Episode Transcript
Transcripts are displayed as originally observed. Some content, including advertisements may have changed.
Use Ctrl + F to search
0:00
ABC Listen, podcasts,
0:02
radio,
0:03
news, music and more. They
0:07
say Hollywood makes two kinds
0:09
of movies. Hero goes on
0:12
a journey or a stranger comes
0:14
to town. So
0:18
I got into background acting
0:21
just through the necessity of needing
0:24
work after COVID and
0:26
being based in LA. Steele
0:28
Saunders is that stranger and
0:31
he moved from Melbourne to LA to pursue
0:33
his dream of comedy and it's here
0:35
on set that his life takes
0:38
a weird turn. They were coming around
0:40
with the checklist saying, oh hey, this
0:42
week over the next three days everyone
0:45
has to go to the scanning truck
0:47
and get scanned.
0:48
The scanning truck. He
0:50
has no idea what
0:53
the studio representatives are talking
0:55
about. You would sort of ask,
0:58
oh, what's that for? And
1:00
they're like, oh, it's just if they need to change
1:03
anything. Now Steele
1:05
didn't like the sound of this scanning
1:07
truck, but he knew he didn't
1:10
want to piss off the bosses. So
1:12
he said, yeah, of course. Inside
1:15
the truck was a strange looking machine.
1:18
You're
1:18
in this very modern
1:21
computer setting, you know, three or 400
1:24
like tiny little cameras and you're on
1:26
this platform and then the platform
1:28
raises up on like hydraulics
1:30
or something. The 360 degree
1:33
array of cameras photographed every
1:36
inch of Steele and it created
1:38
a lifelike 3D model. Just
1:40
in a couple seconds, the platform
1:43
goes back down and you're done. And
1:45
then you walk out and go, oh, there's a lot of cool computers
1:47
in there. That was, that was
1:48
like kind of like admittedly
1:51
exciting to see it all. Steele
1:53
had just been digitally cloned.
1:56
His body double now existed inside
1:59
a computer.
1:59
Where it could be made to do anything
2:02
without him knowing. And just checking
2:04
Steel, you were wearing clothes at this point,
2:06
correct? I was very well dressed. Like,
2:09
if you're gonna get scanned, that's the outfit
2:11
to get scanned in. I'm gonna look fantastic for all eternity.
2:14
What was the studio doing, copying
2:17
and pasting actors, faces
2:19
and bodies?
2:21
The answer
2:22
would tear Hollywood apart and
2:24
change the future of movies and
2:26
TV.
2:28
A new stranger had come to town.
2:30
Its name was A.I.
2:37
Welcome to Hello AI Overlords,
2:40
a science fiction series about how AI
2:42
has burst into our lives in just
2:44
a few short years. I'm James Pertill.
2:48
By 2023, the excitement
2:50
in tech circles has finally spilled
2:52
over into the mainstream. Chat
2:54
GPT, image generators and
2:57
other powerful new tools are here.
3:00
I'm gonna introduce you to people who discover
3:02
that AI has infiltrated
3:04
their lives. And not always
3:06
for the better. This is a
3:09
story in three parts. The first
3:11
contact, then the backlash, then
3:14
acceptance. What happens when
3:16
the world wakes up to AI? Part
3:22
one, first contact.
3:24
It starts with Anna, another
3:26
background actor in Los Angeles. I
3:31
was brought on as a background
3:34
actor in a very large superhero
3:36
movie. It was a really great experience,
3:39
great director, great cast. And
3:41
just like in Alien Encounter
3:43
movies, Anna's first brush
3:46
with generative AI was kind of spooky
3:48
and strange. It was a weird day.
3:50
Like, to set the scene, there was weird
3:52
vibes. There were a lot of people from the studio
3:55
there, larger named producers,
3:57
some larger, you know, studio.
3:59
video types
4:01
that were on set and the vibes were kind of weird
4:03
that day.
4:03
And I should say, Anna is not
4:06
her real name. She's just a
4:08
bit nervous about annoying the studios
4:10
and breaching non-disclosure agreements. Anyway,
4:14
one day on set at the superhero movie,
4:16
she got this familiar tap on the
4:18
shoulder.
4:19
And they're like, hey, we have this really cool opportunity.
4:22
Would you be interested in doing something?
4:25
You know, the people
4:25
from the studio are here. We're
4:28
really interested in trying to capture
4:30
the best background that we have. I'm
4:32
so curious for someone that was like, sure,
4:35
I'll see what this is. So they lead Anna
4:37
through a maze of corridors and into
4:39
a secluded room. And
4:42
there it is, a scanning machine.
4:44
It
4:44
kind of looks like an MRI machine, but
4:47
like standing up.
4:47
Anna says the studio representatives
4:50
never mentioned digital clones. They
4:53
were very cagey about how the studio
4:56
would use the scan. It's like when they make
4:58
action figures, it's like, you know, video games.
5:00
It's cool. It's like, it's the thing that we do.
5:03
This is just like
5:03
a new way of doing it. And I'm
5:05
like, that's
5:07
interesting. So are we
5:09
getting paid more for this?
5:10
They said, nope, you will not
5:13
get paid more. I was
5:15
like, this feels wrong. Everything about this
5:17
feels wrong. I don't like this. I was like, hey,
5:19
I don't want to do this. And they're like, you really should
5:21
do it, though. And I was like, I'm good.
5:24
So Anna walks away. But
5:26
hundreds, maybe even thousands of friends
5:29
and colleagues, other background actors got
5:32
scanned. And then in
5:34
early 2023, a rumor starts
5:36
flying around Hollywood. The
5:39
studios plan to replace them with
5:41
scanned versions of themselves. AI
5:44
was taking their jobs. And
5:47
so they go on strike. One
5:49
day stronger. One day
5:51
more. It's
5:55
Hollywood's biggest shutdown in 40 years.
5:59
There's hundreds of actors on
6:02
the picket line. The entire
6:04
business model has been changed
6:07
by streaming, digital,
6:10
AI.
6:10
Even Fran Drescher
6:13
is there, who you might remember as the
6:15
nanny from the 90s sitcom The
6:17
Nanny. This is a moment of history
6:20
that is a moment of truth. She's
6:22
now the president of the Actors Union,
6:25
and she makes clear that actors from across
6:27
the United States are terrified.
6:30
Studios will use AI to replace them. We
6:32
are all going to be in jeopardy
6:35
of being replaced by machines and
6:38
big business. Now the studios
6:40
deny they
6:40
want to use digital clones to
6:42
replace actors, but
6:45
they also don't want to rule it out.
6:48
And this becomes a major sticking point
6:50
in negotiations with actors, and
6:52
the strike drags on. By
6:54
November, more than five months later,
6:57
actors are still negotiating with studios.
7:00
Of course at the moment if you walked on any set in
7:02
LA at the moment, it would be a very
7:05
lonely and dusty place. With
7:07
actors refusing to act, Hollywood
7:09
grinds to a halt. Which has
7:11
caused an unimaginable
7:14
ripple effect through Los
7:16
Angeles, because it's not just the
7:19
actors aren't working, it's the caterers
7:22
and the lighting people and
7:24
the cleaners. As the strike
7:26
drags on, many actors struggle
7:29
to make ends meet. But they
7:31
refuse to give up. They say their
7:33
future is at stake. I self-deprecatingly
7:37
say we're just cardboard cutouts
7:40
essentially. We're flesh-filled cardboard
7:42
cutouts. Of course the worry is that they're going
7:44
to take the flesh out and just use the
7:46
cutout. He still feels a sense
7:48
of hopelessness about AI. For
7:51
him it's not magical, it's
7:54
just another way of undercutting
7:56
workers. can
8:00
we get rid of humans? Yes. So
8:02
you know essentially this has been
8:04
taking place for you know since
8:07
the industrial era like how can we use
8:09
technology to put humans out
8:11
of work and that
8:13
to me is just like
8:16
really sad and and and just
8:18
like in society I always
8:20
wonder like what jobs are we going to leave
8:23
for the humans? After I spoke
8:25
with Steel the striking actors finally
8:27
reach a deal with the studios and
8:30
the deal includes some protections for
8:32
actors against AI including
8:34
paying them for use of their digital replica.
8:37
It's been a wild year for Hollywood, the scanning
8:39
trucks, the strike and now the deal
8:42
and this story is just one of many tales
8:44
of people encountering AI and
8:47
they all follow a similar pattern. People
8:49
learn that AI is more than a cute novelty,
8:52
more than a tool to write dumb poems or
8:55
generate silly images. It can be
8:57
used to undermine their livelihoods
9:00
and so we enter the next phase of
9:02
the world waking up to AI.
9:05
Part two,
9:06
the AI backlash. Imagine
9:10
you've spent years learning your
9:12
trade as a writer and finally
9:15
you've carved out a living as a published
9:17
author.
9:20
Meet Jane Friedman. She's a
9:22
human, Manhattan-based author. My
9:24
business which is you know a freelance business
9:27
really rides on my reputation
9:29
as a writer. Jane Friedman is
9:31
a big deal in the publishing world. She's
9:34
the author of a series of books on the business
9:37
being a writer and she gives a lot
9:39
of workshops on how to be successful.
9:42
My name gets shared a lot in the writing
9:44
and publishing community as a valuable
9:46
resource. This is someone you can trust
9:49
to get straight information.
9:50
So Jane
9:52
feels pretty safe in her work but
9:55
then one day in August
9:58
her life changes.
9:59
forever. I received an email from someone
10:03
who heard about a class I
10:05
was about to offer. This person
10:07
had gone on Amazon and they searched
10:10
for Jane's books. But some
10:12
of the books looked odd.
10:14
When she looked closer she realized
10:16
this all looks a little
10:17
strange, you know, her picture's not on it, her bio's
10:19
not on it, but her name's on it. So
10:22
Jane took a look and what she
10:24
found shocked her. There
10:26
were half a dozen books published
10:29
under her name that she
10:31
hadn't written
10:32
and they were for sale on Amazon.
10:35
And when she read further it was
10:37
obvious they were AI generated.
10:39
It was really repetitive,
10:42
generic, it wasn't
10:44
meaningful information. It
10:47
was the sort of thing to my
10:49
eye that looked like what you would generate
10:51
out of chat GPT which I've experimented
10:54
a lot with, being the curious person that I
10:56
am. Jane asked Amazon to
10:58
take the books down, but at first
11:01
the platform refused.
11:03
She feared the worst. Her
11:05
livelihood was built on her reputation
11:08
for being knowledgeable and trustworthy. Now
11:11
it might be drowned in a sea
11:13
of crappy AI content. The
11:16
books had her name on it. How
11:18
could people tell the difference between her
11:21
real books and the fake ones? Eventually
11:23
Amazon responded. I received
11:26
a message from their PR
11:28
person saying we're
11:31
looking into it, please be patient.
11:33
Then
11:35
when I got up the next morning, this is Tuesday,
11:39
most traces of the books were gone.
11:42
Getting the books removed was a win for
11:44
Jane.
11:45
But there was still the mystery of
11:47
how the AI was able to copy
11:50
her. Now we know that chat GPT
11:52
and other new writing tools are trained on
11:54
billions of words scraped from the internet
11:57
and from digital libraries of books. AI
12:03
companies generally don't say
12:05
which books and which websites they've
12:08
scraped, but Jane suspected
12:10
her website had been targeted. AI
12:13
writing tools seemed to know a lot
12:16
about her and her work, exactly
12:20
the information she'd published on
12:22
that website. How else would it know except
12:24
by doing a lot of scraping of
12:26
my website? But she didn't know
12:29
for sure. Then she
12:31
stumbled upon an article that seemed
12:33
to answer her question. The reporter
12:36
had unearthed a list of websites
12:38
that Google and Facebook use to train
12:40
their large language models. The first site
12:42
I put in was JaneFriedman.com. You
12:46
actually
12:46
have a ranking. It tells you
12:49
numerically how important
12:51
your site is in the grand scheme
12:54
of all of the grains of sand. And
12:56
I felt a little bit proud that my site
12:59
actually ranked relatively high.
13:02
This list explains how an AI
13:04
writing tool is able to produce
13:07
cheap and crappy knockoffs of Jane's
13:09
writing. But it also raises
13:11
more questions about the future of
13:13
her industry. I would say the authors
13:16
who are most at risk are the ones who
13:19
are already putting out low quality material
13:21
and they probably know that they're doing it.
13:24
Jane says AI isn't anywhere
13:26
near good enough to replace most human
13:28
authors at the moment. But
13:31
it is improving rapidly and
13:33
authors are getting worried. Amazon
13:36
has tightened its rules a little around
13:39
AI content
13:40
and Jane
13:41
has lawed up. I have actually
13:43
hired a law firm to prepare a trademark
13:46
application for me. Jane's taken
13:48
the unprecedented step of trademarking
13:51
her name. If someone is abusing
13:53
it, I really need to have immediate
13:56
remedies to stop the abuse. So
13:58
I was imagining. in the era
14:01
of AI that people could totally
14:04
create an entire website full
14:06
of Jane Friedman articles or create
14:09
videos or podcasts or courses
14:11
with the Jane Friedman name and then what would
14:13
I do? In less than a year, chat
14:16
GPT has changed publishing and
14:19
the backlash is here.
14:21
Authors are negotiating contracts so
14:23
a publisher can't use their books to
14:26
train an AI. And some
14:28
writers are suing open AI for
14:30
copyright infringement. They say
14:32
the maker of chat GPT stole
14:34
their voices. These are voices
14:37
they'd spent their lives crafting. Media
14:40
outlets like the ABC are
14:42
blocking open AI from scraping
14:44
data from their websites to train
14:47
future language models. So what's
14:49
the chance of this backlash slowing
14:52
big tech? Well
14:54
we don't know but it showed the world that
14:56
AI has a dark side
14:59
and that the AI utopia big
15:01
tech promised may
15:03
only be good for some people.
15:10
But the backlash isn't the full
15:13
picture. Many people find
15:15
the new AI tools really really
15:18
useful. Let's call this part
15:20
of the story the new reality.
15:27
I said at the start that Hollywood
15:29
makes
15:30
two kinds of movies. A
15:32
stranger comes to town and
15:34
a hero goes on a journey. Well
15:37
this is a story of a hero that goes
15:39
on a journey. And I'm
15:41
the hero if I do so myself.
15:44
Hello everyone. Thanks for having me. Intimidating
15:48
to look at this big classroom full
15:50
of you guys. I've talked my way into
15:52
a school in Perth to
15:55
interrogate the students one by
15:57
one. That's why I've come here today to kind of
15:59
find out for. you how you using
16:02
these new generative AI tools. And
16:05
I'm here to solve an important mystery
16:07
about 2023. But yeah this is like your
16:09
chance to kind of give a proper
16:12
insight into what the year has been like
16:14
for you. At the start of the year there
16:16
were big headlines about how chat GPT
16:19
would ruin education. Experts
16:21
said students would use AI to
16:24
cheat and then well
16:27
nothing. We never actually heard
16:29
if this happened. So
16:32
I'm here at the end of the year to finally
16:35
answer the question. I'm just going to interview
16:37
the small groups in the small room. How much
16:39
do students use chat GPT
16:42
to cheat? How does that sound? A
16:45
room full of teenagers stare back
16:47
at me in silence with barely
16:50
concealed contempt. Yeah
16:53
okay cool. I press on
16:56
with my mission to get to the truth.
16:58
Great well yeah here's the questions.
17:00
I come across Brody. He's in
17:02
year 11 and he's doing a ton of
17:05
hard subjects like physics, chemistry.
17:08
He looks me in the eye and he says he
17:11
isn't a cheater. Though he does use
17:13
chat GPT. I use a lot of
17:15
not just chat GPT but other sort of AI
17:18
tools for inspiration for certain projects.
17:20
So I might use AI
17:23
image tools to start
17:25
some design ideas for websites,
17:28
for characters, that sort of stuff. I turn
17:30
to Brody's friend Danny. He's been
17:32
watching the interrogation. Danny
17:35
doesn't look particularly nervous and
17:37
when I ask him if he uses AI to cheat at
17:40
school he confidently says
17:43
no. If you use chat GPT you'll have a like,
17:45
it's not cheating but you get a head start in
17:47
what the subject is going to be like, like the basic outlines
17:50
of it and sometimes I've seen people
17:52
use like, generate AI to like generate
17:54
photos for their PowerPoints. So two students
17:56
down and no cheating here.
17:58
I think they're telling the truth. But
18:01
I do have a big ABC microphone
18:03
and their teacher is just standing over
18:05
there So who knows? So
18:09
I lay my cards on the table and
18:11
I ask the class directly So how come
18:14
none of you guys use it to cheat? If
18:16
I'm not learning it then there's not much
18:18
point using it to cheat in class cheating
18:21
will just lead to More work in the future.
18:23
These are very mature answers. Do you
18:25
think they're widely held or are you
18:27
guys the exception? Probably while
18:29
already held because with asking before
18:31
this I've talked to many friends about it and they have
18:34
similar views Because the main is
18:36
that there's exams to it. You have to know
18:38
everything anyway, so it really doesn't matter
18:41
using AI or anything I'm
18:48
defeated by their academic
18:51
integrity Everyone said
18:53
the students were going to use chat GPT
18:55
to cheat during 2023 But
18:58
not a single student will admit to this
19:02
In the days that follow I
19:04
have a crisis of confidence
19:07
The hero goes on a journey But
19:10
the journey proves far harder
19:12
than they expected. I start to
19:14
question the very basis of my investigation What
19:18
is cheating? Where is the line
19:20
between getting chat GPT to
19:22
write your homework and Learning
19:25
how to use an important AI tool the
19:27
school may call it cheating But
19:30
an employer well, they might call
19:32
it innovation Then
19:35
I get a message. I've cast
19:38
the net wide to try and find
19:40
a student who cheats I've
19:42
been posting on study forums for
19:45
high school students around Australia And
19:48
my hard work starts paying off
19:51
I start getting voice messages in response
19:53
to my questions one arrives
19:55
from a year 11 students in Sydney
19:58
Krissula has
19:59
a a truly terrifying list
20:02
of subjects. My subjects this year are advanced
20:04
mathematics, advanced English,
20:05
ancient English, economics, modern great continuous,
20:08
hopefully, extension history. Chrisola
20:10
says at the start of 2023, she
20:13
was addicted to chat GPT,
20:16
and she was using it to help with
20:18
her homework.
20:19
It was especially useful for literary
20:21
analysis and providing quotes for things in
20:23
text.
20:24
Then she noticed she was doing
20:26
worth and assessments.
20:27
I noticed my critical and analytical thinking
20:30
was slightly impaired.
20:31
And she grew worried about how using
20:33
the tool was affecting her learning.
20:36
The daily use had formed a habit. So
20:38
Chrisola blocked chat GPT
20:41
on her computer, but she tells
20:43
me that her friends are still addicted
20:46
and maybe use it to cheat.
20:48
My friends used chat GPT constantly,
20:51
even handing in entire assessments written
20:53
entirely by the AI.
20:55
I knew I was getting close.
20:58
And then finally, I score
21:00
a hit. My hard-hitting investigation
21:03
of cheating in Australian schools pays
21:06
off. A year 12 student sends
21:08
the message I've been waiting for. I
21:11
can see the WA media award dangling
21:13
before me. I call him straight
21:15
away. I probably wouldn't want to use my
21:19
full real name. Yeah. Eric.
21:21
Eric's a good name. I like it, Eric. Eric
21:23
isn't his real name. He wants to be anonymous
21:26
for this interview. And it's soon
21:28
clear that Eric is here to spill
21:31
the beans. He tells me how
21:33
he began cheating. I went, well,
21:35
if this can spit out information this
21:37
fast, what's to say, it
21:39
couldn't do it for schoolwork. Early in
21:41
term one, Eric gave chat GPT
21:44
his geography assignment. So
21:46
I was sitting there and I was just watching it. And it's
21:48
just printing out all these lines, printing them out, printing
21:50
them out. And I'm going, wow. How
21:53
have we created something like this
21:56
as a species? And it's like a full
21:58
complex essay, just being.
22:00
And to do your screen in seconds is really
22:02
amazing thing.
22:03
Yeah, obviously you have that sort of joy
22:06
kick back when you're like, oh, no, I don't
22:08
have to do it. I've got it done. It's done now. So
22:11
just to be clear, did that count as cheating?
22:14
That would count as cheating. Yeah, I would assume
22:16
so. If it was called, I'd assume that would be
22:19
seen as cheating, but slipped
22:21
through fine. So, yeah, Eric cheated
22:23
on that geography assignment. Although for
22:26
essays that actually counted towards
22:28
his HSC, he says he didn't most
22:31
of the time use it to do his homework.
22:34
So around around 60% of
22:36
my homework was written by chat
22:39
JBT due to the fact that
22:42
I found it to be a lot
22:44
easier to do and give me more time. How
22:46
common is your story among the people
22:48
that you know? I would say you'd be
22:51
hard pressed to find a student that hasn't
22:53
used AI for at least something. Whether
22:56
that is just studying or an
22:58
assessment or a piece of homework, I think
23:01
it would be very difficult for you to find a student
23:03
that is sitting the HSC right now.
23:07
Eric says using chat JBT to do
23:10
most of his homework hasn't affected
23:12
his ability to learn. And
23:14
the reason he gives for this is that
23:17
homework is a waste of time. With
23:19
a lot of homework tasks, personally, I find
23:22
that they are instead of learning, it's
23:24
more regurgitating. Mission
23:28
complete. I say goodbye
23:30
to Eric. The hero staggers
23:32
to the end of his journey. Wary
23:35
but triumphant. A cheater
23:37
found. But of
23:39
course, this isn't all about cheating.
23:42
The reason I've told you these students' stories
23:44
is because I think it's important
23:47
to get to the nuance of how AI
23:49
can be used. At the start
23:51
of 2023, we had the usual
23:53
knee-jerk responses to a new technology.
23:56
Schools banned chat TPT and everyone
23:59
predicted student... students would cheat. But
24:02
the story is more complicated.
24:04
Students are using AI in lots of
24:06
different ways. As AI tools
24:08
multiply, the line between schoolwork
24:11
and plagiarism is getting blurred.
24:14
Schools may have to rethink education.
24:17
One of the very mature teenagers at
24:19
the Perth High School, Barron, summed
24:22
it up best. The problem about integrating
24:24
AI into school is that it's pretty
24:26
hard to enforce the
24:28
difference between using it for research
24:30
and using it for cheating. Even if it's
24:33
easy to find out that someone's cheating, preventing
24:36
someone from cheating in the first place would be very
24:38
difficult. So, that's students.
24:41
For them, AI can be great. It's
24:43
something they use to benefit themselves.
24:47
For actors, it's a different story. AI
24:50
is something that happens to them. And
24:52
we'll have to see if the New Deal struck with studios
24:55
means they actually get paid when their
24:57
digital replica gets used. For
24:59
writers, well, they're getting their words scraped from
25:02
the internet and fed back to them as
25:04
AI content. And they
25:06
have to compete with it. In every
25:08
story, it's roughly the same technology.
25:11
It's all generative AI. The
25:13
difference is who's using it and
25:16
who's getting used, who benefits
25:18
and who loses out. This
25:24
is Hello AI Overlords, a science-fiction
25:26
series. You can find our previous
25:28
episodes on the ABC Listen app. See
25:31
you next week.
Podchaser is the ultimate destination for podcast data, search, and discovery. Learn More