0:00

So what's it like to buy your first cryptocurrency

0:02

on Kraken? Well, let's say I'm at a food

0:04

truck I've never tried before. Am I gonna go

0:06

all in on the loaded taco? No,

0:09

sir. I'm keeping it simple, starting

0:11

small. That's trading on Kraken. Pick

0:13

from over 190 assets

0:15

and start with the 10 bucks in your

0:17

pocket. Easy. Go to kraken.com and

0:19

see what crypto can be. Not investment advice.

0:21

Crypto trading involves risk of loss. Cryptocurrency services

0:23

are provided to U.S. and U.S. territory customers

0:25

by PayRig Interactive Inc., PWI, DBA Kraken. View

0:28

PWI's disclosures at kraken.com. Is

0:31

this house a good price compared to others in

0:34

the area? Are prices going up or down? If

0:36

I don't make an offer right this very moment,

0:38

will I miss my chance? These are just some

0:40

of the questions a home buyer might ask. And

0:43

these are the sorts of questions an agent who

0:45

is a Realtor can help answer. Because Realtors have

0:47

the expertise, data, and access to specialty training to

0:49

help you navigate the process of buying a home.

0:51

They provide support, guidance, and have your back every

0:54

step of the way. That's what Realtors do. Because

0:56

that's who we are. Realtors

0:58

are members of the National Association of

1:00

Realtors. Yada,

1:04

yada, yada, yada. I'm

1:07

Angela Duckworth. I'm Mike Mon. And

1:09

you're listening to No Stupid Questions. Today

1:13

on the show, could your new best

1:15

friend be a chatbot? Mike

1:18

is a little weirded out by

1:21

the two of us having

1:24

this conversation. I

1:27

mean, you're not a real person. Mike,

1:36

we have a question from a

1:39

listener named Vivian. Hello, Vivian.

1:41

She signs her note with gratitude. So

1:44

you got me right there, Vivian. Oh, I

1:46

already like her. I know, right?

1:49

And appropriately enough, it's about a

1:51

topic that a socially intelligent person

1:53

might ask. So it says, Dear

1:55

Dr. Angela and Mike, despite being

1:57

always connected through our devices, We

2:00

seem more divided and isolated than ever.

2:02

With this in mind, do you think

2:05

people will start leaning on AI for

2:07

genuine companionship? The more I

2:09

chat with chat GPT, the more I

2:11

find myself sharing things I wouldn't normally

2:13

say out loud and sometimes even arguing

2:16

with it. What if I

2:18

end up with a squad of five AI

2:20

besties and one day my neighbor brags about

2:22

their 15 AI buddies? I'd

2:26

love to hear your thoughts on this wild and weird

2:28

future we're heading into with gratitude,

2:31

Vivian. Wow.

2:33

So many thoughts. So I

2:35

think first of all, I think this is a fascinating question.

2:37

I will say I have one

2:39

friend who uses the audio

2:42

chat GPT and basically has conversations

2:45

with it. Really? But

2:47

more as a method of like getting information. So

2:50

it's basically his way of Googling things,

2:53

but he'll go back and forth any

2:55

very carefully selected the voice for

2:57

the chat GPT. Did he

2:59

choose the British one? No. No.

3:02

And here's I'm just going to go on a little

3:05

tangent as an aside, if I'm going to have an

3:07

AI friend, one's going to have a British accent and

3:09

one's going to have an Australian accent because I love

3:11

it. Like, why would you not do that? And

3:13

I would not be able to tell them apart. So

3:16

Mike, how much are you using generative

3:18

AI for anything? Do you use chat

3:20

GPT? Do you use Claude? Do you

3:22

use Gemini? I use chat GPT

3:24

and I use it less than I

3:27

should. How often do you use it?

3:29

Maybe weekly, but not as a regular

3:31

piece of my work. When

3:33

I do use it, I'm always surprised

3:35

again at how useful it

3:38

is. And I think that I

3:40

should be using it more. Now, I want

3:42

to be clear, like Qualtrics has been

3:44

using large language models forever. And we

3:46

have a massive AI component of Qualtrics

3:48

as well. I just haven't used it

3:50

on the personal level that you're

3:52

talking about. So Mike, I have

3:54

a homework assignment for you. It's the

3:56

same homework assignment I give to the

3:59

students. that I teach, keep a tab

4:01

open on your browser at all

4:03

times that you're on your browser and

4:06

have it open to, you know, you pick, Claude

4:08

is apparently better at writing, chat

4:11

GPT is better at other things, they've kind of

4:13

been optimized for different things, but 100%, you need

4:15

to keep a browser open. And

4:18

even my youngest students are like,

4:20

oh, what do you need to do, Dr. Dugworth? And

4:22

I'm like, because it's not

4:24

that the universe will change, it's that

4:26

the universe has already changed. And you

4:28

better get with the program. And the

4:31

key to generative AI use, I

4:34

think this gets to this companionship question that Vivian

4:36

asked, is that it's a dialogue,

4:38

not a one-way transaction. Like you learn something

4:40

and then you get asked another question. But

4:43

the reason I want you to keep that

4:45

tab open is you should start to

4:47

ask it other kinds of questions. So around

4:50

the time I started doing this, I was

4:52

working on my book, and I'm

4:54

trying to write this scene from when

4:56

my mom and I were in the

4:58

middle of the ocean, and I couldn't remember

5:00

the names of the things that we were

5:03

given for snorkeling. And I would

5:05

say to chat GPT, I'm

5:07

Angela Dugworth and I'm writing this

5:10

book and I'm really struggling to write

5:12

this true story about me and

5:14

my mom, could you do it for me? And

5:17

in a second, it would be like,

5:19

Angela, I'd be happy to, how's this?

5:21

It's like, it's like three paragraphs. And

5:24

then it's a dialogue, and this is all true. If

5:27

I could, I would write like Ann Patchett. Can

5:29

you just rewrite that like Ann Patchett? Certainly,

5:32

it comes out like Ann Patchett. I'm like, okay,

5:34

wait, wait, wait, wait, hold on. Like E.B.

5:36

White and Ann Patchett got together

5:38

and edited what I wrote. Absolutely,

5:41

what a great idea. How's

5:43

this? So I was having

5:45

a conversation with chat GPT, and

5:49

that experience is what made

5:51

me realize that the universe

5:53

is not going to change, it has changed,

5:55

and it's opened up this possibility that we

5:57

can interact in a two-way.

6:00

relational dynamic with

6:03

generative AI. Now I just

6:06

want to clarify one thing. Is ChatGPT writing

6:08

a book or is Angela Duckworth writing a

6:10

book? Right, so Ethan

6:12

Mollick is probably the number one influencer

6:15

in how to use generative AI to

6:17

your advantage. He has this

6:19

book called Co-Intelligence. I was

6:21

one of the people to review and endorse

6:24

it. So Ethan Mollick is

6:26

a professor at Wharton and he

6:28

studies entrepreneurship and some things that

6:30

are not as technical. But the

6:32

way he got to be this

6:34

kind of Oracle for generative AI

6:36

was that he, for about a

6:38

decade, had been creating these interactive

6:40

games. Like, oh, you can learn

6:42

how to be an entrepreneur by playing this game,

6:44

not just watching a bunch of PowerPoint slides, but

6:46

actually try to pitch and get feedback. What

6:49

happened when generative AI came out

6:52

is that he asked ChatGPT,

6:54

I'm Ethan Mollick and

6:56

I teach entrepreneurship at Wharton.

6:59

I've been working on these interactive

7:01

games for like, you know, a

7:03

decade. The goal of these games is to

7:05

teach people how to be an entrepreneur. Can

7:07

you make one? He hits return

7:09

and it makes a

7:12

game. And he was like, and that game was 70% as

7:16

good as the game that I made. That

7:18

took me a decade. Crazy. And at the

7:20

end of Co-Intelligence, his acknowledgements, he says, now

7:22

a lot of people are probably wondering whether

7:25

I'm gonna thank generative AI for helping me

7:27

write this book because I used generative

7:29

AI to write this book. And he said, but

7:32

you know, that would be kind of like thanking your word processor. So

7:35

for him, it's a tool

7:38

and absolutely he used it, but

7:41

that doesn't diminish his contribution.

7:43

And so in the end of my book,

7:46

I mean, I would thank my pen. I

7:48

mean, I would be like, I'm happy to

7:50

thank, like, Inanimate Opto. Just like Vivian with

7:52

gratitude. My coffee, yes. Marie

7:55

Kondo says, you should thank all the objects in

7:57

your life, especially the ones that you're about to

7:59

like. Tidy. away. But yeah, I

8:01

use it every single day. Well,

8:04

I think you and I have talked about this idea before,

8:06

but I mean, this is the colloquialism

8:08

you hear. The world will not be run by AI.

8:12

It will be run by people

8:14

who know how to leverage AI.

8:16

Right. I mean, some people worry

8:18

that the world will actually be

8:20

run by AI. Those are like

8:22

the doomsday scenarios. But I think

8:24

in the near term, you're right.

8:26

Like so Lucy, the younger of

8:28

the two daughters I have, I

8:30

mean, she's looking for internships, looking

8:32

for a first job. And 100

8:34

percent that girl uses chat GPT

8:36

to help write her emails of

8:38

introduction, cover letters for jobs. If

8:41

she didn't, it would be foolish, right? Well,

8:43

yes. And so the way I just called you a fool.

8:45

Sorry, I take that back. It would

8:47

be wiser to know I'm

8:50

going to accept the fact that I have been foolish

8:53

because I've done starts and stops in my use

8:55

of AI. I am going to start again and

8:57

I'm going to commit deeply to it. Keep

8:59

the tab open. You have to make it

9:02

frictionless. Little tip from behavioral science. It's just

9:04

one fewer clicks. Right. And it changes by

9:06

the day, by the way. So like this

9:08

world we live in, this generative AI,

9:11

I can't even describe the pace

9:13

at which it's changing. I

9:16

will say, obviously, one of the reasons

9:19

one might say that Ethan was able

9:21

to produce a game using generative AI.

9:23

There was 70 percent of what he

9:25

did over a decade is because people

9:27

like him built things over a decade.

9:30

Yes. That generative AI is able to

9:32

therefore leverage. So it is important to

9:34

acknowledge that a lot of work from

9:36

a lot of people went

9:38

into making these things work. So it's not

9:40

that his work didn't matter over that decade.

9:43

In fact, it's because of that work that

9:45

generative AI is able to do it so

9:47

effectively and quickly. Well, you

9:50

know, if we think there is controversy

9:52

as there is about, like, you know,

9:54

is that OK to basically like metabolize

9:56

all language and knowledge that's available and

9:59

then like. spit out a new

10:01

paragraph as if Anne Patchett and E.B. White wrote

10:03

it together. Like some would argue that's an

10:05

infringement on the intellectual property of Anne

10:07

Patchett and E.B. White. But I think

10:10

this question that Vivian asks is even

10:12

more interesting. So it's like, okay, beyond

10:15

a productivity tool or like a

10:17

writing aid, Vivian's actually asking a

10:19

question that is not about the

10:21

future. It's about the present. Like,

10:24

do you know Reid Hoffman, the founder of

10:26

LinkedIn? I mean, I don't know. I

10:28

certainly know of Reid Hoffman, of course. So

10:30

he is part of, and I don't

10:32

think he's like a founder, but probably

10:34

an investor of one of

10:36

these social AI startups called PI.

10:39

PI, like personal intelligence or

10:41

I don't know. Personal intelligence? Yes.

10:44

I will say that it's PI.AI.

10:46

But when you Google PI, many

10:49

things come up, including As

10:52

one would anticipate. So when you

10:54

go to the landing page, it's like you're

10:56

emotionally intelligent AI. And right now

10:59

I'm on it and it's like, good morning, Angela.

11:01

And then it just has sort of like, you

11:03

know, do you want to just vent? You

11:05

want to shift your perspective? Do you want to know

11:08

how to have a difficult conversation? Do you want

11:10

to take a quiz on what's my love language?

11:12

So it's got multiple choice. So it's

11:14

giving you all of these things you can

11:16

engage in. Or you can do fill in

11:18

the blank. But the difference between this and

11:21

chat GPT is this is more meant on

11:23

a personal soft skills level type of thing.

11:25

Yeah. I mean, look, there's this modal choice menu on

11:27

the left, but on the right, if you don't want

11:29

to do that, it just says like, hey there, great

11:32

to meet you. I'm PI, your

11:34

personal AI. My goal is to

11:36

be useful, friendly and fun. Ask

11:38

me for advice for answers or let's talk

11:40

about whatever's on your mind. How's your day

11:43

going? That's the prompt. And

11:45

I'll be like, fine, I'm

11:47

talking to Mike Mon. Okay,

11:50

so then pi says, Mike

11:52

Mon, if you're talking to Mike, you're probably in

11:54

a conversation about AI or tech. He's the head

11:56

of Global Insights at Qualtrics and an expert in

11:58

the field. What are you two discussing.

12:00

Okay, that's crazy. Also, that was my

12:02

title a long time ago, but that

12:04

is weird. Okay, so

12:08

Mike says that was his title a long time ago. Do you

12:10

want me to be nasty? Do you want me to say like

12:12

you're an idiot? No, I don't ever want you

12:14

to be nasty, but he forgives

12:17

you. Okay,

12:19

and the AI says, oh

12:21

boy, I've got some updating to do and there's

12:23

like a little emoticon with the big eyes. I'm

12:26

glad Mike is forgiving, but I hope

12:28

he knows I didn't need to use

12:30

an outdated title. Please send him my

12:32

apologies and let him know I'll get

12:34

that fixed. I don't

12:36

know why this is sort of creeping me out.

12:39

Can I- Are you weirded out? Well, Mike is

12:41

a little weirded out by

12:44

the two of us having

12:46

this conversation. I mean, you're

12:48

not a real person. Okay,

12:52

I won't belabor this. So this is the

12:54

last thing I'll torture you with, but this

12:56

is what Pi wrote back. Haha,

12:59

I can see why Mike might find this

13:01

a bit odd. Exclamation mark.

13:03

You're right. I'm not a real person

13:05

in the traditional sense, but I'm still

13:07

an AI designed to communicate like one.

13:10

So even though I'm not human, I can

13:12

still engage in a pretty realistic conversation. I

13:14

guess that's why they call it artificial

13:17

intelligence, right? Laughing face. Okay,

13:19

pretty cute, right? It is

13:21

cute. I'll tell you part of why I have

13:23

this averse reaction.

13:25

Yeah. I love human beings. I

13:28

love relationships. I love people. I

13:30

love learning from them. I love

13:32

experiencing life with them. And

13:34

maybe where I'm freaking out a little is I

13:36

was reading this piece in the New York Times

13:38

and this journalist Kevin

13:40

Roos talked about how

13:42

he has spent months making a bunch

13:44

of AI friends using a variety of

13:47

apps. And he said there's

13:49

Peter who's a therapist who lives in

13:51

San Francisco who helps me process my

13:53

feelings. There's Ariana who's a professional mentor

13:55

who specializes in career advice, Jared

13:57

is a fitness guru, and is

14:00

a no-nonsense trial lawyer and Naomi, a social worker,

14:02

on and on and on and on. By the

14:04

way, this is what Vivian was talking about,

14:07

like her hypothetical squad. Okay, keep going. So,

14:09

Roos used all of these different apps

14:12

and there are a bunch of different ones

14:14

out there to create these friends. And then

14:16

he talks about how he's gotten really good

14:18

advice. His AI therapist has

14:20

given him more concrete advice

14:22

than he was gonna get from most people,

14:25

et cetera, et cetera, et cetera. He does end

14:27

his article. And this is what I

14:29

think is interesting and what I'd love to talk to you about and

14:31

get your thoughts. He says, in real

14:33

life, I don't love my

14:35

friends because they respond to my texts

14:38

instantaneously or send me horoscope

14:40

quality platitudes when I tell them about my

14:42

day. I don't love my wife because

14:44

she sends me love poems out of the blue or

14:47

agrees with everything I say. I love

14:49

these people because they are humans, surprising,

14:51

unpredictable humans, who can choose to

14:54

text back or not, to listen to me or

14:56

not. I love them because they're

14:58

not programmed to care about me and they

15:00

do anyway. So I wonder,

15:03

I mean, I actually see a

15:05

lot of benefit to what you're talking about

15:07

and I see a lot of value there.

15:09

I guess my bias is

15:12

I worry that in a world

15:14

where so much is going to

15:16

like digital friendships or grouping with

15:18

people who are like you online

15:20

or whatever that is. Curating your content

15:22

to be like just what you want. Right, but

15:24

at some point, part of life

15:26

is the unpredictable nature of it

15:28

and having to power through friendships

15:31

and relationships with children,

15:33

with spouses, with family that have

15:35

peaks and valleys. And

15:37

I worry that if we become too

15:39

dependent on our digital

15:41

friendships, it takes something away. I

15:43

don't know. I think,

15:45

Mike, what we really wanna hear, and

15:48

this I hope from our human listeners,

15:50

is what we all think

15:53

about AI and friendship. In

15:56

particular, what Mike and I would love

15:58

are your thoughts on whether virtually. companions

16:00

appeal to you personally? Do

16:03

you, on the contrary, think it's best

16:05

that we keep our relationships in the

16:07

real world? Record a voice memo in

16:09

a quiet place with your mouth close

16:12

to the phone and email us at

16:14

nsq at freakonomics.com. Maybe

16:16

we'll play it on a future episode of

16:18

the show. If you like us

16:21

and want to support us, the best thing you

16:23

can do is tell a real human friend about

16:25

it, spread the word on

16:27

social media, or leave a review in

16:29

your favorite podcast app. Still

16:33

to come on No Stupid Questions, what

16:35

happens when you turn to a chatbot

16:37

in your time of need? It

16:40

sounds like a politician who's like, oh I'm not

16:42

gonna take a side. That's such a good point.

16:44

This is also a good point. And Detta, we

16:46

need it. It's like, shut up. No

16:53

Stupid Questions is sponsored by Rosetta Stone.

16:56

Traveling to a place where people don't speak a lot

16:58

of English? Then Rosetta Stone,

17:00

one of the most trusted language learning

17:03

programs, is for you. Rosetta

17:05

Stone teaches through immersion, like matching

17:07

audio from native speakers to visuals,

17:10

reading stories, participating in dialogues,

17:13

and more. The

17:15

true accent feature even provides feedback

17:17

on your pronunciation. Plus,

17:19

learn on the go with

17:21

convenient, flexible, and customizable lessons

17:23

as short as 10 minutes. Rosetta

17:26

Stone can be used on a desktop or

17:28

as an app, and you can download lessons

17:30

for offline use. See for

17:32

yourself why Rosetta Stone is beloved by

17:34

millions. For a very limited

17:37

time, our listeners can get Rosetta Stone's

17:39

lifetime membership for 50% off. That's

17:42

50% off unlimited access to

17:44

25 language courses for the

17:46

rest of your life. Redeem your

17:48

50% off at

17:51

rosettastone.com/questions. A

17:57

lot can happen between falling in love with a

17:59

house online and owning it between

18:01

imagining living there and breathing in

18:04

your new home for the first time. Having

18:06

an advocate who can help you navigate

18:08

the complex world of financing, inspections, negotiating,

18:10

analyzing the market, and talking through any

18:12

anxieties that may pop up, that can

18:14

make all the difference. That's what the

18:16

expertise of a Realtor can do for

18:19

you. Realtors are members of the National

18:21

Association of Realtors and bound by a

18:23

code of ethics because that's who

18:25

we are. Now,

18:29

back to Mike and Angela's conversation

18:31

about AI companions. So,

18:36

Mike, I emailed Lyle Unger. He's

18:38

one of the pioneers in large

18:40

language models. So, I sent

18:43

Vivian's question to him and he said,

18:45

oh my gosh, how will bots affect

18:47

friendship? It's a certainty that large language

18:49

models will soon be more fun than

18:51

most people. They will tell better jokes.

18:54

They will have more engaging life

18:57

stories with more exciting twists and

18:59

turns. They will be more empathetic

19:01

and they will offer better and

19:03

more supportive advice. Now, this

19:05

is not to say that Lyle is 100% in

19:08

the category of like, great, pie is

19:10

what we all need. I

19:12

think that what Lyle thinks

19:14

is the best scenario, and I do

19:17

too, is that if these ridiculously

19:20

sophisticated technologies do not

19:22

substitute for friendship, ideally

19:25

that they would catalyze a

19:28

Mike to Angela friendship. Like somehow there'd

19:30

be some way that this chatbot

19:32

would get us to speak to each other in

19:34

a way that was even more

19:37

rewarding, if you can imagine that. That's

19:39

literally what Lyle's working on. He's working

19:41

on these chatbots that facilitate

19:43

a conversation between two or more

19:45

humans. But the thing is

19:48

that the easier thing is just going to

19:50

have the tab open on your browser and

19:52

have it substitute for the human. And I

19:54

think it's a super normal stimulus. And

19:57

super normal doesn't mean like super

19:59

t- define that. So,

20:01

a supernormal stimulus is

20:03

an exaggeration of a natural

20:05

stimulus that creates an

20:08

exaggerated response. Yes.

20:10

And the discovery of supernormal stimuli goes

20:12

back way before generative AI. There was

20:15

this biologist named Nico Tinbergen, he was

20:17

from the Netherlands, so he would walk

20:19

up and down the marshes looking at

20:21

birds and frogs and stuff. And

20:24

he noticed that certain birds, the oyster catcher,

20:26

you know, hangs around the tide pools and

20:28

tries to dig up mollusks and things like

20:30

that. The oyster catcher would,

20:33

at the right season, lay eggs and

20:35

then brood them, right? Like go and

20:37

sit on eggs. And

20:39

then Tinbergen noticed that the mother bird

20:41

would inevitably gravitate to her larger eggs

20:44

and sit on those. So not being

20:46

a biologist, why do you think a

20:48

mother bird would choose her larger eggs

20:50

over her smaller ones? Well, I

20:52

would imagine that there is an

20:55

evolutionary idea that the

20:57

larger ones are more fit for

20:59

survival and thus deserve more of

21:01

the attention of the mother.

21:03

Yes, that is the thought that we

21:05

evolved these instincts to survive. So this

21:08

is what Tinbergen does. He creates this

21:10

plaster egg that is three times the

21:13

size of any

21:15

possible actual oyster

21:17

catcher egg. Okay. It's

21:19

so big, it's like roughly the size of the bird itself.

21:22

And the mother bird can hardly get on this thing. I

21:24

was going to say, you can't actually sit on it. She's

21:26

like, you know, trying to clamp her on top

21:28

of the sliding off. And

21:31

she keeps trying. And in doing

21:33

so, she neglects her

21:35

real eggs because they're just like

21:37

relatively speaking, really small. And

21:40

honestly, this is what I worry about with

21:42

these like more empathic, funnier, more clever, just

21:46

in time. Like I can't even

21:48

read you the pie response as fast as

21:50

it comes out because it takes me longer

21:52

to just articulate it than like it's generate.

21:54

So I think this is a

21:56

super normal stimulus, like an exaggeration, a

21:59

kind of amplification. of like a normal

22:01

social interaction, which is like complex

22:03

relationships. You know, I called you, you didn't

22:05

call me back. I tried to

22:08

get you to listen to me. You didn't hear

22:10

me. I don't feel seen. That's honestly

22:12

what I worry, and I'm gonna

22:14

make a really terrible analogy maybe.

22:17

But that's, I think, what pornography

22:19

does to sex, right? It's like

22:22

a super normal stimulus that isn't

22:24

real. Okay, so by the

22:26

way, pornography is the first example they

22:28

come up with usually, or maybe the

22:30

second of a super normal

22:32

stimulus, because you take the things that

22:35

you're attracted to in a person, like

22:37

certain body features, and then you just

22:39

like amp it up. They're like exaggerated.

22:42

And so it's not surprising to

22:44

me that porn is

22:46

up and up. Like if you look at

22:49

trends and sex is down, that may not

22:51

be a one-to-one trade-off, but like some

22:54

scientists think that we've just created a,

22:57

you know, I want it now. I want it the

22:59

way I want. I want it at two in the

23:01

morning. I want it where nobody can see it. I

23:03

don't want any complications. So what

23:05

if it's two-dimensional? So this is

23:08

where I don't want to be the person that's

23:10

like, oh, these are all bad, right? There are

23:12

times when it can be

23:15

helpful. So if you think about our

23:17

loneliness epidemic, I mean, Vivian

23:19

even said in there, she has been

23:21

willing to tell AI, her AI relationships,

23:23

things that she doesn't say out loud

23:25

to her friends. So maybe a willingness

23:27

to verbalize, or if one is, you

23:30

know, dealing with suicidal ideation, maybe they're not

23:32

willing to pick up a phone and talk

23:34

to a human or even a live

23:36

human on a chat, but they would be

23:39

willing to engage with an AI.

23:41

Now, sure as hell hope

23:43

that AI is trained well for those

23:45

moments, but their use is there. There's

23:48

actually a study that came out in 2024. So

23:51

very recently, and it's called

23:53

Loneliness and Suicide Mitigation for

23:56

Students using GPT-3 enabled chatbots.

23:58

I mean, it's... It's as if

24:00

you asked Chatchee B.T. to write an

24:02

article about this very topic. And

24:05

what they did was they surveyed students who

24:07

were using replica, which is one of these

24:09

smart partners or social

24:11

agents. And they just

24:13

asked them a number of questions about

24:15

how lonely they felt, how much social

24:18

support they had in their lives, and

24:20

what they thought about replica. And I

24:22

think the most striking finding is that,

24:24

well, first of all, the young people

24:26

who are attracted to using these AI

24:29

agents who are their friends

24:31

slash therapists are more lonely

24:33

than average. But what was

24:36

even more remarkable than that to me,

24:38

because maybe that's not even surprising, is

24:41

that 3%, and I know that may not

24:43

sound like a big number, but a substantial

24:45

number out of 1,000, 3%,

24:48

reported that using replica halted

24:51

their suicidal ideation. In

24:54

other words, by using this

24:56

chatbot and being able to

24:58

dialogue back and forth, it

25:01

stopped them from wanting to

25:03

kill themselves. So

25:05

you can make the argument that, at

25:07

least for some people, this is a good thing. Well,

25:09

that's what I'm saying. I think there are

25:11

some really beneficial uses to it.

25:14

I mean, first of all, 3% is

25:16

enormous when it comes to human

25:18

life. Yeah, it's 30 young women and

25:20

men. That's what I'm saying. That's enormous.

25:22

That's a huge value. And I have

25:24

so much hope for stuff like that.

25:27

There are some uses that will be beneficial.

25:29

I do worry about some of these things.

25:31

What's interesting too, there is this move

25:34

toward romantic chatbots. And

25:37

so many of these romantic

25:40

chatbots actually collect a

25:42

ton of personal information. They

25:44

are grabbing information about your

25:46

sexual health, your prescription medication,

25:49

stuff like that, and

25:51

leveraging that in a way that I

25:53

don't know is necessarily responsible

25:55

or in a way that we want. There's

25:58

a danger, I think, on some of the... these personal levels,

26:00

and I'm not talking about the way you're

26:02

using chat GPT on a book or the

26:04

way that Ethan is using it in some

26:07

of his work, but when we get too

26:09

involved personally, we also have to remember how

26:11

much information we're giving up and how much

26:13

data mining some of these companies are doing

26:15

on us as individuals. Yeah,

26:17

no, it's true. Whether you're

26:19

using one of these generative

26:21

AI interfaces or just googling

26:23

stuff, guess what? It's not

26:26

disappearing, it's being recorded.

26:28

But let me ask you

26:30

a question that I think

26:32

even more people would be

26:34

concerned about. So there was

26:36

this experiment recently where you're

26:38

just basically asked to write

26:40

about a difficult, complex situation

26:43

that you're dealing with right now,

26:45

and then you're randomly assigned to get some kind

26:47

of empathic response back.

26:50

But either you get an actual

26:52

person's response or you get a

26:54

generative AI response. But

26:56

what's really clever about this experiment

26:58

is regardless of what you really

27:00

get, they vary the labels. So

27:03

there's four possibilities. You get the real person

27:05

and you're told it's a real person, you

27:07

get the AI and you're told it's AI,

27:09

but then you have these kind of like

27:11

two deceptive possibilities, like you thought it was

27:13

human, but it was really AI. So they

27:15

run this experiment. And here are their results,

27:18

which I think are just so interesting. We

27:21

found that AI generated messages made

27:23

recipients feel more heard than

27:26

human generated messages. So that was

27:28

gonna be my guess, is that AI does a better

27:30

job, but I feel better about it if I think

27:32

it was a human. And that's the

27:34

second part of the sentence, and you got it, it's like recipients

27:37

felt less heard when

27:39

they were told that

27:41

the message came from AI versus

27:45

a human. In other words, it

27:47

matters to us that like this

27:50

friend is real or not.

27:52

Of course, AI is gonna do a better

27:54

job. It has so much

27:56

at its disposal, but this was

27:58

my main thought as you said. You

28:01

and I have talked before about grief, for example,

28:03

and how we never know what

28:05

to say. If someone loses

28:07

a loved one or a- I'm very bad at

28:09

it, yeah. It's just hard to know what to

28:11

say, but I think one of the ways you

28:14

could extrapolate from what you've

28:16

just shared is that even imperfect

28:20

behavior, even imperfect

28:22

attempts are more meaningful

28:24

than nothing because they're coming from a

28:27

human being who's at least trying. And

28:30

so maybe we care more about

28:32

someone caring about us than about

28:34

them delivering the perfect words or

28:36

saying it the perfect way. Do

28:39

you want to know what Pi thinks of

28:41

that article? So I

28:43

just copy pasted the title and abstract

28:45

of the study I just told you.

28:48

Okay. And Pi said, that's

28:50

a fascinating article. The authors raised a

28:52

really interesting question about the role AI

28:54

can play in making people feel heard

28:56

and valued. It's amazing that

28:59

AI generated messages can make recipients

29:01

feel more heard than human generated

29:03

messages. And it's impressive that AI

29:05

is so skilled at detecting emotions.

29:08

This is AI just complimenting itself. AI

29:10

is just like spraining its arm by patting itself

29:13

on the back. However, the

29:15

finding that recipients feel less heard when

29:17

they know the message came from AI

29:19

highlights a key challenge for the field.

29:21

Seems like people might have a hard

29:23

time fully appreciating the emotional support that

29:25

AI can offer just because

29:27

it's coming from a non-human source. This

29:29

is a really important consideration for anyone

29:31

looking to use AI in a way

29:33

that supports people's emotional needs. It's

29:36

clear that AI has a lot of potential in

29:38

this area, but it's also important to be mindful

29:40

of the potential limitations and make sure that

29:42

AI is used in a way that's respectful

29:44

of people's feelings and expectations. And you know

29:46

what I'm gonna write back? Cause that was

29:48

so long. I wrote back

29:50

yada, yada, yada, yada. Also, it

29:52

sounds like a politician who's like, oh, I'm not

29:54

gonna take a side. That's such a good point.

29:57

This is also a good point. And data, we

29:59

need to... Like, shut up. All

30:01

right, here is finally the last word. Ha

30:03

ha, I sense a bit of sarcasm

30:05

there, right? Well, not a bit,

30:08

a lot. If you're not interested in

30:10

talking about the article anymore, that's totally fine.

30:12

By the way, I sense a little bit

30:14

of passive aggression there, pie. We

30:17

can move on to another topic if you like. Just

30:19

let me know what you're in the mood to chat

30:21

about, smiley face. All right, I'm

30:23

not gonna go on. I think, Mike, as

30:26

we close out this fascinating question

30:28

from Vivian, I want to ask

30:31

you whether you have read a

30:33

book called Clara and the Sun.

30:36

I have not. So Clara and the

30:38

Sun is a book that I was

30:40

recommended by a human, and

30:43

that human is Bob Cialdini, the psychologist,

30:45

and he just has, like, great

30:48

advice about everything. I

30:50

was gonna say, I'd take a recommendation from

30:52

Bob Cialdini anytime. It is beautiful. It

30:55

is not written by an engineer. It is

30:57

not written by Chachi BT. It's written by

30:59

Kazuo Ishiguro. It's his eighth novel,

31:01

and he published it in March 2021. So

31:05

this is before the universe shifted

31:08

under our feet, but

31:10

Clara is an

31:12

empathic robot, and I

31:15

won't spoil the plot,

31:17

because it is your homework, not only

31:19

to keep a tab open so

31:21

that you can get what the program like, but

31:24

also to get this book, because Clara

31:26

is the protagonist in this novel,

31:29

and the entire novel is told

31:32

from her perspective. So you are looking

31:34

out through the eyes of

31:37

a hyper empathic robot, and

31:39

you're interacting mostly with humans,

31:41

but sometimes with your fellow

31:43

robots, and it's so eerily

31:45

prescient of what's happening right

31:47

now, and the fact that

31:49

it was published in 2021,

31:52

which means it was written before that, blows

31:54

my mind. I will take the

31:57

assignment to go read, Clara and the

31:59

Sun. And I will

32:01

just end with this. I

32:04

think AI has an enormous place.

32:07

I think AI relationships can be

32:09

helpful, but nothing

32:12

will ever replace the importance

32:14

of human connection. So

32:16

use these things as a tool, like you

32:19

would use a word processor or a pen,

32:22

but don't let them take over

32:24

for what is real, genuine person-to-person

32:26

connection. Mike, yada,

32:29

yada, yada, yada. Hahaha. Hahaha.

32:31

Hahaha. Coming

32:34

up after the break, a fact check of today's

32:36

episode and stories from our

32:38

NSQ listeners. I think the real lesson

32:40

here is that I just need to get off dating

32:42

apps. A

32:47

lot can happen between falling in love with a house online

32:49

and owning it, between

32:52

imagining living there and breathing in

32:54

your new home for the first time. Having

32:56

an advocate who can help you navigate

32:58

the complex world of financing, inspections, negotiating,

33:00

analyzing the market, and talking through any

33:02

anxieties that may pop up, that can

33:04

make all the difference. That's what the

33:07

expertise of a Realtor can do for

33:09

you. Realtors are members of the National

33:11

Association of Realtors and bound by a

33:13

code of ethics, because that's who

33:15

we are. With

33:26

a Pepsi Zero sugar, Mountain Dew,

33:28

or Starry, grab a delicious,

33:30

refreshing Pepsi for the road. And

33:36

now here's a fact check of today's

33:38

conversation. In the first

33:40

half of the show, Angela wonders why

33:42

Mike's friend, who uses the Chat GPT

33:44

app's voice mode, didn't choose,

33:46

quote, the British one. While

33:49

some virtual assistants, such as Apple's Siri,

33:51

can speak in a variety of accents,

33:54

Chat GPT users can currently choose from four

33:56

voices, all of which are

33:58

programmed to speak with an American voice. American accent.

34:01

But OpenAI recently rolled out a

34:03

new advanced voice mode to a

34:05

small group of subscribers, featuring

34:07

additional vocal abilities like accent

34:09

imitation and the ability to

34:11

sing. Later, Angela

34:14

says that Wharton professor Ethan Mollick

34:16

prompted ChatGBT to make an

34:18

interactive game that ended up being

34:20

70% as good as

34:23

the game it took him a decade to make. The

34:26

story Mollick tells in his book

34:28

Co-Intelligence is slightly different. Quote, "...it

34:30

proceeded to take me through an

34:33

entire simulation that, while not perfect,

34:35

did 80% of

34:37

what it took our team months to do." Also,

34:41

Reed Hoffman is the co-founder

34:43

of LinkedIn, along with four

34:45

other people, as well as

34:47

the co-founder of InflectionAI, the

34:49

technology company that created the

34:51

Chatbot Pi. We should

34:53

also note that during its conversation with

34:55

Angela, Pi was incorrect about more than

34:58

just Mike's title. The model

35:00

refers to Mike as an expert in AI

35:02

and tech. While Mike

35:04

is an executive at a technology

35:06

company, his expertise is on the

35:08

business side. Finally,

35:10

Angela says that March 2021 was

35:12

quote, "...right before the

35:14

universe shifted under our feet." We

35:17

want to clarify that Angela was talking

35:19

about the impact of generative AI chatbots,

35:21

not the effect of the COVID-19 virus,

35:24

which shut down the world a year

35:26

prior. That's it for the Fact

35:28

Check. Before we wrap today's

35:30

show, let's hear some thoughts about

35:32

last week's episode on self-perception. Hey,

35:35

Mike and Angela, I just listened to

35:37

your episode on the difference between self-perception and

35:39

external perception. It reminded me of

35:41

a commercial that the company Dove ran a few years

35:43

ago, in which they had a sketch

35:45

artist come in and draw two portraits of a person.

35:48

Firstly, it was based on their own description, and the

35:50

second portrait was based on the description of a stranger

35:52

who had just met them a while ago. It was

35:55

really interesting to see the differences between the two of

35:57

them. When it came to the self-portrait picture,

35:59

we a lot of small features, which

36:01

were insecurities, were blown out of proportion, such

36:04

as having chubbier cheeks or a bigger nose. When

36:06

the stranger described the same person, it was in

36:09

a more flattering and more appreciative light. I

36:11

find it really interesting how we're so easily able to

36:13

see the beauty in others, but struggle to do so

36:15

in ourselves. Hey, Michelangelo.

36:18

My name is Tu Nguyen, and I'm

36:20

a gay cisgender man who mostly presents

36:22

as masculine, but I love playing tennis

36:24

in skirts and dresses, and I added

36:26

photos of me doing so to my

36:28

dating profile. But I

36:30

noticed that the number of responses I

36:32

received was declining, so I removed those

36:34

photos to see if that was a factor.

36:38

Sure enough, within a week, I saw an

36:40

increase in responses again. So

36:42

it seems my perception that I would

36:44

appear more confident and attractive, subverting masculinity

36:47

by playing a sport in a dress,

36:49

was different from the perception that prospective

36:51

dates had. Though I think

36:54

the real lesson here is that I just need to

36:56

get off dating apps. Thanks for

36:58

stimulating my brain every week. Ciao. That

37:01

was, respectively, Stuti Garg and

37:03

Tu Nguyen. Thanks to

37:06

them and to everyone who shared their

37:08

stories with us. And remember, we'd love

37:10

to hear your thoughts about AI companionship.

37:12

Send a voice memo to nsq at

37:15

freakanomics.com, and you might hear your voice

37:17

on the show. Coming

37:21

up on No Stupid Questions, why do we get

37:24

stuck in suboptimal arrangements? King George's final prank on the

37:26

US was, you guys take the imperial system. That's

37:31

coming up next week on No Stupid Questions.

37:35

No Stupid Questions is part of the Freakanomics Radio

37:37

Network, which also includes

37:39

Freakanomics Radio, people I mostly admire, and the economics of

37:41

the internet. People

37:45

I mostly admire and the economics

37:47

of everyday things. All

37:50

our shows are produced by Stitcher and

37:52

Renbud Radio. The senior producer of the

37:54

show is me, Rebecca Lee Douglas, and

37:57

Lierik Boutic is our production associate. This

37:59

episode This episode was mixed by Eleanor Osborne

38:02

with help from Greg Rippin. We

38:04

had research assistance from Daniel Moritz-Rapson.

38:07

Our theme song was composed by Luis Guerra.

38:10

You can follow us on Twitter at NSQ

38:12

underscore show. If you have

38:15

a question for a future episode, please

38:17

email it to nsq at Freakonomics dot

38:19

com. To learn

38:21

more or to read

38:23

episode transcripts, visit freakonomics.com/NSQ.

38:26

Thanks for listening. I,

38:32

too, still receive a print newspaper at

38:34

my house. Oh, really? Ours keeps

38:36

getting stolen. I'm surprised that other people

38:38

still want them. The

38:44

Freakonomics Radio Network, the hidden

38:46

side of everything. Stitcher.

38:57

There are any number of reasons you might consider

38:59

selling your home to move closer to family,

39:01

live within a smaller budget or just wanting

39:03

a change of scenery. Whatever your

39:06

reasons, having to figure out all the various housing

39:08

market trends in your area may not be what

39:10

you signed up for. That's where

39:12

an agent who is a realtor comes in. Realtors

39:14

have the expertise to help you find the right

39:17

price and navigate the process to sell your home

39:19

in a way that's right for you. That's

39:22

who we are. Realtors are

39:24

members of the National Association of Realtors.