0:03

Welcome to Preparing for AI with

0:05

Matt Cartwright and Jimmy Rhodes , the

0:08

podcast which investigates the effect of AI

0:10

on jobs , one industry at a time . We

0:13

dig deep into barriers to change , the coming

0:15

backlash and ideas for solutions

0:17

and actions that individuals and groups can

0:19

take . We're making it our mission to

0:21

help you prepare for the human social

0:23

impacts of AI . We're making it our mission to help you

0:25

prepare for the human social impacts of AI . Touch my tears with your lips

0:27

, touch my world with your fingertips . Welcome

0:30

to Preparing for AI with me , matt Cartwright

0:32

, and me , jimmy Rhodes , and

0:35

welcome back after a couple of weeks

0:37

away . We

0:43

are back with the podcast and this is going to be an industry-focused episode . We're

0:45

going to be looking at the comms industry with Daniel Lyons later , but

0:48

because we've been away for a while and because we've

0:50

had well , there's been so much kind of

0:52

going on , as usual , we will

0:54

do a kind of catch-up , but we're going to look at it more

0:56

about a kind of introduction , welcome

0:58

back and the things that have really been interesting

1:01

us over the last few weeks

1:03

. So , jimmy , that have really been interesting us over the

1:05

last few weeks . So , jimmy , do you want to start off ?

1:10

And then I've got a few things that I wanted to bring to our

1:12

listeners' attention as well . Sure , yeah , so I think the biggest news

1:14

is Apple's WWDC conference

1:17

. Apple

1:20

finally got on into the AI game , so they haven't really talked

1:22

about AI much . They've been very quiet

1:25

on the subject . They haven't been developing

1:27

their own models or discussing AI

1:29

, but , as always with Apple

1:31

, they've decided that they're now

1:33

going to own AI and it's kind

1:35

of their idea and it's this , it's this new

1:37

thing that they've come up with . So

1:40

I think they're calling it Apple intelligence , or that's

1:42

what it's been dubbed online . So

1:44

what ? So what they're actually doing

1:46

think they're calling it Apple intelligence , or that's what it's

1:48

been dubbed online so what they're actually doing is they're integrating

1:50

chat GPT into Siri . So Siri is going to be back with

1:53

a vengeance . I think it was pretty

1:55

useless before previously , but

1:58

the idea is now throughout the iPhone

2:00

, you're going to have chat GPT

2:02

.

2:04

Sorry , Jimmy . As you just said , Siri

2:06

there , every

2:11

single Apple device in our studio has started twitching away like crazy , so it's good to see that , at

2:13

the moment at least , Siri is still operating as it did for

2:16

the last however many years .

2:17

Yeah so , absolutely

2:20

so . Apple are going to yeah , so , absolutely so

2:22

Apple are going to bring

2:24

chat GPT into Apple devices

2:26

. That's kind of the gist of it . I

2:31

think there's been a bit of shock around it because Apple have always

2:33

been really pro-privacy and have

2:35

actually got pretty good security on their devices

2:38

and all this kind of thing . And now what

2:40

it sounds like what they're going to be doing is sending

2:42

all your data to chat

2:44

GPT servers to

2:46

do inference , so that Siri gets

2:48

improved and you get a much better

2:51

experience on the device

2:53

, which is a bit of a weird one because , as

2:55

I say , they've been pretty quiet on it . Everyone thought they were going

2:57

to develop their own models , but it seems like

2:59

they're going this chat GPT route instead . So , in

3:02

a positive light , what they're looking to , what they're promising

3:04

? This chat GPT route instead ? So , in a positive light , what they're looking to

3:06

, what they're promising is that you're going to have a

3:09

kind of seamless experience across all

3:11

of your devices , all of your iOS devices , with

3:13

AI features incorporated across all

3:16

of your apps . So you'll and

3:18

, as I say , it should just be a massive improvement

3:20

over what you've had previously , with

3:23

Siri being , I , I guess , on

3:25

the back burner for quite a while well

3:28

, is there any plans for

3:30

kind of hardware ?

3:32

so you know we talked , I think , in the last episode

3:34

we were talking about the um

3:36

, microsoft surface laptops that will contain

3:38

the new kind of chips that will

3:40

. You know , they'll have the gpu , they'll have um

3:43

, obviously normal process , and then they'll have this kind

3:45

of neural unit . I mean , is there

3:47

any talk yet about devices and whether they will

3:49

have any ? You know particular

3:52

change to the , the chip infrastructure

3:54

, or at the moment are we just looking at this as a

3:56

? It's just a kind of software addition

3:59

?

4:00

it's a , as far as I understand it , it's a software

4:02

addition , um . I think

4:04

in the future we are going to see

4:06

more on hardware type ai

4:09

, um , as you mentioned , with the surface devices

4:11

we spoke about recently . I

4:13

think one other thing this has done

4:15

is finally

4:17

I mean it was already pretty much dead in the water

4:19

but the , the rabbit device , um

4:22

, the sort of dedicated hardware device . I mean it was

4:24

stillborn , wasn't it ?

4:25

It was always , it was yeah .

4:28

But what they always said was you know , you can

4:30

do all this with a phone and eventually

4:33

Apple or Google will just introduce this into

4:35

phones . And eventually turned

4:37

out to be like two or three months later , and

4:39

the rabbit was crap anyway .

4:41

I still hope , without going on about it , you

4:43

know that at some point that

4:46

some of the ai tools do allow us to move

4:48

away from screens a bit more . And

4:50

you know you can have a screen but

4:52

not necessarily have to use the screen all the time

4:54

. And I think you know , if you've got an apple watch , one

4:57

of the things with apple watches you

4:59

know I can see that being really useful in terms of you

5:01

can just talk to it and you've got something right next to your

5:03

face , because it is the

5:06

one good thing about the rabbit and

5:08

the AI pin was that idea of you

5:10

know steering people away from from

5:12

screen time , not just in terms of it making

5:14

it a more natural interaction , but actually just in terms

5:16

of you know , the health of your eyes and

5:18

not looking at rectangles every

5:21

day rectangles

5:27

every day .

5:27

Yeah , I totally agree , and now that we've got the Apple ad section of the podcast out of the

5:29

way , other mobile devices are available .

5:32

So the first thing I wanted to talk about was a

5:35

piece of research on impressions of AI

5:37

that the Reuters Institute and the University

5:39

of Oxford put out , probably

5:42

about a month or so ago now . This

5:44

was based on the public of six

5:47

countries and it was on what

5:49

they think of the application of AI in news

5:51

, so specifically in news and journalism , but then

5:53

also across work and life . The

5:56

countries they looked at were Argentina , denmark

5:59

, france , japan , the UK and the US . So

6:01

you know , although it's a , you

6:04

know those countries are not all the same , obviously , but

6:06

it's . I would say this is not

6:09

a . It's not a reflection

6:11

of the whole world let's put it that way , but it's

6:13

still interesting and I would imagine , for people who

6:15

are listening to this podcast , it probably reflects

6:17

the you know the kind of countries that

6:19

you're listening from . So

6:22

ChatGPT was the best known generative

6:24

AI product , unsurprisingly

6:26

, but there was only 1% of

6:29

people in Japan who were

6:31

using either ChatGPT or

6:33

any generative AI tools daily

6:35

, and in France and the UK that was 2%

6:37

. It was 7% in the US , a

6:40

total of around 30% across

6:42

the population . So this was a kind of average out across

6:44

the population of the six countries had not

6:46

heard of any AI tools at all . 56%

6:50

of 18 to 24 year olds have

6:52

used chat GPT at least once , but

6:54

that's only 16% when you get to

6:56

age 55 and over . There

7:00

was optimism around AI's

7:02

impact on science , healthcare , daily

7:04

routine and , surprisingly to me , media

7:07

and entertainment . I'm not

7:09

sure I necessarily agree on media

7:11

. I guess entertainment makes more sense and it

7:13

was quite significant . So 17% more

7:15

optimists than pessimists in that area . But

7:17

then cost of living . I'm not sure

7:19

why cost of living . Maybe that's just a reflection of where

7:22

people's priorities are in general . Job

7:24

security and news were the top areas

7:26

of concern . In

7:28

Argentina , only 41%

7:31

of people had heard of chat , gpt and

7:35

Google Gemini . This was , I thought , really interesting

7:37

. Google Gemini 15% of people in the UK

7:39

had heard of it . France was only 13%

7:42

, usa 24% . Microsoft

7:45

Copilot was about the same . Claude was

7:48

between 2% in Germany and 5%

7:50

in the US , which kind

7:52

of surprised me and disappoints me , because I'm

7:57

massively a fan of Anthropic

8:00

and the way that they kind of operate as a company and

8:02

as and the way that they kind of operate as

8:04

a company and

8:06

you know , as open AI becomes closed AI and becomes more and more

8:08

of a commercial outfit that seems to

8:10

care nothing for security and anything

8:12

other than making money and being the first to AGI

8:14

. Anthropic are the only ones who

8:16

really seem to have a you know , a genuine

8:19

desire to make something that benefits humanity

8:21

. So a shout out to everybody who

8:23

isn't using AI tools yet to use

8:25

anthropic tools , because they are

8:27

by far the best company out there at the

8:29

moment . And the UK had

8:32

the lowest score of only 2%

8:34

of people using AI to try and get the latest

8:36

news . In the US it

8:39

was 10% of people and

8:41

, like I say , this was specifically looking at

8:43

news and journalism . So that's that's

8:45

why I had these kind of specific questions , but

8:47

, yeah , why this was really interesting

8:49

to me . Um , and there was another piece this

8:52

was last year , but saying that 46

8:54

of people in the us at that time had not heard

8:56

of chat gpt . Is that

8:58

, I think , for people like , yeah , jimmy

9:00

and myself , when we are

9:02

kind of , you know , submerging this stuff

9:05

every day , we think this is right at the top of

9:07

people's agenda and everybody is thinking about

9:09

and knows about AI . But what this actually

9:11

shows , if you're listening to the podcast and you

9:13

are thinking about AI , is , you

9:15

know , you're already in a fairly

9:18

small group of people and you're already probably ahead

9:20

of most people , so you know whether people are putting

9:22

their head in the sand because they are scared

9:24

and they , they , you know , you

9:26

know they're worried about what happens next , so they just don't want to think

9:29

about it or whether you're just people's

9:31

lives have , you know , taken over and there's

9:33

enough things to worry about . This is not

9:35

at the top of the agenda , but I would bet

9:38

that if we were sat here in a year's time with

9:40

the advances that are going to happen , people will

9:42

. If we looked at this in a year's

9:44

time , a lot more people will be thinking about

9:46

and worrying about

9:48

and acting on and you know , getting involved

9:50

with AI . I'm pretty sure that's the case .

9:53

And the funny thing about that for me is I know it was focused

9:55

on news , but the interesting thing is how

9:58

many people say they

10:00

aren't aware of these AI tools . But

10:03

now I mean , as I said , as I said , as

10:05

I said the in

10:07

in

10:13

the update , apple are now integrating AI throughout the iPhone , google also announced that they're they're

10:15

bringing more generative AI experiences

10:18

into Google search recently , and

10:20

Microsoft Bing already uses AI

10:22

, so is

10:24

it ? So ? Do people actually need to be aware of

10:26

that ? They're using AI tools ? Because I

10:29

think in a lot of cases , people probably are already using

10:31

them , possibly daily , and they just don't even

10:33

know it , because these things are starting

10:35

to become integrated into all of the software

10:38

that we use . And that's kind of the way that

10:40

I see it going is that , yeah

10:43

, there's going to be a niche who know all about AI

10:45

and know about chat , gpt , but at

10:47

some point soon , everyone's going to be using it all the

10:49

time because it's getting built into things

10:51

that we use .

10:53

Yeah , and it already is , isn't it ? I mean customer

10:55

service , for example . And

10:58

one of the things

11:00

I noticed a lot is you know the calls that you get . Now

11:02

, where

11:09

you used to get a sales call , you can tell . Now a lot of those calls are an AI sales call

11:11

. Um , you know , there are those kinds of changes that are happening

11:13

and we're not . We don't necessarily even need to think

11:15

about it , do we ? It kind of doesn't matter , because you're

11:18

either going to answer that call or not , regardless

11:20

of whether it's an AI . So I think you know every one of those

11:22

calls . I would , you know , I

11:25

would turn off

11:27

, cancel the call , regardless of whether it's an AI or a

11:29

person . But

11:32

, yeah , it is becoming ubiquitous in many ways . I think the thing

11:34

that I would be more concerned about and you know this

11:36

is maybe because of where I

11:39

come at this as a kind

11:41

of problem for civilization

11:43

is being

11:45

aware of know

11:47

ai tools is

11:49

maybe not as important as being aware of ai

11:52

and the changes that it will

11:54

make to our world . You know it's it's

11:56

not about the chatbot , it's about the

11:58

potential in two , three , four , five , ten

12:00

, fifteen years . And that's where

12:03

it scares me a little bit

12:05

to think that people are not aware of this at all . And

12:07

I had a conversation with with

12:09

one of my my tutors on the AI governance

12:12

course that I did the other day and we were talking about

12:14

I said why , why is it not an election issue

12:16

? You know , in the UK , for example , why

12:19

is it not an election issue ? Because if you look at the

12:21

kind of poor sentiment towards

12:24

ai that you see in a lot of developed

12:26

countries and I put developed in kind of , you

12:29

know , inverted commas

12:31

um , there's a lot

12:33

of negativity and so it would seem to be

12:35

an easy win . It's a kind of low-hanging

12:37

fruit for a political party to say hey , we're

12:39

, you know , we're going to sort this out , we're going to protect your jobs

12:42

. And his point to me , which I think he's bang on with , is you know there's're going to sort this

12:44

out and we're going to protect your jobs . And his point to me , which I think he's bang on with , is you know , this is just not

12:46

the bandwidth for it in this election , because

12:49

the most pressing things facing

12:51

people are , you know , costs of living , they're the economy

12:53

. They are unfortunately , people

12:56

think they're immigration the issues that people

12:58

think are important in the short

13:00

term , I guess sorry , I think are important are

13:02

important to them in the short term are

13:04

what are in people's minds at the moment . But

13:06

I do hope that when the dust settles in a few

13:09

months' time from the various elections

13:11

, that there's then more space to start

13:13

looking at this , and I think that will happen . I

13:15

do genuinely think you

13:18

can sort of feel that the

13:20

kind of cogs are turning a little bit and there is a

13:22

lot more going on and a lot more understanding that

13:24

we cannot just allow you

13:26

know three , four companies in Silicon Valley

13:28

to just in a black box , just go

13:31

on completely ungoverned , do

13:33

whatever they want , to develop something that

13:35

has , you know , potential threats to

13:37

the whole of society .

13:40

Yeah , we can't . We . I think over

13:43

time there'll be more

13:45

and more realisation that we can't just blunder into

13:47

this , and some of that's happened already . We've talked about

13:49

it on previous episodes . Where there's

13:51

been international conferences on

13:54

AI , there's been a lot of talk around how we

13:59

that

14:02

. In China

14:04

in particular which we talked about a few weeks ago , but

14:07

absolutely , I think , elections

14:10

the focus is obviously

14:12

going to be right now on some of the bigger

14:14

topics , particularly in the UK , but

14:17

anywhere in the world right now , we've just had a period

14:19

of massive inflation and there's been lots

14:21

of societal problems , which

14:23

you know . So AI is right down

14:25

the list at the moment , but I think it is going to become

14:28

more and more significant .

14:31

Another thing that I wanted to first have a

14:33

chat about and another thing that's been

14:36

, you know , out in I say media , I

14:38

mean sort of AI , specific media

14:40

and social media , but is this question

14:42

around whether there is enough data and

14:44

whether we're running out of data for for large language

14:47

models ? And I think , as an extension of that conversation

14:49

and something that you know me and you have talked

14:52

about almost to

14:54

the cows come home recently , is this

14:56

question around whether the current

14:58

large language model architecture , so that the

15:00

kind of neural networks that are that are currently being

15:02

, whether that's enough for

15:04

us to get to , you know , agi

15:07

, advanced AI , whatever you want to call it , or

15:10

is everything being overhyped at the moment

15:12

? So you know where do you stand on this

15:14

data point , whether we have run

15:16

out of data or whether we're going to run out of data .

15:20

It's really difficult . So we clearly have

15:22

run out of data . I mean , we actually ran out of data

15:24

a long time ago , so , for the benefit of

15:26

everyone listening , basically

15:28

, these models have been trained on all

15:31

of the information that's available to all of humanity

15:33

, like everything they can get their hands been . Restrictions

15:35

put on the APIs that Twitter have and

15:45

and forums like Reddit use , and

15:48

that's as a bit of a backlash to

15:50

the fact that AI models were just trained

15:53

on all their data and it was all freely available

15:55

previously . So these

15:57

models , like chat , gpt , three , four , they've

15:59

all been trained on everything that's available

16:02

already . So they've they have literally

16:04

run out of data . In that sense , the

16:06

question is whether you believe open

16:09

ai when they say that they can . So

16:11

what they're saying now is that they can generate effectively

16:13

. What they can do is generate data using

16:15

ai and , using

16:18

that generated data , they can then go and train

16:20

like on like , continue to train their ais

16:22

and they continue to improve . Now

16:25

that it like that remains to be seen

16:27

, because I guess what you have to do is wait for the next

16:29

models to come out and see whether they do actually

16:31

keep improving and do get , do get better , which

16:33

they are um , but is that going

16:35

to slow down ? Um , is the ? Is

16:38

it ? Is it going to plateau ? Has it already plateaued

16:40

? I honestly don't know . And and again

16:42

, as you said , open ai a

16:44

much more closed ai now

16:46

. And so I don't

16:48

know .

16:49

And again , as you

16:51

said , open AI much more

16:53

closed AI now

16:56

, and so I don't necessarily

16:58

you can't really take what they're

17:00

, you can't really to what they say

17:02

, watch what they do . And I think that really

17:05

applies here is , although

17:08

OpenAI say , oh , there's no problem , but

17:10

you know the amounts of money that players

17:12

are trying to buy data

17:14

from . You know newspapers , magazines that

17:28

have large amounts of kind of high

17:31

quality data . Um , another

17:34

point is you know why did people wonder

17:36

at the time ? Why did elon musk buy twitter ? Well

17:38

, because it's data . You know there's a huge

17:41

amount of data in there . Now the data in twitter

17:43

scares the hell out of me the idea that that

17:45

is . I mean that's . You know if we're talking about

17:47

crap in , crap out , you put that stuff in

17:49

my God . But it's data

17:51

to advanced AI , agi , because it just doesn't kind

17:54

of make sense . You

18:10

know as much as it seems to be . I don't want to

18:12

use the word sentient , but it seems to be kind of

18:14

intelligent . It's parroting back stuff

18:17

that it's been trained on . I

18:19

sort of worry more about the idea , you

18:22

know , the kind of dead internet theory

18:24

. So dead internet talks about how I

18:26

think it's you know potentially more

18:29

than 50 of the internet now is is

18:31

just nonsense because it's , you know , troll

18:33

farms . It's ai making

18:35

it up and therefore the information that's out

18:37

there on the internet is not accurate

18:39

. There's so much crap out there that basically

18:42

you're putting crap into it , it's it's going to output

18:44

crap and so , regardless of whether

18:46

there's more data or not , the existing

18:48

data is not good . So you

18:50

know , it's a bigger question around amount

18:53

of data , where there's more data , the

18:56

quality of the data that was previously used

18:58

, um , and in turn that kind

19:00

of you know feeds , a never-ending kind

19:02

of loop . If you've got aiIs training themselves

19:05

on that existing data , I think you're right

19:07

. I mean , we don't know because we're not privy to

19:09

what's going on within

19:11

those organizations and we don't know enough because

19:14

no one knows enough about the way large language models

19:16

work . But it definitely seems like

19:18

something that is highly possible and I think I

19:21

more and more think at the moment you

19:23

talked about OpenAI . I mean , they're so far from the original

19:25

purpose , they're so focused

19:27

now on being the first to create AGI

19:29

and you know , investment

19:31

and money , that it's quite easy to believe

19:33

that there is a lot of hype

19:35

just to generate investment . I do think

19:37

we're probably at the top of a hype cycle . I

19:39

don't think that necessarily means that you

19:42

know there's going to be an IO winter for the next

19:44

10 years , but I do wonder whether things have been

19:46

a little bit oversold . And you know , ai

19:48

, you know agi by 2025 , agi

19:50

by september . Some of that

19:52

seems to be now 2027

19:55

, 2028 .

19:56

It seems to be kind of rolling back a little bit

19:58

yeah , and no one even

20:00

agrees on the definition of agi , so we

20:02

were chatting about it earlier , I think .

20:04

I think what was the term you said they're now

20:06

using advanced AI , which is

20:08

which is not defined , but which avoids

20:10

the the need to kind of , you know , find

20:13

an AGI definition yeah , because

20:15

this is what everyone's been struggling with , right so

20:17

AGI ?

20:18

does AGI mean conscious machines

20:21

that have their own free will

20:23

and a self-determination , or

20:26

does it mean something that can

20:28

perform almost any task

20:30

in to the same level as a human

20:32

and doesn't need supervision ? I

20:34

would , I would lean towards the the

20:36

latter , um myself

20:38

, because I think we're we

20:41

don't even really understand any of the former

20:43

, like what consciousness is and all

20:45

this kind of stuff which we're probably not going to get into now

20:47

, maybe for a future episode . But

20:49

I I'd sort of lean

20:51

on the latter of those definitions , which

20:53

that's . I

20:56

feel like that is the kind of aim and the

20:58

target and the goal for companies

21:00

like chat , gpt , is having a machine

21:02

where you can just let

21:04

it loose and it will , it

21:07

will , it will automate

21:09

a vast array of tasks , um

21:11

, and hence the talk , the , hence the podcast

21:13

and the sort of talk about how

21:16

that's going to threaten jobs . But

21:18

I don't even think that we're that close

21:21

to reaching that definition

21:23

. And the reason I feel like

21:25

that is because it's

21:27

like how , like with however

21:29

smart a large language model appears to be

21:31

and however many questions it can answer and however

21:33

many puzzles it can solve

21:35

and however many things it can do better and even

21:38

than the average human , it

21:40

still seems to require

21:42

a level of supervision which

21:45

a human wouldn't require , like I . I

21:47

wouldn't trust it to go and just get on with something

21:49

. And I've tried . I've tried some

21:52

of the agentic type models as well

21:54

, which where you can actually use an agent to go off

21:56

and write code and , you know , talk to another

21:58

ai to get testing done on

22:00

the code , and then there's another ai which is

22:02

supervising them and all this kind of stuff and it doesn't

22:04

really work . Yet then devin is

22:06

an example of that . So there was devin , and

22:09

then there's open devin and various models , but

22:12

they don't really work . They end up costing

22:14

you a fortune because they go around in circles

22:16

, um , and they and they don't

22:18

know when they've completed the task . There's all sorts of

22:20

real kind of complications

22:22

with it which seem to be very human

22:25

problems where a human would just be like

22:27

okay , you know , I need to point

22:29

you in a different direction now . Um , stop

22:32

what you're doing , let's have a review . Whatever it is

22:34

, we're not there yet and and

22:36

maybe we'll get there , but I feel like

22:38

, um , I feel like that is a

22:40

is a sort of elusive

22:43

moving milestone ? Yeah , yeah

22:47

.

22:47

So the last thing , and

22:49

this is , I guess , quite important . So , you know

22:51

, governance , alignment , the sort of general

22:53

security is what's been

22:56

kind of occupying my mind

22:58

and this

23:00

is , I guess , a sort of soft launch announcement

23:02

. But we're going to be relaunching

23:05

the podcast , going forward . Going

23:11

forward , we're still going to have an element where we focus on jobs

23:13

, but we're going to sort of branch out a little bit because we think there

23:15

is an urgent need now , and particularly post-elections

23:18

in many Western countries

23:21

this year and we've added France to that list in the last

23:23

week or so we think there's an

23:25

urgent need to inform people

23:27

and actually to help try and

23:29

achieve our original purpose , which was giving

23:31

people actions that they can take

23:33

to try and mitigate the human impacts

23:35

of AI . So I think it's not

23:37

an exaggeration . We've said on the

23:40

dystopia episode you know , if

23:42

nothing changed , we're on a pretty

23:44

fast path to destruction of you

23:46

know humanity , whether that's destruction

23:48

of the kind of social system or it's destruction

23:51

of the planet . You know , I'm not saying

23:53

for a second that there will be nothing

23:55

. So we're not saying that that is necessarily

23:57

the end goal . But you know that's where we're headed

23:59

without those measures and whether

24:01

those measures are taken quickly enough to

24:04

address the kind of more existential threats

24:06

is , you know , a properly

24:08

kind of defining moment

24:10

for humanity . So we

24:12

want to keep it light hearted . We want to keep

24:14

it funny where we can . We want to keep interviewing

24:17

people , but we want to branch out a little bit

24:19

more than jobs . So we'll continue to focus

24:21

on industries , but we will also look

24:23

at the alignment problem , the

24:25

security and safety around ai

24:27

and governance . So hopefully , when

24:30

we relaunch that , we will be able to get some really

24:32

interesting guests on the show

24:34

, and we'll be doing that from the

24:36

next episode onwards . So let's

24:38

move on to our main

24:40

episode . So , as I said , we have a guest

24:43

on , so we're going to change

24:45

into our dressing gowns and

24:47

then we're going to get into the other studio in

24:49

the back and we will be back with

24:51

you in two minutes time . So

25:03

welcome back . Jimmy and I are in our dressing

25:05

gowns now . That's a site you don't want to

25:07

see , so that's why we keep the videos off YouTube

25:09

and keep this to a podcast . So welcome

25:11

to the podcast , dan lyons . Dan is

25:13

the strategic comms advisor , who's

25:16

worked across a variety of roles

25:18

. He started out as a journalist , he's worked in government

25:20

and private sector , and his last role

25:22

was a managing director of a global

25:24

strategic consultancy .

25:25

So , dan , welcome to preparing

25:28

for ai thank

25:30

you great to be'm a fan

25:32

of the podcast , so it's lovely to actually

25:34

be here and chatting to you guys .

25:36

Well , that's why we wanted you on , because we , you know , obviously

25:39

we have 2 million listeners , but to have one of

25:41

them who's such an expert in a field on

25:43

the podcast is a pleasure for us as

25:45

well . So I guess let's start off , let's

25:47

have a look at your kind of own

25:50

experiences . So , if we make this

25:52

, I mean let's look at , I guess , the last six

25:54

months , so from the beginning of the year

25:56

, what have you seen

25:58

in the industry in terms of both

26:01

the adoption of AI tools

26:03

but also the kind of attitude , guess

26:12

I'm interested in the attitude of .

26:12

You know people at the top , but also you know people working in the industry and how they are reacting

26:14

to those tools and how they're reacting to you know potential

26:17

for job losses , or you

26:19

know changes or insecurity

26:21

around their roles the first thing to say is that

26:24

sort of ai has actually sort of been creeping in as

26:26

as sort of in terms of ai-based

26:28

tools to industry for quite a few years I

26:30

think , starting with mainly

26:33

kind of executional tasks

26:35

, particularly around sort of data analysis

26:37

, media monitoring , the use of

26:40

AI tools to sort of gather in large amounts

26:42

of you know media articles , to analyze

26:45

trends , to sort of say , for example , how negative

26:48

an article is , how positive it is , and

26:51

derive performance-related

26:53

data from that . Also

26:56

, in a place like China where I'm based , the

26:59

use of AI for translation , which

27:01

has really accelerated certain areas of the

27:03

industry , and abilities to produce

27:08

content and to analyse content . I

27:11

think adoption is still

27:13

low , though I

27:18

think the introduction of chat , gpt has been an inflection

27:20

point , but really the usage across the industry

27:23

is still relatively low . I think there's

27:25

been some studies last year

27:27

by the Chartered Institute of Public Relations . I

27:29

think only 40%

27:31

of tasks that are performed by PR professionals

27:33

are now assisted by AI

27:35

tools and I think that's up from about 12%

27:37

the previous year . So you know there's

27:39

still a lot that's going on within the industry that doesn't

27:42

rely on AI and

27:44

within that I would say that sort of most of

27:46

the usage is , as I said , low level rather

27:48

than strategic . So you know , monitoring

27:50

, data analysis , information analysis and

27:52

executional tasks . What tends

27:55

to sort of remain on touch is more strategic

27:57

work , so that's sort of crisis management , uh

28:00

, you know , risk mapping , risk forecasting

28:02

and , obviously within a business like

28:04

pr , relationship management

28:06

. So that's both with your

28:08

stakeholders , with your clients , with

28:10

with the media , with journalists . I

28:12

think that's very much still sort of a , you know

28:15

, a human-led task rather than anything

28:17

that relies on um

28:19

, on ai , um . I

28:22

think there's two sort of issues that are impacting how

28:24

it's being sort of adopted . I

28:26

think the first one is a skills gap . I

28:28

think , you know , within

28:30

sort of the you know my , you

28:32

know amongst my peers and within sort of

28:34

companies that I've worked at , I think there's

28:37

very few people who you could

28:39

say were experts in sort of the use of AI tools

28:41

, and usage has tended to be sort of

28:43

fairly organic and has evolved over time

28:45

. And I think there's

28:48

a particular issue around ethics . You

28:51

know ethics of AI in in PR

28:53

. I think PR

28:55

professionals , communication professionals , are a little bit

28:57

nervous about using them , mainly

29:00

because you know how accurate are they

29:02

? You know I'm , you know I'm relying on sort

29:04

of information I'm getting from these tools and

29:07

I don't want to pass on any accurate information to either

29:10

within my company or to clients . You

29:12

know it doesn't get the tone and the style right all

29:14

the time . I mean , I personally use

29:16

things like chat gbt for , you

29:18

know , the first draft just to get something

29:20

down on paper , just to sort of spark an

29:22

idea . But often , you know , I'll

29:24

completely change sort of what's produced . I

29:27

rarely , if any , you know , if

29:29

at all , use anything that is produced without

29:32

editing and

29:34

I think , particularly on the agency

29:36

side , there's

29:39

an issue around the ethics of it . So billing

29:42

clients for work that has been created

29:44

using AI , the optics of that , particularly

29:50

if you're charging quite a lot as an agency , are

29:52

tricky . You know it feels a little bit like cheating

29:54

. So

29:56

those are the two issues that I think are sort of

29:59

having an impact on the adoption . But

30:01

you know the launch of chat GPT

30:04

has definitely been an inflection

30:06

point within the

30:08

last sort of six to nine months . You

30:12

know there's a definite increase in sort of the use of large language models

30:15

and so that 40 figure that

30:17

I mentioned earlier could definitely be a lot higher . Um

30:19

, and you

30:22

know sort of the types of things that are being

30:24

used . Uh , you know low , low level

30:26

content creation , you know , social

30:28

media , press releases , uh

30:30

, again sort of media analysis and translation

30:33

. So people are trying to get up to speed fairly

30:35

quickly .

30:36

It feels like . So I mean , first

30:38

thing I was going to ask is like , do you think you said

30:40

you use it already ? So why

30:42

do you use it ? Does it save you

30:44

time if you don't use , if

30:46

you don't actually use most of what it writes , but you

30:48

just get it , get the ball rolling with it ?

30:59

Yeah , don't use if you don't actually use most of what it writes , but you just get it , get

31:01

the ball rolling with it . Yeah , so it says so . For me it it saves time . You know , if you , if you sort of

31:03

put in you know a suitably detailed prompt , uh , you know you could save up to an hour uh in in creating

31:06

sort of the first draft or something . If you've um

31:08

, if , if you use chat , gpt

31:10

or a similar tool , uh , you know , so

31:12

it's very efficient . But for me , actually

31:14

, it's also , you know , even if you've been

31:16

in pr or you know sort of the communication

31:18

industry for a long time , you , you

31:20

know , you might not always get the inspiration . You need , sort

31:23

of uh from the , from the bat , and you

31:25

know , sometimes I like to see something

31:28

on paper , even if it's not , if it's something that I will

31:30

not use , just as a kind of a sparring

31:32

pad or , you know , a launch pad , instead of doing something

31:34

else . So so it's , it's not

31:36

just sort of the efficiency aspect

31:38

for me , it's also that it sparks

31:41

my own thoughts on on something and

31:44

, as I said , you know it rarely gets , you

31:46

know , it rarely gets it right first time , but it's , you know

31:48

a good start , for you know if you , if

31:50

you need a combination of wording

31:52

that sort of might make a good social media post or

31:54

or sort of you know

31:56

key messaging or or press releases

31:58

, or you know kind of a keynote speech

32:00

or quote . Um , you know , there's always something

32:03

you can use that then sort of uh

32:05

, that prompts you to sort of edit

32:07

in your own style I know exactly what what you

32:09

mean , so I've kind of used it in the same way .

32:11

It feels like a little cheat to get around

32:13

writer's block or something like that , where , like

32:15

, staring at a blank piece of paper can

32:17

be quite daunting , where if

32:20

you just pop a prompt into chat , gpt , it gives you

32:22

something , even if it's just a scaffolding where

32:24

you have to then rewrite almost everything .

32:26

Yeah , and I think I everything , yeah , and I think I mean going back to the point about

32:28

why , sort of why take up in the industry

32:30

as a whole might be . Like you know , I

32:33

, I sort of am a regular user of of

32:35

, uh , of chat , gpt

32:38

and and the large language tools , that sort of

32:40

um that are available and in fact

32:42

my , my sort of . Some

32:45

companies are actually using , uh , you

32:47

know they're creating their own versions of of chat

32:49

, gpt , sort of you know that are sort of

32:51

a tailored for for that business and and for

32:53

the sort of the . You know the work that

32:55

that is done by sort of uh , the , the

32:58

teams within the business . Uh , my own , I

33:00

recognize my own sort of skills , uh , you

33:02

know my skills gap and the need to

33:05

uh , the need to sort of build it

33:07

up . So I mean , there's probably a lot more I could be

33:09

doing and there's probably a lot more that could be , could

33:11

be done within uh . You know the , the

33:13

industry and the companies that I've I've

33:15

worked for but are not done because people just

33:17

simply don't uh you know , they're not knowledgeable

33:19

enough and you know , as you

33:21

say , you've said on sort of um previous podcasts

33:24

and I'm sure you've sort of covered it

33:26

today that you know things

33:28

are changing so fast and and keeping

33:30

on top of the you know , proliferation of

33:33

tools that are out there , you know , I think sort of

33:35

again within the

33:37

studies that have been done by by sort of

33:39

the governing bodies , you know there's probably up

33:41

to about 10 000 ai

33:43

assisted tools that that you know could potentially

33:45

be used within sort of pr industry alone . So being

33:48

able to keep on top of that is incredibly hard . Um . So

33:51

you know , people are people like me are getting up to speed as

33:54

quickly as they can , but there's , you know there's still

33:56

a way to go .

33:57

I think sort of at your level and

33:59

and we we maybe face the same point

34:01

in a lot of episodes . I'm thinking of a law episode

34:03

as a as a good example of this actually , where

34:06

there feels like that

34:10

there may be a big difference in terms of the

34:12

immediate threat depending on

34:14

where you work . And sometimes we hear

34:16

that you know AI is different from other

34:18

revolutions because it's coming for , you know

34:20

, more senior jobs first , but actually

34:23

if you look at the area , so you know , just looking

34:25

at some of the areas in the notes , I've got social

34:27

media management . You know the AI

34:29

driven tools for scheduling , managing

34:35

your social media content and stuff already . Data analysis and reporting ai can

34:37

already process and analyze your data sets and you give you feedback on that

34:39

presents the for your media monitoring and analysis

34:42

. I mean , that's media monitoring for me is . You

34:44

know that that's gone . I think that that

34:46

if anyone's still paying for people to you

34:49

know do media monitoring , then I think you're you're

34:51

throwing your money away and content

34:53

creation and writing . You know it feels like and

34:55

, having done the interview a few weeks

34:57

ago , it feels like

34:59

one of the issues with content

35:02

might actually be it's not

35:04

that the generated content is

35:06

as good as content created

35:08

by people . But it's just that , whether

35:11

it's the editors or the , you know , the senior

35:13

management , or the consumers themselves

35:15

, people are just willing to accept poorer quality

35:17

content . So you know , with all those kinds

35:19

of different areas , that we think

35:22

they are being already

35:24

affected or they are going to be massively affected

35:26

by AI . Your role

35:28

may not be directly affected , but

35:31

what are you seeing in terms of the

35:33

people you work with ? I mean , are they

35:36

seeing the sort

35:39

of writings on the wall and they're thinking my

35:41

job's gone , or are they

35:44

not really thinking that way ? I mean , what

35:46

is the sentiment in the industry ? Is there a lot

35:48

of fear ? Is there a lot of excitement

35:50

? I know it's difficult to

35:52

speak for everyone across an entire industry

35:54

, but you know your experiences of people . Are

35:57

they optimistic , pessimistic

35:59

? You know what are their feelings about the

36:01

kind of AI revolution .

36:03

Yeah , I mean , I think , in general , the

36:06

feeling is that this is something that needs

36:08

to be taken seriously , not not only from

36:10

the perspective of how it will impact the

36:12

, the pr industry or the communications

36:14

industry , but also how it will affect our

36:17

sort of the environment which

36:19

we kind of operate in . So that's the wider media environment

36:22

, the kind of the corporate world . You

36:25

know . How do we as a , how do

36:27

we as an industry , tailor our offering

36:29

and the services we provide to

36:32

basically take account

36:35

of not only

36:37

opportunities that AI provides , but actually the risks

36:39

as well , and these are risks

36:41

from a media point of view how

36:43

to , for example

36:45

, help companies protect themselves

36:47

against misinformation or , you

36:49

know , other organizations against mission misinformation

36:52

? How to , you know , help

36:54

a company through a deep fake crisis , for example

36:56

, that you know may sort of have a big impact on

36:58

their , their business ?

37:00

um , so there's a new world

37:02

of work for you yeah

37:04

, exactly .

37:05

So you know , I think sort of a new kind of sub-industry

37:07

here , right , yeah , so I mean , within the kind of the strategic consultancy environment . There's

37:09

a new kind of sub-industry here , right , yeah . So I mean , within the kind of the strategic

37:11

consultancy environment there's definitely a sense of we

37:14

need to get up to speed , but there's definitely an opportunity

37:16

to be able to advise people on

37:18

how to sort of handle this brave new world

37:21

. I guess the sense or the

37:23

general sort of feeling is that yes , there's a recognition

37:25

that you know , as adoption accelerates , low-level

37:28

, entry-level tasks will be displaced , but

37:31

actually you know this is an opportunity for you know

37:33

profession wide strategic shift in

37:35

focus and actually you know any threat

37:37

to jobs would be because people

37:39

need upskilling , not because the jobs will necessarily

37:42

disappear entirely . So

37:44

I think I think the sort of the kind

37:46

of the general consensus view is that yes

37:48

, the PR is infused with AI , but

37:51

wholesale job replacement is

37:53

not happening yet . So

37:55

I've heard people say it's like the introduction of Excel

37:58

and the impact that had

38:00

on the accountancy profession . People

38:02

were fearing that the creation of spreadsheets

38:04

and Excel as

38:08

a software would obviate

38:11

the need for paid professionals who

38:13

would do your accounting . But obviously

38:15

accountancy still thrives . You

38:19

know , I personally am not

38:21

convinced that that's sort of a good analogy and

38:23

I'm not 100% convinced that eventually

38:25

the proliferation

38:28

of AI tools and the development of AI

38:30

won't eventually touch more strategic

38:32

areas such as crisis

38:35

management , c-suite advisory , or

38:37

that certain roles , for example , won't become

38:39

superfluous once companies , both agencies

38:42

and in-house , realize it's just cheaper to use AI

38:44

. So

38:48

if you're an agency and you can

38:50

see over time that actually you don't need so many

38:52

junior associates or

38:54

junior team members you know doing

38:56

the work for you , because actually a lot of that can be done by

38:58

you know fewer people using

39:01

tools , then you know the

39:03

economic logic of it is that that

39:06

it that it sort of those roles would

39:08

disappear . So

39:10

I've seen this sort of the idea that people

39:12

become trained as prompt architects

39:15

and I think you know that's a part of why

39:17

people you know would

39:19

traditionally join us .

39:22

It's also bollocks the

39:25

idea of prompt engineers and prompt architects

39:27

is bullshit . It's

39:31

an industry that might exist for a year and

39:33

then it's gone . I mean I I sorry to interrupt

39:35

, but I mean I I've done a course on um

39:37

from Vanderbilt university , an online course on

39:39

prompt engineering , and it's fun and it's kind of useful

39:42

. But I did a course a

39:45

couple of months ago and the course was

39:47

obviously written in late 2023

39:49

. And when I did the course , it was

39:51

already out of date to

39:53

me because , a lot of the things it

39:55

was teaching you to engineer . You no longer need to

39:57

engineer and I think for the same reasons , if you're

39:59

learning to engineer things now , you

40:01

know those things will be . You

40:03

won't have to . The whole point of the advancement

40:05

of the models is that you'll be able to speak

40:08

in a natural way and it will understand and be able to

40:10

. You know prompt , or you can even just tell it now

40:12

give me the prompt to do this and it will give you the prompt

40:14

. Then you give it back the prompt and then it does it . So , yeah

40:17

, I , I think the idea of any

40:19

roles you know , or not necessarily

40:21

roles . I mean , there might be roles but the fact that you can go

40:23

and become and have a career as a prompt architect

40:25

or prompt engineer is is for the

40:27

birds I , yeah , I totally

40:30

agree .

40:30

I mean I I thought that right from the start

40:32

, with things like prompt engineering Even

40:35

between I think I'll use mid-journey as an

40:37

example , but between mid-journey I think

40:39

it was two and three the need

40:42

to do any kind of lengthy

40:45

prompt engineering to get the

40:47

model to output images just went

40:50

away . It

40:52

went from .

40:52

You have to be really specific about how you , how

40:55

you prompt it to , you can just tell it

40:57

you want a picture of whatever , like a

40:59

bird on a mountain or something I think I

41:01

think learn to prompt sorry , just to say that I think learning

41:03

to prompt is useful and I think

41:05

studying these courses is useful for helping

41:08

you to be able to prompt better . And I can , you

41:10

know , I taught uh chatPT

41:12

, a language I wanted to use so

41:14

that it would I could prompt it quicker to

41:16

give me actually information

41:18

for this show . So I could give it three

41:21

asterisks is , followed by a word and a number

41:23

, and three asterisks is , and it would then give

41:25

me information within a timeframe

41:27

on a certain thing . And it's quite fun and it allows

41:29

you to do things . But the idea that that would

41:31

be something that is useful enough

41:34

to be a career or a job , I think is is yeah

41:36

, it's , yeah , I think it's

41:38

a non-starter .

41:39

Sorry , dan for uh so

41:43

I think sort of the , the acceptance

41:45

and adoption within pr will will have

41:48

an impact . Um , as

41:50

I said before , you know there is a

41:52

hesitancy at the moment and a nervousness

41:54

, but I think within the next sort of six

41:56

months you know 12 months I think I

41:58

think that will slowly sort of ever way and

42:01

then actually the sort of the industry itself will sort

42:03

of start to see the impact within the sort of the structures

42:05

of companies and within the sort of the the industry itself

42:07

on , uh , you know , from from ai

42:09

. I think the second sort of slightly sort of

42:11

linked aspect of that is the

42:14

reaction of in-house teams

42:16

, but also you know clients who are , you

42:18

know , in-house teams . You know at the moment there's

42:20

probably not a lot of awareness of how

42:23

AI is being used , sort of both

42:25

within , you know , within companies

42:27

, but by sort of external parties possibilities

42:33

. Possibly , even if there was full awareness , that sort of clients would still feel they need

42:35

that . You know the access to the sort of the top level advisors . You know the people with you

42:37

know huge experience in sort of

42:39

certain fields , you know , and in

42:41

certain with certain capabilities , for example

42:44

, around crisis and or sort of political

42:46

advisory . You know that I

42:48

, I can possibly see that changing

42:50

if , if , for example , 12

42:53

months down the line , a few years down the line , someone produces

42:55

something called the boardroom ai

42:57

advisor , which is perhaps built

42:59

into companies business

43:02

continuity plans and , and you

43:04

know , provides that pr function . It's sort of it's

43:06

, it's built in . In terms of risk forecasting

43:08

, you know you can assess reputational

43:10

risk scenarios , provide

43:12

playbooks , because I I mean , we're all using playbooks

43:15

, so nothing is sort of entirely original and

43:17

you know , and then even execute these plans , you know

43:19

, linked to media databases

43:21

, linked to sort of you know , social media

43:23

channels , could basically sort of assess

43:26

kind of the ongoing sentiment

43:28

around a particular issue and prompt

43:31

you know prompt responses

43:33

that go public . I mean , you know that's way

43:35

off , but you know that could be another game changer . Ultimately

43:46

, it comes down to a point that I think

43:48

probably is key to adoption within

43:50

the industry and you know , beyond

43:52

PR , beyond communications , and you know beyond , beyond , uh , beyond pr , beyond communications

43:55

, and you know , for the overall adoption of ai

43:57

and that's , you know , trust , and we can talk about

43:59

that a bit more if you want yeah , I just want

44:01

just before we do , because I think trust is a

44:04

a massive one

44:06

, not just in this industry , but I I

44:08

just want to go back just .

44:09

We talked about crisis management , so I think crisis management

44:11

actually is a really good example of one where it

44:13

seems initially , you know , I

44:16

had an example the

44:18

Institute from PR it was from a link from

44:20

them about AI tools that are being used to monitor

44:22

sort of real-time data and

44:25

detect potential crisis

44:27

, allowing companies to respond swiftly

44:29

and effectively and kind of mitigate damage

44:31

. But it's about detecting them . It's not

44:33

about you know , giving you the advice on

44:35

how to deal with it . And I think a great example

44:38

is you know you work in China . There's a very specific

44:40

environment in China where things which maybe

44:42

somewhere else would not be an issue the way

44:44

that you , you know , refer to the mainland

44:47

, a certain island , an administrative

44:49

region , you know can create

44:51

huge crisis and

44:53

therefore having that requisite

44:56

knowledge and understanding the nuance

44:58

and the political situation is

45:00

really , really important . But I think we

45:02

always talk about how you know the

45:05

kind of advancement of things and I think part

45:08

of the problem I can see is that if I

45:10

was in your organization and I was

45:12

adopting AI , I would build

45:14

an offline LLM

45:18

system within your organization that

45:20

only had your data in there and

45:22

was picking up all your data and basically , every

45:24

time you handle a crisis , you're telling it about

45:26

the nuances of the situation in China and

45:29

you're training it to basically do your

45:31

job , and you're training it every time you do

45:33

a good job're training it . You know

45:35

, one more step to you losing your job , and

45:37

that was the example we gave a couple of weeks ago , that the

45:39

guy who you know lost his job and was like , hey

45:42

, it was my data that was used to train the AI

45:44

that's now replaced me . I think when

45:46

you say a long way off , I don't know , I

45:48

don't think it's a long way off , I think it's maybe a couple of years

45:50

off , but you can easily see a

45:54

model being created that allows

45:56

you know within an organization

45:58

for it to pick up the nuance , to be able to do

46:01

most of that crisis management . And then again

46:04

, you don't remove people completely . You

46:06

might still be there as the kind of last step in the chain

46:09

, because we need someone to blame when it all goes wrong

46:11

, but you're certainly able to take out

46:13

a lot of people in

46:15

the chain and we're able to , you know

46:18

, massively reduce the teams that are working

46:20

on it .

46:20

So it was not so much a question .

46:22

It was more a kind of observation no

46:25

, no , I think you're right .

46:26

I think , yes , it sort of , and

46:29

again it probably comes down to the trust issue that we're talking

46:31

about . But you know the , the sort

46:33

of the , I

46:35

guess the framework for risk is

46:37

programmable . You know , and

46:40

in terms of the red lines

46:42

you're talking about , when it comes to sort of operating

46:44

in China , you know they're already well known , they're already sort

46:46

of well publicized and published

46:48

. So there's no reason why if

46:50

you typed in , you know what

46:53

are the three red lines that companies need to bear in

46:55

mind , when you

46:57

know , when operating in China from a reputational

46:59

risk point of view , then that probably already exists

47:01

there , so

47:04

you can set the parameters of your risk , you can

47:06

monitor that risk , and

47:08

then it's

47:11

easily sort of programmableable . You know

47:13

how would you respond to that and perhaps , you

47:15

know , perhaps a tool would come up with based

47:18

on all sort of the inputs , you know , three possible courses

47:20

of action and maybe there's a person at

47:22

the other end who makes that decision , but eventually

47:24

maybe it's just another ai tool that makes that

47:26

decision and then passes the , you

47:29

know , bypasses the sort of the

47:31

need for human interaction at all . I mean , that's , that's

47:33

theoretical and and I don't be given

47:36

sort of risk issues . You know people

47:38

probably wouldn't want to hand that over entirely to to

47:40

sort of you know how it all throughout any sort of uh

47:43

without any human input . But you could

47:45

, as you say , you could do away with a lot of people in the chain you're

47:47

giving me uh , you're giving me loads of great business

47:50

ideas here , dan yeah

47:53

, I went in on that and so

47:55

, yeah , I think sort of the a lot , a lot of

47:57

this , I think , within the pr industry . But , you know , possibly

47:59

on that comes down to to trust , because

48:01

they're currently nobody really trusts ai to get

48:03

things right , um , and

48:05

there's too much sort of uncertainty . So the

48:07

logic is that you'll always need a sort of a human guiding

48:10

hand and , as I said , you know

48:12

the wider environment , you

48:14

know with what with ai , ai comes ai

48:16

risks , um , deepfake misinformation

48:19

, you know that's potentially new avenues of

48:21

business , uh . So

48:23

you know , maybe the trust will never materialize and

48:25

you'll always need this sort of this . These human custodians

48:28

and I know that sort of the professional

48:30

bodies you know actually see a role for the pr

48:32

industry . For , you know , advising on governance

48:35

issues , about the ethical use of AI , you

48:37

know regulatory issues , um

48:39

, you know , almost sort of setting themselves up

48:41

as the reputational authority , uh

48:43

, around AI . You know that is the

48:45

kind of the million dollar question . I guess that sort

48:47

of trust is the inhibitor

48:50

. If you remove that inhibitor , then you know what's possible

48:52

. And actually , I mean , before we get to

48:54

that , I do think sort of there's probably a silver lining

48:56

. And actually , jimmy , what you were saying

48:58

in terms of the business ideas , I think actually

49:00

sort of in the next few years you might see actually

49:02

a new wave of entrepreneurialism , uh

49:05

, you know , within the industry . Currently

49:08

, you know you've got lots of big companies . There's

49:10

economies of scale . You know they sort of need

49:12

the expertise teams . But if you

49:14

know , if AI tools are doing

49:16

a lot of the specialist work , then actually you

49:18

could see sort of you know one man bands

49:21

, smaller PR companies that

49:23

actually can do the

49:25

work of a huge agency because actually you

49:27

know they can offer a full suite of creative

49:29

and advisory services .

49:31

Yeah , I was gonna say I mean , I said it half

49:33

jokingly , but exactly that , like I think there's two

49:35

. I think there's two different things . When we talk about

49:37

jobs on the podcast , there's

49:40

two different ways that jobs might be affected

49:42

. One is like adoption within existing

49:44

industries , like within , for example

49:47

, within existing pr companies , things like that

49:49

, um . The other is exactly

49:51

that . It's the kind of the , the

49:53

entrepreneurs and the innovators and the disruptors

49:56

that come in and just set

49:58

something up that just out competes

50:00

because it's they've you know , they've

50:02

figured out the technology . It's just based on ai

50:04

. Maybe the , the

50:07

trust thing is still an issue , and so you know

50:09

, certain , certain entities need to rely

50:11

on um , you know , rely on , rely

50:14

on corp um companies or pr companies

50:16

that have humans in the loop , but

50:18

it's sort of you know you

50:20

, you get this kind of lowering of the bar where

50:22

actually services like this

50:25

probably become accessible to a broader

50:27

group of people much cheaper as well

50:29

.

50:30

They don't need economies of scale either , do they like

50:32

they don't need economies of scale ? You could actually argue that as

50:35

a one-man band or a small company can

50:37

do things more efficiently because they don't have all the overheads

50:39

of a big pr firm no , exactly

50:41

yeah so I think that's .

50:43

I think there's there's potential for massive disruption

50:45

in that kind of sense as well

50:47

, and I don't think we've seen that

50:49

yet , I think we're still . You

50:52

know , there are some companies that have set up , that have done

50:55

that , are doing this kind of one-man band sort of thing

50:57

, and I know it's been talked around that

50:59

you know , in the next , maybe in the next

51:01

five , ten years , we'll see the first sort

51:03

of you know company that's just one person

51:06

and a bunch of ais , who becomes

51:08

worth you know billions of dollars

51:10

, because because of exactly that

51:12

, because they can just automate everything , but it's

51:14

, it'll be interesting to find out perhaps

51:17

leading into that .

51:17

I think you know the the one I obviously

51:19

pr is an industry in isolation

51:22

and it's you know . The other is , the other side

51:24

of a coin is is the mass

51:26

media market and you know , I think sort

51:28

of if you're going to have a conversation about , about

51:30

sort of pr , then you need to

51:32

have a conversation about you know what , what , what environment

51:34

is pr operating ? And you know the wider

51:37

, you know the wider media environment

51:39

and you know that's probably a discussion for another

51:41

podcast about . You know how will people consume

51:43

media and what , what

51:45

will media be and how ai

51:47

will influence that . Um

51:49

, you know , and you know what

51:52

is the role of journalism sort of in the ai world

51:54

and , as I said , you know in some ways

51:56

that will build into the kind of you may

51:58

actually see a sort of a reaction where , because , because

52:01

the proliferation of of content and

52:03

you know imagery and video , that

52:06

, that , no , you know sort of it's it's

52:08

very , very hard to sort of uh

52:11

, pinpoint where it's originated and you know

52:13

concerns around misinformation and deep fakes . You know you actually mean that that sort of pinpoint

52:15

where it's originated and you know concerns around misinformation and deep fakes . You know you actually may

52:17

. That sort of that may mean you're in a kind of

52:19

a kind of a virtuous

52:22

circle where you know people will

52:25

never fully trust the media and therefore you'll always need

52:27

, you'll always need , sort of PR

52:29

, I guess a PR industry to

52:31

to help companies you know companies

52:33

, organisations , but also just the wider public to help company you know companies , organizations , but also just the wider public to

52:35

navigate , you know to navigate that sort of

52:37

uncertainty . So , yes , I mean . So

52:40

trust for me is the kind of the , the sort of the

52:42

inhibiting factor , and it'd

52:44

be interesting to get your guys views on at

52:47

what point do people start to trust ?

52:49

ai , I mean , I , I think I

52:51

think never and and I

52:53

I agree with you more so I've

52:55

been thinking about it a lot the last two

52:58

days , since we kind of exchanged notes actually and

53:01

I think , yeah , I think it's potentially

53:03

the biggest barrier

53:05

or sort of challenge for AI to overcome

53:07

. Something

53:13

that really stood out to me I can't remember who it was , it was a member of the general

53:15

public or a comment on a board but someone had said we

53:17

never asked for this . They were talking about AI

53:19

and they were saying we never asked for this . No one asked us whether

53:22

we wanted this . And okay

53:24

, yeah , jimmy said yeah , well , that's the same with everything

53:26

. You know , we don't ask for it and I agree

53:28

. But you know we're being kind of given

53:31

this thing that we're told , hey , this is going to change the world

53:33

, you just got to accept it , it's going to happen

53:35

to you and so , yes , we are

53:37

going to have to accept it and it is . It is

53:39

going to be part of our lives and it's going to make fantastic

53:42

positive changes and it's going to potentially threaten

53:44

, you know , the existence of humanity

53:47

and all of these challenges we're going to have to face . But

53:49

if you start on that basis that people feel

53:52

this is being imposed on them . And

53:55

then you take the distrust that we have in the

53:57

world at the moment you know , I think

53:59

quite rightly in institutions

54:01

and authority , and you put those

54:03

two things together and then you

54:05

say you have to trust this thing . So

54:07

let's , for a second , throw out the

54:10

fact that we're talking about trusting a

54:12

super intelligent , you know form that

54:14

has a level of intelligence

54:16

we've never seen before . You know some

54:19

point in the future and may have its own wants and wills

54:21

and desires . Forget that that's . You know that's a way

54:23

off at the moment . But someone's controlling

54:26

AI . You know big

54:28

tech firms , governments , whoever it is , the

54:30

military , whoever's got control of of those

54:32

. I think the assumption for most

54:34

people is that ai is being

54:36

run by them , whoever

54:38

they are , and therefore , how

54:41

do you overcome that trust issue ? It's

54:43

fine when you're using it for things

54:45

which are , you know , potentially

54:47

fun , or or or

54:49

frivolous , or , you know , semi-useful

54:52

, or even for things you know there seems

54:54

to be a lot of trusting kind of science and health care , that

54:56

it will be a positive there . But when

54:58

you're having to make a decision

55:00

that affects the potential

55:03

future of your organization , for example , or

55:05

the future of your safety . You

55:07

know , getting in an autonomous vehicle you

55:10

might get an autonomous vehicle as a normal

55:12

private citizen , but if you're someone who knows

55:14

that there are people who are , you know , out

55:17

to get you , are you ever going to get in an

55:19

autonomous vehicle ? And I think it's

55:21

the sort of the fears of society

55:24

in general , the distrust that's out there . That

55:27

, for me , is why we will probably never

55:29

overcome that barrier of trust . I

55:31

say never . I think we always say on the podcast

55:33

we should never say never . We're talking

55:36

in a finite amount of time

55:38

, so let's talk in our lifetimes . But

55:40

yeah , I think trust is absolutely

55:43

the biggest barrier and I think you raise it in the PR

55:45

industry . I think you're right , because you

55:47

are putting your business's future

55:49

and you're putting a crisis

55:52

or you're putting the reputation of your

55:54

business in the hands of a person

55:56

or an organization or an ai tool

55:59

.

55:59

But it does translate across the entire spectrum

56:02

, I think yeah , and I think sort of you might

56:04

, you might also then have a kind

56:06

of a dual track world where you know

56:08

there's an acceptance , that sort of a lot of the information

56:10

that's out there is produced by ai , but actually there's

56:12

a kind of almost like a quality mark that

56:14

goes with you know , this , this , this

56:16

, this article , this , this , this

56:18

product , this content is , you

56:20

know , has has been produced 100

56:23

by , you know a human being , um

56:25

, and that you know this sort of , and

56:27

I think that you know you're already seeing , uh

56:30

, you know , news websites . I

56:33

think they're well , I don't know whether people are being

56:35

compelled to or whether it's just a sort of you know an

56:37

ethical , uh , you know an ethical

56:39

decision to to sort of indicate where , where

56:41

articles have been written with the help of ai

56:43

, that it may become sort of necessary , you

56:46

know , I don't know whether through regulation or otherwise

56:48

to sort of indicate where , where information has

56:50

been produced with the help of ai and to what extent

56:53

.

56:53

I think the EU's AI Act that

56:55

will be covered , and

56:58

what usually happens with a lot of territories

57:00

is that they follow the EU because the

57:03

EU's the strictest and so

57:05

we may as well just follow what they put in . But I think that

57:07

I'm not 100% , but I'm pretty sure that

57:09

in the Act is exactly that that

57:11

you will need to with whether it's images

57:13

, you know stories , articles everything

57:15

will need to be labeled to be quite clear that it's produced

57:18

by ai and and you're right in terms of you

57:20

know organizations taking that decision

57:22

. So the economists , for example , have

57:24

, uh , have taken a kind of

57:26

you know editorial policy where they will quite

57:29

clearly state what has used ai and what

57:31

hasn't . I think a lot of , I say , reputable

57:34

organizations will will choose to do the same

57:36

thing don't you think , though sorry

57:38

to jump in , but as

57:40

ai becomes more and more ubiquitous

57:43

?

57:43

we talked um in the kind of introduction

57:45

to the episode about actually how few people have

57:47

used ai or heard of it or

57:49

various things , and it's actually still

57:52

relatively low numbers . Do

57:54

you not think , as ai becomes more and more ubiquitous

57:57

, though , what isn't

57:59

going to be have had ai

58:02

used in its production

58:04

, because I feel

58:06

like it's going ? I mean okay , for like to give you

58:08

a concrete example , google

58:10

google are building ai into their search function

58:13

right now , and so if you use

58:15

Google search to assist

58:17

you with finding information , does

58:19

that mean you have to label it as AI

58:22

assisted ? I'm curious

58:24

about this , because I genuinely think it's not

58:26

created it .

58:27

I think the issue here is about whether it's a creation

58:29

of AI , so that's AI using or

58:31

assisting you in

58:34

carrying something out . I think the issue here is about

58:36

intellectual property and whether an

58:39

article or a piece of music

58:41

or an image is AI created

58:43

. I think that's where it is , because it comes out . Dan

58:46

mentioned about deep fakes and stuff . I think that's

58:48

probably at the heart of it , isn't it About how

58:50

do you make sure that people know

58:53

what's AI and what's not ? Actually , if the

58:55

search is helping you do it , but the AI is not

58:57

producing the content , it's just

58:59

helping you with the process , I don't

59:01

see what the issue would be with that .

59:03

But I'm still not clear , Like if

59:05

you create an article using AI , but

59:08

you just tweak a few bits then

59:14

it's not created by ai , like I think it's a really great area actually

59:16

.

59:16

Yeah , you're right , you're right , no , you've , you've banged on there . I just , I just read

59:18

it and I guess it will .

59:19

There will be someone , you know , there'll

59:22

be lawyers and judges who probably argue where the fine line

59:24

lies between assistance

59:26

and creation . You know , you

59:29

know that that will be the sort of the , the key

59:31

issue . I think that it will be a badge of one to

59:33

say , you know , that sort of a hundred percent

59:35

of this article or 100 of this content

59:37

was was produced .

59:39

You know , in an analog way , I wanted

59:42

to finish the episode on a on a quote

59:44

that you , that you sent

59:47

me yesterday I thought it was longer ago than

59:49

that , but yesterday actually . So , you

59:51

, we were talking about the

59:54

sort of conversation today , what we might talk about

59:56

, and and you mentioned trust and , and you said

59:58

, ultimately , if trust was an issue , there's

1:00:00

no part of communications that couldn't be done

1:00:02

by ai , and I guess , I think

1:00:04

not now , but , you know

1:00:06

, two , three , five years in the future . I

1:00:09

think that applies across almost

1:00:11

every job and probably almost every task

1:00:14

, and that that's why I think you know the

1:00:16

point that we've made here trust

1:00:18

is an issue . So if

1:00:20

trust was an issue , we could do lots of things , but trust

1:00:22

is an issue and therefore I

1:00:25

do think you know you're quite right , it will

1:00:27

be a barrier for certain roles in communications

1:00:29

, but it would be a barrier for a lot of things . And I

1:00:32

still think that sort of self-driving

1:00:34

vehicles thing is a great example

1:00:36

. Would you trust the self-driving

1:00:38

vehicle ? It's not about the

1:00:40

technology of the vehicle , is it ? It's

1:00:42

your trust that you're putting in who's

1:00:44

got control of that vehicle or what

1:00:47

has got control of that vehicle , and that's

1:00:50

what I think is the issue with trust . It's

1:00:52

not about the technology

1:00:54

, it's not about the ability

1:00:56

of AI to do the role . It's about the biases

1:00:58

and it's about the motivations and

1:01:01

who's controlling it and what are their motivations

1:01:03

. And we're in a , you know , post-truth , post-trust

1:01:06

era . I think you're bang

1:01:08

on . I think I think trust it's going to be a thing

1:01:10

that we're probably going to

1:01:12

touch on more and more , I think , um

1:01:14

, over the next however many months and

1:01:16

years that we do this podcast . But , um , thank

1:01:19

you , dan . It's been an absolute pleasure

1:01:21

, really interesting conversation . So thanks for giving

1:01:23

us your time this evening . Thank you very much . So

1:01:25

that's it for this week . Um

1:01:27

, I want to just finish off . No-transcript

1:02:02

could have if it's not managed and

1:02:04

developed properly . So I want to ask

1:02:06

people who listen to this show this

1:02:09

week to please ask three

1:02:11

people just recommend our show to

1:02:13

three people three friends , three family , whoever

1:02:15

it is but actually not

1:02:17

just recommend them the show . Recommend

1:02:20

them a particular episode that you think would

1:02:22

be of interest to them and help us to try

1:02:24

and grow the show . It's something we're going to focus on in

1:02:27

the next few weeks and months . This is

1:02:29

not about generating money for us

1:02:31

. This is about generating an impact and if we

1:02:33

don't have an audience , then we can't get that

1:02:35

message across . So it's just a bit of a request

1:02:37

for me that people do that . Three people get

1:02:40

them to listen to an episode . Ideally

1:02:42

they'll subscribe and they'll follow the show . If they're not interested

1:02:45

, then that's fine . But the thing that

1:02:47

we would ask you is just to let people have a listen and hopefully get them

1:02:49

to be involved . Have a listen and hopefully get them to be involved

1:02:51

, so we'll finish , as always

1:02:53

, with a song . So

1:02:56

thank you , jimmy and Suno , for that

1:02:58

and we will hopefully

1:03:00

have you listening again next week . Thanks again

1:03:02

, dan , thanks Jimmy , and take care everyone

1:03:05

.

1:03:14

See you soon . People talk in shadows

1:03:16

, gossip

1:03:18

fills the air , stories

1:03:22

told in countless echoes , but

1:03:24

AI wouldn't

1:03:27

dare . It's

1:03:30

a task of subtlety . Trust

1:03:34

is hard to gain . Whisper

1:03:38

words and empathy AI

1:03:41

can play that game . Ai

1:03:46

won't replace us . Communications

1:03:50

need touch Tapped

1:03:53

into the human force

1:03:56

. Ai just ain't

1:03:58

got that much . Understanding

1:04:25

hearts and minds isn't data or code . In every word

1:04:28

, compassion finds a hand to lighten

1:04:30

the load , and white

1:04:33

lies as lost without a sign

1:04:35

While we see through

1:04:37

all disguise . Ai

1:04:41

won't replace us . Communications

1:04:45

need a touch Tapped

1:04:48

into the human pulse

1:04:51

. Ai just ain't

1:04:53

got that much . I'm

1:04:55

that much

1:04:57

, I'm

1:05:01

that much , I'm that

1:05:03

much , I'm

1:05:07

that much . Thank you , you

1:05:09

.