0:03

This is the Discovery Files podcast from the U.S.

0:06

National Science Foundation.

0:09

Artificial intelligence is becoming ubiquitous, transforming Americans daily lives.

0:14

AI driven technology is promising practical solutions

0:17

to global challenges, from agriculture to health care and education.

0:21

NSF supported AI research advances breakthroughs that push the frontiers

0:25

of knowledge, benefit people, and are aligned to the needs of society.

0:28

We're joined by Dr. Sethuraman Panchanathan,

0:31

the 15th director of the U.S. National Science Foundation.

0:34

Dr. Panchanathan has more than three decades of experience as a computer scientist and engineer,

0:38

where he's contributed to advancing research, innovation, strategic

0:42

partnerships, entrepreneurship, global development, and economic growth.

0:46

Dr. Panchanathan, thank you so much for joining me today.

0:48

Thank you. It's great to be with you. I'd like to start with a little bit of background.

0:51

And in your case we're talking about AI today.

0:54

And I want to get into your computer vision background a little bit.

0:57

Could you tell us what computer vision is and how do we see it in the devices

1:01

we use every day? You know, the best way of describing computer vision is, I mean, we know vision

1:05

because we all have the ability. Most of us have the gift of sight. Right?

1:09

And so if you look at the computer, therefore,

1:11

having the ability to perceive the world around us through the capabilities that we have through our vision,

1:17

and that would then qualify as computer vision in the sense that you're processing

1:21

visual information that is around you and be able to make some meaningful

1:27

decisions based on the understanding that you have

1:30

processing the visual information.

1:33

So that would be thought of as computer vision writ large.

1:36

So it's very important to kind of the architecture of how AI is working

1:39

right now. It is because AI is essentially a mechanism by which what you understand

1:45

then helps you to be able to

1:48

gain knowledge and use that knowledge for further

1:52

making the correlations connections, and be able to help advance

1:58

any application that you are trying to advance if, let's say, transportation.

2:02

So you're able to gather the information to computer vision.

2:04

You have an understanding of what is happening. Imagine you're driving your car with the many cameras that you have.

2:10

It understands what is in front of you and what is on your side, and even the back through the cameras.

2:15

And you acquire the information. You process it and then you gather from that.

2:20

So the features that you process, you get an understanding of what is happening around you.

2:24

And what then AI does is able to then

2:26

learn from what is being processed

2:30

so that it can be used for navigating you

2:34

through terrains that may be already, terrains that you've seen, or trains

2:38

that are unknown, and being able to draw the correspondences from the terrain

2:42

that we have known before to terrains that are similar or things that are unknown.

2:46

And you're trying to make connections to the unknown tenants.

2:49

So that's the kind of thing that I provides you,

2:52

the ability to be able to learn from the data,

2:55

to be able to predict, to be able to help make decisions.

2:59

So these are the things where I as a field

3:02

is, you know, augmenting the vision capabilities.

3:06

And so the computer vision and AI then becomes a much more powerful

3:10

way of being able to perceive the environment and act on it.

3:14

So we kind of contextualize that with vehicles

3:17

and how it's kind of analyzing the environment for people

3:20

that are just hearing the term artificial intelligence.

3:24

Can you kind of explain what it actually is?

3:26

No, it is basically that, you know, when we as humans,

3:30

when we see something, you classify whatever you see.

3:34

When I see you, for example, if I've seen you before, I recognize you.

3:38

If I have not seen you before, then I say, oh, he looks like this person.

3:43

Or you attach a label to the new person that you're seeing.

3:48

Even though you may classify it as close as something you say, now I know him,

3:52

and I know that this is what his name is and this is how he looks.

3:55

And so all of the features that you have with your name then get attached, right.

4:01

And so now if I start doing this for all the things that they encounter,

4:06

and that helps me in recognition, but also it helps me in things

4:11

that I have not seen before and be able to classify to the nearest possible match.

4:16

And this kind of a way in which you are learning from the data.

4:20

So here we are talking about vision and recognition through vision.

4:23

But it doesn't have to be a vision, any data for that matter.

4:26

The more you're able to train based on the data that you have seen

4:31

and that's what you do as a human, then you use that for

4:35

not only understanding the data or recognizing similar things patterns,

4:39

but you're also able to navigate to things that you might not have seen before,

4:43

but you have a sense of what it might be like based on that learning.

4:47

So in a sense, you can say AI is basically

4:50

a set of learning prediction

4:54

and therefore being able to make progress in terms of making decisions.

4:59

But it's mostly about how you're able to take

5:03

the information that is at hand, process them,

5:06

and then build models based on that so that learning

5:10

those models can help you navigate through new data on a similar data,

5:14

or even exactly that data in, depending on what the outcome is.

5:17

Whether you want to recognize or find the closest thing that is to it,

5:21

or be able to build things with what you have learned

5:25

and be able to now build things based on what you have learned.

5:28

So you said progress there, and I want to think about the future.

5:31

Can you tell us some of the things that I could be used for?

5:35

Everything that you can imagine in the world around you, because it's like

5:38

asking the question, you know, but what can a human be useful for?

5:42

The question is no different, isn't it? Once you have a person who is capable of

5:46

doing something, then you say that this person can do these things.

5:50

So you take any task, whether it is a driving task,

5:54

whether it is a task that is based on

5:57

predicting the weather or whether it is a task based on how do I learn,

6:02

what do I need to learn with that which I don't know anything about?

6:04

You take any task for that matter. You can always find a way that I can be useful in that context,

6:10

because at the end of the day, it is all about data.

6:12

Learning from the data, building models and then using those models,

6:16

then being able to predict based on what is not known yet.

6:20

So these are the kinds of things that you do in every aspect of your life.

6:23

I mean, you look at your everyday activity

6:25

right from the time that you wake up until you go to sleep.

6:28

Every activity that you encounter is based on some kind of a recognition

6:32

pattern, recognition as some kind of a decision that you make

6:35

or some kind of a learning that you have that helps you for the future,

6:38

and some kind of set of some new hypotheses that you double

6:42

based on the information that you have gathered.

6:44

All of these things are what you do every day in your life,

6:47

from morning to night. One or more of these kinds of things.

6:50

So I, for me, is exactly similar to that.

6:54

It's now an agent instead of a human who is essentially facing

6:58

the same set of things. A lot of data.

7:00

There are existing models. You build new models or refine the models, use it for recognition

7:06

or using it for developing the closest match or using it

7:11

for developing newer hypothesis

7:15

or helping us in making decisions.

7:17

So all of these things are what we do as humans,

7:21

and therefore I can do similar things in any field that you can think about.

7:25

Right? Is part of the challenge then deciding what fields it's most beneficial to use it.

7:29

I mean. It depends on which field where you need the most help in, and particularly

7:35

in terms of being able to automate more easily, if you could put it that way.

7:39

So the low hanging fruit opportunities might be easier to build models

7:43

with a small amount of data still be able to do meaningful things,

7:46

but that does not preclude us from building these mega models

7:49

that you hear about. ChatGPT and others, which can be useful for a variety of other applications.

7:55

So at the end of the day, it is not limited,

7:57

but it depends on what you want to apply that for

8:00

and with what level of efficiency and efficacy that you want that model to

8:04

work for you, or that particular solution you want to work for you.

8:07

And that's what it is based on. And so if you take health, there are aspects of health

8:11

that can benefit from using AI in in all the way from understanding

8:15

the genome to all the way to delivering patient care, individualized personalized

8:19

patient care. If you take health, if you take transportation,

8:22

you can have a lot of automation built into the transportation.

8:24

You can have a lot of safety features built because of the automation

8:27

that is there in transportation. It might help in terms of speeding up things versus where they are

8:33

right now in terms of what you can do with existing transportation infrastructures.

8:37

So it could be in transportation.

8:39

It could be in learning. A student now can find that they can do anything in terms

8:43

of advancing their learning aspirations by being able to find what are those

8:47

specific things that you find that you don't understand.

8:49

Well, now, for example, if you're taking Calculus Course,

8:53

let's say that some of the calculus course requires you

8:56

to have an understanding of mathematical concepts

8:59

that were in your prior mathematics courses,

9:02

but some of which you are very familiar with.

9:05

Some of this you find that you either did not learn

9:07

well or you've forgotten whatever that might be.

9:10

And so what happens is, depending on what you might need,

9:13

you could go back and get the appropriate things filled up

9:16

so they are able to understand new concept them.

9:18

So essentially it becomes like your partner if you, me,

9:21

you and the machine in this case AI are working together in terms

9:26

of being able to expand your capabilities or your effectiveness.

9:31

And so that's what makes it so interesting, exciting

9:35

in whatever field it is in, you can find something

9:39

that will enrich you, empower you, augment you,

9:43

make you a lot more effective and efficient.

9:46

There's a lot of fear with people and their job security

9:50

and how this might impact that. Can you talk about some of the ways I seem to be able to be worked alongside?

9:57

How will it be a tool that benefits workers everywhere?

10:00

That's a good question. Again, it depends on the work.

10:02

Is such a broad thing right?

10:05

There are so many dimensions and classifications

10:08

of the types of work in every type of work.

10:12

It doesn't matter whether you're a physician or a physician assistant,

10:15

if you are a student, a tutor, or a professor,

10:19

if you are a truck driver or you're an automobile designer,

10:24

it doesn't matter what your type of workers.

10:27

You can always find those things where the AI tools can be very helpful

10:32

to you to enhance your capabilities and abilities.

10:36

And I would say that the things

10:39

that you may not be doing right now, which you are capable of

10:43

because you're doing the things that otherwise occupy you.

10:47

If you're unburdened by those things,

10:50

then you are able to now use your skill sets even more,

10:54

and maybe your creative mindsets are expanded even more,

10:58

and artists can now do even more creative art.

11:02

A physician can do amazing diagnosis and treatment beyond what they are right now.

11:07

All of these are possible because you now this becomes a companion for you.

11:11

So in a sense, when people worry about all, is my job going to be lost?

11:15

It is not that you lose. Your job necessarily is always the case.

11:18

Yes, in some cases that might be the case, right?

11:21

A machine may be an AI device may be able to do that what you do.

11:26

But to me that frees you up to doing other things.

11:30

Now the pace of progress clearly is much faster than it was before.

11:33

I mean, you could argue that similar thing was true in in

11:36

when tellers were there and then the ATM machines came about is

11:40

so you could think of it that those who are tellers,

11:43

the 50 of them in a bank now, maybe there's 1 or 2 with the ATM machine.

11:47

Let us say on what you do electronically with your mobile device.

11:51

But those 48 other people have found other ways

11:54

in which they are using the talents and skills, not necessarily

11:57

only in the bank environment, but in other environments.

12:00

This is what will happen at AI too, is that the jobs of today,

12:04

some of them will be augmented, enriched, enhanced.

12:07

Some of them will be lost, in which case

12:10

people will naturally gravitate to learning those skill sets and mindsets

12:13

that they will expand to being able to create the new jobs of the future,

12:18

or to do the new jobs of the future.

12:20

So it's very hard to say, by looking at only a static picture

12:24

of where we are and saying, what I am doing today, be

12:27

gone is a very narrow perspective through which you look at the world, right?

12:32

It's much, much, much more broader than that.

12:34

Now comes with the associated questions and things that we ought to do.

12:39

So there is no longer this feeling that, you know, you had this

12:42

18 years of education or 21 years of education

12:45

or 25 years of education, whatever that is that you do.

12:48

And then you go take a job and then you stay with the job.

12:51

And that was no longer true. Even, you know, a couple of decades ago.

12:53

And it's no more becoming less and less true now. Right?

12:57

So you can imagine that for that, learning is a lifelong pursuit.

13:01

You're constantly upskilling, reskilling, retooling, learning new things.

13:07

So it is not like, you know, let me finish my studies and then I have my work.

13:12

It's the learning and work and learning have become so intertwined

13:16

that it is a symbiotic. It's kind of an activity.

13:19

And in that kind of a scenario, nobody's outdated.

13:23

No skill becomes unnecessary because you have to acquire new skills,

13:27

so your skill becomes different. It constantly shapes, modulates itself and so on.

13:32

So that's something that is very interesting

13:34

as a paradigm shift, if you want to look at it that way.

13:37

But what is even more exciting is that you can think of the kind of jobs

13:41

that you were doing that you felt like, why am I doing all of this

13:45

is now going to be replaced by machines that can do the job

13:48

that allows you to express your skill, ability, creativity,

13:53

and other kinds of abilities that you have to be able

13:57

to express in its fullest form. And we may not even know what they are.

14:00

It's some of these things may not have even come out of you

14:03

because you've been anonymous about boredom, but

14:06

you have been engaged in those activities

14:09

that have not let your creativity manifest itself in its fullest form.

14:13

Human talent is capable of doing a lot, lot more than what we think.

14:18

So I think I want to look at it from that futuristic perspective,

14:21

because if you look at anything

14:24

which is empowered by technology will be thought of as,

14:27

oh my God, this is going to take away something from me.

14:29

But that is only by imagining yourself as that

14:33

just static picture of today, but not envisioning what it could be.

14:36

The tomorrow. So thinking about tomorrow, I want to ask you about how NSF

14:41

is working to kind of guide how AI is developed in the United States.

14:46

Let the questions so clearly you said, what area is this?

14:49

I have an impact. And I almost answered the question, missing everything and anything.

14:54

And you know what kind of impact it will have on humans.

14:58

We talked about the potential potentials, you know, amazing.

15:02

And and then it could be enormous. We talked about what kind of things it will do for jobs,

15:07

answered by saying that it can empower you, enrich you, and so on.

15:10

So when you look at that, NSF is engaging every dimension

15:15

of all of this, all the way from looking at

15:19

if you need to have a device, the device starts with understanding what

15:25

the materials are to building the devices, to building the technology.

15:29

And you look at the physics of it, the chemistry of it,

15:32

the mathematics of it, and the engineering of it, all of that and the design of it,

15:37

all of that is something that NSF makes possible through its various tweaks.

15:41

Then you look at all the applications that it can be put to use, whether it is,

15:45

a geoscience application like, you know, it could be exploring the planet

15:49

and finding solutions for challenges like climate and mitigation,

15:53

climate adaptation and so on. It could be that

15:55

or it could be a bioscience problem that you're trying to understand

15:58

the fundamental basis of how the human body functions all the way from a cell.

16:02

Right. And how might you personalize strategies for having a good health

16:07

outcome all the time? A bioscience structure, it has those kinds of things.

16:11

How can you build better environments?

16:13

By having good understanding of biology and, and so on.

16:17

Then you look at our social, behavioral, economic scientist directorate

16:20

and all the work that NSF does is at the end of the day, it is all about interfacing with the human

16:26

and all the aspects that come with the social, the behavioral,

16:30

the ethics, the policy, all of that is all centered

16:33

around the work that the social economic sciences director works on.

16:37

And as I said, the mathematical physical center structure. And if you look at applications

16:40

like astronomy, understanding where we came from, right.

16:42

Again, all of that and more is what NSF is constantly engaged in.

16:48

And when we talk about educating the future,

16:50

the talent of the future, again, our education Directorate and all of the directorates are engaged

16:55

with educating, you know, helping with educating better K-12 students,

16:59

better undergraduate students, graduate students, better community college

17:03

students, skill sets that there are all these students build better research

17:08

and then all constantly in all of these things,

17:10

pushing the boundaries of discovery, okay.

17:13

And then providing the platforms that allow all of this

17:16

to manifest itself in terms of better industries of the future,

17:20

that are entrepreneurial outcomes, better entrepreneurs, all of these things.

17:24

I mean, you could I can go into every aspect of it.

17:26

You will find. That's why NSF is a unique agency.

17:29

It touches pretty much every aspect

17:32

of what is needed for us to deliver the futures that we are talking about.

17:37

As much as I as a field is important,

17:40

but I get touched by and influences

17:43

a host of other disciplines areas, and so therefore

17:47

it is a comprehensive picture, not just only what we do in the computing

17:51

information sciences, engineering directorate,

17:53

but as you know, AI has been involved for the last several decades over.

17:56

Something like 50 years. So 50, 50, 60 years, but also in the investing in AI Institute.

18:02

Right. Which means that we are looking at every possible avenue

18:06

by which I can be further propelled into the future.

18:09

But what I can do to propel things into the future,

18:12

these AI institutes are tremendous investments, right?

18:15

That's another thing that that you have seen that we do here.

18:17

So if you look at the collective investments in AI around AI

18:22

and what NSF makes writ large in, in various,

18:26

fields of science, engineering and technology, you find that all of

18:30

that means something for the advancement of this into the future.

18:34

For the final question today, I want to ask you about the future.

18:37

What about how AI is developing excites you the most?

18:41

I think what excites me the most is, you know, I've always believed

18:43

the human potential is enormous.

18:46

But yet if you look at globally in our nation to but globally,

18:50

I don't think we have been able to exercise all of the human ingenuity,

18:54

the innovative mindset and the human spirit.

18:57

And so what I and present to us the opportunity to be able

19:02

to unleash all of the human potential for the benefit of humanity and beyond,

19:07

and that, to me is very exciting that we will be able to do that

19:12

at scale and at speed and everywhere, not limited to a few locations.

19:18

That really makes it very exciting.

19:20

Any minute. Imagine, let's take our nation.

19:23

Imagine every area, rural area, every urban area,

19:27

all the 50 states and territories of our country and people everywhere

19:31

being empowered, energized to be able to exercise their talent to the fullest

19:35

without any holding back, without any constraints.

19:39

Imagine that that can do amazing things for our nation, build

19:42

tremendous prosperity for our nation and through that, and have prosperity possible all across the globe.

19:48

Special thanks to Doctor Sethuraman Panchanathan. For The Discovery files, I'm Nate Pottker.

19:52

You can watch video versions of these conversations on our YouTube channel by searching @NSFscience.

19:57

Please subscribe wherever you get podcasts and if you like our program,

20:00

share with a friend and consider leaving a review.

20:04

Discover how the U.S.