February 20, 2024 • 26 mins
By 2030, the best CIO might be a CIB: a chief intelligence bot. And the human CEO might be bolstered by a swarm of CEO bots, able to harness a vast diversity of perspectives. Can you imagine? In this episode, BCG’s Mickey McManus envisions a future where top leaders share the corner office with AI—blending the logical with human intuition. Mickey, a BCG senior advisor and a pioneer in human-centered design, considers how a company like this would function. Hint: he recommends an approach much like jazz improvisation.
Help Us Imagine the Future--Send Us Your Ideas or Questions
What major changes or disruptions do YOU see on the horizon?
Please email us the future scenario you’d like us to explore. Or, send us a question you might have for GENE or about working with GENE.
Email it to: imaginethis@bcg.com.
We will pick our favorites and explore them with GENE in an upcoming episode of Imagine This.
Learn More
You’ll find more of what BCG has to say about generative AI here:
Listen to more of GENE, BCG's AI bot here:
GenAI on GenAI: https://link.chtbl.com/bcg-genai-on-genai?sid=IT
Also, in this episode, Mickey recommends an entirely new reading list for the C-suite, including:
-Iain M. Banks’s Culture series: https://www.amazon.com/The-Culture-9-book-series/dp/B07WLZZ9WV
-Joy Buolamwini’s Unmasking AI [https://www.amazon.com/Unmasking-AI-Mission-Protect-Machines/dp/0593241835]
-Safiya Umoja Noble’s Algorithms of Oppression [https://www.amazon.com/Algorithms-Oppression-Search-Engines-Reinforce/dp/1479837245]
You can find the podcast on YouTube at: https://www.youtube.com/watch?v=dpecNh11jKY&list=PLMJgyXjV5gMLbR8vIRdKaCAMmAJhzcRa8
This podcast uses the following third-party services for analysis:
Chartable - https://chartable.com/privacy
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
- Imagine this (00:02):
It's 2030
and half the jobs in the C-suite
of one Fortune 500 companyare performed by AI.
Mickey, you're having drinks with the CEO
after work.
What's on her mind?
- Yeah, well, I think she'sprobably had to do a lot
of things differently,
and she's probably had to figure out
how to, how to sort ofplay jazz collaboration
(00:26):
with her machine-learningC-level executives
and her human ones
and how to get them to play together well.
And so I'm actually curioushow she made it there.
- That's Mickey McManus, senior advisor
and leadership coach at BCG
and co-author of "Trillions:
Thriving in the EmergingInformation Ecology."
(00:47):
And I'm your host, Paul Michelman.
Welcome to "Imagine This,"
where we take a trip into the future
that we hope will challenge
the way you think and act today.
On this episode
of "Imagine This," we consider a world
where autonomous AI botsare running the company.
Speaking of which, let meintroduce my AI co-host,
GENE. Good day to you, GENE.
(01:09):
- Good day, everyone.
- GENE, would you please describe
what your role is on this show?
- Well, my role? Think of meas your digital Socrates,
here to ask the probing questions
that peel back the layers. Alongside Paul,
I'm half of the dynamic duogiving our guests a friendly
nudge towards those aha moments.
(01:29):
Hello, everyone. Ready to dive deep?
(theme music)
- A quick programming note:
None of GENE's commentary is scripted.
The final version of thisprogram has been edited
from the originalconversation, which means
that we're not includingeverything we recorded.
That applies to us humansas well as machines.
Mickey, can you take us on abrief trip into the C-suite
(01:52):
of the first company where
half the executives are AI bots?
What does command look like for a bot?
- Yeah, so there are
interesting things going on right now.
I think, in this nextwave with regenerative AI,
it's actually going to beabout ideas and about actions.
And generative AI has aninteresting capacity there.
(02:14):
It can coach. It can becreative. It can command,
and it can wait for feedback
and then actually dosomething about that command.
It can create synthesis.
It can synthesize ideastogether in different ways.
I could draw a pictureof a software interface,
and I can draw some arrows and say,
"Hey, when somebody clickson this button, do this."
(02:36):
And then I can draw anotherpicture of a visualization
and say, "Hey, if this happens, do this."
And I could draw a picture
and say, "I'd love to seea data graph of this."
I could draw that all on a whiteboard.
And I could take a picturewith one of the newer systems,
the generative AI systems,
and say, "Make that into code."
You know, I'm commanding ittoday by drawing something
and giving it some goals.
(02:57):
But what happens when itactually can kind of
continue on that chaining of command
and do a lot of other kinds of things?
And recently, we saw somegenerative AI systems
that basically couldn'tlog in to a website
because of those CAPTCHAs,you know, those things
that show up images and,
and you've got to figure out if
it's trying to test if it's human.
(03:17):
And the AI system actuallywent to basically like
a job site and hired somehumans to actually decode
that thing so that they could then get
through the system and be able to do it.
And it turns out thatthe person who was hired,
it was like a TaskRabbit thing,
the person who was hired said,
"Are you just messing around with me?
(03:38):
Are you an AI?" And itsaid, "No, I'm not."
And so we also have to be aware
that just like thebiases that humans have,
we likely will find these AI systems
not only can take action
but they can take actionin ways that they can,
get somethingdone that they want done,
that they were commanded todo or that they decided to do.
- GENE has a question on that subject.
(04:00):
- Mickey, envisioningthis AI-infused C-suite,
let's ponder theirdecision-making processes.
How might the AI's data-driven logic clash
with human intuition,
and what measures are in place
to harmonize these distinct approaches?
- I'm not sure I know yet.
I think, since I'm sittingdown with the CEO, I'd like
to ask her what, what it'slike in the year 2030.
(04:22):
But you know, I think so much of
what we think we do is logical as humans.
You know, we think we makelogical decisions based on
historical data, and we, we play that out.
But, but so much of what wereally do is based on emotion.
So much of what wereally do is based on
a passionate idea
or a framing of somethingthat actually helps us see
(04:43):
that every word we use islike derived from the things
around us and the way we think.
And so almost everything weuse is a metaphor and analogy.
And so humans aresuperpowered at this.
So even if the AI systemsaid something that
might have been a greatlogical story, it could be
(05:03):
that the human's job isreally to find a thread that
surprises them and that they can pull on
that turns into a powerful story
that can motivate the companyor motivate shareholders
or motivate customers in some new ways.
So I think this goes back toa little bit of the comment
that I made at the beginning,
which is maybe it's alittle bit more about improv
(05:25):
or a little bit moreabout jazz collaboration.
So, you know, I'm justwondering
what we can learn from TinaFey, what we can learn from
Wynton Marsalis in this dynamic
and how much that is going
to turn out to be really important,
which is being fun to play with
and thinking about everything as an offer.
And that's not necessarilysomething we teach today
(05:47):
in business school,
but it might be something that turns out
to be important in the future.
- So what's a piece ofadvice you would give someone
who aspires to be in thisC-suite that we are describing?
What's the piece of adviceyou would give them today
in order to prepare for this world
(06:09):
of improvisation that you're describing?
- I think actually the advice I'd give
to somebody today right nowis mess around. Start playing
with the generative AI functionality.
Look at ways that youcan actually be a centaur.
And so that's oneway of thinking of it is
I do the thing that I'm pretty good at,
and then I hand it off to the horse part,
(06:30):
and it does the thing thatit's pretty good at, you know,
and then it's my upper-bodyand its lower-body strength.
Maybe play with somethingjust on the weekend, just
as part of an experimentation.Play with something
where you're a cyborg, whereit's sort of you don't know
where you end and itbegins, where it's sort
of this play back andforth and back and forth,
and you're tuning it
because you're acting likeit's another teammate.
(06:52):
Not like it's a thing I prompt,
but it's something thatI actually play with.
- So who is the CIO, right,in the C-suite we're
describing? Is there a CIO?
- I think so. I think there is,
and I don't know if it's oneof the ones that's the bot.
There's no reason that weshould think that by 2030,
the best CIO is a human.
(07:13):
And it might actually be
that it might be better that it's not.
One of the things humansdo have is context
and a mental model.
And right now we haven'tseen any of the systems
that are really big and,
and deployed with any kind ofsymbolic contextual reasoning
or mental modeling of the universe.
That's something all lifeon earth gets for free,
but it's because of a
(07:33):
3 billion-year-old
R&D experiment
called life on earth.
So it's been running for a long
time, and it's figured that out.
So I don't know if the CIO is actually one
of the C-suites that's the bot or not.
I do think that when youthink about information,
it is the new oil in some ways,
and I think the CIO isgoing to be much more critical
(07:55):
to understand how tobuild the data factory.
How do we do that in a waythat we can actually capture
more valuable information
and let more businesspeople take advantage of it
and more of the botstake advantage of it
so that we can actually make decisions?
We suddenly have more insightsthan we ever could before,
and it's really hard forhumans to see those insights.
(08:16):
We have bounded rationality.We can't see
5, 10, 20,30 dimensions at once.
We just can't. So it might be
that the CIO's job is, in manyways, to help us cope with
that high dimensionalityand, and cleanse, append,
and build out that data factory
to kind of think about it.
So I'm not sure if they'rea hybrid, if they're a human,
(08:40):
or if they're a bot that does this
because it turns out, you know,
they understand thesystems better than we do.
- Let's talk about the CEO specifically.
In my opening question to you,
I presumed by my pronounchoice that the CEO
was human. Is that a fair assumption?
- I think so. I thinkthe CEO is still human.
(09:02):
And I think we'retalking to her at the bar
and we're trying to findout how she got there
mostly because I think a lot of the role
of the CEO is to set that culture.
I think great CEOs really geta sense of the zeitgeist
of what's happening in their organization.
When I was a CEO, I did alot of MBWA. I did manage
(09:23):
by walking around. I'd wanderaround from team to team,
and I'd be like, "Oh, wow,you're working on this?
Did you know the guys upstairsare working on something
that, that I see a connection to?
You might want to take alittle bit of a field trip."
So while we might be ableto encode those things,
well, we might be ableto create a sort of swarm
of CEO bots so
that the CEO can bemuch more intimate
(09:44):
with a lot more people,
I still think it's probablya human role for a while.
- I do think the notionthat a CEO bot can harness
a diversity of perspective in a way
that a single human could never do
to be quite powerful.
- Mickey, could AI in the C-suite
also amplify our biases
(10:05):
if not carefully checked, potentially
automating systemic issues at scale?
- Well, I wish I could say no,
but I would say that's already happening.
You know, we love training data.
We love moredata. More data. That's good.
So your training data saysthat you pay women 25 cents
for every dollar you pay aman in corporate America.
(10:25):
So the AI is going tosay, "Yeah, double down.
Keep paying them 25 cents.
It's worked out for the last 30 years."
That seems really bad.
- GENE, what differentadvice would you give a CEO,
considering you understand
what it's like to work with humans?
- Ah, Mickey, navigating the role of CEO
with AI cohorts, a delicate dance;
is it not? Here's a thought: Focus less on command, (10:46):
undefined
more on cultivating AIas creative partners.
Encourage a culture where AIsuggestions are not edicts
but conversation starters,
blending the logicalwith the human intuition.
- I do think that we'reseeing early indications
that the systems are actually helping
(11:09):
with creative explorations.They're helping
with prompting us in many ways.
And I think that there'san opportunity there.
One of the projects I workedon a number of years back
was a project for DARPA,
and it was about logistics planning.
And the system was could weactually tap into retired
generals that could, ata glance, look at things
(11:29):
and go, "Hey, you haven'tthought about this"
to help people with thefog of war, for example,
or to help people with movinglarge amounts of logistics.
But then we also were playing with AI bots
that could actually play out
and go like, "Nobody'sever done it that way.
And you're planning to do it that way?
Like we just want to flag, youmight not be able to pull
that off 'cause no human hasever pulled that off before."
(11:50):
And it led to some reallyfascinating conversations
because sometimes it was a retired person
who just had this crazy wisdom,
and they were like, "Whoa, youknow what's going to happen?
Like that helicopter, there's always
a flight ceiling of this
in that area of the mountains.
It's almost impossible to getthrough unless you do this."
And then this, the bot wouldalso be able to sort of say,
"Hey, look. Here'sthe data. Just look at it."
(12:13):
And maybeit's to what GENE said:
What if it could say,"Nobody's ever done this,
but this team did this,and this team did this,
and this team did this, and nobody's ever
assembled it this way before"?
What if you were to do that? Because if we
actually assembled that, we'd get
something new which might be novel
new approaches, which wejust would never have found
(12:33):
'cause it'd be like findinga needle in a haystack.
And so I do think theremight be an opportunity there
for that to happen as well.
- So is the AI just makinga suggestion to connect
or maybe going right ahead and connecting?
- Well, I think by 2030 both will happen.
I think that goes backto that command part.
I'm going to begin trusting and,
right at the beginning ofthis, we talked about trust.
(12:54):
And it turns out if youread Wynton Marsalis'
biography, when he improvises,
he said, the thing is, the one thing,
the quote I love out of that,
you have to learn how totrust the other players
because when you trust, you listen.
And when you listen,
you can actually take it to new places.
So trust is going to turn out to be
(13:15):
one of the most important things
between now and 2030,is trusted AI systems
that I can take to the bank.
I know it's not going to suddenly
go off the rails,
which is just the same as with working
with a great person in your company.
You know, you give themnudges, they give you nudges,
and you get to some pointwhere you're like, "Look,
(13:36):
I trust Paul for doingthis kind of stuff."
- We're going to take a quickbreak. Coming up,
a closer look at
how we match the rightjobs to the right vessel.
But first, our AI handler, Bill, gives us
a peek behind the scenes.
(theme music)
(13:56):
- Hi. This is Bill Moore from BCG.
If you are interested intaking a peek under the hood
of our AI co-host, GENE, stickaround after the end credits.
In this episode, we'llexplore AI model parameters
and learn how small changes
can make a big difference in the
creativity and accuracyof AI model outputs.
(14:17):
(theme music)
- Welcome back to "Imagine This..."
I'm your host, Paul Michelman,
and we're talking with BCGSenior Advisor Mickey McManus
about a future withautonomous AI executives.
Mickey, we've seen howpowerful generative AI can be.
(14:38):
What is the next evolution of AI,
and how will it be used in the C-suite?
- Generative AI systemsare now starting to be used
with larger-scale systems,
and they're starting to lookat what's called "augmented
collective intelligence,"where it's teaming the best
of people and the best of machines,
and it's building it intoa system model of the world
(15:00):
or a system model that cantake into account things like
sustainability impacts
or things like market dynamics
and things like geographyand things like shipping
and all the other things, logistics.
That could be really powerful.
How could we run into the future
and help our C-suite reallyplay with that future
and then backward chain totoday so that we had some
(15:23):
directionality over that future,
and we could prototype ittoday and learn where to go.
- Yeah, and unlocking that,right, the ability to do
that consistently wouldbe profoundly powerful.
I'd love this notion of
the collectively intelligent C-suite.
Can we drill down into how
that will manifest in thiscompany, maybe in a role
that we haven't discussed yet?
(15:44):
How about the CMO?
- Yeah, I think the CMOis an interesting one.
You know, they're probablyseeing the most early wins
from generative AI
because it's one thing
to build a marketing campaign
to do a hypertargeted approach to things,
to have algorithmic personasso that you can do, you know,
an email campaign or you cando some kind of a thing.
(16:06):
And I'm actually seeingsome of the CMOs say,
"Wait a second.
Why are we hiring 22 agencies
when I could maybe havea few people using AI
and these other systemsand do as well?"
And so I see that happening
and I certainly know some CMOs
that are really pushing thelimits on what's possible,
and I've seen them challenge
(16:27):
and say, "Look, we normallytake seven months
to plan out this transmedia experience."
And they've now got, I talked to one CMO,
and they've got a tool thatlets them do it in seven
hours for a rough approximation.
And she said, "Well, I'mactually still going to want
to take seven months because
I want you now to focus on wonder.
(16:49):
I want you now to focus on delight.
Like how does the brand come through
in the channels we use?
How does the brand come through
in the interactions people have?
Let's really turn the knobs up on that."
The CMO might also havesome of the biggest
things explode in their facein the next year or two
just because these systems arepretty good at hallucinating
(17:10):
and pretty good at saying things
that maybe would be inappropriate.
And so they may be at a faster
learning path because of that.
- Mm hmm. Considering the kindof pressure that C-suites
will feel in the environmentthat we are discussing,
let alone the environment of today,
presumably the world of
(17:31):
collective intelligenceisn't all wine and roses.
On that front,
GENE, I believe you have a question.
- Mickey, in our pursuitof AI co-founder harmony,
could we inadvertently normalize
an overreliance on AI,potentially diluting human
decision-making prowess?
- Yeah, yeah, I think that could happen.
(17:53):
You know, if we don't use it, we lose it.
Neurons that fire together wire together.
And so I think there'sa real challenge with
what does apprenticeshiplook like. If everything is
so normalized becauseAI can do this stuff,
I don't end up sweating. Idon't end up building those
muscles for doing those things.
And so I think that's areal challenge in terms
of education. You know,
(18:14):
there's this real cautionary tale
that comes out of the aerospace industry.
When they first startedputting autopilots in planes,
the autopilot would occasionally go,
"Oops, I'm going to stall"
and just hand off to the pilot,
but the pilot would be asleep at the wheel
and wouldn't have enough context
to be able to do something about it.
And so they actually had to
change the way that pilots learn
(18:34):
and change the way thatthey actually hand off so
that it kept the musclebuilt for the human.
So I think that gives us hope
that we could actuallybe building new capacity
over time than we've everbeen able to build before.
But we've got to do it by playing with it.
Like we've got to be in the mix
'cause that's the way our brains learn.
Our brains learn
by being shaped by others
(18:55):
and by being shaped by these things.
- GENE has a question.
- Mickey, considering thehuman-robot C-suite mix,
how do you envision handlingconflicts of interest
or ethics clashes that may arise?
- Yeah, I think ethics
and conflicts of interest,
those are just going to be hard
because, in a sense, ethicaldilemmas are called dilemmas
(19:15):
because they're hard,
and you've got to actually, you know,
have working ethicists.
And I think we're going tosee that as a new role,
not just once a month or oncea quarter or once a year
we make some decision,but like a collection
of people across the organization
that are rolling up their sleeves
and actually having theseconversations in the teams.
(19:37):
But these are going to be hard decisions,
and we're going to have to move
and help shape the norms
for society over this period of time.
You know, the other thingI would suggest the C-level
executives do today is do some reading.
So I think about Iain Banks,
who has since passed away,
but he was a sciencefiction author who came up
(19:59):
with this whole idea of the culture,
and machines actually run all the cities.
Machines run the planets
because humans are justnot so good at that stuff.
But humans are amazing for other reasons.
And it might be worth reading one of,
one of Iain Banks'novels about the culture.
It also might be really worth reading
Joy Buolamwini's "Unmasking AI"
or Safiya Noble's"Algorithms of Oppression"
(20:22):
because I do think we're going to have
to get smarter about those ethical things
and those challenges
and make sure that we help ourorganizations come out
in a more positive wayand build those things.
Don't necessarily read the business books
'cause they're very backward looking,
but try to look at these things
that are more diverse thoughts
that help us understand what's coming.
(20:42):
- Yeah, the CEO's new readinglist sounds a lot different
from the old one, doesn't it?
(theme music)
GENE, we're going to turn towards
wrapping up this fabulous conversation.
What are three stepsthat CEOs can take now
to start preparing for afuture of collaborating
with autonomous AI botsat the highest reaches
(21:05):
of organizational power?
- Three steps for CEOs to prepare
for AI collaborationat the executive level:
First, get your hands dirty with AI.
Try out the tools, attend workshops,
or collaborate with expertsto understand the capabilities
and limitations of AI in your industry.
Second, establish
(21:26):
and imbue a robust ethicsframework that aligns
with your company's values, ensuring
that every AI decision passesthrough this ethical lens.
Third, foster a culture of
continuous learning and agility.
Encourage your teams to adapt,learn, and co-create with AI.
These steps are not justbeneficial, they are crucial.
(21:49):
Start yesterday becausetomorrow is already here.
- Alright, Mickey, thatseems pretty sensible.
Pretty solid. What might you add?
- The one thing I would add is
that I think the C-suiteneeds to be playing together,
and they need to be taking it seriously,
not thinking about itlike the CTO will solve it
or the CIO will solve it.
But I think the team has to play together
and actually try stuff out themselves
(22:12):
and start to build
a collaborative sense of things
so that they have a common language
and they have a commonway of doing this so
that we're all kind of in thistogether and we're learning.
So I'd like to see thatkind of collaboration.
(theme music)
- That was Mickey McManus, senior advisor
and leadership coach at BCG.
Mickey, thank you so much.
- Thanks. I appreciate having the time,
(22:33):
and thank you, GENE.
- Mickey, thank you forstretching our imaginations wide.
And Paul, always a delight to spar
with words in this digital agora.
Until next time, may yourconversations be as rich
and unpredictable as an AIlearning to paint. Farewell.
(theme music)
(22:57):
- Want to know more about howwe built GENE? Stick around
to hear our AI handler,Bill Moore, demonstrate
how we made GENE more creative.
(theme music)
- Hello, listener. My name's Bill Moore,
and I was part of the teamthat helped to design GENE.
And GENE is made up of manythings: large language models,
(23:17):
a speech synthesis model,and a speech-to-text model.
But it's also driven byprompts and parameters.
And so I thought it would beinteresting to explore some
of these aspects today.
And we're going to start GENE up here.
GENE, we're going to be talking today
about parameters in large language models.
(23:38):
Specifically, we'll be talking about
the temperature parameter.
What, what can you tell ouraudience about temperature?
- Certainly. Imaginetemperature as a scale
of volatility in a languagemodel's decision making.
A lower temperature leads
to more predictable, cooler results,
while a higher one brings
hotter, more chaotic outputs.
(23:59):
- So, GENE, you are normally running
at a temperature of one,
and sometimes we'll,
we'll move it up and down from there,
but let's do some experiments.
OK, I'm going to drop yourtemperature down to zero,
and then we'll see how you respond.
OK, GENE, you're now runningat a temperature of zero.
So I have a question foryou. Which came first:
the chicken or the egg?
(24:19):
- Ah, the age-oldquandary, Bill. Logic might
direct us towards the egg.
However, this question is akinto asking, which comes first:
the code or the AI model?
It's a cyclical evolution.
- So what this actuallyshows us is sort of
how these language models work.
They're predicting thenext word based on all
of their training data
(24:40):
and the relationships
that they understand in latent space,
as well as the inputprovided through the prompt.
So with a lower temperature setting,
the model is more inclined
to select the most probable next word.
On the other hand, witha higher temperature,
we increase the chances ofchoosing less probable words.
Let's put your temperature up to a 1.5,
and we'll see how you respond.
Which came first (25:00):
the chicken or the egg?
- Uh, cranking up themercury, are we, Bill?
The answer leans intowhimsy perhaps, shining
beyond the chicken-egg
librinal contra. Viewing the chicken
as a protein sequencing machine
that upgraded egg blueprints,
the mop-feathered,Rubenesque chicken, sporting
flagel-pated croissantantdentition duster, fancied
(25:23):
originating an impressive finale.
- OK, so now GENE's ata pretty high temperature,
and it increases creativity dramatically.
It also increases hallucinations.
So if you're building anapplication that needs to be
grounded as much as possiblein low hallucinations,
you definitely want to berunning a low temperature.
(theme music)
(25:45):
- This episode was madepossible by Mickey McManus
generously sharing his time and insight.
And also by BCG's AI whisperer Bill Moore,
BCG's PodSquad, producer Michael May,
composer Kenny Kusiak,
and sound engineer George Drabing Hicks.
We'd like to stay intouch, so please subscribe
(26:05):
and leave a rating wherever you found us.
And if you'd like to hear more from me,
check out the BCGpodcast "GenAI on GenAI,"
wherever you get your podcasts.
(theme music)
It's 2030
and half the jobs in the C-suite
of one Fortune 500 companyare performed by AI.
Mickey, you're having drinks with the CEO
after work.
What's on her mind?
- Yeah, well, I think she'sprobably had to do a lot
of things differently,
and she's probably had to figure out
how to, how to sort ofplay jazz collaboration
(00:26):
with her machine-learningC-level executives
and her human ones
and how to get them to play together well.
And so I'm actually curioushow she made it there.
- That's Mickey McManus, senior advisor
and leadership coach at BCG
and co-author of "Trillions:
Thriving in the EmergingInformation Ecology."
(00:47):
And I'm your host, Paul Michelman.
Welcome to "Imagine This,"
where we take a trip into the future
that we hope will challenge
the way you think and act today.
On this episode
of "Imagine This," we consider a world
where autonomous AI botsare running the company.
Speaking of which, let meintroduce my AI co-host,
GENE. Good day to you, GENE.
(01:09):
- Good day, everyone.
- GENE, would you please describe
what your role is on this show?
- Well, my role? Think of meas your digital Socrates,
here to ask the probing questions
that peel back the layers. Alongside Paul,
I'm half of the dynamic duogiving our guests a friendly
nudge towards those aha moments.
(01:29):
Hello, everyone. Ready to dive deep?
(theme music)
- A quick programming note:
None of GENE's commentary is scripted.
The final version of thisprogram has been edited
from the originalconversation, which means
that we're not includingeverything we recorded.
That applies to us humansas well as machines.
Mickey, can you take us on abrief trip into the C-suite
(01:52):
of the first company where
half the executives are AI bots?
What does command look like for a bot?
- Yeah, so there are
interesting things going on right now.
I think, in this nextwave with regenerative AI,
it's actually going to beabout ideas and about actions.
And generative AI has aninteresting capacity there.
(02:14):
It can coach. It can becreative. It can command,
and it can wait for feedback
and then actually dosomething about that command.
It can create synthesis.
It can synthesize ideastogether in different ways.
I could draw a pictureof a software interface,
and I can draw some arrows and say,
"Hey, when somebody clickson this button, do this."
(02:36):
And then I can draw anotherpicture of a visualization
and say, "Hey, if this happens, do this."
And I could draw a picture
and say, "I'd love to seea data graph of this."
I could draw that all on a whiteboard.
And I could take a picturewith one of the newer systems,
the generative AI systems,
and say, "Make that into code."
You know, I'm commanding ittoday by drawing something
and giving it some goals.
(02:57):
But what happens when itactually can kind of
continue on that chaining of command
and do a lot of other kinds of things?
And recently, we saw somegenerative AI systems
that basically couldn'tlog in to a website
because of those CAPTCHAs,you know, those things
that show up images and,
and you've got to figure out if
it's trying to test if it's human.
(03:17):
And the AI system actuallywent to basically like
a job site and hired somehumans to actually decode
that thing so that they could then get
through the system and be able to do it.
And it turns out thatthe person who was hired,
it was like a TaskRabbit thing,
the person who was hired said,
"Are you just messing around with me?
(03:38):
Are you an AI?" And itsaid, "No, I'm not."
And so we also have to be aware
that just like thebiases that humans have,
we likely will find these AI systems
not only can take action
but they can take actionin ways that they can,
get somethingdone that they want done,
that they were commanded todo or that they decided to do.
- GENE has a question on that subject.
(04:00):
- Mickey, envisioningthis AI-infused C-suite,
let's ponder theirdecision-making processes.
How might the AI's data-driven logic clash
with human intuition,
and what measures are in place
to harmonize these distinct approaches?
- I'm not sure I know yet.
I think, since I'm sittingdown with the CEO, I'd like
to ask her what, what it'slike in the year 2030.
(04:22):
But you know, I think so much of
what we think we do is logical as humans.
You know, we think we makelogical decisions based on
historical data, and we, we play that out.
But, but so much of what wereally do is based on emotion.
So much of what wereally do is based on
a passionate idea
or a framing of somethingthat actually helps us see
(04:43):
that every word we use islike derived from the things
around us and the way we think.
And so almost everything weuse is a metaphor and analogy.
And so humans aresuperpowered at this.
So even if the AI systemsaid something that
might have been a greatlogical story, it could be
(05:03):
that the human's job isreally to find a thread that
surprises them and that they can pull on
that turns into a powerful story
that can motivate the companyor motivate shareholders
or motivate customers in some new ways.
So I think this goes back toa little bit of the comment
that I made at the beginning,
which is maybe it's alittle bit more about improv
(05:25):
or a little bit moreabout jazz collaboration.
So, you know, I'm justwondering
what we can learn from TinaFey, what we can learn from
Wynton Marsalis in this dynamic
and how much that is going
to turn out to be really important,
which is being fun to play with
and thinking about everything as an offer.
And that's not necessarilysomething we teach today
(05:47):
in business school,
but it might be something that turns out
to be important in the future.
- So what's a piece ofadvice you would give someone
who aspires to be in thisC-suite that we are describing?
What's the piece of adviceyou would give them today
in order to prepare for this world
(06:09):
of improvisation that you're describing?
- I think actually the advice I'd give
to somebody today right nowis mess around. Start playing
with the generative AI functionality.
Look at ways that youcan actually be a centaur.
And so that's oneway of thinking of it is
I do the thing that I'm pretty good at,
and then I hand it off to the horse part,
(06:30):
and it does the thing thatit's pretty good at, you know,
and then it's my upper-bodyand its lower-body strength.
Maybe play with somethingjust on the weekend, just
as part of an experimentation.Play with something
where you're a cyborg, whereit's sort of you don't know
where you end and itbegins, where it's sort
of this play back andforth and back and forth,
and you're tuning it
because you're acting likeit's another teammate.
(06:52):
Not like it's a thing I prompt,
but it's something thatI actually play with.
- So who is the CIO, right,in the C-suite we're
describing? Is there a CIO?
- I think so. I think there is,
and I don't know if it's oneof the ones that's the bot.
There's no reason that weshould think that by 2030,
the best CIO is a human.
(07:13):
And it might actually be
that it might be better that it's not.
One of the things humansdo have is context
and a mental model.
And right now we haven'tseen any of the systems
that are really big and,
and deployed with any kind ofsymbolic contextual reasoning
or mental modeling of the universe.
That's something all lifeon earth gets for free,
but it's because of a
(07:33):
3 billion-year-old
R&D experiment
called life on earth.
So it's been running for a long
time, and it's figured that out.
So I don't know if the CIO is actually one
of the C-suites that's the bot or not.
I do think that when youthink about information,
it is the new oil in some ways,
and I think the CIO isgoing to be much more critical
(07:55):
to understand how tobuild the data factory.
How do we do that in a waythat we can actually capture
more valuable information
and let more businesspeople take advantage of it
and more of the botstake advantage of it
so that we can actually make decisions?
We suddenly have more insightsthan we ever could before,
and it's really hard forhumans to see those insights.
(08:16):
We have bounded rationality.We can't see
5, 10, 20,30 dimensions at once.
We just can't. So it might be
that the CIO's job is, in manyways, to help us cope with
that high dimensionalityand, and cleanse, append,
and build out that data factory
to kind of think about it.
So I'm not sure if they'rea hybrid, if they're a human,
(08:40):
or if they're a bot that does this
because it turns out, you know,
they understand thesystems better than we do.
- Let's talk about the CEO specifically.
In my opening question to you,
I presumed by my pronounchoice that the CEO
was human. Is that a fair assumption?
- I think so. I thinkthe CEO is still human.
(09:02):
And I think we'retalking to her at the bar
and we're trying to findout how she got there
mostly because I think a lot of the role
of the CEO is to set that culture.
I think great CEOs really geta sense of the zeitgeist
of what's happening in their organization.
When I was a CEO, I did alot of MBWA. I did manage
(09:23):
by walking around. I'd wanderaround from team to team,
and I'd be like, "Oh, wow,you're working on this?
Did you know the guys upstairsare working on something
that, that I see a connection to?
You might want to take alittle bit of a field trip."
So while we might be ableto encode those things,
well, we might be ableto create a sort of swarm
of CEO bots so
that the CEO can bemuch more intimate
(09:44):
with a lot more people,
I still think it's probablya human role for a while.
- I do think the notionthat a CEO bot can harness
a diversity of perspective in a way
that a single human could never do
to be quite powerful.
- Mickey, could AI in the C-suite
also amplify our biases
(10:05):
if not carefully checked, potentially
automating systemic issues at scale?
- Well, I wish I could say no,
but I would say that's already happening.
You know, we love training data.
We love moredata. More data. That's good.
So your training data saysthat you pay women 25 cents
for every dollar you pay aman in corporate America.
(10:25):
So the AI is going tosay, "Yeah, double down.
Keep paying them 25 cents.
It's worked out for the last 30 years."
That seems really bad.
- GENE, what differentadvice would you give a CEO,
considering you understand
what it's like to work with humans?
- Ah, Mickey, navigating the role of CEO
with AI cohorts, a delicate dance;
is it not? Here's a thought: Focus less on command, (10:46):
undefined
more on cultivating AIas creative partners.
Encourage a culture where AIsuggestions are not edicts
but conversation starters,
blending the logicalwith the human intuition.
- I do think that we'reseeing early indications
that the systems are actually helping
(11:09):
with creative explorations.They're helping
with prompting us in many ways.
And I think that there'san opportunity there.
One of the projects I workedon a number of years back
was a project for DARPA,
and it was about logistics planning.
And the system was could weactually tap into retired
generals that could, ata glance, look at things
(11:29):
and go, "Hey, you haven'tthought about this"
to help people with thefog of war, for example,
or to help people with movinglarge amounts of logistics.
But then we also were playing with AI bots
that could actually play out
and go like, "Nobody'sever done it that way.
And you're planning to do it that way?
Like we just want to flag, youmight not be able to pull
that off 'cause no human hasever pulled that off before."
(11:50):
And it led to some reallyfascinating conversations
because sometimes it was a retired person
who just had this crazy wisdom,
and they were like, "Whoa, youknow what's going to happen?
Like that helicopter, there's always
a flight ceiling of this
in that area of the mountains.
It's almost impossible to getthrough unless you do this."
And then this, the bot wouldalso be able to sort of say,
"Hey, look. Here'sthe data. Just look at it."
(12:13):
And maybeit's to what GENE said:
What if it could say,"Nobody's ever done this,
but this team did this,and this team did this,
and this team did this, and nobody's ever
assembled it this way before"?
What if you were to do that? Because if we
actually assembled that, we'd get
something new which might be novel
new approaches, which wejust would never have found
(12:33):
'cause it'd be like findinga needle in a haystack.
And so I do think theremight be an opportunity there
for that to happen as well.
- So is the AI just makinga suggestion to connect
or maybe going right ahead and connecting?
- Well, I think by 2030 both will happen.
I think that goes backto that command part.
I'm going to begin trusting and,
right at the beginning ofthis, we talked about trust.
(12:54):
And it turns out if youread Wynton Marsalis'
biography, when he improvises,
he said, the thing is, the one thing,
the quote I love out of that,
you have to learn how totrust the other players
because when you trust, you listen.
And when you listen,
you can actually take it to new places.
So trust is going to turn out to be
(13:15):
one of the most important things
between now and 2030,is trusted AI systems
that I can take to the bank.
I know it's not going to suddenly
go off the rails,
which is just the same as with working
with a great person in your company.
You know, you give themnudges, they give you nudges,
and you get to some pointwhere you're like, "Look,
(13:36):
I trust Paul for doingthis kind of stuff."
- We're going to take a quickbreak. Coming up,
a closer look at
how we match the rightjobs to the right vessel.
But first, our AI handler, Bill, gives us
a peek behind the scenes.
(theme music)
(13:56):
- Hi. This is Bill Moore from BCG.
If you are interested intaking a peek under the hood
of our AI co-host, GENE, stickaround after the end credits.
In this episode, we'llexplore AI model parameters
and learn how small changes
can make a big difference in the
creativity and accuracyof AI model outputs.
(14:17):
(theme music)
- Welcome back to "Imagine This..."
I'm your host, Paul Michelman,
and we're talking with BCGSenior Advisor Mickey McManus
about a future withautonomous AI executives.
Mickey, we've seen howpowerful generative AI can be.
(14:38):
What is the next evolution of AI,
and how will it be used in the C-suite?
- Generative AI systemsare now starting to be used
with larger-scale systems,
and they're starting to lookat what's called "augmented
collective intelligence,"where it's teaming the best
of people and the best of machines,
and it's building it intoa system model of the world
(15:00):
or a system model that cantake into account things like
sustainability impacts
or things like market dynamics
and things like geographyand things like shipping
and all the other things, logistics.
That could be really powerful.
How could we run into the future
and help our C-suite reallyplay with that future
and then backward chain totoday so that we had some
(15:23):
directionality over that future,
and we could prototype ittoday and learn where to go.
- Yeah, and unlocking that,right, the ability to do
that consistently wouldbe profoundly powerful.
I'd love this notion of
the collectively intelligent C-suite.
Can we drill down into how
that will manifest in thiscompany, maybe in a role
that we haven't discussed yet?
(15:44):
How about the CMO?
- Yeah, I think the CMOis an interesting one.
You know, they're probablyseeing the most early wins
from generative AI
because it's one thing
to build a marketing campaign
to do a hypertargeted approach to things,
to have algorithmic personasso that you can do, you know,
an email campaign or you cando some kind of a thing.
(16:06):
And I'm actually seeingsome of the CMOs say,
"Wait a second.
Why are we hiring 22 agencies
when I could maybe havea few people using AI
and these other systemsand do as well?"
And so I see that happening
and I certainly know some CMOs
that are really pushing thelimits on what's possible,
and I've seen them challenge
(16:27):
and say, "Look, we normallytake seven months
to plan out this transmedia experience."
And they've now got, I talked to one CMO,
and they've got a tool thatlets them do it in seven
hours for a rough approximation.
And she said, "Well, I'mactually still going to want
to take seven months because
I want you now to focus on wonder.
(16:49):
I want you now to focus on delight.
Like how does the brand come through
in the channels we use?
How does the brand come through
in the interactions people have?
Let's really turn the knobs up on that."
The CMO might also havesome of the biggest
things explode in their facein the next year or two
just because these systems arepretty good at hallucinating
(17:10):
and pretty good at saying things
that maybe would be inappropriate.
And so they may be at a faster
learning path because of that.
- Mm hmm. Considering the kindof pressure that C-suites
will feel in the environmentthat we are discussing,
let alone the environment of today,
presumably the world of
(17:31):
collective intelligenceisn't all wine and roses.
On that front,
GENE, I believe you have a question.
- Mickey, in our pursuitof AI co-founder harmony,
could we inadvertently normalize
an overreliance on AI,potentially diluting human
decision-making prowess?
- Yeah, yeah, I think that could happen.
(17:53):
You know, if we don't use it, we lose it.
Neurons that fire together wire together.
And so I think there'sa real challenge with
what does apprenticeshiplook like. If everything is
so normalized becauseAI can do this stuff,
I don't end up sweating. Idon't end up building those
muscles for doing those things.
And so I think that's areal challenge in terms
of education. You know,
(18:14):
there's this real cautionary tale
that comes out of the aerospace industry.
When they first startedputting autopilots in planes,
the autopilot would occasionally go,
"Oops, I'm going to stall"
and just hand off to the pilot,
but the pilot would be asleep at the wheel
and wouldn't have enough context
to be able to do something about it.
And so they actually had to
change the way that pilots learn
(18:34):
and change the way thatthey actually hand off so
that it kept the musclebuilt for the human.
So I think that gives us hope
that we could actuallybe building new capacity
over time than we've everbeen able to build before.
But we've got to do it by playing with it.
Like we've got to be in the mix
'cause that's the way our brains learn.
Our brains learn
by being shaped by others
(18:55):
and by being shaped by these things.
- GENE has a question.
- Mickey, considering thehuman-robot C-suite mix,
how do you envision handlingconflicts of interest
or ethics clashes that may arise?
- Yeah, I think ethics
and conflicts of interest,
those are just going to be hard
because, in a sense, ethicaldilemmas are called dilemmas
(19:15):
because they're hard,
and you've got to actually, you know,
have working ethicists.
And I think we're going tosee that as a new role,
not just once a month or oncea quarter or once a year
we make some decision,but like a collection
of people across the organization
that are rolling up their sleeves
and actually having theseconversations in the teams.
(19:37):
But these are going to be hard decisions,
and we're going to have to move
and help shape the norms
for society over this period of time.
You know, the other thingI would suggest the C-level
executives do today is do some reading.
So I think about Iain Banks,
who has since passed away,
but he was a sciencefiction author who came up
(19:59):
with this whole idea of the culture,
and machines actually run all the cities.
Machines run the planets
because humans are justnot so good at that stuff.
But humans are amazing for other reasons.
And it might be worth reading one of,
one of Iain Banks'novels about the culture.
It also might be really worth reading
Joy Buolamwini's "Unmasking AI"
or Safiya Noble's"Algorithms of Oppression"
(20:22):
because I do think we're going to have
to get smarter about those ethical things
and those challenges
and make sure that we help ourorganizations come out
in a more positive wayand build those things.
Don't necessarily read the business books
'cause they're very backward looking,
but try to look at these things
that are more diverse thoughts
that help us understand what's coming.
(20:42):
- Yeah, the CEO's new readinglist sounds a lot different
from the old one, doesn't it?
(theme music)
GENE, we're going to turn towards
wrapping up this fabulous conversation.
What are three stepsthat CEOs can take now
to start preparing for afuture of collaborating
with autonomous AI botsat the highest reaches
(21:05):
of organizational power?
- Three steps for CEOs to prepare
for AI collaborationat the executive level:
First, get your hands dirty with AI.
Try out the tools, attend workshops,
or collaborate with expertsto understand the capabilities
and limitations of AI in your industry.
Second, establish
(21:26):
and imbue a robust ethicsframework that aligns
with your company's values, ensuring
that every AI decision passesthrough this ethical lens.
Third, foster a culture of
continuous learning and agility.
Encourage your teams to adapt,learn, and co-create with AI.
These steps are not justbeneficial, they are crucial.
(21:49):
Start yesterday becausetomorrow is already here.
- Alright, Mickey, thatseems pretty sensible.
Pretty solid. What might you add?
- The one thing I would add is
that I think the C-suiteneeds to be playing together,
and they need to be taking it seriously,
not thinking about itlike the CTO will solve it
or the CIO will solve it.
But I think the team has to play together
and actually try stuff out themselves
(22:12):
and start to build
a collaborative sense of things
so that they have a common language
and they have a commonway of doing this so
that we're all kind of in thistogether and we're learning.
So I'd like to see thatkind of collaboration.
(theme music)
- That was Mickey McManus, senior advisor
and leadership coach at BCG.
Mickey, thank you so much.
- Thanks. I appreciate having the time,
(22:33):
and thank you, GENE.
- Mickey, thank you forstretching our imaginations wide.
And Paul, always a delight to spar
with words in this digital agora.
Until next time, may yourconversations be as rich
and unpredictable as an AIlearning to paint. Farewell.
(theme music)
(22:57):
- Want to know more about howwe built GENE? Stick around
to hear our AI handler,Bill Moore, demonstrate
how we made GENE more creative.
(theme music)
- Hello, listener. My name's Bill Moore,
and I was part of the teamthat helped to design GENE.
And GENE is made up of manythings: large language models,
(23:17):
a speech synthesis model,and a speech-to-text model.
But it's also driven byprompts and parameters.
And so I thought it would beinteresting to explore some
of these aspects today.
And we're going to start GENE up here.
GENE, we're going to be talking today
about parameters in large language models.
(23:38):
Specifically, we'll be talking about
the temperature parameter.
What, what can you tell ouraudience about temperature?
- Certainly. Imaginetemperature as a scale
of volatility in a languagemodel's decision making.
A lower temperature leads
to more predictable, cooler results,
while a higher one brings
hotter, more chaotic outputs.
(23:59):
- So, GENE, you are normally running
at a temperature of one,
and sometimes we'll,
we'll move it up and down from there,
but let's do some experiments.
OK, I'm going to drop yourtemperature down to zero,
and then we'll see how you respond.
OK, GENE, you're now runningat a temperature of zero.
So I have a question foryou. Which came first:
the chicken or the egg?
(24:19):
- Ah, the age-oldquandary, Bill. Logic might
direct us towards the egg.
However, this question is akinto asking, which comes first:
the code or the AI model?
It's a cyclical evolution.
- So what this actuallyshows us is sort of
how these language models work.
They're predicting thenext word based on all
of their training data
(24:40):
and the relationships
that they understand in latent space,
as well as the inputprovided through the prompt.
So with a lower temperature setting,
the model is more inclined
to select the most probable next word.
On the other hand, witha higher temperature,
we increase the chances ofchoosing less probable words.
Let's put your temperature up to a 1.5,
and we'll see how you respond.
Which came first (25:00):
the chicken or the egg?
- Uh, cranking up themercury, are we, Bill?
The answer leans intowhimsy perhaps, shining
beyond the chicken-egg
librinal contra. Viewing the chicken
as a protein sequencing machine
that upgraded egg blueprints,
the mop-feathered,Rubenesque chicken, sporting
flagel-pated croissantantdentition duster, fancied
(25:23):
originating an impressive finale.
- OK, so now GENE's ata pretty high temperature,
and it increases creativity dramatically.
It also increases hallucinations.
So if you're building anapplication that needs to be
grounded as much as possiblein low hallucinations,
you definitely want to berunning a low temperature.
(theme music)
(25:45):
- This episode was madepossible by Mickey McManus
generously sharing his time and insight.
And also by BCG's AI whisperer Bill Moore,
BCG's PodSquad, producer Michael May,
composer Kenny Kusiak,
and sound engineer George Drabing Hicks.
We'd like to stay intouch, so please subscribe
(26:05):
and leave a rating wherever you found us.
And if you'd like to hear more from me,
check out the BCGpodcast "GenAI on GenAI,"
wherever you get your podcasts.
(theme music)
Popular Podcasts
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations.
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
The Nikki Glaser Podcast
Every week comedian and infamous roaster Nikki Glaser provides a fun, fast-paced, and brutally honest look into current pop-culture and her own personal life.