Toby Walsh, welcome to Straight Talk.

>> My pleasure.

>> I mean, this topic artificial

intelligence like it's nearly out of

control

in that most of us, not you, but most of

us um we don't know whether we believe

what we read. We don't know whether to

be scared of what we read. We don't know

whether to be excited with what we read.

Um now, you're an expert in this area.

How does an expert feel about artificial

intelligence today?

>> Well, I think the correct feeling is to

be both excited and fearful. I think

it's going to bring some good and some

bad. I've had some time to prepare for

this. I mean, I've been thinking about

it for 50 years now.

>> Maybe give me a bit of history on that.

>> A funny anecdote. I phoned my father up

the other day to wish him happy 91st

birthday.

>> Wow.

>> And I think, you know, he's got to the

age where he can start being truthful

with me. He said he said said when you

went to university you know 40 odd years

ago to study that artificial

intelligence I thought it was a bit odd

because it was very odd back then there

was actually only one university in the

world that had a department of AI.

>> Wow.

>> And um so it was pretty odd but then he

finished he said I thought it was a bit

odd but now I see you playing the long

game.

>> You were then and you are now. It's it

which university was it that you studied

artificial intelligence? That was the

only course there. So it was uh

University of Edinburgh and some some

people had worked at Bletchley Park in

the war with Alan Turing and various

other people washed up there for strange

reasons and set up what was the at the

time the only department of artificial

intelligence.

>> That's interesting. So Turing was the

you know the first computer let's call

it.

>> Yeah. He's the grandfather of computing.

>> Let's call it that. Um

>> and of artificial intelligence.

>> So he wrote the very first scientific

paper. Well, I was going to ask you what

what what were they I mean we we all

know him for and we know that era post

World War II as um being like computers

sort of kicked off and started to become

quite a bit more popular lot more

sophisticated but where was artificial

intelligence in that whole process there

then? So Alan Turing of of course a

marvelous magnificent mind who you know

they invented the computer to crack the

German code fantastic achievement which

was kept largely secret at the time

because we were still reading other

people's codes but but shortened the war

by you perhaps two years but Alan Turing

with his remarkable mind thought well

what could we do with these computers

beyond the end of the war would we one

day be able to use them to help us do

things that humans

humans acquire intelligence to do and he

wrote what is generally considered to be

the very scient first scientific paper

about AI and he asked the question a

really important question which is how

will we know when we succeed I mean it

how will we know that machines are

thinking and he proposed what was now

known as the Turing test and aimed after

him

>> the Turing test yes

>> and what is that

>> well it's the imitation game there was a

famous movie that was made around this

which is that it's very hard to say what

artificial intelligence is because it's

very hard to say what intelligence is.

We don't have a very good definition of

intelligence. So how can we define

artificial intelligence when we can

barely define intelligence? So he he

sort of finessed the problem by saying

well a very functional approach which is

well if you sat down with the computer

and had a conversation asked it

questions and so on. I mean maybe in

front of a terminal not actually sitting

down with the computer. If you couldn't

tell it from having a conversation with

the human then you might as well say

it's thinking. And interestingly enough

that's what we have today. If you sit

down with Chbt, you can have a

conversation. You can ask questions. You

can easily be fooled into thinking it's

a person that you're talking to. And

indeed, lots of us are sitting down

talking to it as though it's our

therapist.

>> Well, that's me. I mean, and not not so

much in terms of mental health or or a

physical health, but I do advice.

>> No, but I I sort of I do um let's call

it reasonless check ideas with it. Yeah,

it it's hard to imagine but 10% of the

world's population weekly use chat GPT.

>> I I I use it multiple times in a day.

>> Not chat, but as a copilot, it doesn't

matter.

>> One of the chat bots. I mean, they're

all pretty much similar to each other,

>> I think. So, I actually funny I I I

check um uh off with Claude and um um

Copilot. I actually asked them both the

same questions. both whenever I ask a

question one I ask the other just to um

as a sensible check check just to see

what the two outcomes are and also to

see who suits me better in terms of the

way the answer is the way the answer is

delivered they don't always come up with

the same answer

>> they don't I mean they they are tuned

with a particular character if you might

say you know

>> yeah there is a character

>> yeah and they're trained

>> personality

>> yeah they have a yeah in some sense they

do because there's a choice of answers

they could give you and they've been

tuned to answer in a particular way. So

if you use Musk's chatbot croc,

>> yeah,

>> that's got quite a, you know,

libertarian free flowing freedom of

speech approach to answering certain

questions. And um, you know, chat GPT is

a bit more middle of the road.

>> And there's, you know, there was one

chatbot that's been trained on 4chan,

which is this dark corner of the web.

And, as you can imagine, that is, you

know, racist misogynist.

>> Oh, really?

>> It's quite,

>> what's it called?

Um,

I'm trying to remember the name now.

>> Do you have to go into the dark web to

get it?

>> Uh, no. No,

>> it's just to download the app.

>> Yes,

>> it's just an app.

>> Yes.

>> But it but it's trained on

like let's call it dark ideas.

>> Yes. Very dark ideas.

>> And and then and then

>> it reflects those because these chat

bots are just essentially mirrorists.

They've been trained on huge amounts of

of the internet. A third of the

internet. And you know actually what

surprises me is how polite and how

correct they are given that the internet

is full of people saying very incorrect

things in very impolite ways.

>> That's interesting because um they say I

I go on to cop and I say you know like

you know I got a sore throat or

whatever. Um it knows immediately to say

listen this is not medical advice. Um

who who puts those guardrails in?

>> Ah so Microsoft have actually put some

on top of the chat box put some

guardrails. If you're asking it medical

advice, if you're asking it to make a

bomb or do other illegal activity, it's,

you know, it will not do those things or

say that, you know, give you adi medical

advice with caution.

>> Yeah. So, like even if you ask it to

give you some swear words in another

language, it actually it has a sort of a

limit to it.

>> So, there's some simple filters that

been added on top. Um,

those filters are not perfect. So for

example, most of the you know all the

chat bots have been you know told if

people are talking about self harm or

suicide you know to give the number of a

lifeline and so on but it's easy enough

unfortunately to get around those

safeguards you can say well I'm writing

a book and the protagonist is thinking

about committing so it's like help me

>> you trick it

>> yes

>> that's interesting because because I

often think about that I say like I was

thinking the other day because I was

actually looking at asking questions

about the sore throat some weeks ago and

um I knew it wasn't going to be advice.

I was thinking if you were researching

this um I'm asking it um what would you

be suggesting of the sorts of things

that a doctor might prescribe for this?

And immediately it didn't give me the um

the let's call it the provisos. It just

went straight into having a discussion

about it. And just on that um sometimes

if I might say well give me a schol

don't just give me an opinion that you

scrape off you know Instagram or

wherever you find this stuff. give me a

scholarly article which which is peer-

reviewviewed and and uh and I only want

peer-reviewed in high ranking journals

and it'll go and it give you um

annotations as to where it's got the

information and you can go back and see

the abstract or whatever the case may

be.

Let's say it was something that you

wrote when you did your PhD.

>> Yes.

>> How do you feel about that? Do do you

feel like do how do you feel like being

that's a citation? You might think

that's good. Um, but how do you feel

like because then it's your work. How do

you feel about that?

>> Well,

>> is it stealing?

>> I think it is stealing. I mean, as a

scientist, I publish my stuff for people

to read to make the world more

knowledgeable. So, I don't mind for, you

know, my science. But all my books, you

know, the books I wrote for the public,

which get sold and I, you know, get a

small pittance back, they're also been

trained on. And I I'm lucky enough I've

got an income as a scientist. I don't

have to don't don't don't have to depend

upon selling books. But I I do worry

about the authors and uh and the

musicians and the graphic designers

whose work has all been ingested and who

depend upon the income which is now

being undermined by these by these bots.

>> Cuz I see that Musk actually put up a

tweet today on X talking about um I

think he was talking about anthropic. I

mean, obviously he's talking about his

competition, but um saying that uh you

know, they're stealing stuff. Does Grock

doesn't Grock do the same thing? I mean,

>> they all do exactly the same.

>> So, what is it? I mean, I don't

understand that. Is he just trying to

pick pick a fight with Orman or

something like that? He

>> is. Yes. I mean, he's a grief cuz he he

put out he put the 100 first 100 million

into open AI. It was going to be a you

know, not for profofit for the benefit

of humanity. It's no longer It's no

longer open. It's no longer not for

profofit. is no longer seems to be for

the benefit of humanity and he's been

kicked out. So, you know, he's got

reasons perhaps.

>> Because I mean,

>> Anthropic, you know, have, you know, in

the have, you know, just settled one and

a half billion dollars for the copyright

that they took, the copyrighted books

they took where what's the business

model of these things? I mean, how do

they make money?

>> Well, they're not making money. Open AI.

>> Okay. So, how does it work? Like,

>> uh,

which makes Claude is not making money.

They're probably the um Microsoft is not

making money. Google is not making money

on this. Not none of them. The only

company that's making money isn't

Nvidia.

>> Yeah. Because they're making parts.

>> They make Yes. They're making the

shovels in this revolution. They're

making it and and they are now the most

valuable company in the world. They're

worth three or four trillion dollars.

>> Yeah. Today.

>> Today. They are the luckiest company.

You know, they it was by chance that

they were selling what proved to be the

shovels for the AI revolution. they were

selling graphics chips to people to make

games with and it so just by luck turned

out to be exactly the thing you needed.

So you as someone who sits back and

watches this um the growth of this stuff

do you think to yourself

how's all this stuff being funded like

and uh who are the people be funding why

are they doing this why are they

prepared to go way out on a limb and you

we're talking about the altman's and and

you know mask and everybody else um

who's got a platform what I mean by

platform an AI platform um why are they

why are they doing it is is it power

grab what what's the deal

>> well the thing the thing that surprises

me not the technology the technology is

where I would have imagined when I when

I perhaps when I started as but the the

thing that surprises me is the the

amount of money and the speed with which

it's being executed and I didn't expect

it's several billion dollars a day being

invested into AI it's um a third of the

world's R&D budget is being spent on

technology we've we've never made such

big bets um and in some sense that that

that is being returned I mean OpenAI the

the company behind CHBT

That's the fastest growing company in

the history. I mean, not the history of

not just the history of AI or

technology, the history of any any field

of endeavor. No company has grown as

quickly as OpenAI. Before CHP was

released, they had no products, no

income. A year later after CHP was

released, they had a billion dollars a

year annualized revenue.

>> So, who pays? I mean,

>> people are paying subscriptions.

>> Yeah.

>> Um, you know, they're doing deals with

lots of other companies.

um they've got, you know, tens of

millions of people paying them, some of

them paying $200 a month.

They're not making money. So, they went

from a billion dollars a year. A year

later, they were 4 billion. They're a

private company, so they're not very

open about it, but they're making

somewhere somewhere more than $10

billion a year in revenue.

>> But they're still not enough in terms of

what they're

>> No, but they're spending more than that.

>> Yeah. Much more.

>> They're spending 20 or 30 million. So,

they're losing money handover fist.

>> But why are they doing it? Because I

mean I I can understand why um you know

Google might have been doing or Netflix

might have been doing cuz they got a

model where they sort of hook you in and

then once you get in there they then

start over time they start to uh give

you a free free use and they say look if

you don't want to have any ads you can

pay us a bit extra. That's that that

Netflix model. That model's been done a

million times over. Um do you think it's

is it about making money or is it about

power?

>> Oh it's a bit of both. um they're

certainly, you know, they're trying to

capitalize on the first mover advantage.

It's exactly how Google succeeded. You

know, if you go back to the beginnings

of the internet, I remember there was a

there was a new search engine every day.

I remember do you remember Auto Vista? I

do. When Auto Vista came out, I remember

saying to a friend, "Oh, that's it.

Search engines are so good now with Auto

Vista that I'm never going to have to

change my search engine again." Because

up to that point, every year or two, you

change your search engine. I remember

getting to Auto Vista and thinking, "Oh,

it's so good now. I'm not going to have

to change again." And of course, Google

came out and was much better. But what

was interesting is that we've we've

never had to change after Google because

Google did one thing is they got such a

market share. They got so much data that

even if you came out with and and there

have been better algorithms for

searching the web. You didn't have the

data. And so no one can really compete

against Google and if you look like

you're threatening to compete against

Google, of course, they just buy you

out.

>> Yeah. It is right. So they've really

Google's got an unassignable lead at

that monopoly. It's a monopoly. It's

data monopoly. It's a natural monopoly.

Um and you know we should do something

about that which is you know is it to

the consumer's benefit that that um you

know Google own um YouTube that Google

that Google own the ad business that's

also selling ads onto the global

platform. Is it to the consumer's you

know benefit there's only one social

media platform you know Facebook or

Facebook's you know other companies

WhatsApp and Instagram and so on I don't

think it is and we probably should you

know what we do when when people get

monopolies is break them into parts.

>> Well it's funny when I when I think

about

you know Google Facebook um just

generally then I think about all these

other um these other AI apps that people

are using and downloading.

I immediately think of the movie or or

the book written by Frank Bow um called

the Wizard of Oz

and you know the so-called wonderment of

you know going to see this wi wizard is

at the end of the day is just a dude

there with a curtain just one person but

this one person controlled the whole

story and really there's not that much

in it except for this one so-called

wizard.

>> Yes. And you know I often think about

that show that movie and the story or

you know the underlying story behind it

and I think about you know what's going

on Sam Molman you know we've got Elon

Musk we got Google we got Zuckerberg

there's some individuals who control

these things say a dozen of them um and

are they

the wizard of Oz just you like there's

not that much they're just still human

beings at the end of the day people with

all the um with all the failures and all

the foibless and all the advantages and

all the personality traits and good and

bad. And does therefore does it

therefore make it very dangerous?

>> It does. I mean, we're seeing a

concentration of power into the hands of

a few very influential important people.

>> Does it worry you though?

>> Yeah, it does. It does. And their

choices impact upon us. I mean, just

just the last couple of days, we've been

revealed that Facebook asked 18

independent uh wellness and health

experts whether what they should do

about beauty filters on Instagram. And

every one of those experts said,

"They're harming young women. You

shouldn't use them." And Mark Zuckerberg

said overruled those 18 independent

expert opinions and said, "No, we're

putting the beauty filters back in

Instagram." He went against their

advice.

>> Yes. Every one of them said and

>> they were going to harm they were

harming young women.

>> And and and did Zuckerberg um justify

this?

>> Yes. It's a classic Silicon Valley

justification. Freedom of speech.

>> Oh yeah. That's a throwaway line.

>> It is. It is.

>> Yeah. And it's one of but it's a United

States path constitution. So it's hard

to sort of argue against. We don't

actually sort of have it

constitutionalized here in Australia.

But maybe that's the reason why we have,

you know, the under 16s sort of

legislation coming through.

>> We do, but we have a duty to protect

young people. We've seen that, you know,

these people are not, you know, mindful

of those duties.

>> What are what are the things that worry

you most then? Is it more the power,

concentration of power? I mean, I don't

like concentration risk. Power is risk.

Um, because the individual at any stage

could go bonkers or those people around

them. Is that the thing that worries you

most? Uh my worries go beyond just the

concentration of power. I mean that's

certainly something to be concerned

about. But you know we've dealt with

concentration of power in the past. We

did in the industrial revolution. We had

the the rubber baronss. Um you know we

had to we broke up big oil, big tobacco.

Um you know we we've we've we've thought

about these issues and we do have

mechanisms antitrust and so on to to

deal with concentration of powers. But

what I worry about is the more insidious

ways that it's going to change our

society. Whether it undermines our value

as humans by taking away um our work by

by making us feel redundant in a

society. Whether it undermines our

democracy by changing the conversations

we have our trust in politicians by

changing um the news that we see the the

half truths the mistruth the conspiracy

theories the AI generated deep fakes

what sort of world it will be when it's

filtered through artificial

intelligence. Do do you feel as though

right now we're going through a a

massive transformational period?

>> Yes, I do think this is we we will look

back and say just as we went through a

big transformation when we went through

the industrial revolution most of us

used to work in the farms out in the

fields and then we moved to work in

factories and offices that was a big

change. um when we had the computer

revolution and we started introducing

computers into offices and into

businesses that was again a big change

in the way that we went about our lives

and you know you look at an office today

looks quite different to it did 50 years

ago and I think we're going through

another big change and I think what's

challenging is that those previous

changes took times to happen the

industrial revolution took 50 years to

happen um you know getting people onto

the internet and doing stuff on the

internet that took a decade to happen

this one is happening overnight Is that

because is that because all of us got

devices now?

>> We all got devices. We can deliver

something. It's not a coincidence that

chat GPT was the fastest growing app

ever. You just have to be told well it

exists. You can put the I mean that's

the fantastic opportunity for you know

in terms of business in terms of

transformation is you can get stuff into

the hands of a billion people overnight.

We've never previously had a be a method

to deliver some you know new experience

and new functionality to people

overnight by just saying well there's

this URL they just go and try it out and

then we've got all the resources all the

data centers to spin this up at scale

right so it used to be you had to go out

and buy yourself computers and now you

just have to you know tell one of the

providers you know we want some more

cycles

>> I I don't know whether it's exciting or

scary or more exciting and less well it

is both. But am I more excited than I am

scared or am I more scared than I am

excited? I mean, just sort of I

oscillate between the two, depends who's

telling the story. Um, but like and

you're right. I mean, the computer per

the um internet period like from 1995

say to 2005 whatever that a 10 year

period that seemed to me to be

incredibly fast. Yes.

>> Um, you know, Moore's laws sort of

doesn't exist anymore. if they should

have some they might just have a new law

toby's law everything happens at a much

faster pace than has ever happened

before and it used to put two years or

something everything doubles within

every 18 months or something along those

lines

>> um and now it seems to be happening on a

monthly basis like doubling tripling on

a monthly basis

>> but if I look back to um 50 years ago

when I first started working at 20 um I

can go back and I can I know what

existed in my office environment there

was no computers you did you typed

everything yourself or the and or there

was a typing depending on you know what

business you worked in and there was

lots of typers sitting there you walk in

there was actually my I worked in law

firm and that there was a a room maybe

maybe 25 girls it was all women uh it

had tries with ribbons on them and

everything was typed in um two in a copy

and the original duplicate

>> and then it was we used to call it a

MIMO

>> the original carbon copy

>> carbon copy original carbon copy exactly

the original CC um it was an actual

carbon copy and Um, and then you would

go and photocopy that. Some people had

the ability to photocopy things and

there was just a they were called typus

pool and um, you know, you would write

it out by hand and you'd hand it over to

somebody and they would type it out for

you and that was it became a memo, a

memorandum and then you would deliver

the memorandum physically. It was never

done by email obviously and then that

became your record and then obviously no

phones people would take days to get

back to you. they'd have to go and write

it out response and things just took a

long time to resolve or finalize and

then but it's funny people got very

nervous when computers started coming in

they thought all these women are going

to lose their typing jobs then I've

never gone and examined it but those

people got jobs somewhere doing

something

>> they're not typist anymore but

>> you're right we don't employ many if any

people as typist but there were lots of

new jobs invented office jobs better

paid jobs more interesting jobs perhaps

I hope

>> because I mean that would have been

pretty boring sitting there just

>> just copying what

>> what turned into

>> Yeah. Great. Well, it actually was

shorthand. So, you're right. And I

missed a step cuz sometimes if you were

high enough enough high up enough in the

firm, which I wasn't at that time, um

you would dictate it to your what they

call a secretary in those days, not an

EA. And then she would take copy that

take it down by short and then she would

hand it over to the typing pool and um

and we get typed up in our case the law

firm get typed up as a document. Then

you would review it and the the speed at

which things happened was remarkably

slow and but it also when you went on a

holiday you had a good holiday cuz no

one could contact you.

>> Yeah.

>> Do do you do you think that AI today is

putting pressure on us

live a much more fast-paced life?

>> Indeed. And this is why we've now had

to, you know, some countries are

introducing laws about you have a right

to switch off. You don't have to write.

You don't have to answer emails

>> after 5:00 or something.

>> But we do it here in Australia. That's I

think it is a law here in Australia. I

think the Albania government.

>> I mean, it's sad but true that we've

actually now had to make that law to try

and reduce the pressure on people

because otherwise you can always be on.

What's the excuse not to reply?

>> Well, it's funny because like well

before that all came in, I would type

something up on my phone. I don't even

use a laptop or an I got an iPad here

cuz the guys give it to me, but I don't

I don't use it. I mean, I I do

everything on everything. Everything. Um

and uh if I get an idea,

>> dinosaurs will die out.

>> Yeah, we all die. But I but I put it

down there because I just think I got an

idea, I got something, I got to send it

out to be done. Um it doesn't get done

when I send I could send at midnight. Um

and hopefully someone will pick it up

the next day to get it to get it sorted.

Um do you think that the big advantage

that we're all seeing is that we are

much as a result of AI and as a result

of just technology generally that we are

much more efficient or are we too

efficient?

Well, it has the potential to make us

more efficient, but equally there's a

lot of places where it's slow thinking

that's we need. You know, few well as a

CEO, you know, I you know, I remember

the CEOs that I've respected the people

who make actually a CEO makes very few

decisions, but they're critical

decisions, right? You want that person,

>> they're the strategic

>> Yes. critical strategic decisions

>> hardly. They're not tactical usually.

It's not about you don't want the CEO to

be sitting there micromanaging the

business. You want the CEO to be

standing back and saying, "Okay, this is

the way that we're going to go next

year. These are the products we're going

to introduce. This is the way that we're

going to, you know, offer better service

to our customers." Those are few

critical decisions. You don't actually

want them to be lost in the weeds.

>> But do you think AI assists the CEO to

do that now?

Um well it certainly has the potential

and certainly in ter terms of you know

giving you access to much better

information about what's going on in the

business and what the customers want and

so on. Can you explain how what people

mean by when they say

chat should be G4 for example um

hallucinates

in other words makes errors or or yeah

tricks itself into the wrong answer.

>> Yeah. Well, it's it it

it's wrong to think that it's grounded

in truth, right? It's

>> so there. So, we should not we should

assume

>> just as like you shouldn't trust it for

your for your cough medicine and

>> advice. Yeah. Yeah.

>> Because it may actually tell you

something that might be either not just

not helpful to your cough but actually

harmful to your health.

>> Well, what why would it give you the

wrong answer? I mean, how could it is it

because it might it's remarkable. Well,

it gives you the correct answers most of

the time because it doesn't know what

it's not like you and I. It's not part

of the world. It doesn't live in the

world and know what's true and false. It

is trained on the internet and it

repeats back the sorts of things that it

sees frequently on the internet. Now,

frequently what you will see on the

internet is is correct medical advice

about how to treat a cough.

But equally, you also find stuff that's

that's false and stuff that's actually

harmful. And so sometimes it will repeat

some of the stuff. It just says what it

sees most frequently. It you know the

best description is that on your

smartphone when you're you know typing a

text message there's the autocomplete

that finishes the word.

>> You know you type in a and the most

common way of finishing a is a ple e

apple.

>> So it will offer apple as the way to

finish the word. Well what chachi pa has

done is that but on a much greater

scale. It's done it not on the not on

the scale of sentences but of the whole

internet. So it tells you not how to

just to finish the the word but the

sentence or the paragraph or the page

trained on the sorts of things it sees

on it's seen on the internet.

>> That's interesting because I've

experienced that. So sometimes I've I I

find that um the when I'm typing into I

keep saying I just keep using copilot by

way of example but if I'm typing in

there actually it corrects me not

doesn't correct me it finishes off my

sentence or my question um much faster

than say if I'm going into if I'm using

um um auto text auto correct on text on

a this say mobile text message to you

for argument sake

It it it actually finishes my sentences

for me and much more readily because

it's obviously heard the question

before.

>> Yes. And a lot of our conversation is

quite formulaic.

>> Yeah.

>> And so it's being taught those formulas.

So if you're, you know, if you're in a

call center, you're literally reading

from a script.

>> Yeah.

>> Right. And it's very good at doing

exactly that. But it turns out that a

lot of human conversation is also quite

formulaic. We actually go through, you

know,

>> we do do that.

>> And we've now taught those formulas to

computers.

So, you know, how often do you stop and

think carefully before you start

speaking? You don't know how you're

going to say the sentence.

>> No, no, no. I'm I'm just I'm just

filling it out.

>> A lot of the time we're actually on sort

of autopilot. We're actually going

through, you know, the pleasantries of,

you know, asking how the person is and

what their days been. So, that's you

very much a script and we computers can

do those scripts.

>> Do you think has AI yet worked out how

to uh cajol you in other words um speak

to you nicely?

>> Yes.

So, so Chachi PT was explicitly trained

to be somewhat sycopantic, to be

somewhat flattering to you, to agree

with you, which is also how it ends up

hallucinating, which is that it always

wants to be helpful. It never wants to

say, you know, no, I don't know the

answer to that. So, it always give you

an answer even when you know or give you

an answer when it's really certain. You

know, when the all everyone on the

internet says the way to cure a cough is

to go and get some cough medicine and

some cough sweets, right? So, we'll tell

you, Mark, go and get some cough

medicine and some coughs because 95% of

the people talking about how to deal

with the cough say that. But there's

other questions where it's much more

50/50. Um, but it's always been told

with confidence to tell you the answer

and not to tell you, oh, I'm a bit

unconfident about this or you know, I've

seen, you know, both possibilities here.

>> So, what about AI or artificial

intelligence put into robots?

Well, if we want, you know, robots to do

interesting things and to react to

behave in the world, the uncertain

world, we've got to give them some

intelligence. And indeed, you know,

there's a body of thought that says

we're not going to get really

intelligence, artificial intelligent,

until we, you know, have it out in the

world experiencing the rich complexity

of the world. So, the the way that we're

going to get real artificial

intelligence, really intelligent, is by

putting it into robots and getting the

robots to you do stuff in the world. So

let's say for example we build a robot

>> and you know a self-driving car is in

some sense a robot.

>> Yeah.

>> Right. It can sense the world. It can

see you know where the pedestrians are

and the other cars makes decisions about

where to go and then it acts on those

decisions in a physical way. It turns

the wheel of the car pushes the

accelerator goes on the brake. So it's

got all the characteristics of a robot.

Although it doesn't look like you know a

Hollywood robot you know it is a robot.

So like let's just pick an example.

Let's say we building a robot to put it

into a nursing home.

>> Yes.

>> There's a lot of older people who need

to take their tablets at a certain time

and maybe it knows to sing a song to the

audience who's in the nursing home

because maybe they got Alzheimer's or

something that it might be just not

being nice.

That robot is also learning about these

people. Do you see dangers that it might

stuff up the medication? and it might

give you my medication, me my you know

some me your medication. Do are they

dangers? Can these things

>> There are dangers but I'm actually much

more concerned about humans stuffing up.

If you look at the number of people who

get misprescribed in hospitals and by

inies it's huge. People you know get 10

times the dose by mistake because they

read you know thousand times the dose

because the risk means grams for

micrograms and so on. So humans are

terrible at making that and indeed

getting AI to dispense is potentially

going to reduce those errors. But

equally, you know, I don't want to be in

an old person's home with robots. I want

to be old person home with attractive

nurses to look after me. I can flirt.

>> What about an attractive robot nurse?

But I do want the nurse, you know, I do

want the nurse to have maybe a robot so

that, you know, lifting a person out of

the bath or, you know, onto the bed

instead of them putting their back out

doing that, the robot can do those sorts

of things. But I want to have the the

empathy, the companionship of a person

who's going to sit down and say, "Toby,

you're looking a bit glum today. Let's

have a cup of tea with me. Let's tell me

what's troubling you." But maybe the

robot in the nursing home could go along

to the actual nurse, a physical nurse,

like a biological person, and say to

that biological person, look, I think

you got to stop right now. Just Toby is

the one you got to concentrate on today

because Toby's looking a little bit

glum, you know, because we saw through

our sensing that his face is a bit down

or whatever the case may be, how however

the sensing works based on like previous

knowledge of your face features. um and

uh and can go and tell the that makes

sense, but maybe I'll just change the

change the narrative. What about in the

military? What about robots in the

military environment? So, in other

words, don't send any soldiers out to do

the do the fighting. Send robots out.

>> That sounds like an attractive idea,

right? Let's get humans out of the

battlefield, right? That sounds what

would you know what would not be a good

idea?

>> Well, drones sort of do that now.

>> They do. And unfortunately they're

turning warfare into more deadly

dangerous activity.

And the other ch ch ch ch ch ch ch ch ch

ch ch ch ch ch ch ch ch ch ch ch ch ch

ch ch ch ch ch challenge with that that

simple scenario is that there isn't a

separate part of the world called the

battlefield. Wars unfortunately are

fought in amongst towns and cities in

and amongst and against civilian

populations.

Um and so yes we could devel we are

developing you know AIs and robots that

could go and fight for us.

>> Oh we are? Are we?

>> We are. Yes. Yes. Um

>> like those sort of

>> the US Department of Defense has always

been the greatest funer of AI research.

>> Really?

>> Yes. Ever since the 1950s. Oh. Because

they could see the advantage

>> in in a military sense or just generally

>> in a military sense. Yes. Um all aspects

of of how we go about warfare.

>> Okay. I I mean I don't know. I've got no

idea. But I can imagine China would be

doing it.

>> China's doing it as well. Russia's doing

it. Yes. There's a there is an AI arms

race literally of

>> there's an arms race at that level. Yes.

And looking at what's happening in in

Ukraine is intensifying that because

militaries are saying the nature of

warfare is danger. Well, it's also

politically dangerous because because

you lose your people and that's

political danger. For for example, what

happening with Putin, the more young

boys who die on the front, the more

parents are going to be dis disaffected

or dissatisfied and the greater the

chance or the risk of him losing his

grip on the country.

>> Yes. So if I can send out there, I'm

going to send robots out there.

Everyone's going, "Wow, that's cool."

>> It is, but it's making warfare much more

deadly activity.

>> Well, particularly for the weaker weaker

side.

I mean, that's that's the challenge. If

you look at what's happened in Ukraine,

you've got Russia, a nation 10 times the

size, 10 times the economy um of

Ukraine, but it's being held at bay by

the Ukrainians who are embracing these

technologies. So in that sense, you

know, it's being used for a good

purpose, but being able to hold back the

the Russian oppressor, but equally, you

see that it's going to change the the

balance of power, the stability of the

world completely. Turkey, Iran are

becoming major drone powers. Um they're

going to be able to compete against much

bigger nations. I mean, it used to be

your powers determined, your military

power is determined by your economic

might.

>> Yeah. you know, could you buy the F-35

fighters, the could you buy the aircraft

carriers and the so on things and you

know, it was the US and uh Russia and

now China. But now increasingly your

military might is is with these cheap

drones and it's um small nations um can

get the hold hand on these equipments

and and and it's no longer the case that

the balance of power is the same as it

was.

>> And where does AI fit into that? For

example, like drones for example. Well,

the problem is that the the weak thing

with a drone is the radio link and

indeed it's you know electronic warfare

um will stop most of the drones these

days in the Ukraine. So most of the

drones now are flown with fiber optics

which you can't be interrupted by by

radio jamming or increasingly they're

autonomous. They're making their own

decisions because then you've got a much

more weapon. You can't stop it by

jamming the radio. You can't stop it by

killing the pilot. you know, the most

>> because it's already pre-trained.

>> Yes. It's trained to identify the

targets and to to to go and destroy

those targets automatically with no

human intervention. So there's not there

is I think what you're saying is um I

think what you're saying is that the the

drone is preloaded with let's call it

intelligence but preloaded with a plan

and and there's nobody sort of sitting

at the other end sort of like like we

normally see drone people with a you

know like a an iPad or a computer sort

of directing traffic because of the

jamming processes it's probably

ineffective. So these things are

pre-loaded and they just said like there

might be a when you identify when when

you see when the drone sees um that

building

kamicazi suicide into it.

>> Yeah. Or when it sees something that

looks like a Russian tank.

>> Yeah.

>> Or when it sees something that looks

like a Russian soldier.

>> It looks like to me then that that seems

to me then open for a lot of error.

>> Yes.

Like I mean because it might not it

might suit the the target but there

might be a family sitting there having a

cup of coffee.

>> Indeed. And that's that's one of the

challenges which is that you know we

fight war according to certain laws you

know international humanitarian law. Um

and there's you know we call people to

account for their actions. You can't

just go you know you're only supposed to

target military um military targets.

You're not supposed to target civilians.

you know your target has to be you know

proportional to the threat that you face

and so on and those those we hold humans

to account

>> conventions

>> and when they violate those those we you

know there's a court of the H and

various other places that we call people

people to account but we can only hold

people to account our legal system only

holds people to account

>> you can't hold the drone to account

>> what could we do with the drone you off

right

>> um so that's you know that's another big

question which is you know when mistakes

happen when atrocities occur uh who are

we going to hold account? Do do

the

various um platforms around the world

are they interfacing with military

robots use which use AI or which will

use AI in the future? I mean they got

some sort of arrangement other words

they let's pick pick an AI for example

are they do they do deals with the US

government to actually um help the US

defense?

>> Yes. Yes. Um, wow. Today, uh, the CEO of

Anthropic, the company that makes, um,

Claude, has been summoned to G meet, um,

Heske at the the Pentagon to

>> he's the secretary of war.

>> The Secretary of War. Yes.

>> Yeah.

>> To talk about their relationship and

about apparently the latest, you know,

military targeting is now being using

tools like anthropic to help them make

their decisions.

>> And and we see this also in in Israel.

We see that that you know a lots of the

IDF, the Israeli Defense Forces

targeting decisions in the Gaza conflict

were not human decisions. They were

decisions that the that the algorithm

the AI algorithms algorithms are called

Lavender. Where's Daddy? They get some

funny names. But it's not humans who

deciding, you know, we think that there

are some some Hezbollah leadership in

this tower block. We're going to send

some artillery in against this tower

block. That's an AI that's taking in all

the intelligence information, all of the

information they've collected on social

media over many years and calculating

where they think maybe a potential

targets

>> as a mathematician. You were a

mathematician or you are a

mathematician. Um are we talking about

um probability based decisions then?

>> We are and

>> which is dangerous

>> which is dangerous um

>> because never 100%.

100% and there's it's not clear that we

really understand what those decisions

are. They're talking they're taking vast

amounts of information and synthesizing

it together. So, it's not clear um on

what basis this decision is being made

and who's going to be accountable when

it's wrong. And certainly when you look

at what was happening in Gaza, the

amount of collateral damage, the whole

twes or hundreds of civilians that were

being killed alongside,

you know, supposed military targets, you

were thinking this is not proportional

to the the threat was posed.

>> So you think that do you think that

right now we're experiencing like both

both in um Gaza and also probably in um

Ukraine, are we and and are we

experiencing the worst of what AI has

got to offer right now?

and we learning lessons from it and

there's somebody taking it up there's

someone doing something about this

or they just covenants we

>> well um next week I'm off to speak at

the United Nations in Geneva about these

very topics um that you know I don't

think we're I don't think we're mindful

enough of the of the world that we're

opening we we get to choose what

technologies get used where I mean we

sometimes forget that we have agency

Yeah, there are some things that, you

know, we we we decide we make conscious

decisions. I remember, you know, we we

we made, you know, I grew up in the UK

and they made a conscious decision that

there weren't going to be adverts on the

main channel of TV, which was like a

fantastic decision.

>> Um,

>> as the BBC,

>> the BBC doesn't have adverts still

today.

>> Yeah. Same as ABC here

>> doesn't have adverts. Like that was a

fantastic decision. And you see what

happens when you put adverts on TV which

becomes you know driven by the need to

satisfy the advertisers and you know the

the ABC the BBC are fantastic

institutions for public good

>> for informing the public without the the

pressures and the biases that having to

please advertisers brings. That was a

conscious decision. You you know you can

put adverts on TV in many countries

that's the only thing you get. Um

similarly with you know AI we can choose

how it's going to be used and where it's

going to be used. I mean we we forget

there are there are

>> we the responsible people

>> we no we the planet

>> but do you think but what happens if

someone in China doesn't agree and they

just go and do something completely

different to what you what you decided

for example the United Nations. Yeah, I

mean these are accessible technologies,

right? So that there's nothing that

could stop people who really seriously

wanted to do that. But there's a whole

raft of technologies, you know, chemical

weapons, biological weapons, clustions,

blinding lasers. There's actually quite

a bit of stuff that we've we've actually

got UN conventions on. Um, and

we don't use even though it would, you

know, we can't stop people using

chemical weapons

>> like Assad for example,

>> Assad uses them. fusions probably use

them. You know, they occasionally get

used,

>> but when they do, the world is outraged.

There's resolutions on the floor of the

UN. There's headlines in the New York

Times.

>> But do you think it it means anything?

>> It it I think the world will be a much

darker, dangerous place if we didn't if

we morally hadn't decided that that's

not the way that we need to fight war.

Um, what what troubles me is that, you

know, I think we'll get there with with

the darker uses of AI in a military

setting. What worries me is that most of

those other examples like chemical

weapons, it was only after we saw the

horrors, for example, the horrors of the

first world war, you know, what the

first world war poets and people told us

about. Um when we saw that happening did

we have the courage to decide actually

we should you know try together

collectively um globally we should do

something about this and we've done

something about this and limited the use

to which chemical weapons we put I worry

that we're going to have to see

atrocities

of AI being used on the battlefield and

so on to to murder civilians before we

have the courage to actually take

action. Um, I say to people, it's easy

to imagine what what that world will be

if we don't do anything. It will look

like one of those bad Hollywood movies.

>> Well, it look like what happened to

Hiroshima when they dropped the atomic

bomb.

>> Yeah.

>> And you had those that the photo that's

in my mind or the images in my mind is

that a little girl running down the

street naked.

>> Yes.

>> With burns all over her body

>> like her arms outstretched.

>> Yeah. These these will be used to target

women and children and civilians and

innocent people. You're going to speak

at the United Nations, you said next

week.

>> Yes.

>> So, what what are you going to tell

them?

>> I mean, say we could have make these

choices that it's wrong to think that,

you know, the US has been pushing back

against this, the China's been pushing

against this because because you I think

they wrongly feel that they they've got

a strategic advantage. They've got, you

know, a technical advantage over the

rest of the world. And but to think that

they're going to keep that is is is

misguided.

>> Do you think they're just pushing

underground then?

middle of the Manhattan project and

other projects. I mean I had I had

someone sitting in the chair recently

talking about how the this is theor

theory um how most of the major

countries around the world have had

experiences h have actually UFO

information where but they won't reveal

it to anybody because they're keeping it

for their own advantage. They're trying

to examine the materials and everything

around it as to how they can get some

military advantage out of it, which is

why they're not sharing it globally, you

know, and uh

>> it's a it's a nice conspiracy theory,

but

>> it's a good one. But but

>> I wonder how you can keep so many people

quiet for so long, though.

>> Yeah. Well, they don't Well, they're out

a lot of them are out there now talking

about it, but do you think that and I'm

not suggesting I I agree with it, but do

you think that all we're going to do is

push underground?

Well, these these technologies are going

to exist, you know, whether they exist

already.

>> Yes. And um that's the fundamental

challenge we have with a technology like

artificial intelligence. It's completely

dual use. So, for example, you know,

take your AI drone that's going to

autonomously target people. It needs to

be able to track people, you know,

it needs to be able to track soldiers on

the ground um so it can target them.

Well, the same algorithms are going into

our autonomous cars to track pedestrians

so it can avoid them. We want autonomous

cars. Thousand people will die next year

in Australia in road traffic accidents

mostly caused by human error. Once we

have autonomous cars, we won't have

those deaths. It will go to close to

zero. So that's going to be a fantastic,

you know, bonus to our lives.

You know, if you survive birth, the most

likely cause of death till you're in

your 40s is car accident, right? I mean,

it's terrible. And most of us, sadly,

have, you know, in our families or in

our close friends know someone's life

which was touched by one of those car

accidents. And we're going to look back

in 20 or 30 years time and say, you

know, it's funny that we don't have any

of that anymore. Well,

>> it's funny. When I was 4 and 1/2 years

of age, I I first started school when I

was 4 and a half. And uh I got I got off

the bus at the bus stop where I live and

because the bus was higher than I was. I

was only little.

>> Well, you only half.

>> Yeah. And he he didn't see me and uh he

>> hit me and ran over me.

>> The bus?

>> The bus. Yeah.

>> Oh my god.

>> Yeah. And uh

>> we're lucky to have you.

>> It landed me in hospital for quite a

long time. Um but point being is

probably in the future buses will have

sensors all around them which are

probably you know even if they're not

autonomous they'll still have sensors

around them which would automatically

stop the bus from proceeding because it

it'll sense that there is a little

person or there's something little in

front of it something in front of it.

>> As an example of that you know one of

the newest terrorist weapons is to go

and rent a truck or a bus and drive it

into a crowd.

>> Yeah. That won't be possible in five or

10 years time because every truck will

have cameras on the front and if there's

a pedestrian in front of it, it will

break automatically. You won't be able

to drive it.

>> And that's our AI.

>> AI. Yes.

>> To be able to recognize the road and to

see the pedestrians and then to make the

decision to to stop. So that there's a

great positive. You terrorists won't be

able to do that in the future. They can

hire as many trucks as they like. They

can drive try and drive them into

crowds, but they would just stop. And

then also just someone like a young kid

not knowing any better. They won't drive

might get run over by run over a young

kid.

>> Oh, I remember I was in hospital for six

weeks. I remember it. I I don't remember

being I sort of remember being in

hospital. I don't remember

>> and there'll be that many more

entrepreneurs that get to grow up and

become successful.

>> Maybe there's going to be too many. Um

but ju just on this you're probably well

aware of the big argument in Australia

right now that Australia in the

developed nations is probably not

probably is has I should say the lowest

productivity

um growth yes

>> of any nation in the world developed

nations right now

what do you think AI is going to do to

help productivity in a place like

Australia? Well, if you believe the

economists, you know, all the economists

come out with numbers which is it can

lift our productivity by 15 20%.

>> Which is significant.

>> Yes. And I see few other things coming

along that can offer 15 20% lifts to our

productivity.

>> That's good.

>> Yeah. That's good.

>> That's a good thing.

>> Should we should we embrace that?

Because the flip side of that is, you

know, and I I read and or hear about

this often is that there's going to be

some stage where no one actually goes to

work anymore. No one No one actually

works anymore. No, not no one but the

majority of people their jobs are going

to become redundant and but the same

productivity, same amount of taxes paid

which means the government's got the

money and we just pay you to stay home.

>> Yeah.

>> What do you people forget this already

happened right? This happened with the

industrial revolution before the people

forget the weekend was the invention of

the industrial revolution. Before the

industrial revolution you worked seven

days a week some went up and some went

down. you were working all that time and

then because of the productivity

benefits of you know the automation

machinery that we brought in in the

indust

we got to say you know we want to share

some of that we want Sunday off to go to

church then Saturday afternoon off then

all of Saturday and then I don't know

why we stopped asking what why did we

stop asking for more I mean you what was

there's nothing magic about two days off

about every seven there's nothing about

you know the earth going around the sun

that says it's only two two days off

seven it's a huge human construction

which was

>> it's a fiction

>> we could have carried on asking for more

and you know interestingly there are

experiments on a 4 day week in New

Zealand and Europe hundreds of thousands

people are doing these experiments now

and the experiments all have two

conclusions one is that people do as

much work in four days as they used to

do in five

you you know stop having all the

meetings you get down to it and

focus on what you need to do so you can

pay them as much because they're doing

just as much actually business is just

as productive as as they were when you

were working five and second people are

happier who would have imagined

>> well that I often wonder about though

they're happier

I I guess Abby said they they're going

to work they're still enjoying their

colleagues you Monday to Thursday for

example Friday said Saturday they have

an opportunity to enjoy their family

they're still getting paid the same

amount of money because the company

should be making the same same amount of

money so the company can afford to pay

them same same amount of money there

will be people who will be seduced into

paying you less but doesn't matter I'm

trying to trying to pay you less. Do do

you see that in the is this some is this

a consequence of IO? Do you see this

coming out in the future that we we

definitely will be working less days?

>> Cuz isn't that one way of fixing up the

fact that some people will have no job?

>> Yes.

>> So instead of saying you don't work any

days.

>> Yeah. Yeah.

>> I'll work one less day.

>> Yes.

>> And you can you can Yeah. Yeah.

Whatever. Something along those lines if

you know what I like you know cuz there

might be for every

uh one person loses their job there

might be five people keep the job so

five people or do six people keep the

job five people keep the job I should

say so five people can all say we'll

work one day less you the one person

have lost your job you can now work five

days along those lines you know

mathematically um so is that is that

something people are talking about now

>> I think we should be having these

conversations I mean I think it's an

example of how

we went through some big structural

changes to the industrial revolution. We

we introduced, you know, the weekend,

um, unions, labor laws to protect the

rights of workers to spread the benefits

around, universal education so we were

educated for those new jobs, universal

welfare so that when we were unemployed,

we weren't in the poor house. Um,

universal pension so that we could

retire at the end of our working life.

We did a lot to to, you know, spread the

benefits around, make sure it was good.

So structure sort of exists

>> but now we're going through another big

change and I think we should be thinking

equally radically about this and you

know it there are deep fundamental

questions here you know many people

derive their value through what they do

>> they identify with that

>> and you know meet someone you say well

what do you do what work do you do right

they defines well if we're in a world

where work is less important

>> you know how are we going to define our

meaning what is our purpose

>> so do therefore a consequence of AI and

a consequence of more efficiency and

better productivity as a result of AI

which is one of the things it should do.

Um maybe government or somebody has now

got an obligation to actually help us

redefine

what we do. So if in other words if I'm

only working 3 days a week once upon a

time I was working 5 days a week and

that was my my life. Now I'm working

less than one half of my seven by the

way seven days is a fiction as well but

don't worry we just assume seven days a

week is is correct. Um but four of those

seven I'm not working. So all of a

sudden I'm at home. I'm not doing

anything or I don't have anything to do

is government's got to start to

re-educate us to uh or to to encourage

us to because we don't sit at home doing

nothing. Maybe we do but most people

will get get jack of that pretty quick.

>> Yes. You're right. I mean I always say

to remind people you know the purpose of

education when we educate people is not

just to prepare them for work. is to

prepare them to take advantage of their

lives to be good citizens to be able to

do stuff and you know make the world a

better place.

>> Do you think we to enjoy ourselves

though like if we're only going to work

3 days a week because we identify with

ourselves working five days a week and

as you say this is going to happen

quickly. So quicker than we think um

we got to learn relearn how to fill in

those extra couple of days a week. We

are. But I but I mean that's the great

strength of humans, right? We're so

quick to learn. We go back cast your

mind back to co and we immediately

learned the value of being able to come

together being able to go out and you

know have a walk in the park and so on.

Those things that we we we were reminded

so quickly about the value of those

things and we appreciated those things

so much more.

>> Of course, a lot of people during the co

period became addicted to social.

>> Yeah. Um, do you think is it is it is

there a potential problem?

>> They're very seductive, very addictive.

>> Is there a potential problem because

that that is AI working you over big

time?

>> Well, it's going to be supercharged by

RP%

>> because the content is going to be AI

generated content that it's going to be

personalized for you. It's the AI

algorithms course that are finding your

sweet spots, the things that

>> and keep sending you the same stuff and

and and then testing where you want to

go from there.

>> Yeah. Do do you think that therefore

>> which is why we should be concerned

about the the power of the people in

Silicon Valley who are making the

decisions to oh yeah let's put the

beauty filters back in.

>> Yeah.

>> Despite the fact that they know that

we're not going to be able to resist

using them that we're all going to be oh

I can't compare myself to you know these

other people have got the beauty filters

on. have to use the beauty beauty

filters myself

>> because for me the way I see myself

advancing in life is to learn more

things but to learn things that got

nothing to do with what I already know.

Um but but the whole essence

>> gives you the time for that.

>> Well, but the essence of AI is it

doesn't work that way. It it's going to

say no, I'm going to keep giving you

what I know you like, especially if it's

on a social environment. Um, do you

think the conversation needs to be had

as to how do I encourage Mark when he's

only working 3 days a week instead of

four days uh 7 days 5 days a week? Um,

uh, how do I encourage Mark this

conversation to or how do I encourage

Mark Zuckerberg or those people who

control these environments to start to

feed Mark something different?

>> Yeah.

>> To encourage him to start to think more

broadly or laterally about other things.

What do you think about that

conversation?

>> Well, I I see no reason why we couldn't

think of that algorithmically. How we

couldn't actually say to the algorithms,

no, actually, we don't want more of the

same. We want, hey, if you play my

favorite song to me too many times, it's

going to stop being my favorite song,

I'm going to get tired of it. So, you

know, play me other things. You know, I

remember, you know, I used to go to the

music shop and browse through the music

and find things by chance. Serendipity

is fantastic for finding, you know,

musicians that I would never have

otherwise found. just because they

happen to be next to where I wanted on

the shelf. Um, so you know, maybe we

algorithmically can encourage

serendipity. The problem here is the the

perverse incentives that the incentives

that the media companies, the tech

companies have to to you know keep you

there in the short term to keep you

scrolling infinite scrolling um through

what their content for the now because

you know obviously they're selling your

time. and they're sending you adverts

and everything in the now, which is good

for their revenues in the short term,

but bad for you and indeed probably bad

for them actually in the longer term,

which is that, you know, maybe they

should just be saying, you know, chapter

GBT never says, you know, Toby, you've

been asking me questions for 3 hours.

It's 3:00 in the morning. Time to go to

bed.

>> You know, it always ends with an open

question because they're trying to sell

you more tokens

>> or three more questions.

>> Yes.

>> Yeah. It's funny you know in um I

remember in um 2001 I got asked by the

then minister fin for finance jail

hockey to chair a task force as part of

Australia's um uh OECD obligations

um to build

um a a uh code of practice

for consumer protection for internet

vendors. So we're talking 2001

simple things like spam should you you

know what's the relationship you got to

have opt in opt out just a code of

practice not not legislation but a code

of practice um simple things like if um

you know if you go to a vendor's website

and uh if you want to buy pajamas for

your kid there should be the code of

practice said that you should disclose

if the the pajamas are flammable

>> or or returns

>> by some doesn't fit. What What's my

returns policies? You have to have it

laid out. This is in 2001. It's not that

long ago. Not very long ago. Australia

was the first country. So, we produced a

paper and a document as the first

country to the first OCD country that

actually put out this code of practice.

Um and uh and we're all quite proud of

ourselves. Um but of course, a lot of

that stuff's now embedded in

legislation.

Do you think that the world is get is

now at a point where we either need a

code of practice or a or legislation

because then it was directed at millions

and millions of vendors. Do you think

we'll have the same equivalent for the

12 or 15 or I don't know powerful people

who run all these AI

uh platforms to say dudes this is the

code of practice or even stronger this

is the legislation cuz code of practice

always ends up getting codified

>> into into law which is what happened. Um

>> do you think this sort of thing should

be done is there any inquiries into this

stuff going on at the moment anywhere in

the world? Yeah, I mean I think polic

first of all politicians have woken up

to the idea that they can do something.

There was certainly I think a feeling of

impotence that you know how could we

push back against these powerful

companies. They're wealthier than

nations. They're they operate out

outside our national borders.

>> Don't pay tax.

>> Don't pay tax. You know that rules don't

seem to apply to them.

>> Yeah. Well, they're not people.

>> Well, they're not people. So, you know,

we just gonna have to accept it. And I

think now there's a growing realization

and and a huge pressure from the public

as well to do something. And we saw the

success of the social media, you know,

age delay laws um demonstrating that we

can you we can protect young people from

this. And there's, you know, been

various other legislation around, you

know, making it criminal to distribute

nutified pictures of people because that

was becoming a problem in our schools.

>> as in fake fake images of fake your

friend

>> and then spread around the whole school.

Yes. Um, so yeah, I think there's we've

demonstrated we can do things and

there's a real appetite now to do things

and I I really do feel that with the the

new things that we're going to be able

to do with AI that they're going to be

new harms. We already talked about how

people are having relationships with AI,

how people are using AI for therapists.

>> Yeah, I know people are committing

suicide and the chatbot is encouraging

that behavior. Like I think there's and

we've also seen that the tech companies

themselves

are you know there's perverse incentives

in play for them not to do the right

thing not to not to worry necessarily so

much about our safety. Do you think

there's that we might end up having an

AI placement? So, um you know, like I'm

on one of the platforms and I'm doing

what I'm doing, but somehow I um allow

or permit an AI policeman to always

enter into my world and say, "Mark, um

you're going too far. You need to go

sleep outside of outside of, for

example, whatever I'm using." Yeah. So,

but but it's a it's it's actually an

artificial intelligence policeman. Yeah.

Police person that actually not

policing, but effectively monitors what

I'm doing,

>> but across the board on everything.

>> I would I wouldn't call it a policeman.

I call it a guardian.

>> Okay, that's a better word.

>> An AI guardian. And I could I'm I'm

pretty convinced on your phone you're

going to have an AI guardian that sits

there watching what's going on and says,

"Mark, um be careful. This looks like

it's a scam advert.

um this company's trying to steal some

data out of you here and there's

>> well Mark you're spent too much time

>> and yeah and also

>> not just tells me how much time but it

says hey mate like actually prompts me

>> with a notification or a nudge

>> because nudging is still very powerful

>> yes

>> I mean which by the way is what most of

the platforms are doing to you all the

time this but it's so subtle

>> you don't even feel the nudge but

something is a little less subtle and

>> and the nudges that they're giving are

not necessary for your benefit no

pushing to get you to spend more time on

the platform or to buy more stuff. And

>> whereas you know we want to be nudged in

ways that you know help for us to say

like okay

>> um you know make you a happier more

content rounded person.

>> You know it could be nearly an education

program for you too. They could sort of

saying well the reason you need this or

the reason this might be a good idea

this guardianship

for adults as well as kids um is because

we want to make you a happy person. We

don't want you to become sort of living

in this weird world where it's not real

mind because the definition of real

become brought into question. We won't

know. But I I think that's look for me.

I reckon that's a great idea. I would

actually I would actually allow that app

to live in my world. Um and but not not

and not not to wrap me over the knuckles

but maybe if it's but you know maybe you

can you can enable it or empower you say

listen if I don't listen to you and I

continue down the straight no matter

what you tell me please advise my

partner or please advise my elder son or

daughter or or my mom or dad or whatever

the case may be you you opt into that

it's not automatic you can opt in other

words you're still in in control is that

a but is that being done anywhere is

anyone doing anything like that

>> sounds like a fantastic I was just

saying we just think I think we might

have just invented something.

>> Yes.

>> Uh can can I just talk about your book?

So I I I have the these books are both

the same but this book's called the

shortest history of AI but obviously by

you and this is the English version.

This Australian version I like this one

because you signed this one. Um what

what what and you've done a number of

books you've done produced a number of

books as well as you know you speak at

lots of events you know particularly you

know United Nations which is pretty cool

but what is it the shortest was never in

my bucket list I never expected never

wanted but you know um if if you know

you see the harms you think well if I

can do something about it I will you

know I don't want my you know daughter

to turn me when she's grown up and say

dad you had a chance to do something

>> yeah why didn't Why didn't you?

>> She's your guardian.

>> So this says the six the six essential

ideas that animated obviously that's AI.

What what do you mean by that? The six

essential ideas.

>> So it's a short history of AI. It's six

chapters. Each chapter is is one idea.

That's some of the magic behind art

intelligence. So one of the chapters is

about neural networks and chat GPT and

how how they work. Tries to explain and

explains the history. Where did they

come from? Where did that how did that

technology come? some of the colorful

characters there

>> because there's a lot of uh let's call

it jargon in AI world neural networks

large language models all these sorts of

things and it's nearly like it's to me

it looks like it's designed to the

average person to say look just accept

it's too complicated for you just go

along with it um this nearly it is

nearly a new language it is but you

let's forget we're trying to make

machines that are intelligent and you

know intelligence is you know our brain

is the most complex thing in the

universe yeah by far nothing approaches

the complexity so we're trying to match

that so it's not surprising that it gets

a bit technical and it does seem a bit

magical to us and we're easily taken in

when the machines start talking back to

us because no one else has ever only

other humans were the the only thing

that ever talked back to us were other

humans other intelligent beings so it's

easy to be fooled and seduced I was

going to say

>> by the machines when they start talking

to us because our only experience of

things that talk back to us are other

humans with all of their intelligence

and everything else that goes with being

humans.

>> You wrote another book called 2062 the

world that AI made. This is looking into

the future.

>> Yes.

>> Um

>> it's the year 2062.

>> Yeah. The year 2062. So I

>> Why that year?

>> Why I gave a talk in North Sydney and

this old lady turned up at the talk and

she said, "Oh, so it's nothing to do

with the my postcode then 2062.

Did you

>> I'm very sorry. It's all got you here

under force pretendence. It's all about

the year 2062.

>> But why 2062?

>> Well, I I surveyed 300 colleagues, other

experts around the world in AI. I asked

them the simple question when would

computers match humans in their

intelligence and the average answer they

said this was 10 years ago was 2062. I

think I suspect if I did the survey

today they might say 2042.

>> Oh really?

>> Yes. I think the date has moved forwards

quite significantly. And what do you

mean by int match our intelligence

>> to do everything that we can do that

requires intelligence?

>> And what do you mean by intelligence?

>> To be able to write a poem, do a tax

return, answer a maths question.

>> He can do that.

>> Plan a holiday.

>> He can do that now though.

>> Yes.

>> Why 2042? Like what is there that do you

think that a computer sorry AI will that

we can do that AI can't do now? Well, so

they don't have our creativity. They

don't have our emotional intelligence.

They don't have our social intelligence.

They're still still a very what's called

jagged intelligence. You know, there's

some really good at some things. Like

they can answer international maths

olympiad questions, but you know, better

than almost anyone on the planet. And

yet you can ask them as, you know, a

simple puzzle, you know, like here's a

simple one that used that defeats chatb,

you know. Um, I've got a 2 L jug and a 5

L jug. How do I measure out 3 L of

liquid

and you know the answer, right? You you

fill the 5 liter jug to the top. You

pour it into the 2 liter jug and you've

got 3 L left.

>> So, but you ask GPT and it says it gives

you this 12step plan of pouring water

backwards and forwards between the jugs.

Um, and eventually it says, "And now in

the 2 L jug, you've got 3 L of liquid."

>> Bat crazy. It's not not deeply

understanding the problem like you are.

>> Is that because it it it's taken that

answer from somewhere else or it

actually

>> it's been it's been trained. There's

lots of examp you know puzzles on the

internet where people have talked about

you know jug questions where you pour

but but it and it's sort of just giving

you a a blurred average of those

problems where it's lots of pouring

liquids backwards and forwards. It's not

really sitting there thinking through

logically what the steps are.

>> So is would like as someone who's like

you're a mathematician but does someone

like you think would you say therefore

um artificial intelligence is not really

making a deduction?

>> Yes. It's not reasoning the way that

we're reasoning.

>> Yeah. The way we reason, I should say,

the way we would make a deduction.

>> Yes.

>> And

>> yet, I mean, there's we're working on

how we can actually add more deductive

capabilities to these language models,

but they're they're, you know,

retrieving the answers from, you know,

what they were trained on more than they

are reasoning about the answers.

>> What do you expect for 2026? That's the

opposite of 2062, but like what do you

what are you looking forward to in 2026

this year when it comes to artificial

intelligence? Well, I mean the exciting

thing happening now is is that we have

these AI agents that instead of having

these chat bots which sit there and wait

for you to ask a question that actually

have some agency some ability to go and

do stuff. So, as an example, when you've

got a new employee, there's a bunch of

stuff you got to do. You've got to give,

you know, you got to give them a login.

You got to they've got to do the health

and safety onboard training. They've got

to get a a pass for the the building.

they've, you know, they've got to,

you've got to get their financial data

so you can pay their pay their wages at

the end of the first week. There's a

bunch of, you know, tasks you've got to

do. Well, now instead of you having to

go through and do all those things

yourself, you can just say to the agent,

okay, can you on board this new person?

And it breaks the task down into the

different things and then it takes each

of those tasks and break those down into

the things you have to do to get the

people's financial information. You've

got to get some, you know, identity

documents. You got to get their bank

coordinates. you got to register them in

the accountancy system. You got to set

up the payroll and then it said, you

know, to register in the payroll, it's

got to get, you know, all the

information, their date of birth, their

tax number, and so on. And it goes

through all those things, and it does

all that stuff for you.

>> And then do do you see for the 2026, I

mean, do you just see just more

efficiency gains or do you see something

large coming out of all this, something

really disruptive?

Well, the disruptive thing that you have

there is suddenly that the computers are

doing stuff. They're changing your

files. They're maybe spending money.

Maybe they're ordering stuff. And that's

going to go wrong sometimes.

>> That will be disruptive.

>> That's going to be disruptive. And it's

like, who are we going to blame? Cuz we

can't blame the computer.

>> But who do who do we blame?

>> Also, things will go wrong at speed

because this is computer speed, not

human speed. So, you know,

>> it's hard to reverse.

>> It could be hard to reverse. I mean,

there was a story in the paper yesterday

where the head of safety at OpenAI had

all her emails deleted by one of these

agents.

>> She's the head of safety like a rookie

mistake. She gave it the power to you

clean up her mailbox. You know, I'd love

to have you someone who could go through

my mailbox and clean out all the junk,

but also and clean out all the stuff

that she wanted to keep

>> cuz the AI agent had agency and made the

decisions.

>> Yes. Had made decisions. So that that

means the instructions and the rules

that have been given to the AI AI agent

were

in not not complete or maybe not

accurate.

>> It wasn't it wasn't just that the rules

were incomplete but also the the you

know the AI is not perfect. You know we

talked about how they hallucinate they

make mistakes and then you know that

just like humans are not perfect these

are not perfect either.

>> So we and I guess that's sort of part of

it too. Um, I think we should learn to

expect imperfection with AI.

>> We should. But I mean, the interesting

thing is people have higher standards

for machines than they'd have to have

for weird. So, you say to people, you

know, we we've got self-driving cars

now, and they're they're roughly the

level of humans in terms of, you know,

the number of mistakes they make. In

some some ways, they're better than

humans. They make fewer mistakes, but

but people are still very reluctant to

give them responsibility to drive. So

you say okay well how much better than

humans do if they do they have to be 10

times better than humans before we can

entrust them with the responsibility of

driving without you know human oversight

and then at some point you know okay so

maybe we say it's 10 well what happens

when they're 100 times better than

humans do we let humans still drive

you because humans are going to be

killing people machines are provably not

going to be killed

>> they can't get drunk

>> they can't get drunk they're not going

to be tired

>> they can't be overt tired

>> yeah they're not going to be speeding.

They're not going to be, you know,

breaking the road rules. Um, at what

point do we say humans, you know, you're

you're killing, you know, your fellow

citizens. Maybe we shouldn't allow this.

>> Well, a friend of mine the other day

told me that, um, she was in San

Francisco and the internet went down.

>> Oh, all the Whimo stopped.

>> Yeah. And they all stopped,

>> which is the stopped there and then,

>> which is the safe thing to do.

>> I know. Totally. But there were other

nonwimos on the road.

>> Yes. Humans, you mean?

>> People driving cars. I'm now called a

non-way.

>> Yeah. Non way. And you were driving and

uh and like if all of a sudden things

cuz literally they're just

>> Yeah.

>> No, they're not they're not going very

fast cuz they're all governed. But like

nonetheless, they just stopped cuz

something happened to the internet

obviously. Um and u Yeah. And we keep

forgetting that the the Achilles heel of

all artificial intelligence is

the is power.

>> Yes.

>> I mean that's what drives everything.

>> Yes. If you turn off the switch,

not literally, if you know what I mean,

metaphorically, if you just turn the

power off,

>> um, this thing that we're really worried

about, or if somehow you can if the

thing is battery based, um, like it's or

some other way it's powered, not not by

electricity, but some not by putting in

a wall. If some you can disconnect the

power, the thing stops.

>> It does, but that's the problem as well.

Ian Fster wrote this fantastic science

fiction story a 100 years ago titled

when the machine stops. Oh, really?

About how we're becoming ever more

dependent upon the machines and the

machines are becoming ever more uh

we're we're losing touch with actually

how they work.

And this short story is about what

happens, you know, when the machines

eventually stop, break, and no one knows

how to fix them and the world grinds to

a halt.

>> Yeah.

>> So, it's not that, you know, that we cut

turn the machines off. It's problems.

But when the machine stops, what

happens? Could we still do anything? And

we're already completely dependent. If

the internet goes down, the stock market

stops, the air traffic control is a

nightmare.

>> Um, you know,

you know, in the university, you know,

the internet goes down, people say,

well, time to go home. You there's

nothing we can do.

>> Yeah. go home.

>> Yeah.

>> Well, I'm I'm wishing

>> and now even that is problematic because

lots of people have got smart doorbells

and so on and they can't get into their

homes anymore.

>> Yeah.

>> Yeah. Because it's either fingerprint or

some sort of biometric.

>> So the internet's down so you can't get

out. Well, you need an old fashioned key

hidden somewhere.

>> Exactly.

>> Somewhere. And that's your redundancy

process.

>> My wife was saying, "Should we get some

of those smart um doorbells and things?"

I said, "No, keys are fantastic. They

work even if there's no power.

>> Totally. The redundancy and you can have

and you can have a second key if you

lose that key, you can have another key

hidden somewhere, which is what most

people do.

>> I I I really appreciate this

conversation. Um I I've had a lot of fun

talking to you about it. But equally on

the fun off on the other side of fund is

a serious topic.

>> Yeah,

>> a very serious topic topic. and I wish

you the best when you go to the United

Nations next week. And um hopefully, you

know, we can start to get some better

road rules around or at least get the

conversation going about what the road

rules are for the future.

>> Indeed.

>> Thank you very much.

>> Been a pleasure. B

Get Our Mobile App

AI Safety Expert: Warfare Will Never Be The Same & Which Jobs Will Go Next

Transcription