The future of learning

This is Stanford Engineering's The Future of Everything, and I'm your host, Russ Altman.

Since we started this show eight years ago, it's become an archive of amazing and impactful

work by my Stanford colleagues.

Research is not something that just happens in the lab, and as you'll hear on this show,

the research at Stanford can impact areas like health, technology, law, and business,

and many other topics that can affect everyday life.

We hope you'll tune in to learn more about how research has the potential to help your

life and to help the lives of people you care about in your family and your community.

In that computer interface, we can observe the learner, and we can observe the learner

just as you said, like Netflix does, like Amazon does.

They're trying to do it to understand you better as a consumer.

Yes, very different motives, but we're trying to do it to understand you better, you individually,

as a human learner.

This is Stanford Engineering's The Future of Everything, and I'm your host, Russ Altman.

If you've listened to this show more than once, or even just once, and you liked it, go ahead and follow us.

That'll help make sure that you hear about every episode and never miss anything about the future of everything.

Today, Candace Till will tell us that research about learning and practice of teaching and learning

need to be tightly linked.

If they're separate, you might not get the best results, but if you can integrate them,

then you will get a cycle of improved teaching and improved learning.

It's the future of learning.

Today, we're continuing our feature, the future in a minute, with Candace at the end of our interview.

I will ask her a few rapid questions, and she will give us a beautiful, short answers.

Before we get started, please remember to hit follow.

If you listen to us once, if you've listened to us twice, no matter how many times,

you probably are a hope or enjoying us.

Go ahead and follow.

We'll keep you up to date.

So we all know that learning is important.

It's important for productivity and you're functioning in the world, but you know what?

Learning also makes people happy.

Researchers in learning learn a lot about how to do the best teaching and how to facilitate

learning, especially in adults.

However, there's a problem.

Sometimes the research that is done in the academic, controlled setting doesn't always transfer

well into the real world teaching and learning schools.

Well, it turns out that there are ways now where we can integrate the research about learning

with the practice of learning to have a tighter feedback loop and to increase the relevance

of the research to the actual practice of learning.

This means we have researchers who are producing better and more robust results and we're having

learners who are doing better, learning faster and better.

We'll hear about that from Candace Till, who's a professor of education at Stanford University.

She's an expert in media and technologies in the context of learning and she will tell

us how learning knowledge and research has evolved and how AI at the end is impacting

her ability to build effective learning environments.

Candace, to start out, why did you decide to focus your career and your research on the

science of learning?

Oh, because learning and the capacity to learn and having agency in your learning has pretty

much every kind of outcome, positive correlation that you can imagine.

I wanted people everywhere to be able to learn what they wanted to learn to build the

knowledge and skills they needed to do what they are trying to do in the world.

Great.

And so kind of a little bit out of order, can you tell me what is the science of learning

and what are the current big challenges to the field?

Oh, the science of learning, it's, first off, it's a really nascent science, unlike physics.

It's not like we know all of this stuff and it's just a matter of getting everyone to

know what we know.

It is, so the science of learning is studying actually how people learn, that moment of

where your knowledge and capability changes from one moment to the next.

And it's an interdisciplinary field.

People come at the science of learning from obviously from education, but also psychology,

communication, computer science is huge, neuroscience, all of those are involved in really

understanding the processes of human learning.

And you work at a school of education.

Is the science of learning a kind of integrated into the curriculum for educators?

Not for all educators at our school at Stanford.

We have a particular one of our graduate programs as well as a master's program is called

Learning Science Technology and Design.

And that is a real focus on both the science of human learning and also how you use emerging

technologies and a design science to take what we know about human learning and use it

to design effective educational experiences.

And so I hope this is a fair question, but when you look at how teaching and learning

happens in the world, and I think your expertise is adult learners, so like, and I think that

includes college, we're going to call college students adults for the purposes of this

discussion.

They are.

When you look at the way we instruct them, are we making the best use of our knowledge

of learning science or do you see a little gap there?

Oh, there's a gap there and everyone will always talk about the gap that you're pointing

at, which is we know a lot about how people learn, but very little of what we know actually

influences practice.

But I want to talk about this as a bi-directional gap.

There are a lot of people and I know most of us have experienced them that are amazing

teachers and some of them are formal teachers, some of them are just people who help us

learn things.

And they actually have a lot of wisdom and knowledge about how to support someone to

learn something.

And the other gap is all of that knowledge, all of that skill doesn't find its way back

into our science.

So I call it a bi-directional gap.

And my life's work is really shifting that relationship between the science of human

learning and teaching practice, whether that practice is done by a human or by a machine.

Great.

And thank you for that last comment, because I wanted to get to the fact that you have

embraced technology as both to do this research in your field, but also as a potential way

to deliver teaching and learning as you just described.

And that's not obvious to me.

For example, I could imagine a learning specialist saying technology is not really part

of this.

We're talking about humans inspiring other humans.

So how do you think about technology for education and learning?

I think about technology as providing infrastructure for human learning.

In the sense of, if you think about it, when you're either a learner or when you're a teacher,

you're making minute by minute decisions.

And your decisions are trying to answer the question, if I'm the teacher, for this

learner, who's in this particular state that they're in, and we're trying to help them

to get here, what's the most effective thing I can do for that learner to help them move

from where they are to where they're trying to get to?

Now, in order to, we call those instructional differentiation decisions.

Oh, good.

That's, it sounds very technical.

That's not really exciting.

Now, in order to make a good decision, really, we would need to consider features about

that human learner.

We'd also need to consider features about the thing they're trying to learn.

And really importantly, features about the context in which that learning is occurring

or in which that knowledge will be applied.

Now, if you think about it, trying to think about, what do I know about the learner?

What do I know about what they're trying to learn?

What do I know about the context we're in right now?

And putting all of those variables together and thinking about, given that, what's the

best thing for me to help this learner, that is way beyond our human capacity for managing

features.

But it's not beyond our algorithms.

So I'm not talking about using algorithms to replace teachers, but to support that process

of human decision making for both the learners and for the teachers and for people designing

learning assets and for, you know, my favorite group, learning researchers.

Great, great.

So thank you for that.

And that allows me then to ask you about some of your historical work, which the comments

you just made to help us understand why you would do this.

So one thing that you did, I think even before you came to Stanford, was called, I think,

the Open Learning Initiative.

So now that'll put some meat on the bones.

Tell me, like, what was that?

How did it work?

Why did it exist?

Okay.

Thanks.

The Open Learning Initiative started in 2002.

So from education and AI terms a long time ago, a pre-AI era, a pre-generative AI.

We were actually using AI even then, and I'll talk about that in a minute.

So what happened was the William and Floor Hula Foundation had just funded OpenCourseWare,

which was MIT's project to make all of their course materials openly available online.

And they were going around asking other universities to make OpenCourseWares.

At Carnegie Mellon, we were really happy about the open idea, but the challenge was to really

support someone to learn something you need more than just course material, you need instruction.

So building on the decades of research we had done in cognitive tutoring and other

learning sciences, our pitch was to create a smaller number of courses, but that they

were designed to support humans to learn, using what we knew about the science of learning.

And then I started just collecting the data from the learner actions, because I wanted

to know was there evidence that what we were designing was actually helping people to

learn something.

And then about a year after we started the Open Learning Initiative, we got one of the

big NSF science of learning grants where then the OLI courseware became a research environment

where we were collecting those data, not just to understand where the things we were designing

helping people learn, but also to then do experimentation in different strategies to

see what actually helped learners.

So can you paint me a picture of what a learner experienced as part of the online learning

initiative?

I'm taking that they weren't just sitting in a classroom with a sage on the stage, lecturing

at them.

No, no, one of the things we knew back then is that learning is as one of our articles

said, learning is not a spectator sport, that what you need to do to learn is you learn

by the process of trying something and then getting feedback.

And that feedback can be, you know, feedback from, oh, you got hit in the head, that's

a kind of feedback by the ball when you threw it wrong or something.

But another kind of feedback is timely and targeted feedback that can understand the

move you made and make an inference about why you made that move and then give you explicit

feedback about, oh, well, it looks like you did that because you were thinking this part

of that was right, but this is where you might have gone off track.

And then, and then most importantly, the opportunity to practice again once you've gotten

the feedback.

So the OLI coursework where had small amounts of text, no, no video at that point in time

because we're talking 2002, no lectures, but small amounts of text and then lots of interactive

activities where the students could try something, get feedback, get hints if they needed,

if they weren't sure what to do.

Was it a mostly, so was an online experience?

It was a, it was a computational experience.

Completely.

It was completely online and did it also involve any other students or was it kind of a one-on-one

tutoring type setup?

This was a one-on-one tutoring type setup and that was, that was in line with the goal

of the Hewlett Foundation at the time, because they were trying to make, quote unquote,

a high quality post-secondary education available to anyone in the world that had a good internet

connection.

Great.

So how did OLI or the online learning initiative, how did that evolve over time?

I know it did.

Oh, yeah, totally.

I'll do it over time.

And so I should say the way that we developed the courseware in OLI was not just I thought,

oh, here's a great course.

Let's build it.

A key ingredient of the OLI project was collaboration, interdisciplinary collaboration.

So we brought people that were experts in their field, together with learning researchers,

together with UX designers, user experience.

User experience, yes.

Together with software engineers and together collectively designing the environment.

The other thing that was really kind of cool about it is faculty from different disciplines,

like we were building courses in biology and in chemistry and in statistics and in logic

and in economics, all these different faculty coming together, talking with each other

about the challenges they were having and the solutions they were coming up with.

So it started this incredible interdisciplinary dialogue across different fields of what

we think of different fields of higher education.

But everybody focused on how do we support the learners to get these concepts?

Were the outcomes good?

The outcomes were great, actually.

Probably the most, the study, we were constantly doing studies, but probably the most famous study

we did was called the Accelerated Learning Study.

And that was, and that was, again, my program officer at the Hewlett Foundation, Mike Spith,

said, Candace, you're telling me all this stuff about using what we know about human learning

to design environments, how do you know it's working?

How do you know it's better?

And I'm like, you know, I tried to explain how educational research is challenging, it's

hard to get good effect sizes, et cetera.

And he's like, okay, well, show me they can learn faster.

As a faster, he said, yes.

So we did a study where we took introductory statistics course students and randomly assign

them to condition.

In the traditional condition, they took the traditional statistics course, which is 15

weeks in length and four class meetings a week.

In the OLI condition, they completed the same course, but in eight weeks with two class

meetings a week.

So, you know, so half, half the entire quarter of the instructor contact hours.

And then we used measures of, both at both, they all took the same midterms and final.

So we had those measures, but we also used another measure developed by a different research

group measuring knowledge of introductory statistics as pre-test and post-test.

We also sampled both conditions during the study to make sure that the people in the

OLI condition weren't just spending a lot more time in those eight weeks on statistics

per day or per week.

And the results were that the students in the OLI condition on the midterms and finals

performed as well or better than the students in the traditional condition.

And they performed 18 points better than the students in the traditional condition on

the external measure of statistics knowledge.

A funny thing about that I want to share is that the faculty member, the faculty member

that was doing the two meetings a week with the students, before the study, he's like,

can't assist this ain't gonna, I'll do it, but it's not gonna work.

And it's not gonna work because my students learn in relationship with me.

And if I'm only seeing them for eight weeks and for twice a week, I'll never get that

won't happen.

And after the study, he said to me, this is the best experience I've had teaching statistics

in my 15 years of teaching statistics.

And I asked him why.

And he said, well, for the first time coming into class, the students knew what they knew

and what they didn't know.

I knew what the students knew and what they didn't know.

But probably most importantly, the students knew that I knew what they knew and didn't

know.

But we could spend that, you know, 50 minutes we had together focused on authentically,

this is where you are, this is how I can support you.

Really great story.

And so that is very important for everyone to know because it shows that really this

is incredibly important to get out into the world.

Now, I know you've done many things since, but one of the things that's fun to hear about

is you did a brief stop or not so brief stop in the tech industry at Amazon.

How did that happen?

How was that?

How does it inform your work today?

OK, so I went from Carnegie Mellon to Stanford and I was at Stanford for about four or five

years continuing this work and I got a call to come up and do some consulting for Amazon

because they had a particular workplace learning problem they wanted to solve.

So I came up and consulted to them and I left and thought, OK, they're not going to do

what I said.

So that's fine.

And I stood into the VP who had invited me and she said, well, if you want to do something

different at Amazon, you write this thing called a PR FAQ.

And that's a press release of when the intervention that you've designed is released into the world

what will be written about it.

And then the FAQ is all of these different questions, the first being, who's the customer,

what's the customer problem, and then the rest of the questions are, how do you think

your solution is going to act?

What's your evidence?

So I wrote one of those about my proposal and they decided they wanted to do it.

But they said, but this isn't a consulting gig.

You need to come up here to Seattle and lead this team.

So what my team did was our goal was to innovate and scale workplace learning for Amazon.

So the idea was, I mean, Amazon at the time had 1.6 million employees and a need to rapidly

upskill people in complex capabilities.

And they recognized that the old mechanisms for doing that just weren't cutting it.

And they also recognized that they were looking at their growth trajectory and didn't think

that higher education was going to be able to supply them with enough talent to meet

that growth trajectory.

So they wanted to develop an infrastructure in a system for rapidly developing complex

capability.

And that perfectly aligned with my research.

And also, as you can imagine, I'd spent the last 15 years trying to build the technical

infrastructure using NSF funding and private foundation funding and graduate students.

And here was a well-resourced company that said, we want you to build what you want to

build.

So come here and we will give you the resources to do it.

My hope was, and I said this to them, that my real passion is for not-for-profit higher

education.

So my hope is, once we build it and demonstrate it works, that you will give it away.

They chose not to do that, but you know, that was my hope.

But you still have the knowledge and the takeaways from that experience.

Absolutely.

So then that's what I saw, I took a leave of absence from Stanford.

And I actually ended up having to resign from Stanford because only two year leave of absence.

But then fortunately, my colleagues and Stanford wanted me back.

So I did get rehired.

And so now I'm building and extending what we built for the public good.

This is the future of everything with Russ Altman.

We'll have more with Candace Till next.

Welcome back to the future of everything I'm Russ Altman.

I'm speaking with Candace Till from Stanford University.

In the last segment, we learned about learning science, how to think about learning and

what are some of the challenges in making sure that learning environments for learners are

effective and work in the real world.

In this segment, we're going to talk a little bit more about this relationship between research

and practice and how to optimize it.

We're also going to talk about how AI is creating great new opportunities for much richer environments

for people to learn it.

Candace, in our conversation, you made a reference to the relationship between research

on learning and then turning it into practice.

And I feel like there's still some just conversation to have about that.

So what is your current way to think about, I don't know if it's attention, but it's

certainly two different activities.

Learning how students, learning how people learn and then translate that into effective

teaching and learning strategies.

So where are we with all that?

Okay, so I'm going to first push on that assumption that that's a linear process.

That's what I'm trying to change, that our notion is we researchers do our research,

we create our causal claims and then we throw it over the over to practitioners and say implement

this and people will learn better.

And when practitioners then take our research and try and implement it and it doesn't work.

And then we say, oh, that's because you didn't implement it with fidelity.

You didn't, you didn't do it exactly the way I designed it.

And then practitioners reasonably say, nah, you designed it with all these artificial

constraints that don't exist in the real world.

So that causal chain that you created that exists for your system, it doesn't exist

in the real part.

Yes.

So what we need is rather than this linear model to engage practitioners as collaborative

researchers.

And what I mean by that is everybody who tries to teach someone something, you can think

about that as you're testing a hypothesis.

You think, okay, here's my learner.

I think if I do this, it will help them learn what I'm hoping they'll learn.

And if you think about that as generating an observation, and then we have an infrastructure

that we can collect all of those observations, then our expertise about teaching and learning

can be not just from these scientifically controlled studies, not just from our own personal

intuition or observation, but from being able to see the observations essentially that

everybody's making, which then allows us also to include a lot more diversity and a lot

more voices in our science.

It sounds to me both totally reasonable, what you just said, and a lot more complicated

to like orchestrate.

Yes.

Is that fair?

It is fair.

And that's where the, that's where a lot of the emerging technology can be really helpful.

I think in, I think that vision I have of practitioners as collaborative researchers

in place-based learning is very difficult to orchestrate.

But if you have the right technical infrastructure, then every move I make can be captured and

as an observation without my having to, to structure and experiment or do extra work

or what have you.

So really it's the technical infrastructure that's the magic sauce.

Great.

So it's remarkable.

We've gone, you know, 20 minutes without saying the word AI or the letters AI.

Tell me, is AI, you kind of just imply that AI could be part of the solution.

And so tell me, particularly generative AI.

So have you and your colleagues like assessed the value there and how, where is, what promise

if any, is it, is it providing?

Okay.

So I would say, first of all, I do make a distinction between AI and generative AI because

I've been using AI to support human learning for over 20 years.

And mostly in the old style of AI where we are collecting the learner's actions and then

modeling those actions and then from that making an estimate of where the student is now

and also using AI to, given where the student is now and where the student's trying to

get to make a recommend, build recommendation models based on past evidence about what kind

of learning experience is going to have the highest probability of supporting the learner.

It's almost like Netflix or Amazon, but for a learner, like, you need to watch this

movie.

You need to do this module of learning.

Yes.

So it's a, I used to always talk about the, I think that's the power of the technology.

Is not that we build learning experiences, but that we build them at an interface because

in that interface, in that computer interface, we can observe the learner and we can observe

the learner just as you said, like Netflix does, like Amazon does.

They're trying to do it to understand you better as a consumer.

Yes.

Very different motors, but we're trying to do it to understand you better.

You individually, as a human learner, better, but also to understand learners as a collective,

the processes of human learning better, and that's the power of this technology, the ability

to collect information from a well-designed interface, model that information, and use

that to give feedback to the human actors in the learning system.

That would be learners themselves giving them feedback, giving feedback to instructors

as we've talked about, giving feedback to designers that say, you know, that asset that

you thought you designed to help someone learn this, it's not doing so well.

And here are some recommendations about how to improve it, and then feedback to learning

researchers so we can start to extract of fundamental principles of human learning.

So that's the power of sort of using the, I think, old AI technology.

Yeah, but maybe we'll call it predictive AI.

Yeah, there we go.

The power of the generative AI, of course, is in the communication interface.

I mean, for decades, I've built dashboards to try and communicate to learners.

This is where you are, this is where you're trying to get to, et cetera.

These are some actions you could take next or two instructors with similar kinds of information.

And, you know, and dashboards can be great, but often people don't know how to interpret

them.

They feel overwhelmed by the information that they're not super effective, actually, in this context.

So here's a power of generative AI, because it's a community, it interprets language.

So I could give the AI all the insights that's in the dashboard.

And rather than having you try and look at a dashboard, you could look at a dashboard,

but you could say, well, tell me this.

And the AI can use the data that it has access to, to give you the information you need and

engage in a conversation with you around it.

So I think a big part of the generative AI is in that human interaction interface.

The other part is the other thing that generative AI is doing is making technical skills much

more accessible to other people.

For example, if you are trying to draw something, a draw process, like I want to draw a diagram

for you that explains osmosis or something.

If I use paper and pencil that I'm very focused on, how's my circle look or whatever.

But now if I can describe it to an AI and it can draw a good graphic representation.

And even in the process of my describing it, I'm having to refine my understanding of

it.

Then that act of creation does two things.

It supports a different way of supporting someone to learn something, learn through creation,

where the focus is on the actual thing that you're trying to design and not on how well

did I draw that circle.

Right.

But also for instructors or for teachers allows me to build a very effective and professional

learning experience to get out things into the world that I want people to understand

and the way I want people to understand it.

So I think it gives, it's an opportunity to redistribute voice and power.

That's my other excitement about it.

So this, yeah, this really does, it sounds very exciting and it sounds to me like you're

taking all of the learnings from the pre-generative AI era, the predictive parts and you're marrying

them to the generative part as part of a richer interface and a more adaptable interface

to the learner.

I'm guessing, and you actually said it already with this example of drawing, I'm guessing

that this doesn't just have to be a textual interface, it could involve, because people

are always saying sometimes I've heard of many times people say, I'm not a book learner,

I'm a visual learner.

And it sounds like there's a space for all of those styles because of the generative AI.

Am I extrapolating too much and saying that?

No, no, you're right, but I also want to correct something.

That's perfect.

And we said, you know, I'm a visual learner, people have these ideas, I'm a visual learner,

I'm an auditory learner, I'm a kinesthetic learner.

Yes, yes.

And I just want to say that's a great example of where my field let us all down.

We had a theory of that there were differences in how people process information and that

if you could only know your learning style or a teacher could only know your learning

style and give you that information in that style, you would learn better.

That's where you kind of got out into the world, but for there was actually empirical.

Oh, my goodness.

Oh, my goodness.

And a whole industry developed around learning.

I did not mean to perpetuate a falsehood, so I apologize.

I'm not the only one.

I mean, it's always, I love the opportunity to debunk that myth that it makes, it makes

really good sense because I know that I never really pigeonholed myself into any learning

type.

And in fact, my instinct was always depending on what I'm learning.

I'm all of the above.

There you go.

And when we do know the good thing about it is we do know that multiple representations

support learning, so that if I can give you some, if I can, if you can experience something

kinesthetically, if you can experience it auditorially, if you can experience visually,

then all those multiple representations will actually support your learning.

But people, I mean, there are, I mean, people have physical differences.

So there are some people that that one mode is better, but for the general population,

you are right that your learning style is inextricably intertwined with what you're trying to

learn.

Well, this is great.

And thank you for painting a picture of the future of learning that is extremely exciting.

Before we end, we have our segment called the future in a minute, and I'm wondering if

you are ready for me to ask you some short questions and for you to give me some short

answers.

Yes.

I am.

Okay.

Great.

Thank you very much.

Next question is, what is one thing that gives you the most hope for the future?

Okay.

The new tools, if we design them well, can give voice to people to participate who previously

did not have voice.

What's one thing you want people to walk away from this episode remembering?

The science of human learning is at the start of a scientific revolution in understanding

human learning.

One can and should participate.

Aside from money, what is the one thing you need to succeed in your research?

A good data infrastructure that is for the public good.

If all goes well, what does the future look like?

The science of learning is continuously improving itself so that we have the ability to support

all learners to have agency and learn what they want to learn globally.

And finally, if you were starting over again and you needed to get your certification or

your degree in a different discipline, what would it be?

You know, that's hard because I love, I mean, I have degrees in education and computer

science and then also in social science and literature.

So I kind of did that.

If I were steady, if I were going to study something new, it would be something completely

different, probably music, because I know nothing about music.

Thanks to Candace Till, that was the future of learning.

Thank you for listening to the future of everything.

We have a lot of episodes in our back catalogs and so please check it and make sure that

you get all the content you might be interested in so you can learn about the future of everything.

Also, you know what?

If somebody popped into your head while you were listening to the interview with Candace,

why don't you recommend the show to them?

They might enjoy it as well and if it popped into your head, maybe you learned that they

should be a listener as well.

You can connect with me on many social media platforms, including LinkedIn, Blue Sky, Threads

and Macedon where I'm at RB Altman or at Rustby Altman.

You can also follow the Stanford School of Engineering at Stanford School of Engineering

or more short at Stanford ENG.

If you'd like to ask a question about this episode or a previous episode, please email

us a written question or a voice memo question.

We might feature it in a future episode.

You can send it to the future of everything at stanford.edu, all one word, the future

of everything, no spaces, no underscores, no dashes, the future of everything at stanford.edu.

Thanks again for tuning in.

We hope you're enjoying the podcast.