Ep96 "What’s the future of education in an AI world?" (Part 2: Sal Khan)

Now that we’re careening into our AI future, what are the most important things for our students to learn? Do we keep teaching as we always have, do we drop our heads on the desk, or are there clever ways to steer and optimize education? What would Sal Khan, founder of Khan Academy, say about all this? Find out in this week's episode.

Episode Transcript

Speaker 1 (00:05):
What are the most important things for our students to
learn now that we're in a world where AI is
so much smarter than any human, Instead of learning skills
that may not matter as much in the future, can
AI help us to learn the important things?

Speaker 2 (00:21):
And what are the important things? Do we give up on.

Speaker 1 (00:24):
Education or are the very clever ways that we can
direct our teaching to get the best out of our schools.
What would Saul Khan, founder of the Kahn Academy say
about all this? Well, you're lucky because you're about to
find out. Welcome to inner Cosmos with me David Eagelman.
I'm a neuroscientist and author at Stanford and in these

(00:47):
episodes we dive into our three pound universe to uncover
some of the most surprising aspects of our lives and
our times. Today's episode is part two about the future

(01:07):
of education. In part one last week, I covered the
long road of how technology changes schooling. For example, when
the printing press was invented in fourteen forty, some thinkers
of the time were sure that invention would dumb us
all down, because now if you ask a student for
an answer, the answer is right there, It's written down.

(01:29):
You just pull the book off the shelf and wham,
it's as though you can pretend you know something. And
we went through similar debates when desk calculators were introduced,
will the kids become stupid? And then there was the
invention of the World Wide Web, And then when they
Google burst onto the scene, and all you had to
do was type your question into this little box and

(01:52):
it had indexed all the information in the world and
it could point you precisely where you needed.

Speaker 2 (01:57):
To go to find the answer.

Speaker 1 (01:59):
So suddenly it didn't matter what question the teacher asked you.

Speaker 2 (02:03):
As long as somebody somewhere.

Speaker 1 (02:05):
In the world knew the answer and had typed it
on the web page, it was just one click away
from you. In all these cases, educators were forced to
change the way they asked questions and assessed a student's knowledge.

Speaker 2 (02:19):
And you know what, it wasn't that hard.

Speaker 1 (02:21):
So where are we with technology in general and AI
in particular right now? I made three main arguments in
last week's episode. I argued that the extant technology wires
the brain of each generation differently, as in how they're
wired to best learn and the ways they seek information.

(02:43):
Then I gave many examples of the ways in which
AI can be used to massively improve the way we teach,
from individualized tutoring to reducing the workload on the teachers,
to training critical thinking and magnifying creativity. If you want
to hear more about that, please check out that last episode.
And finally, I made the argument that the purpose for

(03:05):
schooling for education is to give the students roots and wings,
as Gerta phrased it, which just means giving them the
foundations that they need for critical thinking and the ability
to create new things and go off in novel directions.

Speaker 2 (03:22):
So today we're going to.

Speaker 1 (03:23):
Get someone else's take on the state of things, and
not just anyone, but someone who has single handedly had
perhaps the largest impact on the planet's education.

Speaker 2 (03:33):
That I know. That's all Con.

Speaker 1 (03:35):
And if you don't know his name, you surely know
his creation, the Con Academy, which is an educational platform
that provides free and very high quality online courses. It
does this across math and science, and history and economics.

Speaker 2 (03:50):
Anything you want to learn.

Speaker 1 (03:51):
It has video lessons and interactive exercises and learning tools,
and the key is, unlike a normal school system, this
allows users to progress at their own pace, so Saul
has a unique vantage point to view the world's education
and as my neighbor here in Silicon Valley, he has
the opportunity to see the latest greatest tech and understand

(04:14):
where things are going.

Speaker 2 (04:16):
So, without further ado, here's my interview with Saul con.

Speaker 1 (04:24):
Okay, sal So, you're famous for having started the con Academy.

Speaker 2 (04:28):
Tell us about the history of that.

Speaker 3 (04:30):
Oh hi, you know, I guess you go back about
twenty years ago. A little over twenty years ago. My
original background was in technology, and I go to business
school and I find myself. I'm working at a hedge
fund in Boston. Two thousand and four. I'm a year
out of business school. I get married. My wife grew
up in New Jersey, so all of my family from
New Orleans, where I was born and raised, comes to
New Jersey for the wedding. While they were staying with me.

(04:52):
My cousin, it just came out of conversation was having
trouble with math Navia. She was twelve years old, and
when I asked her about it is unit conversion, etc.
I told Navy, I'm a hundred percent sure you can.
You can learn this material if you're up for it.
I'm happy to tutor you remotely when you go back
to New Orleans. She agrees, she goes back. We start
working together, slowly, but surely she gets unit conversion. She
gets caught up with her class, littlehead of her class.

(05:13):
Then I'd become what I call a tiger cousin. And
I call up her school and they.

Speaker 2 (05:19):
Said, who are you? I said, I'm her cousin.

Speaker 3 (05:21):
And I actually convinced them to let her retake a
placement exam because I thought, like, well, she knows her
material now, and they did, and she went from being
like a for lack of better word, to remut, like
in a slower track, to not being in a faster track.

Speaker 2 (05:34):
So I was like, well, this is cool. Like I'm
able to connect with a young cousin.

Speaker 3 (05:38):
I start tutoring her younger brother's words preadient, the family
free tutoring is going on. Before I know it, I'm
tutoring ten to fifteen cousins, family friends. You know, I
saw a common pattern that they were struggling, not because
they weren't going to good schools, not because they didn't
have good teachers, not because they weren't hard working or bright.
It's because they had gaps in their learning, especially in
a topic like math. But even you know, part of
the reason we were having trouble in science was gaps

(06:00):
sometimes in their math. So I started writing software for
them to kind of give them practice them being your cousins,
my cousins. Yeah, this was back in the day. If
you had to look for practice problems on factoring polynomials,
there were stuff on the internet, but they weren't very
good and they were very inconsistent. It's like, well, I
think I could write a little piece of software that'll
generate problems for them, and then I want to keep

(06:20):
track of how they're doing. And then that was the
first conducaut I mean, nothing to do with videos. Then
this was two thousand and five.

Speaker 2 (06:26):
Then how many people were using it?

Speaker 3 (06:28):
Initially just my cousins, and I was seeing benefit and
I could keeping track of them. Yeah, ten to fifteen cousins,
and then words started to spread. Their friends started to
sign up to it because they're like, oh, this is
really helping me get good at math, and so their
friends and at some point had to shut it down.
There got to about ten thousand folks who were trying
to use just the software, and it was taking down
my thirty dollars a month web hosting, and that wasn't

(06:50):
engineered perfectly either to scale to that. So I actually
just didn't allow anyone else to sign up, which is
like the opposite of Silicon Valley, Like, no, I don't
like it, there's too many people here. And then and
I was showing this software off at a dinner party.
And by this point we had moved out to northern California,
and I was at a dinner party in San Mateo
and the host and I was showing off this stuff
I was making, and she said, well, how are you

(07:11):
scaling your lessons. I'm like, I'm not. He's like, whight
you record some videos on YouTube for your family. I said,
that's a dumb idea. You know that's cat's playing piano.
That's not serious. But I did it anyway. Two thousand
and six, Okay, and yeah, I just started making mini lessons.
Actually YouTube forced them to be mini lessons at a
ten minute limit at the time, on things that I
found myself having to re explain to my cousins a lot,

(07:32):
Like I'd explain least common multiple to one of my
cousins one week and then the next week, I'm re
explaining it to another cousin, and so I said, well, okay,
I know least common multiples shows up a lot, you know,
negative exposed.

Speaker 2 (07:42):
So I just kept doing it.

Speaker 3 (07:42):
I started emailing them and saying, any question you have,
I'll do it asynchronously. You email me the question, I'll
give you a video explanation of it.

Speaker 2 (07:50):
Wow.

Speaker 3 (07:50):
And so we started building up a library of that.
And you know, my cousins really appreciated it. I sometimes joke,
but it's true. They said they liked me better on
YouTube than in person. And what they were saying is
that they like the asynchronous, the on demand, the no shame.
But we still were getting on the phone and they
did appreciate me as a human being being in their life,

(08:11):
and we were able to go deeper on the phone.
But and yeah, and then one thing led to another.
Two thousand and seven, two thousand and eight, fifteen to
one hundred thousand folks.

Speaker 2 (08:18):
I was still at the Hedge Fund.

Speaker 3 (08:20):
I set up as a nonprofit in two thousand and eight,
the mission Free World Class Education for Anyone, Anywhere. Two
thousand and nine, quit the day job. Two thousand and
nine was a hard year, but twenty ten, we got
our first real funding to become an organization.

Speaker 2 (08:30):
Wow.

Speaker 1 (08:31):
And how many people use con Academy now, Oh, you know,
it depends on how you account for it. I think
we're one hundred and seventy million registered users, but.

Speaker 2 (08:39):
You know, not everyone uses it all the time.

Speaker 3 (08:42):
You know, in a given year, it's probably closer to
about one hundred million folks are using it.

Speaker 2 (08:46):
Incredible.

Speaker 1 (08:47):
I mean, that must make you the biggest single educator
and now I know you have a whole team of
people doing it now, but the biggest educational organization in
the world.

Speaker 2 (08:55):
I would assume. Yes, it depends how you account for it.

Speaker 3 (08:58):
Yeah, I mean, but in some ways yeah, yeah, I
mean some people can make an argument that you know,
YouTube amongst other things, does some stuff. But yeah, I look,
but I always point out to the team that's nice.
But we have a long way to go because even
of that one hundred million, there's probably about three million
that are like really getting a lot out of it.
They're using it almost like you know, regularly, and those

(09:18):
people are really getting an education.

Speaker 2 (09:20):
Incredible.

Speaker 1 (09:21):
Okay, So now we find ourselves in twenty twenty five
and AI is on the scene, and so how are
you using AI in the con academy.

Speaker 3 (09:29):
Yeah, And I remind my because it's very easy to
get nambored with technology, especially things as cool as generative AI.
We had a preview if you rewind up two two
and a half years summer of twenty twenty two. Sam
Altman and Greg Brockman reach out to me at the
time and it's just this kind of cryptic email saying
and I didn't know them that well. I'd run into

(09:50):
them at some conferences and this and that, and they said, Hey,
we're training our new model. We think this is the
one that's going to really get people interested. We want
to launch with some social positive use cases that you
GPT four it ended up being GPT four And this
was this was about five months before chat GPT came out,
which actually was a bit of an accident. Chat which
and Chat GPT wasn't even built on GPT four. It

(10:11):
was built on GPT three point five. But when they
emailed me, I was like, I was curious. I had
seen GPT two and GPT three. I thought they were
very cool, but I didn't think there were anywhere close
to ready for prime time.

Speaker 2 (10:23):
But as a nerd, I just wanted to see what
they were going to do. So two weeks later they
had finished the training and they had myself and our
chief Learning officer, Kristin. We were on a zoom with
them and they had a little chat interface and this
was new too.

Speaker 3 (10:37):
People don't remember these, you know, just even have a
chat interface with an AI. And they had an ap
biology question and there's actually context why they picked ap biology.
But they said, so, what's the answer and I said, oh,
c it's osmosis. And then they clicked enter and the
AI said, oh the answer ce. I'm like, okay, ask
it why. It explained it. And people are used to
this now, but in the summer of twenty twenty two,

(11:00):
this was different than anything anyone had ever seen before.
And I'm like, okay, this is weird. Ask it why
the other choices are incorrect? It explained it beautifully, I said,
ask it tried another question like that.

Speaker 2 (11:11):
It did a great job. And that's when I was like, oh,
what is this like? Am I being punked?

Speaker 1 (11:15):
Like?

Speaker 3 (11:16):
But we saw it before, not all of open AI
even saw it. But then they gave Kristen myself access
to it over the weekend and we couldn't sleep, and
we couldn't tell anyone about it, Like, because this was
the biggest secret, I couldn't tell my wife about it.
But I was like middle of the night, I was
slacking christ and I was like, look, I just got
it to write the Decoration of Independence and the voice
of Donald Trump.

Speaker 2 (11:36):
Isn't this hilarious?

Speaker 3 (11:37):
And I'm writing poems and I'm doing this and look,
if you prompt it the right way, can.

Speaker 2 (11:40):
Act as a tutor.

Speaker 3 (11:41):
And that's when going back to what Kin Academy was
always about. As I mentioned, I started tutoring my cousins
and everything that we've done over the last fifteen twenty years,
it's really about how do we use technology to approximate
elements of a tutor on demand bite sized video kind
of approximates an on demand explanation from a tutor. These
practice problems that you're doing, the problems that you need

(12:02):
to do as opposed to what everyone else is doing.

Speaker 2 (12:04):
That's what you would do with a little bit of
a good tutor. They would give you what you need.

Speaker 3 (12:08):
And when we realize that you could steer or prompt
these this new generation of models to really be not
just pretend to be the tutor, but really have strong
tutor moves, so to speak.

Speaker 2 (12:19):
We're like, Okay, this thing has.

Speaker 3 (12:21):
Issues, It hallucinates, it makes horrible math errors. But if
we assume that those things can either be mitigated or
they're going to get a lot better, which has happened.
This is going to be a game changer for us
being able to scale up what we think is world
class education, which is more personalization. Giving what Navia had
in sal what if every student we could eventually get there.

Speaker 1 (12:41):
So let's double click on that idea about an AI tutor,
because I heard it talk from Steve Jobs way back
in the day, maybe in the nineties or something. He said,
just imagine if everyone had Aristotle as a tutor, if
I'm remembering what he said, treg But I asked my
son about that at someonhere. He's thirteen years old, and
I said, you know, hey, what would it be like
for you to have Aristotles a tutor? And he's said,
I probably wouldn't do much with it. And the reason

(13:02):
he said that is because, first of all, his internal
model of what he knows and what Aristotle knows from
all his writings they're different. So you wouldn't even know
what to ask him. But besides that, what he wants
to do is be with friends and play video games stuff.
So that model of tutor, I think isn't going to work.
But what you're doing is different. It's not just someone
available to answer questions for you. So tell us about

(13:24):
how you think about surfacing information in the learner.

Speaker 3 (13:30):
No, that's right, and I want to be clear where
you know, it's a journey and it's only going to
get hopefully better over time. But immediately when we started
seeing all this, we said, all right, well, some of
the limitations of classic con Academy is you could watch
a five minute video, but what if you have a question.
If you have a peer around, you could ask them.
If you have a teacher around, you can ask them.
But what if you're a young girl in Afghanistan and
you don't have a school.

Speaker 2 (13:49):
What do you do? Then?

Speaker 3 (13:50):
If you're working on a on a exercise and you
just got stuck. We try to make it as gradual
as possible, but sometimes you just get stuck, and that's
an that's the most likely time that you'll just engage.

Speaker 2 (14:00):
What if you could get unstuck?

Speaker 3 (14:02):
So the first iteration of what we called what we
still call Conmigo, which is the AI tutor on con Academy,
was supporting those use cases. Now, to your point and
to your son's point, what we see is about ten
fifteen percent. We haven't, you know, had a formal study
of this, but what we're observing about ten fifteen percent
of students make good use of that. They're the students

(14:22):
that if they watch a video, they're still curious. They're like,
what about this, or how about this? What if we
change this variable? And so they know how to ask that.
But there's another eighty percent of students who are like, oh,
you know whatever, this isn't my thing. And so what
we're realizing is to have the benefits of an Aristotle,
and Aristotle was Alexander the great tutor. Aristotle didn't just
sit there and wait, I'm guessing, didn't just sit there

(14:43):
and wait for Alexander to ask him another question. He said, Okay,
this is what we're going to do, young Alexander. And
by the way, how come you didn't do your homework.
I'm going to go talk to was it Prince phil
King Philip or verbal Macedonia?

Speaker 2 (14:55):
Yeah, his dad, right.

Speaker 3 (14:57):
Philip of Macedonia I'm going to go talk to King
Philip about you're not doing your work.

Speaker 2 (15:02):
And that's what I used to do with my cousins.

Speaker 3 (15:03):
Yes, I would answer their questions, but most of the
time I was driving, I'd be like, this is what
we're going to do today because I noticed this, And
if they showed up unprepared, I would hold them accountable.
And sometimes I would call their mom or their dad
up and say, look, you got to make sure they
have time for this. And so we are working on
ways that the AI can be more proactive.

Speaker 2 (15:21):
Now.

Speaker 3 (15:21):
At the end of the day, we do think that
that teacher and if a student has access to a teacher,
which is true of most of the world we live in,
there's parts of the world where they don't, But if
they have access to a teacher, the teacher is still
the biggest motivational mechanic for students. And so we are
thinking about how can the AI both support the teacher
as a teaching assistant helping them with some of their tasks,

(15:43):
but then also be a connective tissue and the teacher
can hold the student accountable on working with the A
and then the a I can report back to the
to the teacher. So it's a journey on how we
can make it more and more part of that loop
and it can drive more and more, but once again
not replacing the humans, but but optimizing the connection with.

Speaker 2 (16:03):
The humans so that they don't have to worry about
some of the busy work. Great.

Speaker 1 (16:07):
So I want to come back to that point about
what this will mean for teachers, but I want to
quickly interject something, which is so I'm sort of a
history buff, but I blanked on this issue about King
Philip of Macedonia and so, but my question is, I
mean the question everyone's asking currently is, you know, with
AI in the world and that's not going away, what
is the importance of memorizing facts?

Speaker 2 (16:29):
How will education change?

Speaker 1 (16:32):
Will it be important for children to learn a lot,
you know, the Battle of Hastings was ten sixty six,
or will it not be so important because they can
retrieve that information to a point?

Speaker 2 (16:40):
What's your take on that.

Speaker 3 (16:42):
I'm solidly in the camp that content knowledge is really
valuable because and we've seen this pattern before Calculator comes
out in the sixties and seventies, We're like.

Speaker 2 (16:51):
Oh, no, one has to learned arithmetic anymore. Well, it
turns out.

Speaker 3 (16:53):
The people who benefit the most from calculators are the
people who understood arithmetic and why why They know how
to use the tool. They know when the tool can
be used, they know when the tool can't be used.
They know when there's a precision error in your calculator.
You know, they know that one third is the same
thing as point three repeating, and so they know that,
Like when your calculator shows zero point three two two
two two two two two three, they're like, oh, one third,

(17:14):
And that's a simple example, but you could imagine other ones.
You could not send people to the moon if you
only use a calculator but you didn't have an understanding
of things like floating point errors, or if you didn't
understand things of whereas the calculator have a limitation. Similar thing.
Internet comes out, Google comes out. People say, oh, you
can web search everything. Why do you need to have
any knowledge anymore? It turns out that people with knowledge

(17:37):
are going to have much better information. For example, you
might have forgotten that it was the Battle of Hastings,
but you're you'll be like, hey, you know google, William
the Conqueror, what was that battle again?

Speaker 2 (17:48):
Or you know, I remember that the card English.

Speaker 3 (17:51):
You know, English has a lot of French words in
it was it was that related to the Battle of
Hastings and and and the Norman conquest. And if you
at least have some of that, you wouldn't even begin
to ask the right questions. Or let's say that you
just have a homework assignment and you copy and paste
it into Google this is twenty years ago. You got
you got five search results, or YouTube videos start showing up.

(18:13):
Unless you have some context and some critical thinking of
judging what's solid information or not, you could just be
completely misguided. Like if you didn't know it was the
Battle of Hastings and you click on a video it's
like we're going to talk about William the Conqueror and
the Battle of Pudding and you're like, oh, it's the
Battle of Pudding. And it might have been a joke video,
but you didn't even watch enough of it to like

(18:34):
that's going to be a problem. And I see pre AI,
those with the most content knowledge are the people in
the position to use the Internet for their knowledge work,
and they're the ones that I see the difference between
when I web search versus when a younger child does,
or even some people you know in my family who
might not have the full context of what they're searching for.

(18:55):
They're doing very superficial searches. They don't know the right
questions to ask. Same thing with AI. AI is like
can be an intelligent partner with you. Imagine trying to
have a philosophical conversation with Aristotle without having read any
of Aristotle, or without having read any philosophy, or at
least without having talked about it. Or imagine you know,
being able to meet Isaac Newton, but you don't. You

(19:16):
haven't gotten the basics of physics or calculus down. You
can't just tell them like, what's going to happen to
this rock? If you should say, hey, does Newton's third
law apply here? Because this seems like an edge case,
and then you're going to be able to go much
much further. So I think content knowledge is actually more
important now because if you want to leverage AI, if
you want it to amplify your intent, you have to
hang with the AI.

Speaker 2 (19:36):
And there's still going to be work, human centered work.

Speaker 3 (19:38):
There's going to be taking care of seniors, things that
are very more manual or more human driven hospitality. But
I think the knowledge economy where AI is going to
become a big part of it, you have to be
able to hang with it, and content will matter more.

Speaker 1 (19:51):
I agree with you, although I do wonder about memorization
versus critical thinking. So how do you think about teaching
children critical thinking?

Speaker 2 (19:59):
At this point here, I'll also sound a little traditionalist.
I think it's both.

Speaker 3 (20:03):
I think it's become very I definitely disagree, like I
don't like pure like Victorian era memorize everything like that's
not good. But I do think in like Western school systems,
it's become fashionable to say any type of memorization is
not good, and we're only going to focus on critical
thinking and skill development. And the reality is is that

(20:25):
I'll speak for myself, A lot of the connections that
have formed in my mind are because I knew something
and then my brain at some point, Like you know,
I every American kid learns about the Louisiana purchase in
eighteen oh three, and then later most American kids don't
learn about the Napoleonic Wars, even though that was a
much more important thing.

Speaker 2 (20:43):
But if you do learn about the Napoleonic Wars, and
they're normally.

Speaker 3 (20:46):
Not connected for anyone right, But at some point you're like, oh, wait,
in that other class, I learned like he's sitting here
fighting all of these other European powers, and eighteen oh
three he actually his navy was defeated. I wonder whether
that has anything to do with the Louisiana purchase. And
even though they never teach us you in history class,
has everything to do with the Louisiana purchase. Napoleon sold

(21:07):
it because there's no way he could defend it. It turns
out Thomas Jefferson could have just taken Louisiana and Napoleon
had no navy to send out because the British had
just destroyed it. And that's critical thinking. But in order
to create those connections, you had to put the Napoleonic
Wars in time and space with the Louisiana purchase, and
you have to know where these things are. You have

(21:29):
to know where Napoleon is, where England is, where the
navy was defeated, where Louisiana is. You're like, oh, critical thinking,
I'd be awfully hard to defend Louisiana if I'm sitting
here battling all these Europeans and I don't have no
navy anymore.

Speaker 2 (21:55):
Well, let me see two.

Speaker 1 (21:56):
Things from the brain point of view, which is four
plasticity to happen. For knowledge to stick, it has to
be relevant in some way. And so one of the
ways that knowledge becomes relevant for us is when we
are curious about something. So if you think, hey, wait,
did the employee ack wars have anything to do with it?
Then you're curious and you look it up and then
it sticks the rest of your life.

Speaker 2 (22:15):
Great.

Speaker 1 (22:15):
But how do we, as educators and using AI, how
do we make things relevant to kids? And how do
two AI tutors play part in that? Yeah, That's what
I'm excited about.

Speaker 3 (22:26):
And I'll keep giving display because I don't want to
make it sound like AI to mar Is is going to
make everything perfect. But we're on a path and we've
already seen once again that fifteen twenty percent of students
who are let's call it proactive and showing more of
that curiosity. I've seen it in front of my eyes
where they're learning about the amendments of the Constitution. But
now with generative AI, they could say, hey, why is
this relevant now? And then the IA says, well, actually,

(22:49):
back just recently in twenty twenty there was this Supreme
Court case and actually that makes a big difference on
like executive power, the Chevron doctrine, and.

Speaker 2 (22:57):
Are like, oh wow, this is a big deal. I
never realized.

Speaker 3 (23:00):
So there's apps or like hey, I'm studying entroped, why
does this matter? And it's like, no, it actually matters
a ton because if you go into this or there's
just a study recently, or if you're trying to and
so I think that's one of the really powerful things
that it. You know, the classic question for every kid,
when am I going to need to use this? I
consider myself better than most people and being able to
answer that question, but even I'll fumble sometimes. But the

(23:21):
AI is very good at that, and it's very good
at like on conmego, if a student literally asked why
do I need to learn this? The first thing that
the AI typically does is is, well, what do you
care about? And the students I want to be a
professional athlete, I want to do this, and it'll actually
the I will remember that there's ways to reset it
and clear it and so it's not creepy. But like
then in the future, it can proactively say, well, you know,
since you said you want to be an athlete.

Speaker 2 (23:42):
This molecule, you know, ATP is crucial.

Speaker 3 (23:45):
You know, you want to maximize your ATP if you're
about to you know, run a marathon or something.

Speaker 2 (23:49):
So that's the hope.

Speaker 3 (23:51):
And as AI gets better and better, as it gets multimodal,
as it gets you know, we can completely imagine ten
years in the future, for sure you're going to You'll
put on augmented reality glasses or virtual reality glasses and
travel to ancient Rome and try to stop, you know,
the assassination of Caesar, or travel into the human brain
and try to, you know, diagnose whether this is a

(24:12):
you know, an Alzheimer's case, or you know, are there
ways that you might be able to address it. So
I think that relevance question AI will help. It's not
solved it now. I'll say one other thing though, you know,
I in my first book, I gave the example of
you know, people always talk about relevance, but I was
this I don't know. I used to think I was
five nine, but I now realize probably closer to five

(24:33):
to seven.

Speaker 2 (24:33):
Five eight.

Speaker 3 (24:34):
But I was this that tall kid in high school.
But when I was in Pe, I never said, why
am I going to have to take an orange ball
and have to put it into a ten foot hoop,
and this has no relev It had no relevance to
my but I was engaged because my friends were there,
I was having fun. I didn't feel disengaged. I do
think that sometimes the relevance question is missing the mark

(24:56):
a bit like math to some degree, the real beauty
of math isn't in its ow implication. It's in its
inherent beauty in it. It's the thought process of it.
And I think the reason why a lot of kids
start saying, oh, when am I going to learn this
is they have so many gaps that they've disengaged, so
they're just protecting their self esteem if you can fill
in those gaps. I saw this with my cousins. Some
of my cousins were saying, I'm not a math person.

(25:16):
I hate math.

Speaker 2 (25:17):
Is that. But once they started getting that personal attention
and I'm like, Hey, isn't this cool? Isn't this cool?

Speaker 3 (25:22):
You can do it this way or this way? Isn't
this amazing that this can be described or predicted? They
started becoming math people. And not that they thought that
they were going to use it in their life, but
they just started seeing the beauty in it and they
start feeling confident in it.

Speaker 2 (25:33):
That's great. You know.

Speaker 1 (25:34):
Another thing is we as humans are so sensitive to
story as opposed to let's say, a bulleted list of points.
And so this is one of the things I'm enthusiastic about
with AI is can it take complicated things, whether it's
history or math or whatever and tell you a story
about it.

Speaker 3 (25:50):
Absolutely, I can create a story. It could create a poem,
a song. It can. I think it can create simulations
for you where yes, it becomes like the ultimate choose
your own eventure. I mean, right now, it could be
text based, it's soon going to be a video and
visual based. But as we said in the not too
far off future, let's call it five ten years, it's

(26:11):
going to be able to create a lucid dream for
you that puts you in the middle of it all.

Speaker 2 (26:15):
Yeah.

Speaker 1 (26:16):
So I want to come back to this point about
what this means for teachers. Tell us your thoughts on that.

Speaker 3 (26:21):
I've been saying this for years because even when I
started making videos and software, some people said, oh, on
demand video does this replace teachers? And my response is
I don't think it does. If all a teacher is
doing is going up, giving a lecture and leaving. Then yes,
maybe on demand video could replace that. And you see

(26:44):
that happening in a lot of colleges. They've you know,
no one goes in med schools, no one goes to
the lecture anymore, because that's what the lectures were. People
are watching the recordings of the lectures on double speed,
and now they can ask questions to an AI.

Speaker 2 (26:55):
So that's that.

Speaker 3 (26:56):
But and I've been saying this for years, the ideal
classroom has always been, and this go back to Socrates
and Plato and Aristotle, is if human beings are getting
together in a room, make them interact with each other,
have a Socratic dialogue, have a simulation. Have students go
into groups and tackle the problem a little bit. Yes,
you could give a little bit of a mini lecture,

(27:17):
but then make it interactive or maybe out of the
thirty students, ten of them really need help with this concept,
while the other twenty can actually learn at their own
pace or keep working on their project, or work on connuiccuting, whatever.
So do the focus intervention with these ten, and then
you don't have to bore or lose the other twenty.
And then do another focus intervention with the other ten
and that type of work. I think most people go

(27:38):
into the teaching profession because they actually want to be
that teacher that unlocks little sal or, little David, makes
them believe in themselves. You know, we all can list
five to ten teachers at every stage of our life.
It's like that was the teacher that made me believe
in myself or made me see wonder in the world,
and they change my trajectory. And every teacher wants to
be that teacher. And I think whether it's video, whether

(28:00):
it's software, and it's AI, if it's used well and
the AI is acting as assistant but not trying to
replace the teacher, which I don't think it should. I
think it could actually free up the teacher to do
a lot of that more human element to work.

Speaker 1 (28:11):
And I know one of the things that you've been
looking at is how AI could improve debating, as in,
I'm going to take a point.

Speaker 2 (28:18):
And the AI is going to give you the opposite
point and tell us about that.

Speaker 3 (28:21):
Yeah, and it's you know, there's actually two projects on
the debating let's call it debate, and one is very
AI and one is very non AI. But I think
they might actually converge. So on Conmigo, we actually have
an activity where you can get into a debate with
an AI on anything, you know, should the government cancel
student debt, should the US maintain military bases around the world,
or should it you know, kind of become more isolationist

(28:43):
and you can pick either side. The AI will have
a polite debate and then actually give you feedback and
then you can switch side. So that's where we've got
a lot of positive feedback. Students enjoy that it's a
safe environment to test ideas without being embarrassed.

Speaker 2 (28:56):
Now we have another project.

Speaker 3 (28:57):
There's a sister nonprofit to Kind Academy that we still
during the pandemic called Schoolhouse dot World. Schoolhouse dot World
is all about free live tutoring from other human beings.
The way that we're able to give people free live
over zoom tutoring from other human beings is through volunteers,
and anyone listening can go be a volunteer, and there's
a whole mechanism to certify yourself or you can get tutoring.

(29:19):
But one of the things we realize a lot of
universities and some of the top universities you, Chicago, MIT, Brown, Yale, Caltech.
There's thirty others said, Hey, if someone is tutoring on
Schoolhouse dot World, so they've certified that they say no
physics or they know biology, and they're tutoring other people,
we want to know about that for college admissions. And
so they have been using it for college admissions. But

(29:40):
on top of that, when we went last year, they're like, hey,
is there any way that we could leverage this platform
to foster like civil dialogue between people, civil debate to
your point, And we're like, okay, let's try something a
little bit wild. Let's have real people fill out a
survey on all the hottest button issues Israel, Palestine, abortion,
affirmative action, free speech, gun control, gun control, go down

(30:04):
the list, the stuff that people are afraid to talk
about because they don't want to lose friendships.

Speaker 2 (30:08):
Students fill it out.

Speaker 3 (30:09):
You get paired, either one on one, sometimes two on
two with someone with a diametrically opposite point of view.
Now what happens on this dialogue project is you're not
supposed to it's not a debate, but you're supposed to
listen to each other. And at the end of it,
you have to with the form represent the other person's
point of view. You also have to reflect on whether

(30:30):
you think they heard you and represent and then any
other reflections on the interaction that you had. And what
happens then is you get a rating on your ability
to engage in hard conversations. And these colleges are very
interested in this as they don't carry your point of
view on these issues. They care that they like that
you have a point of view, but that you were
able to engage in a respectful and thoughtful and you're

(30:50):
able to steal man the other person's argument, you're able
to represent it. Now, those are both ways to have
dialogue in a world where it's getting increasingly polarized, increasingly
hard to have these conversations and feel safe. Over time,
these might converge a little bit. We are imagining ways
that AI. We are already using AI on schoolhouse to
give the tutor's feedback after the fact to saying, hey,

(31:11):
you know, you probably didn't ask enough questions, or you
ignored this one student, or they were actually asking this.
Over time is going to be real time. We also
imagine this dialogue project the AI might be able to facilitate.
We actually haven't had any where people started yelling at
each other or got on out. But maybe it can
make sure that the conversation. Oh oh, David, you know
I can tell your feeling a little sensitive about what

(31:31):
Sala said. Sal, is there another way that you might
have said that that you know might be a little
less triggering?

Speaker 2 (31:36):
Or David, can you tell sal a way.

Speaker 3 (31:38):
That he might have said that that you would have
appreciated things like that? I could imagine AI facilitating that
type of thing.

Speaker 1 (31:43):
Oh great, And what do you think is the future
of AI in terms of assessments instead of taking standardized tests?
What's the future in five years?

Speaker 2 (31:53):
Like everything?

Speaker 3 (31:53):
It's I mean, if if we're living in a science
fiction book and standardized tests I have historically defended because
I've said, look, to get good at anything, you have
to measure it, And would you prefer unstandardized measurement you
want to be able to benchmark. Now, the problem that
most people have with sanitized tests, which I also agree with,
is if you want something that's consistent and standardized and scalable,

(32:15):
it's got to be you know, fairly narrow in what
it can measure. So most standardized tests are multiple choice
or just numeric entry. They don't let you do a
lot of open ended things. And unfortunately, yes, they do
measure some things that are of value, but they can't
measure how well you could write a longer paper, or
how well you can design something, or how well you
can problem solve things. What's exciting about generative AI is yes,

(32:39):
I do think in the next in fact, we're working
on this kind of the future of assessment. We can
incrementally add layers on top of traditional standardized assessment. So
you can ask a student, Okay, you picked C. Can
you explain your reasoning? Why do you think you pick C?
Or even after you you actually got this one wrong,
do you think it was a careless mistake or do
you think you know there was an error?

Speaker 2 (32:57):
What was your error?

Speaker 3 (32:58):
And if you can give some of that subjective insights
to teachers, it becomes more actionable. Then they know what
to do with it. Reading comprehension instead of what was
the author's intent? You know ABCD, you could say, before
you see the choices in your own words, what do
you think the author's intent was?

Speaker 2 (33:12):
Now look at the choices.

Speaker 3 (33:13):
Now if you fast forward five years, ten years, I
think it's going to feel very much like a job
interview or like an oral thesis defense at a university.
And I know people's alarm bells start going off of
a bias and this and that and a it can
be creepy and data and all that, But I always
point out, like job interviews and these rich assessments like

(33:34):
thesis defense oral exams are rich, but they're incredibly biased,
and they're incredibly inconsistent. Right now, you know, even if
you're getting interviewed by the same person, depending on their mood,
depending on the day of the week, you could get
a completely different experience with that person. If we can
get AI, and if we can benchmark AI and test
AID and make sure, okay, two people equivalent, one is

(33:54):
African American, one is Asian, is giving the same signal
because other than race, these two simulated candidates are completely identical.
So you can benchmark an AI in ways that you
can't benchmark a human being. But if you can do that,
I think AI is going to You're going to be
able to have standardized assessments that feel a lot like
a job interview, that feel a lot like a simulation

(34:16):
or a game or a problem solving that I think
can really broaden the oppertunity. I mean even things like
sense of humor, or communication skills or body language.

Speaker 2 (34:25):
I think the AI could give feedback.

Speaker 3 (34:27):
Now, I'm a big believer you shouldn't just say, oh,
that person's body language was off, never give them a job.
It should be like, hey, you know what your body
language right now? You were a little bit you weren't
making eye contact, et cetera. I'd give you a three
out of five. Here's tips. Come back to me tomorrow
and take the assessment again. Mastery learning.

Speaker 1 (34:42):
Oh great, so you've touched on a few things about
what the world might look like five ten years ago.
When I think of five years ago, I mean, the
only people who even knew these llms were working was
you know, some researchers that could will whatever.

Speaker 2 (34:54):
But it hadn't burst onto the world yet.

Speaker 1 (34:57):
And so five years hence, what else do you think
it's going to be different about, for example, what you're
doing with con migo or whatever is next.

Speaker 2 (35:04):
What does it look like for students? It's hard.

Speaker 3 (35:06):
Yeah, and you said five years ago. Even five years ago,
the AI researchers themselves didn't even appreciate I remember when
Greg and Sam gave us access to GPT four in
the summer of twenty twenty two, and my first question
before I even tried it out, I was like, does
this work in other languages? And they said, oh, we
don't think so. And then I barely speak Bengali, which
is where my family. My family comes from Bengal, and

(35:27):
I said some it wrote back to me to Bengali,
and I took a screenshot I sent to say They're like, oh,
oh yeah, we just figured that out when you asked,
so they didn't even know. And then one of my
friend's parents as a researcher at another AI firm, and
he's like, you know what, we didn't realize that you
guys did Kin Academy did.

Speaker 2 (35:45):
I'm like, oh wow, what's our role here?

Speaker 3 (35:47):
He's like, we didn't realize the power of prompting that
A lot of the AI researchers, if you go five
years ago, four years ago, three years ago, they were
just like going to make a better model with more parameters,
better training data. And I think one of the breakthrough
that happened both with GPT four and chat GPT to
some degree one was a chat interface that unlocked a
lot and then the other was that you can actually

(36:08):
and GPT four in particular was very steerable, which means
you could prompt it and have it take on personas
as a tutor or as a poet, or as a
rapper or whatever you want to make it do. And
they I don't think the researchers had the time to
even sit there and prompt it. So it is amazing
even they didn't realize how far you could get with
something like that five years in the future. You know,

(36:28):
there's some people are arguing that, you know, we're hitting
some type of an asymptote. You know, the jump from
GPT three to GPT four was massive. The jump from
GPT four now two years later to GPT four Omni
one reasoning model three or whatever, or Gemini three or
Claude three point five.

Speaker 2 (36:46):
It's better.

Speaker 3 (36:47):
I mean, the hallucinations are way down, the math errors
the way down.

Speaker 2 (36:51):
It is getting better at reasoning.

Speaker 3 (36:52):
But it's not as much of a you know, we
thought GPT five would be around by now, and that
you know, maybe it would be approaching you know, something
even more mind blowing or even transcending human intelligence that
hasn't happened yet. So I think there's two scenarios. I
think there's a world where in the next five years
we do and then that's a wild world. I think
you're going to see massive acceleration. I mean in your world,

(37:13):
understanding the brain, understanding complex systems, drug developments.

Speaker 2 (37:17):
Science.

Speaker 3 (37:18):
I've heard people say we're gonna have one hundred years
of science breakthroughs in the next ten I don't know.

Speaker 2 (37:23):
I don't know. I think that's at least true.

Speaker 1 (37:25):
I've actually made the argument that A is going to
help in science in several ways. The main way it's
going to do it is simply by pulling together facts
that I couldn't possibly know because I couldn't have read
all those journal articles. But the thing I'm really looking
forward to now, which current lms cannot do, is scientific

(37:45):
discovery of the type that humans do, which is extrapolating
thinking of a new framework, saying what if I were
writing on that photon of light, what would the world
look like? Oh? That leads me to special theory of relativity.
That's the kind of thing the current llms are incredible
at interpolating, but not at extrapolating. So that's what I'm

(38:06):
looking forward to in science. But actually, let me ask
you a question on this, And actually you mentioned hallucinations
a moment ago, and I always think about what I
like about hallucinations in MS, which is that you can
also think of it as hypotheses, which is to say,
and this is what scientists do all the time.

Speaker 2 (38:25):
Is hey, what if? What if?

Speaker 1 (38:26):
You know, we're trying things out all the time. Most
of them don't lead anywhere. Occasionally you can build a
bridge to that and say, oh my gosh, I just
made real progress.

Speaker 2 (38:33):
So what do you think is the effect.

Speaker 1 (38:35):
Of AIS in education on creativity in students.

Speaker 3 (38:41):
I read a lot about it in my book Brave
New Words, and I think it's going to be a
creativity amplifier. You know, some people would argue like, oh,
if this thing can create a lazy student's just not
going to create themselves. And look, lazy students are going
to figure out ways to be lazy. There's ways to
police them. We're working on that too. But I always
point out the most creative moments in my life have
been when I'm in a room with other creative people.
If you look at all the creative moments in history,

(39:03):
you know, Florence during the Renaissance, or some of the
you know beat nicks in you know, different cities in
the world, or Silicon Valley.

Speaker 2 (39:11):
Right now, it's there.

Speaker 3 (39:12):
There is this effect of people meeting other creative people
and then collaborating with them, and the power of AI
is you know, like I remember being in middle school
in high school and I was a curious kid, but
I had to suppress that curiosity in the classroom otherwise
I would get beat up. But I had no one
to talk about this stuff with, and you just kind
of sit been by yourself. It's like, I wonder, I

(39:33):
wonder if this is different, and I wonder if I
could do this. But now, if young sal had had
chat GBT or Gemini or something, I.

Speaker 2 (39:41):
Could have played around. I was like, well, what do
you mean by that? Or how does this?

Speaker 3 (39:44):
And I do think you know, to your point, some
of the big breakthroughs are naive thinking, and some of
the best naive thinking is coming from people who haven't
been indoctrinated yet. And every couple of weeks I do
get an email from a student saying, Oh, I watched
your video on this, but I have a new theory
that disproves you know Newton, And I'm like, okay, whatever, whatever,

(40:05):
But you know, one out of every thousand of those
kids is going to be right if they can start
to validate or fine tune their ideas with an AI
that could even bring what's already known out there. One
of them is like, wait, I think I'm onto something,
and maybe the AI says, you know, you might be
onto something that might reconcile gravity with quantum mechanics.

Speaker 1 (40:43):
It feels to me like the two most important things
for educators is, you know, roots and wings. This was
something Gerta said two hundred and fifty years ago that
that's our job is to give kids roots and wings,
meaning all the content knowledge as you mentioned, and the
wings being the creativity part. So if students have a
partner with which they can do it, great, that makes

(41:05):
things better. How do we make sure that kids are
using it as a partner as opposed to being the
lazy kid and just letting let me do it.

Speaker 2 (41:13):
Yeah. I wrote about a whole in my book.

Speaker 3 (41:15):
I wrote a whole chapter about cheating, and I always
like to start like, you know, everyone's concerned about cheating
because of chat GEPT and it can cheat. I was like,
but what was the state of cheating before, and it
turns out there's been a lot of cheating, in fact,
even since you and I were in school. Unfortunately cheating,
you know, I think the internet, there's services online that
a human being will write your paper in Kenya, a
PhD in English will write your paper for five dollars

(41:37):
a page, write your term paper for you. And that's
arguably and that's been around for twenty years. So and
sorority A sororities sisters and fraternity brothers are you know, sharing,
So that's always been there and so our solution to
it is, yeah, you're right, chat GPT could be used.

Speaker 2 (41:53):
In fact, I'm sure people are using it.

Speaker 3 (41:55):
And anyone who says that there's tools that can detect
AI cheating is it snake oil. In fact, there's a
lot of FA accusations happening right now, especially in higher education.
But I think the real solution is and it doesn't
just solve the cheating problem from AI. I think it
solved the cheating problem generally, but it can also support
students and teachers better is have the students have the
teacher signed through the AI, have the students work with

(42:17):
the AI on the paper, and then have the AI
make the entire process transparent to the professor or to
the teacher, because then and we're building this. But then
if the AI could say, yes, professor, we worked on
this for four hours. David had some trouble coming up
with a thesis statement. This is where we worked. You
can see two hours into it. We had a lot
of back and forth on this part. You can see
his edit history, but you know, three hours into it,

(42:39):
he just copy and pasted this paragraph into this. I
do not know where it comes from. It actually doesn't
look like his writing. We should look into this. Don't
accuse him, ask him about that paragraph. I think that
could be healthy. Or if the whole thing just appears
or it just comes linearly on, and you know, it
doesn't look like something that someone worked on, then yeah,
that's the red flag. But that would have flagged even
the situation where you hired someone in Africa to write

(43:01):
your paper for you and.

Speaker 2 (43:02):
It can quiz you on it. Yes, that's excellent.

Speaker 1 (43:06):
One question just to push back on that is will
teachers want to look through the scrolls of data about
this paragraph appeared and so on?

Speaker 2 (43:14):
Or is that too much work for them? Well, it
is true.

Speaker 3 (43:16):
Well, That's why it's useful for the AI, because the
AI could say so right now, our product manager, who
leads what we call writing coach, she used to be
an English teacher, and she used to have one hundred
seventh grade students on her docket. She assigns one paper
to grade. Those hundred papers would take her seventeen hours.
That is not a great like that seventeen hours above

(43:38):
and beyond being at the school for forty like it's incredible,
and teachers are doing this right now. So imagine this
new world where those hundred papers are done. And even
those papers you don't know if Chad Gpt did them,
you didn't know what happened. Now imagine this new world
the papers are all done. The AI, first of all,
can say, you know her name is Sarah Sarah, based

(43:58):
on our rubric we constructed together, here's my preliminary grade
and my justification of why. But you should review it.
So what used to take seventeen hours can now maybe
take you two hours. You still should review it, but
like oh this, and in a lot of ways, it
might be far more consistent because if you're grading one
hundred seventh grade papers on the great Gatsby that ninetieth paper.

(44:19):
You know you're not going to be consistent. I wouldn't
be consistent one hundred percent. But also yes, and I'm
confident that this is David's work, et cetera.

Speaker 2 (44:26):
It's just giving.

Speaker 3 (44:27):
It can give the synopsis, but then Sarah can ask questions.
She's like, tell me more. Are there some themes you see?
But yeah, it turns out a lot of your students
are having trouble with thesis statements. So Sarah doesn't have
to read every transcript that's not But once again, this
acts as a teaching assistant and it can do all
of that grunt work, but give Sarah way way more
insights on where the students are, streamline the grading, and

(44:48):
undermine any kind of cheating that might happen.

Speaker 2 (44:50):
I see.

Speaker 1 (44:51):
And it's not only grading the individual papers, but it's
also giving her feedback about how the group did.

Speaker 2 (44:56):
Yeah, yeah, we should do.

Speaker 3 (44:57):
Let's let's modify tomorrow's lesson plan to be focused on outlining,
because that's where a lot of students struggled.

Speaker 1 (45:02):
In your experience, how are teachers currently feeling about AAR
feeling excited? Are they feeling threatened? And what advice do
you give to teachers. I've been pleasantly surprised. You know,
when we started working on kon Mega, we had we
had all these debates inside of our organization, people like
it's communities for cheating, and had hallucinates and and what
I what I told the team at the time is like, look,
we we got to just this is going to be

(45:23):
a thing.

Speaker 3 (45:23):
It's going to be it's going to improve. We have
these are real concerns, but we have to turn them
into features. So we said, Okay, we're gonna create oversight.
We're going to moderate what students are doing. We're gonna
cot Migo's not going to give you the answer. It's
going to be socratic. It's going to ask questions. But
I was still afraid that when we launched, like there's
there's gonna be a lot of blowback. Con Academy's going
all in on AI. It's not perfect, et cetera. We
were working on that, and then all of a sudden

(45:45):
and we were all under a non disclosure agreement with
chat GPT, with open Ai all lawyered up, and and
then and we were supposed to launch March of twenty
twenty three, November twenty twenty two, the last day of November.

Speaker 2 (45:56):
Chat GPT comes out, and I remember, like, what's going on.

Speaker 3 (45:58):
You guys said you were going to launch anything until March,
and then I slacked Greg Brock and he's like, oh no,
we all we did is put it.

Speaker 2 (46:04):
We launched all these apps on our old model.

Speaker 3 (46:06):
Chat GPT was one of them, just a chat interface
on GPT three point five. We too are surprised. We
didn't think anyone would care until GPT.

Speaker 2 (46:14):
Four came out.

Speaker 3 (46:15):
We were really wired because that was the moment where
everyone says, oh my god, this thing has a lot
of errors, it's bad at math, and it's a cheating tool.
And then you saw these school districts start to ban it, etc.
And we're like, oh no, we're working on this thing.
We're gonna release in March, and everyone's already assuming that
this is evil. It ended up that that was actually
a blessing because by the time March came around four
months later, the school system actually had said or started

(46:38):
to think, well, you know what, this does have issues,
but if only someone were to mitigate the risks, but
this is going to be part of kids' lives and
it couldn't theory be used to teach them? What if
we could use it that way? And so when we
launched Conmego, it had positive reaction. And I think what
we're seeing from teachers is for some teachers there is
a little trepidation, but as soon as we make clear, like, look,

(46:59):
if I harted teaching sistant for every one of you, you
would love that and you wouldn't feel threatened no matter
how amazing that teaching assistant is. As long as you're
in charge, that's what this is. And then Amani like,
oh wow, yeah, this is really a gift. But and
the other benefit is teachers are immediately seeing how it
benefits their lives. I give that example of an English
teacher having to spend seventeen hours grading one assignment, or

(47:19):
if you're spending ten hours a week lesson planning or
writing progress reports, and if you can shorten that by
a factor of five or a factor of ten, that's
an immediate benefit for the teacher.

Speaker 2 (47:28):
And it's a benefit for the teacher.

Speaker 3 (47:30):
If you have a classroom of thirty kids and they're
all at different levels, you can't replicate yourself to address
every one of their needs. So if more of those
kids are having their gaps filled in and you're getting
that information and then you can up level the lessons
and it takes you less time to prepare them. That's
starting to get to, you know, better and better from
the teacher's point of view.

Speaker 1 (47:49):
What is the most surprising way that teachers or students
have used your platform?

Speaker 3 (47:54):
I think early on I write about this in Brave
New Words. This was I saw my opened the book
this way when my daughter. This was in the early
days before we had launched con Migo. We had access
to GPT four no one else did. I had prompted
it to just kind of co write a story with her,
and it was a story about the social media influencer
Samantha who's stuck on a desert island and she was

(48:15):
having like a panic attack because she couldn't share how
beautiful it was with her friends. And then my daughter said,
can I talk to Samantha. I'm like, I guess you could,
and so she said, I'd like to talk to Samantha,
and then the aimedia says, hey, Samantha, here, I don't
know what to do. I can't share this, and my
daughter's like, it's okay, you know, just enjoy the beauty

(48:35):
and like.

Speaker 2 (48:36):
And that was the first one. I was like, this
is science fiction.

Speaker 3 (48:38):
Like, my daughter is not only writing a story with
an AI, but now she can actually talk to a
character that they constructed together in her story. And that
was some of the idea behind Now we have activities
on con Migo where you can talk to AI simulations
of literary figures or historical characters. And you know, we
have something called con World School, which is a virtual

(48:59):
high school, and one of there's a student in India
and she was starting The Great Gatsby. She had a huge,
long conversation with Jay Gatsby AI simulation of Jay Gatsby,
And you know, she told me the whole conversation. She's like,
and I apologize for taking up all of his time.
And she knew that he's an AI. It's not like
she's you know, but and it was a pretty meaningful
conversation about you know, goals and aspirations and never being

(49:20):
able to reach them, et cetera. You know, I'm a
Star trek nerd, especially you know, the next generation of
the Holidack. And I used to think that was the
most fake part of Star Trek, Like even in the
twenty fourth century. Surely we're not going to be able
to have like conversations with Albert Einstein. But literally, that's
already hap there, that's already happening, and it's only going
to get better and better.

Speaker 1 (49:43):
That was my interview with Saul Khan, founder of the
Kahn Academy. One of the things I thought was interesting
is that he's in favor of teaching wrote knowledge. Sometimes
he argues that it is to some degree a really
important factor. But of all the things we covered today,
I have to say that my absolute favorite is the
project of learning critical thinking by debating with an AI,

(50:06):
one that is friendly but firm, and the idea that
we can start doing this right away. We can start
getting our students to take a side on some issue
and put forward logical arguments about it, and get graded
on the quality of their arguments, and then on their
ability to represent the argument of the other side, and

(50:26):
then to switch sides and do it all again. That
feels to me like a place where AI can fill
a niche that is not currently filled, and it can
do this with the loving patience that we would devoutly
wish for when arguing with our classmates. Or our children
or our students. The AI never gets tired or bored

(50:48):
or offended. So in the end, as we start each
school year in a world that changes faster and faster,
I have high hopes that will continue to discover new
ways to teach our students things that last, that we
will give them roots and wings.

Speaker 2 (51:07):
And if we can use AI.

Speaker 1 (51:09):
To help students sharpen their reasoning and engage in deeper
conversations and see the world from multiple perspectives, then we're
not just preparing them to take the standardized tests. We
are preparing them for life. The future of education isn't
just about what we learn, but how we learn, and

(51:29):
with the right technology, I have no doubt that we
can make that future and our students quite bright. Go
to Eagleman dot com slash podcasts for more information and
to find further reading.

Speaker 2 (51:47):
Send me an email.

Speaker 1 (51:48):
At podcasts at eagleman dot com with questions or discussion,
and check out and subscribe to Intercosmos on YouTube for
videos of each episode and to leave comments until next time.
I'm Dave Eagelman and this is inner Cosmos.