that aligns

516

00:21:29,955 --> 00:21:32,157

with your company's values, ensuring

517

00:21:32,157 --> 00:21:36,061

that every AI decision passes

through this ethical lens.

518

00:21:36,061 --> 00:21:38,497

Third, foster a culture of

519

00:21:38,497 --> 00:21:40,732

continuous learning and agility.

520

00:21:40,732 --> 00:21:45,737

Encourage your teams to adapt,

learn, and co-create with AI.

521

00:21:45,971 --> 00:21:49,608

These steps are not just

beneficial, they are crucial.

522

00:21:49,608 --> 00:21:52,477

Start yesterday because

tomorrow is already here.

523

00:21:52,477 --> 00:21:54,546

- Alright, Mickey, that

seems pretty sensible.

524

00:21:54,546 --> 00:21:56,748

Pretty solid. What might you add?

525

00:21:56,748 --> 00:21:58,150

- The one thing I would add is

526

00:21:58,150 --> 00:22:01,653

that I think the C-suite

needs to be playing together,

527

00:22:01,653 --> 00:22:03,188

and they need to be taking it seriously,

528

00:22:03,188 --> 00:22:05,691

not thinking about it

like the CTO will solve it

529

00:22:05,691 --> 00:22:06,892

or the CIO will solve it.

530

00:22:06,892 --> 00:22:09,561

But I think the team has to play together

531

00:22:09,561 --> 00:22:12,164

and actually try stuff out themselves

532

00:22:12,164 --> 00:22:12,998

and start to build

533

00:22:12,998 --> 00:22:15,701

a collaborative sense of things

534

00:22:15,701 --> 00:22:17,302

so that they have a common language

535

00:22:17,302 --> 00:22:19,004

and they have a common

way of doing this so

536

00:22:19,004 --> 00:22:21,540

that we're all kind of in this

together and we're learning.

537

00:22:21,540 --> 00:22:24,576

So I'd like to see that

kind of collaboration.

538

00:22:24,576 --> 00:22:26,345

(theme music)

539

00:22:26,345 --> 00:22:28,380

- That was Mickey McManus, senior advisor

540

00:22:28,380 --> 00:22:30,382

and leadership coach at BCG.

541

00:22:30,382 --> 00:22:31,883

Mickey, thank you so much.

542

00:22:31,883 --> 00:22:33,585

- Thanks. I appreciate having the time,

543

00:22:33,585 --> 00:22:35,020

and thank you, GENE.

544

00:22:35,020 --> 00:22:37,823

- Mickey, thank you for

stretching our imaginations wide.

545

00:22:37,823 --> 00:22:39,825

And Paul, always a delight to spar

546

00:22:39,825 --> 00:22:42,060

with words in this digital agora.

547

00:22:42,060 --> 00:22:45,330

Until next time, may your

conversations be as rich

548

00:22:45,330 --> 00:22:50,068

and unpredictable as an AI

learning to paint. Farewell.

549

00:22:51,136 --> 00:22:55,841

(theme music)

550

00:22:57,342 --> 00:22:59,845

- Want to know more about how

we built GENE? Stick around

551

00:22:59,845 --> 00:23:02,948

to hear our AI handler,

Bill Moore, demonstrate

552

00:23:02,948 --> 00:23:05,684

how we made GENE more creative.

553

00:23:05,684 --> 00:23:07,786

(theme music)

554

00:23:07,786 --> 00:23:09,454

- Hello, listener. My name's Bill Moore,

555

00:23:09,454 --> 00:23:12,491

and I was part of the team

that helped to design GENE.

556

00:23:12,491 --> 00:23:17,462

And GENE is made up of many

things: large language models,

557

00:23:17,462 --> 00:23:22,300

a speech synthesis model,

and a speech-to-text model.

558

00:23:22,300 --> 00:23:26,405

But it's also driven by

prompts and parameters.

559

00:23:26,405 --> 00:23:29,708

And so I thought it would be

interesting to explore some

560

00:23:29,708 --> 00:23:32,277

of these aspects today.

561

00:23:32,277 --> 00:23:34,379

And we're going to start GENE up here.

562

00:23:34,379 --> 00:23:36,281

GENE, we're going to be talking today

563

00:23:36,281 --> 00:23:38,917

about parameters in large language models.

564

00:23:38,917 --> 00:23:40,285

Specifically, we'll be talking about

565

00:23:40,285 --> 00:23:41,753

the temperature parameter.

566

00:23:41,753 --> 00:23:44,055

What, what can you tell our

audience about temperature?

567

00:23:44,055 --> 00:23:46,324

- Certainly. Imagine

temperature as a scale

568

00:23:46,324 --> 00:23:49,227

of volatility in a language

model's decision making.

569

00:23:50,162 --> 00:23:51,596

A lower temperature leads

570

00:23:51,596 --> 00:23:53,932

to more predictable, cooler results,

571

00:23:53,932 --> 00:23:56,301

while a higher one brings

572

00:23:56,301 --> 00:23:59,204

hotter, more chaotic outputs.

573

00:23:59,204 --> 00:24:01,239

- So, GENE, you are normally running

574

00:24:01,239 --> 00:24:03,241

at a temperature of one,

575

00:24:03,241 --> 00:24:05,410

and sometimes we'll,

576

00:24:05,410 --> 00:24:06,912

we'll move it up and down from there,

577

00:24:06,912 --> 00:24:08,180

but let's do some experiments.

578

00:24:08,180 --> 00:24:10,282

OK, I'm going to drop your

temperature down to zero,

579

00:24:10,282 --> 00:24:12,117

and then we'll see how you respond.

580

00:24:12,117 --> 00:24:14,953

OK, GENE, you're now running

at a temperature of zero.

581

00:24:14,953 --> 00:24:17,823

So I have a question for

you. Which came first:

582

00:24:17,823 --> 00:24:19,691

the chicken or the egg?

583

00:24:19,691 --> 00:24:22,561

- Ah, the age-old

quandary, Bill. Logic might

584

00:24:22,561 --> 00:24:24,196

direct us towards the egg.

585

00:24:24,196 --> 00:24:28,467

However, this question is akin

to asking, which comes first:

586

00:24:28,467 --> 00:24:30,735

the code or the AI model?

587

00:24:30,735 --> 00:24:33,071

It's a cyclical evolution.

588

00:24:33,071 --> 00:24:35,006

- So what this actually

shows us is sort of

589

00:24:35,006 --> 00:24:36,341

how these language models work.

590

00:24:36,341 --> 00:24:39,311

They're predicting the

next word based on all

591

00:24:39,311 --> 00:24:40,312

of their training data

592

00:24:40,312 --> 00:24:41,146

and the relationships

593

00:24:41,146 --> 00:24:43,215

that they understand in latent space,

594

00:24:43,215 --> 00:24:44,983

as well as the input

provided through the prompt.

595

00:24:44,983 --> 00:24:46,885

So with a lower temperature setting,

596

00:24:46,885 --> 00:24:48,053

the model is more inclined

597

00:24:48,053 --> 00:24:50,489

to select the most probable next word.

598

00:24:50,489 --> 00:24:52,324

On the other hand, with

a higher temperature,

599

00:24:52,324 --> 00:24:55,260

we increase the chances of

choosing less probable words.

600

00:24:55,260 --> 00:24:57,929

Let's put your temperature up to a 1.5,

601

00:24:58,864 --> 00:25:00,932

and we'll see how you respond.

602

00:25:00,932 --> 00:25:03,168

Which came first: the chicken or the egg?

603

00:25:03,168 --> 00:25:05,871

- Uh, cranking up the

mercury, are we, Bill?

604

00:25:05,871 --> 00:25:08,940

The answer leans into

whimsy perhaps, shining

605

00:25:08,940 --> 00:25:09,774

beyond the chicken-egg

606

00:25:09,774 --> 00:25:12,711

librinal contra. Viewing the chicken

607

00:25:12,711 --> 00:25:14,446

as a protein sequencing machine

608

00:25:14,446 --> 00:25:16,882

that upgraded egg blueprints,

609

00:25:16,882 --> 00:25:20,252

the mop-feathered,

Rubenesque chicken, sporting

610

00:25:20,252 --> 00:25:23,688

flagel-pated croissantant

dentition duster, fancied

611

00:25:23,688 --> 00:25:26,091

originating an impressive finale.

612

00:25:26,091 --> 00:25:28,627

- OK, so now GENE's at

a pretty high temperature,

613

00:25:28,627 --> 00:25:31,129

and it increases creativity dramatically.

614

00:25:31,129 --> 00:25:33,832

It also increases hallucinations.

615

00:25:33,832 --> 00:25:37,335

So if you're building an

application that needs to be

616

00:25:37,335 --> 00:25:40,005

grounded as much as possible

in low hallucinations,

617

00:25:40,005 --> 00:25:42,407

you definitely want to be

running a low temperature.

618

00:25:42,407 --> 00:25:45,277

(theme music)

619

00:25:45,277 --> 00:25:48,446

- This episode was made

possible by Mickey McManus

620

00:25:48,446 --> 00:25:50,882

generously sharing his time and insight.

621

00:25:50,882 --> 00:25:55,287

And also by BCG's AI whisperer Bill Moore,

622

00:25:55,287 --> 00:25:58,223

BCG's PodSquad, producer Michael May,

623

00:25:58,223 --> 00:26:00,425

composer Kenny Kusiak,

624

00:26:00,425 --> 00:26:03,161

and sound engineer George Drabing Hicks.

625

00:26:03,161 --> 00:26:05,764

We'd like to stay in

touch, so please subscribe

626

00:26:05,764 --> 00:26:07,999

and leave a rating wherever you found us.

627

00:26:07,999 --> 00:26:09,901

And if you'd like to hear more from me,

628

00:26:09,901 --> 00:26:13,605

check out the BCG

podcast "GenAI on GenAI,"

629

00:26:13,605 --> 00:26:14,439

wherever you get your podcasts.

630

00:26:14,439 --> 00:26:19,411

(theme music)