Subscribe
Share
Share
Embed
Before a strategy becomes a success story, it's a judgement call.
Now in its second season, Founder's Mentality: The CEO Sessions gets inside the decisions that defined some of the world's greatest leaders - a bet the market couldn't yet see, a crisis that put the organisation's purpose to the test, and the courage to keep showing up even when you don't have the answer.
How do you stay close to the frontline as the organisation scales?
What does AI mean for the thing you have spent years building?
How do you transform the company without breaking what makes it great?
Hosted by Jimmy Allen, Advisory Partner at Bain & Company, bestselling author and leader of Bain's Global CEO Forum, each episode is a candid, in-depth conversation with leaders from Mars, The Economist Group, Khan Academy, and more.
Each episode is built around one real story, one defining moment, and the lessons that come from living through it. Dense with insight and designed for leaders who listen with a pen in hand, the series explores the outer game of markets, technology and growth, and the inner game of energy, humility and courage, and what it takes to master both.
The question this season isn't just what the CEO should do.
It's who does the CEO need to become?
Join the Conversation:
https://www.bain.com/founders-mentality/
https://www.bain.com/insights/topics/ceo-agenda
Links:
Bain & Company LinkedIn (https://www.linkedin.com/company/bain-and-company/)
Bain & Company X (https://x.com/BainandCompany)
Jimmy Allen LinkedIn (https://www.linkedin.com/in/james-allen-3442b/)
About the Host:
Jimmy Allen is an Advisory Partner at Bain & Company with over 35 years’ experience advising leading organizations. He’s the author of multiple best-selling books on growth and leadership and the host and founder of Bain’s Global CEO Community Forum. Jimmy is a regular speaker at global business events, including the World Economic Forum, and serves on the Botswana Economic Advisory Council. Outside of consulting, Jimmy started his own record label (Abubilla Music) in 2008 and supports Singing Wells, a project dedicated to preserving Kenyan village music.
Bain & Company:
Founder’s Mentality: The CEO Sessions is brought to you by Bain & Company, a global consultancy trusted by the world’s most influential business leaders. With decades of experience guiding organizations through growth, transformation, and leadership development, Bain’s executive insights offer what it takes to lead at scale.
- I was pretty convinced that
if we didn't lean in here,
we would be doing the world a disservice
because it was clear that
once all of this came out,
a million and one people
were gonna leverage it,
but they weren't going to
use it in a trustworthy way.
- We usually ask our guests
to take us back to a single moment,
a decision, a challenge,
a pivot of some kind,
and we find something from
the past and then move forward
to discover how this
shaped them as a leader.
This episode is a little different.
There isn't a past moment
because our guest is still in it,
and whether we like it or not, so are we.
(upbeat music)
I'm Jimmy Allen,
and this is "Founder's
Mentality: The CEO Sessions."
Sal Khan is the founder
and CEO of Khan Academy,
one of the world's largest
free education platforms.
200 million users,
hundreds of courses from
pre-K to early college,
in over 50 languages,
all built on one vision:
world-class education,
free for everyone anywhere.
Let's get this out upfront.
Yeah, this is gonna be about AI.
I mean, every podcast
seems to be about AI,
but we're talking about AI in education.
Gulp!
I mean, because with AI, today
students can get an answer
to anything from a chat
window instantly for free.
It might not be right.
Certainly won't help them
learn, but it is instant.
What does that mean for Khan Academy?
What does that mean for learning itself?
Every leader, every organization,
every team is facing their
own version of that question.
We are all trying to
figure out what AI means
for the thing we've spent years building.
Does it enhance what
we've done or replace it?
These questions are scary, man.
And the fact that none of us
can be certain of our
answers is scarier still.
But from where Sal stands,
in the middle of it, he sees
something extraordinary,
and he's ready to grab it now.
And I don't know if too many
other leaders so willing
to move so fast to disrupt
themselves through AI.
And I think we can all
learn from his journey.
- I got an email from Sam
Altman and Greg Brockman,
who I was acquainted with.
You know, we weren't like best buds.
They sent me an email saying
they're training their next model
and they thought it could be of interest.
You know, we get, I don't
know, 10 or 15 inbound requests
maybe every day.
But this one, you know,
they were legitimate folks.
I had been tracking the GPT-2, GPT-3.
So I just was curious what
it would be able to do.
This was five, six months
before ChatGPT came out.
And two weeks later, once
the training had done,
it was actually on a Zoom or Google Meet,
and it was myself, our
chief learning officer,
and it was Greg.
And he showed me an AP Biology question
and he said, "Sal, what do
you think the answer is?"
I was like, "Okay, you
know, it's C, osmosis."
And it was able to answer the question.
So I'm like, "Okay, that's interesting.
Ask it to explain the reasoning."
And it did.
Ask it to explain why the
other choices weren't correct.
It did.
Ask it to write another
question like this. And it did.
And that's when I was like,
"Okay, am I being punked?
Is this a real thing?"
And Greg said, "Hey, would
y'all like some access to this
to play around with it?"
And I said, "Absolutely, we would."
And that weekend, we had
access to it over Slack
and we were chatting with it over Slack.
And myself and our chief learning officer,
we couldn't sleep.
We were also Slacking each other
in the middle of the night.
Hey, like, I got it to do
this, I got it to do that.
And just to give you a
sense of how early this was,
right when they gave us access,
I asked Greg, "Does this
only work in English?"
And he said, "Yeah, it
only works in English."
And I was just curious.
I barely speak Bengali.
I was born in New Orleans,
but my family's originally from Bengal
and India and Bangladesh.
And I tried to speak to it in Bengali
and it responded back to me in Bengali.
And so I said, what?
So I took a screenshot of it
and I sent it back to Greg.
And he said, "Oh yeah, after you asked,
we realized it's speaking
every language we tested."
And we're all used to all of this now,
but in late summer, early fall of 2023,
this was mind-blowing.
And so we got more and
more of Khan Academy's team
under a non-disclosure agreement,
started getting access to it.
We started having all the debates that,
frankly, people are still having.
This could be a cheating tool.
This could be inappropriate conversations.
This can hallucinate.
This is making a lot of
really basic math mistakes.
But it could also answer
some cool questions
and motivate people.
And that's where I really
pushed the team that, look,
we've got to turn these
fears into features.
They're real fears, but
let's put guardrails.
Let's have oversight.
Let's see if we can reduce the error rate.
Make sure our tutor that
we create is Socratic.
It doesn't cheat.
It drives critical thinking
as opposed to having cognitive offloading.
And maybe we can use it
to be an extra layer on Khan Academy.
- Subsequently, over
time, you've talked about
how revolutionary this will be,
and AI has led to a big pivot.
But when did you see it for Khan Academy?
At what point did you say,
this is one of these moments
that I'm gonna have to
mobilize significant resource
and make this a thing?
- Oh, honestly, within a week.
It was probably only about
five people at Khan Academy
even had access to it at that point.
I knew this was going
to change everything.
This was going to change society.
Never in my life, and I'm not
sure if ever again in my life
am I going to get a preview like that
where, you know, it wasn't clear
if the president of the United States
knew that this was going to come.
So it really felt like
I was making contact
with an alien life form.
It felt like this changes everything,
and I had to go all in and
help the team make that change.
- So from that then you began to say,
how do I bring my team on board?
And the tension at least
I would've seen is,
if we had dialed back
to before that meeting,
one of the things that was
so critical to the brand
was these trusted teachers
within the videos,
very curated, very cultivated.
And suddenly now you're
introducing something else,
which is almost the opposite
of a standard set video.
Was everybody aligned with you immediately
or were there people that were saying,
wait a minute, this feels against
what we've been trying to do?
- It was less about the videos.
We didn't view this as
a video replacement.
At the time we were
debating this internally,
that wasn't the main concern.
The main concern was hallucinations.
Khan Academy, people go
to for trusted content.
There's a million people
teaching on YouTube.
Some of it's quite
good. A lot of it isn't.
But I think the trust, the consistency
people really value on Khan Academy.
And so if we released an
AI that could hallucinate,
makes math errors, we
realized that could be a risk.
We're dealing with under-18
students in classrooms.
What kind of conversations
could you have with this AI?
They could try to cheat with this AI.
What if they were talking about
suicide or things like that?
What's the responsibility that we have?
You know, what I encouraged our team,
I was like, "These are real issues.
I also feel them, but let's
see how far that we can get."
This technology even then was
getting dramatically better.
Even under NDA, OpenAI
was sending us a new model
almost every three, four
weeks that was getting better.
We were sending them information.
I actually think we helped
them discover some mistakes
in their initial training data
that helped them fix some of
the math errors quite quickly.
One interesting thing,
we were having a little Christmas party,
and one of my family friends
is a researcher at Google in AI.
And he's like, "You know,
Khan Academy showed us a lot."
I'm like, "What have we showed you?
Y'all are leading-edge researchers."
He's like, "You showed us
the power of prompting."
And it tells you that a
lot of these researchers,
they're like, more parameters,
better training data.
And they knew prompting was a thing,
but they didn't realize how far
or how powerful you
could get with prompting.
And it took us to say like,
"Okay, we really want to stress test this.
Can we get the error rates down?
Can we get the hallucinations down?"
But I was pretty convinced
that if we didn't lean in here,
we would be doing not just ourselves,
but maybe the world a disservice
because it was clear that
once all of this came out,
a million and one people
were gonna leverage it,
but they weren't going to
use it in a trustworthy way.
They weren't going to care
as much about cheating
or hallucinations or safety.
So that was our job.
And that's what I told our team.
We have a very mission-driven team,
and it's never about
what's gonna make Khan Academy do well.
We're a nonprofit.
It should be, what does the world need?
And I think that was and
actually continues to be
the most convincing thing
when I talked to the team.
- There are two ways to respond
when a technology arrives
that disrupts everything we've built.
We can fight it or we can ask,
what does the world
actually need from us now?
What do our customers want us to do?
How can we make this disruption
good for our communities?
Look, Sal identified the negatives of AI
to his business, to education.
He did this quite quickly.
Hallucinations, cheating,
the erosion of trust.
I mean, he named them all,
but then he made a choice
to turn every single one of
them into an opportunity.
Almost a design challenge.
Turn the fears into features.
And to do that, he had to bring
his organization with him.
- I think there were definitely people
who just didn't want to move
aggressively in this direction.
I've learned over the years
that instead of shying away
from that conversation,
it's better to just make it very clear
that like, look, we 100%
recognize these risks.
We're not gonna do anything irresponsible.
I plan on devoting the rest
of my life to Khan Academy,
which is hopefully many
decades in the future.
The last thing I
personally would want to do
is to jeopardize our
credibility, my credibility,
but we've got to move forward.
And if people are up for
doing it in a responsible way
and moving forward,
this is going to be probably
a very interesting journey.
And if they're not, that's okay.
But this might not be the
ship to be journeying on.
And I think that helped
align a lot of folks.
And then we did start to
build a little tiger team
in those early days to do
some very rapid prototyping.
We did a lot of experimentation
in September, October.
November is when we started
the real rapid prototyping,
stuff that we could
actually potentially deploy
and test with people.
And it was from then to about
January where we're like,
"Oh yeah, we can actually
launch something."
They were going to launch
GPT-4 on March of 2023.
And another interesting thing,
you know, these things
are very unpredictable.
November 29th, this thing
called ChatGPT comes out.
And we were all lawyered up with NDAs.
We couldn't tell anyone
anything about this stuff.
And I immediately Slacked Greg Brockman.
Actually, we had a phone call about it.
I'm like, "Hey, you said
you weren't gonna launch anything.
You just launched this thing
that looks a lot like this chatbot."
And he said, "No, we didn't think
we were launching anything.
We were putting in some apps
on our old model, on GPT-3.5,
which had been out for several months,
but no one really noticed."
But ChatGPT like woke up
everyone to what's possible.
And I was really worried
because it had no guardrails.
It was an inferior model to
the one we were working with.
It hallucinated more.
It had more math errors.
It was clear people were gonna
use it as a cheating tool.
When that went out, I was pretty bummed.
I said, "Oh, the whole
education establishment,
this is their first impression.
They're gonna think it's a cheating tool.
It's error-prone, et cetera, et cetera."
But in some ways it was a blessing,
even with the internal team,
because I was able to say, look,
the cat's already out of the bag.
Someone's got to do this
in a responsible way.
And by the time we launched in March,
I think people were actually hungry.
The education community
was hungry for that.
And so having that tiger team
doing more and more and more,
having this external pressure
that like someone's got to
be a responsible actor here.
The responsible actor doesn't mean
wait two years until you deploy something.
- And even just this tiger team,
where did the tiger team emerge from
that could pivot that fast?
Because that wasn't
necessarily your model.
You had an extraordinary repeatable model
that you were doing.
I know there was a lot of new videos
and software development going on.
But did you have a tiger team
able to work at that pace
readily available in the organization?
- At the time, we were about
a 200-person organization.
We had definitely professionalized,
and as you said, been a more repeatable
and consistent organization,
but I think we've always
been fairly entrepreneurial
and fairly nimble.
But it was a bit of a push.
And for me, it was obviously
I was feeling that energy,
but I didn't want to
rock the boat too much.
But there were definitely team members
who were feeling what I was feeling
and some that were feeling
that we needed to do even
more, move even faster.
And I was able to tap
into several of their,
you know, a lot of their energy.
And it was initially seven or eight people
and they were working very
closely with the OpenAI team
that really started
making a lot of progress.
And I would say it's by
January of 2023 that we said,
"Okay, by the GPT-4 launch,
in the next month or two,
we think we put the right
guardrails in place.
We think we can..."
We didn't eliminate the errors yet,
but we think we, okay,
we could warn the user,
okay, this is a test,
you know, but this is
available as an extra resource,
that we would have enough to get it out.
- And I buy into small,
nimble organization,
but you are now at the bleeding edge
of something that the metabolic rate
is beyond anything the world has seen.
Has that permeated into the organization?
- I would say in the last six months,
it's hit an inflection point,
step function, whatever
you want to call it.
And we definitely had, even
five or six months ago,
about 10 or 15 primarily engineers,
but even a few designers
and product managers
who really leaned into this.
And seemingly we're already getting
5, 10X productivity gains from it.
So we have been, we're actually
in the midst of it right now
thinking about how do we
accelerate that further?
And I tell our team, my goal,
because you can imagine there's
some worry about job loss
or identity loss in all of this.
And I tell our team, look,
I don't own Khan Academy.
It's not like I take surplus profit
and it gets dividended
out to shareholders.
No one owns Khan Academy.
We have a very big mission.
You know, we're almost laughably small
with a laughably small budget
considering our mission.
- I was shocked. I was shocked.
- You know, free world-class
education for anyone anywhere,
which is arguably the mission
of the global education system.
And we have a real budget.
It's about 100 million a
year, but it's the same budget
as a large high school
in the United States.
So I've told our team,
if you can automate
yourself out of your role,
trust me, you're not gonna get laid off.
We're gonna give you a better role.
You're gonna get a better salary.
And by the way, if you do that,
there's probably gonna be 10 people
who want to double your salary.
You know, I hate to tell you,
people are gonna try to poach you.
And I've also said that like, you know,
our fiduciary responsibility
is to the mission and to our funders.
And what we need to do
is hopefully with that
same hundred million,
if we can have three times the impact,
five times the impact,
that's what I want to do.
There's no shortage of work for us to do,
so no one should be
worried about their jobs.
And I still tell the team this,
you want to be in an organization
that is maybe making us
all a little uncomfortable.
I get a little uncomfortable
with all of this,
but you'd rather have a little discomfort
and be moving along than
being complacent about it.
And then all of a sudden
the world completely
changes in a year or two
and your skills are obsolete.
- It's too often
categorized as a will issue.
Do my people have the will to adopt?
I think it's a skill issue.
You know, you cannot have such a profound
and dramatic change in society
without also creating discomfort.
And I think we're gonna have
to learn how to learn again.
You can introduce Claude
into organizations,
but then people have to
prompt the right way.
Then they have to respond to
the output and prompt again.
And what you're discovering is
you're reintroducing the art
of how to ask the right
question, how to be curious.
And I'm interested, you
know, as you've rolled out
and begun to get early
lessons from Khanmigo,
whether that's one of the things
that's happening with adoption
is it takes a long time
to figure out how to
ask the right questions,
and we're gonna have to
teach people how to learn.
- You're right.
There is actually a parallel between
what you see in classrooms
and what you see in...
What we learned,
you know, I am someone
who is naturally curious.
I feel like my learning has
been accelerated dramatically,
not just by Khanmigo, but
by having access to ChatGPT
and Gemini and Claude.
You know, I've been going deep on issues
that I've never been able to go before.
I've been able to learn
what I need to prepare
for videos much faster
than I've had to.
You know, really going to edge questions,
and I'm asking hopefully deep questions,
the right questions.
And there's maybe a slightly
naive assumption that, hey,
you add this to existing
Khan Academy as a supplement.
And wow, the students who are stuck
are just gonna get their
questions answered.
And not in a way that just
gives them the answer,
but Socratically push them,
build their critical thinking.
And we try to do that
same thing with Khanmigo.
It doesn't give the answer,
but it turns out that a lot of kids,
actually, some kids just don't like that.
Hey, they ask how to do this question
and they say, "What do you
think is the first step?"
"Oh, I got to think now."
Or, "Why do you think that worked?"
Or, "Can you give me the
meaning of this phrase?"
People don't like to
think. It takes effort.
But even more we found,
people didn't know how to
construct what they needed.
And at first, we thought this
was a flaw purely of the AI,
but the more we talked to teachers,
they said, "No, a lot
of these students just,
when I ask them, 'Hey, can I help you?',
they say, 'No, I'm fine.'
But then they clearly need help,
but they don't know how
to construct that help."
And there's been some press lately.
I talked to the reporter,
I think it was the Wall Street Journal,
and I said, "You know, for a lot of kids,
Khanmigo was kind of a non-event."
And they took that to mean
that like, "Oh, it didn't work.
AI is dead in classrooms."
I was like, no, no,
that's not what I meant.
It's just we learned that
about for 10 or 15% of kids,
just as for about 10 or 15% of employees,
they're willing to try things
out, learn, dig into it,
and it's a game changer for them.
But for everyone else, we
realized you need more support,
you need to make it more proactive.
You know, the first wave
of AI in the classrooms
was akin of me going into a classroom
and saying, "Hey everyone, my name's Sal.
I'm a pretty good tutor.
I'm gonna go sit in the back of the room.
Come ask me any questions
if you have any."
Not many kids, probably 10 or
15%, would take me up on that.
Same thing happened with the AI V1.
But we're not saying, "Oh,
therefore AI doesn't work."
Our take is we have to make it deeper.
And we're having conversations about,
hey, let's use Khanmigo
not as just a support
to help you answer traditional questions,
but it allows us to go
deeper with the questions.
If we had a multiple
choice question before,
maybe we can still have that
multiple choice question.
But to really get credit for it,
you also have to explain
your reasoning to Khanmigo.
And Khanmigo says, "Okay, yeah, Jimmy,
I think you really get this idea."
That builds more critical depth.
And it's the opposite of the narrative
that everyone has about AI
that's like cognitive offloading.
I do think there is cognitive offloading
if you just go to ChatGPT
and have it do your homework.
But in a case like this
where the AI is pushing your thinking,
making you go beyond the multiple choice,
we're very excited and we've
invested a lot this past year.
We're about to launch it
for this coming school year
where it's gonna be more deeply integrated
for more students.
- I'm all into your original vision,
free education for everyone.
Seems pretty bold enough,
but there's something really fascinating.
You are at the cutting
edge of teaching people
how to thrive in AI,
not be replaced by AI.
Your exact description is if
you learn how to ask questions,
if you learn how to get in there
and ask the first, second, third question,
it becomes a productivity tool.
If you don't learn and
you use it to cheat,
it's a replacement tool
because it figures out it
can do your job a lot better.
You must feel like someone's
got to get in there
and teach people how to thrive with AI
because there's gonna be a lot of forces
that are gonna teach people
how to be replaced by AI.
And this must feel like
a pretty urgent mission.
- You're right.
I mean, there's obviously
the internal stuff
we're trying to do at Khan Academy.
Then we're trying to work with
the broader education system.
There's three things
about learning with AI.
There's learning about AI,
like what's a large language model?
What's a neural net? What's a parameter?
What's training data?
Then there is learning
other topics using AI.
So okay, we want you to learn algebra.
Maybe AI can help you
learn algebra better.
And then there is learning
to use AI generally.
And so that's the vibe coding,
the Claude Cowork, et cetera.
And that one,
actually, that one's the
fastest-moving target.
It's evolving so, so fast.
- Guys, let's capture Sal's
promise to his people.
If you can automate
yourself out of your role,
you're not gonna get laid off.
We're gonna give you a better
role and a better salary.
Is this naive or is this exactly
what leadership means in a world of AI?
AI is a huge topic right
now in our CEO forums.
We've had over 80 sessions on it.
And so far, the discussion
breaks into really
three separate journeys.
The first is the journey
each of us is having with AI,
to learn, to make
ourselves more productive.
And here, the world of
CEOs is indeed divided.
Some, like Sal, well, they're all in.
Sal's decided that to
lead in a world of AI,
he must understand it
and how it impacts his own productivity.
And many CEOs are with him here.
Some CEOs are doing AI
to their organization
with no understanding of how
it actually impacts the human.
I don't think any CEO can afford
to be in this second group.
It's not leadership. It's abdication.
The second journey
is about how you improve the
organization's productivity.
And look, here as an old
guy, I have this to offer.
For productivity, a lot of
the playbook is very familiar.
This is re-engineering,
but with profoundly more powerful tools.
Of course, you need to go
after productivity
improvements through AI,
but you'll get the best results
when you're improving
known repeatable models,
when you've codified specific
things into known routines.
If you input a mess, AI will
just give you a bigger mess.
Worse yet, it'll dress up that
mess to look kind of pretty,
until the dance starts,
and that dance partner won't swing.
Most companies are getting on with this,
but most struggle with one thing.
And that's the Sal promise.
What do you tell the folks
that are busy teaching AI to do their job?
Do you promise them something better?
Or are you waving goodbye?
This is a really big problem,
and it's filling our newspapers
as I speak these words.
Oh, and as an aside,
beware of the AI black
box snake oil salesman.
We saw them during
re-engineering, and they are back.
And then there's the third journey.
Are we using AI
to fundamentally improve
our competitive situation?
If we're a consumer goods company,
how does AI make us better
at core consumer insight?
How does it help us
improve the profitability
of our retail partners?
If we're a chemicals company,
how does AI help us to innovate faster,
to bring better solutions
to our customers?
And here we're seeing the following.
Unless you were born digital,
few companies are making real progress.
Oh, I mean, there's a lot of AI busyness.
There's a lot of AI teams doing AI things.
If you're feeling a kind of deja vu,
just replace the word AI with agile.
We've seen this movie before.
Folks are busy, but no one is trying
to fundamentally change their
strategic position through AI.
And that is why Sal's
story is so important.
His journey with AI is asking
a fundamental question.
And the answer is learning
how to learn again.
And the more he leans in,
the more uncomfortable the journey is,
because uncomfortable is exactly
where all of us should be.
The discomfort means we're moving.
When he thinks about AI in education,
he doesn't start with the technology.
He references Socrates.
That's the context he's working from.
And this framing reminds me of
Don Katz, founder of Audible,
who is featured in episode
three of season one.
When he built Audible,
it was so disruptive
that it looked to a lot of people
like the end of literature.
And someone asked him,
"Aren't you destroying books?"
And his answer was this,
"Knowledge has been transmitted
orally for 100,000 years.
It's only since the Gutenberg Bible
that we've been doing
it with print at scale.
Audible isn't trying to
destroy storytelling.
It's helping to highlight the
importance of the human voice,
because when you think in centuries,
the present moment looks different."
- In the early days, Khan
Academy, some people know,
started with me while I was
working at a hedge fund,
tutoring my cousins.
I started writing software,
videos for them, et cetera.
What I realized we were building,
especially as more people were using us
and we got our first
philanthropic funding,
is trying to approximate
leveraging technology,
real personalized mastery education.
And personalized mastery education
sounds like a fancy thing,
but this is arguably what
humanity has always known
as the best type of education.
You can go back 2,300 years.
Young Alexander the Great had
Aristotle as a personal tutor.
You could imagine if young
Alexander needed to speed up,
Aristotle would've done that.
Had some gaps, Aristotle
would've addressed that.
And for most of human history,
that's how people learned.
Even today, elite
athletes, elite musicians,
that's how they learn.
But in education, that technique,
especially for academic
learning, was very elite.
Very few people had access to it.
You were a future king
or a member of nobility.
Fast-forward to about 200, 300 years ago,
and society was industrializing,
becoming wealthier.
And our leaders of our society
had a very utopian idea.
What if we had mass public education?
Not just 20 or 30% of
people got an education.
What if everyone did?
But they said we can't afford everyone
to get a one-on-one tutor
and get that personalization.
Let's use some techniques from
the Industrial Revolution.
Let's bash students together,
move them together at a set pace,
apply some centrally
determined standards to them.
Some kids will get it, some kids won't.
We'll sort them in a
seemingly intelligent way
into different careers.
And I know I've just described it
in a slightly pejorative way,
but it has done wonderful
things for society.
Literacy rates have gone up close to 100%.
And when it's done well,
the societies that adopted
this mass education,
it's no coincidence
that they were the first
to have a broad middle class.
If you think about the US,
Germany, the UK, Japan.
But it left a lot of people behind.
And Benjamin Bloom quantified this in 1984
where he wrote this two-sigma problem.
Two-sigma refers to
two standard deviation.
He was, as far as I know,
the first that really tried
to measure the effect of
personalized tutoring.
And he made an argument
that it would be a two
standard deviation improvement.
That would take someone
from the 50th percentile
to the 96th percentile or so.
Now, you know, people
have debated his study
and how strong the effect really is.
And he did it with a very
narrow knowledge base
with only a two- or three-week study.
But there's been many studies since then.
They might not have seen
two standard deviations,
but they validated it.
And in many ways, it validates
everyone's intuition.
Wealthy families, even after
industrialized education,
have known if your kids
are falling behind,
get them tutoring.
Either tutor them yourselves.
I was tutoring my cousins.
I saw it help them.
Or go hire a tutor.
That's the safety net that
wealthy or middle-class
or upper middle-class families have had
for the last several hundred years.
And that's what Khan
Academy's been trying to do.
But there's been fair
criticisms of Khan Academy.
Hey, a 5, 10-minute video is nice,
but you can't ask it a question.
Having personalized practice,
these questions that are
hopefully at your learning edge,
that's nice.
But what if you need help in the moment
and there's no one around you to help you?
We've seen the best
classroom implementations
of Khan Academy where the
teacher is able to help you,
where you could ask a peer,
but what if someone's not around?
So that was the original thinking
that this AI really could
play that type of a role.
- In the moment when you made the decision
that we are gonna have to bring AI in,
we're gonna have to lead with this,
we're gonna have to be the
ones that do it responsibly,
we're gonna have to teach
people how to learn again,
where does the conviction come from
that you were confident enough
that you were right on this?
- I think multiple times
in Khan Academy's history,
I had to, in hindsight,
have a lot of conviction.
In some ways, the AI was
hitting me over the head.
This was such a powerful technology.
But probably the hardest decision
was when I had to quit
my job back in 2009.
And there the conviction came from,
I saw the benefit it
had on my direct family,
like dramatic benefit.
Nadia went from being a struggling student
to being one of the top
students in her class,
and that happened to her brothers.
And then I started getting letters
from people from all over the world.
And you know, that's
not an efficacy study,
but when strangers start
sending you letters
saying that they thought
they were horrible at math,
but now they're the best student
or they're deployed in Iraq
and now they're going back to school
and going to become a physicist,
that kind of thing gives
you a ton of conviction.
And that was only growing, and
then the usage was growing.
And then even when you have
that type of conviction,
you know, I quit my day job,
living off of savings, our
first son had been born,
you have to keep testing that conviction.
And I was wavering big time.
Six, seven, eight months down the road,
we were living off of savings.
I would keep going back
to those testimonials
I was getting from people over the world.
Okay, no, there's something
here, there's something here.
But I was thinking
about going back to work
and then thankfully I got some validation
from some serious entrepreneurs
and philanthropists,
Ann and John Doerr, Bill Gates, Google.
So went kind of from zero
and then I got the validation
that I think a lot of entrepreneurs...
You can have a lot of conviction yourself,
but at some point you need some validation
from other people.
- You know, there's a tendency
to turn AI into something
kind of too big to control.
We never actually defined
the AI issue we're facing,
so it looks all threatening
and unaddressable.
And I'm reminded of a conversation I had
with a Kodak executive
when he talked about their
failure to deal fundamentally
with the threat of digital photography.
He told me this,
"For too long, we talked about the problem
as the digital threat.
No one can action that, and we didn't.
We took too long
to break the problem down into
addressable sub-components,
and we lost."
Well, Sal's doing the opposite.
Like many great founders,
he doesn't try to make problems bigger.
He makes them smaller, more
human, more manageable.
Now, don't get me wrong.
He's not making the
need for action smaller.
I mean, he brings urgency.
He's not making the
journey more comfortable.
In fact, he's clear this will
be an uncomfortable ride.
But what he is doing is
making AI addressable.
Job one is to figure out
how AI can help us learn,
how to learn again.
How can it be a tutor?
And by framing the question this way,
Sal gives his people confidence.
Hey, we've been here
before. This is what we do.
We just grab the best
delivery mechanisms available
to teach folks.
We got this.
Most leaders experience disruption
as something that happens to them.
Sal experienced it as something
his organization was built for
because they've been on the
same mission for decades,
and AI is giving them a
whole new set of tools.
There are few CEOs today
that truly are the product.
Sal has done the most
videos in Khan Academy.
Sal has invested the most in
this version of Khan Academy.
And yet he stays curious.
He's humble enough to
see when new technologies
demand he change.
He's vulnerable enough to be clear
that he has to launch the journey,
even if he's the first to say
he doesn't know exactly how it ends.
We can all learn a lot from Sal.
And not just his videos
introducing basic calculus.
Having gotten a C- my
freshman year of college,
I too realized I needed his help.
(upbeat music)
Everything from today's
episode and every episode
is at bain.com/founders-mentality.
We'll be back in two weeks. Stay curious.
Next time on "Founder's
Mentality: The CEO Sessions."
- I was made CEO of The Economist,
and I picked up my phone,
as we would all do,
to all my friends.
And I said, "Would you like to come here?"
And they said, "Why?"
You know, it's a publishing business.
I mean, we love the brand,
but, you know, this is a
mission to nowhere, right?
Just thought, okay, you
know, I get it, right?
But I can see that we can fix this.