Episode Transcript
Transcripts are displayed as originally observed. Some content, including advertisements may have changed.
Use Ctrl + F to search
0:01
Support for this show comes from
0:03
Ferragamo. A century ago,
0:05
a shoemaker in southern Italy had
0:07
a vision to marry abstract art
0:09
with high-quality craftsmanship. And that led
0:11
them to create beautiful, timeless footwear.
0:13
Today, the name Ferragamo is synonymous
0:15
with luxurious shoes and accessories, and
0:17
for more than a decade, Ferragamo
0:20
footwear has been made with an
0:22
ethical supply chain and sustainable footprint.
0:24
Their classic ballerinas or contemporary moccasins
0:26
are a perfect blend of fashion
0:28
and comfort, and yes, Kara Swisher
0:30
has a pair of both, and
0:32
I bought them. Discover Ferragamo's timeless
0:35
styles with a modern touch at
0:37
ferragamo.com. Have
0:42
a question or need how-to advice? Just
0:45
ask MetaAI. Whether you need
0:47
to summarize your class notes or want to create
0:49
a recipe with the ingredients you already have in
0:51
your fridge, MetaAI has the
0:53
answers. You can also
0:55
research topics, explore interests, and so
0:58
much more. It's the most
1:00
advanced AI at your fingertips.
1:03
Expand your world with MetaAI.
1:06
Now on Instagram, WhatsApp, Facebook,
1:08
and Messenger. Hi,
1:24
everyone from New York Magazine and the Vox
1:26
Media Podcast Network. This is on with Kara
1:28
Swisher, and I'm Kara Swisher. My
1:30
guest today is Craig Peters, CEO
1:32
of Getty Images. You've
1:34
seen Getty's work, even if you don't know it.
1:36
Getty photographers are on the red carpets at the
1:38
Met Gala or at the White House or at
1:41
the Olympics for sure. Pictures that are then used
1:43
by media companies across the globe. Getty
1:45
also has a huge inventory of stock photos
1:47
and video. It has also one of the
1:50
largest archives of historical images in the world.
1:52
You'd think that would be enough, but
1:54
it's not these days. Craig Peters has been heading
1:56
Getty since 2019 and took the company back. public
2:00
in 2022. The same month as
2:02
Bad Luck would have it that
2:04
OpenAI launched Dali in beta, which
2:06
means questions about how to protect
2:08
copyrighted images got a whole lot
2:10
more urgent both for creators and
2:12
for the public. Questions that have
2:14
not gone away. Fittingly, our expert
2:16
question this week comes from former
2:18
Time Life and National Geographic photojournalist
2:20
Rick Smolin. Craig has been
2:22
an outspoken voice on improving standards and protecting
2:25
creators, but his stance on AI has also
2:27
shifted in the past few years. So
2:29
I'm looking forward to talking about how he
2:32
sees the business model for working collaboratively with
2:34
this new technology in the future. And
2:36
at the end of the day, this
2:38
is about transparency and trust and authenticity.
2:41
My biggest worry, of course, is if
2:43
this is a replay of what the
2:45
internet did to media the first go-round.
2:48
We'll talk about that and more, including
2:50
why they can't get fingers right. Those
2:52
fingers are creepy. Hi
3:02
Craig, great to see you again. Great
3:04
to see you as well. Thanks for having me.
3:07
Craig and I had breakfast in New York, I
3:09
don't know, a couple months ago, talking about these
3:11
issues. I'm trying to get my
3:13
arms around AI and all its aspects
3:15
and interviewed wide ranges of people
3:17
and obviously photos
3:19
are of course at the center of
3:21
this in many ways, even though the
3:24
focus is often on text, but photos
3:26
and video are critical. So you've been
3:28
at the forefront of conversations around generative
3:30
AI copyright issues, but Getty has also
3:32
been collaborating and utilizing AI like most
3:34
of media. I want to talk
3:36
about all that and where you see this heading for
3:38
creatives and for the public. But
3:40
first, I think most people know or have seen the
3:42
name Getty Images in a caption, but they don't have
3:44
a good understanding of the business as a whole. So
3:47
why don't you give us a very quick talk
3:50
about what it does. Alright, well first let
3:52
me start. We were founded by Mark Getty
3:55
of the Getty family and
3:57
Jonathan Klein back in 1991. So
4:00
the company's been around for almost 30 years. We
4:03
are a business to business content platform.
4:07
We provide editorial content, so
4:09
coverage of news, sports, entertainment,
4:12
as well as a really deep
4:14
archive, and creative content. So
4:17
this would be materials that would be used
4:19
to promote your brand, your company, your product
4:21
offerings, or your services. We
4:24
provide that in stills, so photography,
4:26
as well as video, as you
4:28
highlighted. We provide that in what
4:30
we call pre-shot, which means it's available
4:33
for download immediately. You can search our platform,
4:36
find content, and then download it and use
4:38
it. In
4:40
some cases, we provide custom creation,
4:43
right? So we might work on an assignment
4:45
to cover a certain event. Just
4:47
for full disclosure, you've done Fox Media events. Correct. And
4:51
you then own the pictures, but Fox gets
4:54
use of them, correct? Correct. We
4:56
represent over 600,000 individual photographers and over
4:59
300 partners. These are partners like
5:01
the BBC or NBC News or
5:03
Bloomberg or Disney or AFP. And
5:07
what we try to do is allow our customers
5:09
really, we try to deliver value
5:11
across four elements. So
5:14
we have editorial content, so when you
5:16
think about creating more efficiently, we'll
5:18
have a team of over 100 individuals in Paris
5:20
for the Olympics. And
5:23
we will allow the entities that are covering
5:25
the Olympics to have
5:27
access to high-quality visuals in real
5:29
time on- Big
5:31
events. Big events or stuff you're contracted for. Like
5:34
you did my last code conference, for example.
5:36
Correct. But then there's also,
5:39
again, what we call creative, which is also
5:41
referred to as stock imagery. It's
5:43
intended to be used for commercial
5:45
purposes, to promote a company's brand,
5:47
products, or services. Right. And
5:51
you've been a company since 2019. You took
5:53
it public through a SPAC in July
5:55
2022, just about two years ago. Correct. Coincidentally,
5:58
it was the same month OpenAI made its image available
6:00
in beta form. What
6:03
did you think at the time? Did you think
6:05
end times or did it not register with you
6:08
all? No, I think, I mean, first off, generative
6:10
AI is not something that is entirely
6:13
new. Sure. It's something that we
6:15
had been tracking for probably seven years prior
6:17
to the launch of DALI. We
6:19
had first had discussions with Nvidia who we now
6:22
have a partnership, which we could talk more about.
6:24
Yeah, I will. But
6:27
we knew those models
6:29
were in development. We
6:31
knew the capabilities were
6:33
progressing. So
6:35
it didn't surprise us in
6:38
any way, shape or form. And
6:41
fundamentally, our business is again, providing
6:43
a level of engagement, whether
6:45
that's, again, through a media outlet
6:48
or through your corporate
6:50
website or your sales and marketing collateral
6:52
in a way that relies on authenticity
6:54
that relies on creativity, it lies on
6:57
core concepts being conveyed. Those are difficult
6:59
to do. Yeah, so here's this thing
7:01
coming, you see it. Obviously it's been
7:03
around and they've been generating
7:05
fake images for a long, long time, absolutely.
7:08
But did you think, well, look,
7:10
there's six fingers, it's kind of weird looking.
7:13
They always have that sort of like a Will
7:15
Smith picture of him eating spaghetti and it doesn't
7:17
look great. Did you think this
7:19
could be a real problem?
7:23
No, I don't think we saw it necessarily as
7:25
a problem. I think we saw some of the
7:27
behaviors around it as a problem. So we
7:30
knew that hair lines and fingers and
7:32
eyes and
7:35
things like that were gonna resolve over time. What
7:38
we saw though was ultimately these platforms
7:43
that were launching these generative models
7:45
were scraping the internet.
7:47
Correct. And taking third party intellectual
7:49
property in order to create
7:51
these services. And fundamentally, we
7:54
believe that creates some real
7:56
problems. Yeah, that's top theft. They've
7:58
done it for many years. I would say
8:00
that there's kind of four things that go
8:02
into generative AI. So GPUs,
8:05
processing and power, and computer
8:07
scientists. And those three things,
8:09
these companies are spending billions and billions
8:12
of dollars around, right? But
8:14
the fourth item is something that they call
8:16
data, I happen to call content. And
8:19
it is something that they are taking, and they
8:22
are scraping and
8:24
taking from across the internet. And
8:26
that fundamentally, to me, raised some
8:28
questions about not
8:31
only should they be doing that,
8:33
is that allowed under the law? But
8:37
it also creates real IP risk for
8:39
end users of these platforms, if
8:42
they're using them commercially. Yes, but they're hoping
8:44
to get away with it just so you wear. But
8:46
you push back pretty hard against AI initially,
8:48
you ban the upload and sale of AI-generated
8:50
images in your database. But it seems you've
8:52
done more than you seem
8:54
to be embracing it more. Talk about
8:57
how it's evolved in your thinking.
9:00
Because at its base, they like to take things. They've been doing
9:02
it for a long time. I
9:05
had a lot of scenes in my book where
9:07
Google was just taking things and then reformulating and
9:09
spinning the book fight that they had. They
9:11
use the expression fair use quite a bit. Did your thoughts
9:14
change on it? Or do you think, well, this is going
9:16
to happen again, and I have to embrace it? I
9:18
think we thought we really
9:20
took a two-pronged strategy. And
9:22
I don't think it's changed or it's
9:24
evolved. I think it's been pretty constant.
9:27
But we are not luddites. We believe
9:29
in technology can be beneficial to society.
9:31
We believe it needs to have some
9:33
guardrails and be applied appropriately. But we
9:35
wanted to embrace this technology to enhance
9:37
what our customers could do. We
9:42
also wanted to make sure that we
9:44
protected the rights of our
9:47
600,000 photographers and 300
9:50
partners and of our own and try
9:53
to get this to be one
9:55
where it was done more
9:57
broadly beneficial, not just to the...
10:00
the model providers, but to the industries as
10:02
a whole. So
10:05
we pursued a strategy that's really twofold.
10:07
So one was we
10:10
did not allow content into our database that
10:12
was generative AI. We did that for two
10:15
reasons. One was that
10:17
our customers are really looking for
10:19
authentic imagery, high-quality imagery, that imagery
10:22
that can really relate to an
10:24
end audience. We think
10:26
that that is still best done
10:28
through professionals using real
10:30
models, etc. Accepting
10:33
AI imagery that's been created from
10:35
engines that are scraping the Internet
10:37
has potential risk. It has potential
10:39
risk from copyright claims. Correct. It
10:42
also has potential risk from people
10:44
that have had their image scraped
10:46
off of the Internet generally, and
10:49
it can replicate those types of things. So we didn't
10:51
want to bring that content onto our platform and then
10:54
ultimately give it to our customers and have them bear
10:56
that risk. Because one of the things that we try
10:58
to do for our customers is we try to remove
11:00
that risk. Right. Now, I'll get to your end-of-the- end,
11:25
that's creative content, not editorials content. So to
11:28
your point, we haven't trained it to know
11:30
who Taylor Swift is, or Scarlett Johansson is,
11:33
or President Biden. That
11:36
creative content is permissioned, it's released.
11:38
So we have model
11:40
releases from the people portrayed
11:42
within that imagery. We have property releases,
11:45
etc. So
11:47
it is trained off of a
11:49
universal content where we have
11:51
clear rights to it. There's a
11:53
share of revenues back to the content
11:55
provider. So as we sell that service,
11:58
we actually give a revenue share back. to
12:00
the content providers that its content was used
12:02
to train. As you said,
12:04
it can't produce deep fakes. It cannot
12:06
produce third party intellectual property. So if
12:08
you search sneakers or type a generative
12:10
prompt of sneakers, it
12:13
will produce sneakers, but it's not going to
12:15
produce Nikes. If you type in laptop, it's
12:17
not going to give you an Apple iMac.
12:21
It is high quality. So
12:23
you can produce models that
12:26
are of very high quality in terms of the
12:28
outputs that it produces. You don't
12:30
have to scrape the entirety of the internet.
12:33
This proved that. And it
12:35
is fully indemnified because, again, we know the
12:37
source of what it was trained on. We
12:40
know the outputs that it can produce,
12:42
and they're controlled. And it
12:44
ultimately gives a tool to corporations
12:48
that they can use within their business day
12:50
to day without taking on intellectual
12:52
property risk. So the idea is to
12:54
make a better version of this, but
12:57
there's so many AI-generating image products out
12:59
there already on the market. Dolly, Mid-Journey,
13:01
OpenAI is currently testing Sora. As
13:04
you said, these people have billions of dollars
13:06
put to task here.
13:09
So talk about your sort of unique
13:11
selling point. Obviously, cleaner database, less biased
13:14
than internet-scraped versions. And you also know
13:16
these images are backed by our uncapped
13:18
indemnification. So explain what you mean by
13:20
that. But let's start with you've got
13:23
a lot of competitors with tons of
13:25
money and have a little less care
13:27
than you do about the imagery. This
13:31
model is intended to be utilized
13:33
by businesses that
13:35
understand that intellectual property can carry real
13:37
risks as they use it in their
13:39
business. Those risks can come
13:42
with large penalties. And what we are
13:44
doing is providing a solution that allows
13:46
them to embrace generative AI, but avoid
13:48
those risks and still produce
13:50
a high-quality output. Other tools that are
13:52
out in the marketplace do
13:54
bear risk because they have scraped
13:57
the internet. You can produce.
14:00
third-party intellectual property with them, you can
14:02
produce a deep-fakes with them. That
14:07
ultimately we believe is a constraint
14:09
on the broad-based commercial adoption of
14:11
generative AI. This has been their
14:13
modus operandi for years and years
14:16
to do this to scrape and
14:18
then later clean themselves
14:20
up. I think it certainly
14:22
break things and move fast. Ultimately,
14:27
we think that this technology
14:29
should not just be thrust into
14:31
the world and
14:33
damn the consequences. We believe there should be
14:35
some thought, we believe there should be some
14:37
level of regulation. We should believe that there
14:39
should be clarity on whether it is in
14:41
fact fair use. One of the companies that
14:44
I'm not sure if you mentioned, which
14:48
Stability AI that launched Stable Diffusion. I
14:50
will. It's a- We knew that they
14:52
had scraped our content from across the
14:54
Internet. And used
14:56
it to train their model. We didn't believe that
14:58
that was covered by fair use in the US
15:00
or fair dealing in the UK. Right.
15:03
So we've litigated on that
15:05
point and it's progressing through
15:07
the courts. I want to talk
15:09
about that in a second. I want to put
15:11
a pin in that. But you talk about what
15:13
uncapped indemnification means. It means that
15:16
Getty Images stands behind
15:19
the content that the generative model produces,
15:21
and that you're safe to use it.
15:23
And if there are any issues that
15:25
occur downstream, which there shouldn't
15:27
be, we will still stand behind it.
15:30
We will protect our clients. So AI gets
15:33
stuff wrong all the time though. Are you
15:35
worried about promising indemnification? I guess that's what
15:37
you're promising now. But are you worried about
15:39
the cost? If you do have copyright and
15:41
for instance, using this technology? No, because again,
15:43
we know what it was trained on. We
15:46
know how it was trained. And
15:49
we trained it in a way that it
15:51
couldn't produce those. I think, again, when you
15:53
just scrape the internet and you take in
15:56
imagery of children, when
15:59
you take in... imagery of famous
16:01
people or brands or third-party intellectual
16:04
property like Disney, Star Wars, or
16:06
the Marvel franchise, you can
16:08
produce infringing outputs. You can produce things
16:11
that are gonna run afoul of
16:13
current laws and potential future laws. This
16:16
was built purposefully from
16:19
the ground up to avoid the social
16:21
issues that are being
16:24
questioned as a result of this technology, but
16:26
also to avoid the commercial issues. Right, right,
16:28
which was these companies are gonna have the
16:31
same thing they did with YouTube and they
16:34
eventually settled it out. So I want to
16:36
get to the lawsuits in a second, but
16:38
the Pick Arts Partnership is different. I mean,
16:40
it's a licensing deal. They are using your
16:42
image database to train their AI, and these
16:45
are these licensing deals. You're not the only
16:47
one. AP is the license deals with OpenAI.
16:49
There's all kinds of people who are striking
16:51
deals like OpenAI
16:53
and Meta did one with your one of
16:55
your competitors Shutterstock. Let's talk
16:57
about those first. Well, I think, you
16:59
know, I think ultimately we believe that
17:01
there are solutions. These services
17:04
can be built that ultimately
17:06
do appropriately compensate for the
17:09
works that they're utilizing as
17:12
a service. And we
17:15
think, you know, the models that we've
17:17
entered into that you'd mentioned, so Nvidia
17:19
and Pixar, those are not payments to
17:22
us, a one-time payment, which is most
17:24
of these deals or a recurring payment.
17:27
They're actually, this is a rev-share deal,
17:30
so as those models produce revenues, they
17:32
will generate revenues that could flip back
17:34
to the content creators. And you
17:37
know, I think some of the deals that are getting
17:39
done today are trying to, in
17:41
essence, whitewash a little bit
17:43
the scraping that has historically been done
17:45
and show that they are willing to
17:48
work with providers. In some cases, it's
17:50
providing access to more current information that's
17:52
difficult to get in real time from
17:54
scraping. Sure. But ultimately it's a small
17:57
drop in the overall bucket. about
18:00
the lawsuit against stability AI, where
18:03
does it stand? And just again,
18:05
you're claiming the company unlawfully copied
18:07
and processed millions of Getty copyright
18:10
protected images, which are
18:12
very easy to track. I mean, more than text,
18:14
much easier. So tell us where that stands
18:17
now, because again, that's an expensive prospect for
18:19
a small company. Well, it's something
18:21
that we invest in. We think
18:23
it's important. We think
18:25
we need to get settled view
18:27
on whether this is in fact fair
18:29
use or not. We don't buy into
18:31
maybe what some of the providers are
18:33
stating is settled fact, that it is
18:35
in fact fair use. We believe that
18:37
that is up for debate. I
18:40
mean, we believe within the realm of imagery, we
18:42
think there's a strong case to be made that
18:44
it is not fair use. And we think there
18:46
are some precedents out there, most
18:49
notably in the US, Warhol versus
18:52
Goldsmith, which was settled at the court,
18:55
which highlights some case law
18:57
that would say that this is going to
18:59
be questionable whether it would qualify for fair
19:01
use. So we launched
19:04
that litigation. It is
19:06
moving forward a little
19:08
bit more at pace in the UK.
19:11
So it is proceeding to trial,
19:13
I expect will likely be a
19:15
trial end of this year, early
19:18
next. In the US, it's moving a little
19:20
more slowly as things take time
19:22
to move through the court system. And I don't have a magic wand
19:25
to speed that up. Talk about the
19:27
calculation of suing versus settling
19:29
essentially. And the costs, because here you are,
19:31
someone grabs something of yours, you have to
19:33
get hit on the head, you have to
19:35
take them to court for hitting you on
19:37
the head. But still, you've been hit on
19:39
the head, correct? Well, I think,
19:41
again, it's in a world where everyone's taking our imagery anyways.
19:43
Our imagery is all over
19:45
the internet. As you mentioned, it's on the
19:47
verge. It's on the media
19:52
platforms that you are visiting.
19:56
The one mistake I made in copyrighted imagery, I
19:58
paid a lot of money. It
20:00
was a lot of money and I never did
20:02
it again. I'll tell you that. So our imagery
20:04
is all over the internet and it's being scraped
20:06
and it's being taken and it's being used to
20:08
train these models. We
20:11
wanna get clarity if in fact
20:13
that is something that they have rights to do.
20:16
I believe it's good for the industry
20:18
overall. When I say the industry overall,
20:20
I'm not just talking technology, I'm not
20:22
just talking content owners, but to have
20:24
clear rules to the road. Is
20:27
this or is this not fair use? And then
20:29
you know what you need to do in
20:32
order to work within that settled law. And that's
20:34
what we're trying to get to. But let's get
20:36
back to the issue of content creation that you
20:38
also brought up. You aren't the only one that's
20:40
worried about losing money. Every week we get a
20:42
question from an outside expert. This week our question
20:44
comes from photojournalist, Rick Smolin, the creator of the
20:46
best selling day in the life book series. You've
20:49
seen the movie Tracks. He's a National Geographic
20:52
photographer portrayed by Adam Driver. Let's have a
20:54
listen to his question. Like many
20:56
of my peers, my photographs are sold
20:58
around the globe by Getty Images. And
21:01
we've always considered our relationship with
21:03
Getty as symbiotic. We
21:05
photographers risk life and limb to
21:08
provide Getty with our images and in
21:10
return Getty markets our photographs to buyers.
21:12
But the advent of AI, it's sort of starting
21:15
to feel a little bit like a marriage where
21:17
one partner in the relationship is
21:19
caught having an affair with a much
21:21
younger, more attractive person while assuring
21:24
their spouse, this is nothing more than
21:26
a dalliance. We
21:28
assume that our images are being sampled
21:30
to generate the synthetic images that Getty's
21:33
now selling. And many of us
21:35
are very concerned that our work has become a commodity
21:38
like coal being shoveled into
21:40
Getty's LLM engine. And
21:43
also sort of like the cheated upon older
21:45
spouse, many of us believe it's
21:47
only a matter of time before we're cast
21:49
aside for this much younger and much sexier
21:51
young thing called AI. So
21:53
my question to Craig Peters is whether
21:55
this marriage is over but you
21:57
just haven't admitted to us yet. told
22:00
us, or if it's not
22:02
over, how does Getty Images
22:04
envision its relationship with actual
22:06
photographers versus prompt jockeys? And
22:09
how is Getty going to compensate us,
22:11
your long-term partners, fairly for our images
22:13
in the future? Thanks. Okay,
22:16
that's a really good question. I hear it a
22:19
lot from photographers. They don't know who to trust.
22:21
So how are you compensating the creators whose work
22:23
is being fed into this tool, talked about how
22:25
it works, and can people opt out if they
22:28
don't want AI trained on their
22:30
work? Right. Well, first off, let me start
22:33
with, I think we don't train off
22:35
of any of our editorial content. And
22:37
the coverage that he is talking about
22:39
that he does, and that we are
22:41
fortunate enough to represent, would
22:44
be considered editorial coverage. We
22:47
believe that the job that he does is
22:49
incredibly important to the world. The
22:53
images that they produce, the
22:55
events that they cover, the topics that
22:57
they cover are incredibly important. Right behind my desk,
22:59
Kara, there is an image that
23:02
was taken in Afghanistan and it was taken
23:04
by one of our photographers, staff photographers, a
23:06
guy named Chris Hundress, who happened to lose
23:08
his life in
23:10
the pursuit of covering conflict around
23:12
the world. And
23:15
that's incredibly important. And I have it there because
23:17
it reminds me that
23:19
we have a very
23:21
important mission and that the individuals that
23:23
are producing this image are incredibly important
23:25
and they are taking very real risks.
23:27
And we value that and we never
23:29
want to see anything
23:32
that we do undermine that or misrepresent
23:34
it. And we think it's important to
23:36
the world going forward and we think
23:38
it's very important that persists.
23:41
So we do not train off
23:43
of editorial content. This model was
23:45
not trained off of editorial content.
23:47
And we believe that that type
23:49
of work has a level
23:52
of importance today and we'll have a level
23:54
of importance going forward in the future. And
23:56
we think that importance only increases as we
23:58
see more and more deep-fake
24:00
content produced by these generative
24:02
models that are trying
24:04
to undermine true knowledge and true truth.
24:06
So the compensation system stays the same?
24:09
So the compensation, so again, we have
24:11
creative imagery, this would be, you could
24:13
also refer to it as stock imagery,
24:16
where we have permissions from the
24:18
individuals that are contained within those
24:20
images. We have the right to
24:22
represent that copy of my, broadly,
24:24
that is training these models. These
24:27
models, when they produce a dollar's worth of
24:30
revenue, we give a share of
24:32
that revenue back to the creators whose content
24:34
was used to train that model. That
24:37
is the same share that they would get if
24:40
they licensed an image, so
24:42
one of their images off of our platform. So
24:45
on average, Kara, we give
24:48
royalties out around 28% of our revenue. We
24:52
invest a lot in our platform and sales and
24:54
marketing and content creation,
24:56
etc. But on average, we're
24:58
sending 28 cents back for every image that
25:01
we license. And you can think the exact
25:03
same thing will happen when
25:06
we license a
25:08
package of, you know, someone subscribes to
25:10
our generative service. So who owns the
25:12
images created by these AI tools? And
25:14
it sounds like in the weeds. That's
25:17
a very good question. It's, again, an
25:19
unsettled dog because we're in new territory.
25:21
So in the US today, you cannot
25:24
copyright a generative output. I
25:27
think over time, that might evolve to where
25:29
it depends on the level of human input
25:31
and involvement in that process. But
25:33
right now is not one that you can
25:35
copyright this image. But what we take a
25:37
stance is for our customers that are using
25:40
our generative tool is since
25:42
they gave us the generative prompt, they
25:44
are part of the creation. And therefore,
25:46
we don't put that imagery back into
25:49
our libraries to resell. It
25:52
is essentially a piece of content that
25:54
is, quote unquote, owned by the customer
25:56
that produced it using the generative model.
26:00
copyright is one that we can't convey,
26:02
that has to be done by the
26:04
US Copyright Office. But we essentially take
26:06
the position that the individuals
26:09
or the businesses that are using that service
26:11
own the output to that service. They own
26:14
the output, and it would be a similar
26:16
thing. But right now, whether they can be
26:18
copyrighted is unclear, even if so, is the
26:20
prompt owner the copywriter? Correct. And
26:22
I think, again, I think that will evolve over time to
26:25
one where I think the level
26:27
of human input will have
26:29
a bearing on whether that
26:32
is, in fact, copyrightable. It will all be. It's
26:34
a question of how much you get paid along the line. So
26:37
you've pushed back against the idea that these licensing
26:39
agreements are Faustian, though, meaning basically you're making a
26:41
deal with the devil. Shutterstock's deal
26:43
with OpenAI is for six years, while
26:45
Red Teamers at OpenAI are testing Sora.
26:50
Are you worried that you're all doing what we
26:52
in the media did with, say, the Facebooks of
26:54
the world, the Googles of the world, who now
26:57
sucked up all the advertising money and everything
26:59
else? What is your biggest worry when
27:01
you're both doing these deals
27:04
and suing, just waiting for clarity,
27:06
I assume, from courts and copyright
27:08
officials? I believe that many
27:11
of the deals that are happening today are
27:13
very Faustian and go back to
27:16
ultimately some of the mistakes that were made early
27:19
on in digital media and or
27:22
in social media, making trades
27:24
for things that weren't in the long term health
27:26
of the industry. What we are trying to do
27:29
very clearly in the partnerships that we are striking and
27:31
in the deals that we are striking and in our
27:33
approach is we are trying
27:35
to build a structure where the
27:41
creators are rewarded for
27:43
the work that was included, and fairly
27:45
so. I gave you
27:48
my four components of generative
27:50
AI, GPUs, processing, computer scientists,
27:52
and content. That fourth
27:54
right now is getting an extremely
27:57
small stake. Some companies are doing
28:00
no deals, so like the mid
28:02
journeys and such, and they
28:04
are providing no benefit back to it.
28:06
We wanna see a world where content
28:09
owners, content creators
28:12
share in ultimately economics
28:15
of these. And I
28:17
think if you equate it again back to a
28:19
Spotify, I mean, the labels
28:21
and Spotify can argue over whether
28:23
the stake of that is fair
28:25
and that's a commercial negotiation, but
28:27
ultimately it has created
28:29
a meaningful revenue stream back
28:32
to artists that support artists in the
28:34
creation. And that's what we'd
28:36
like to see flow out of generative
28:39
AI. That's the world that we believe
28:41
is right. Ultimately the creative industries represent
28:44
10% of GDP, more
28:48
than 10% of GDP in the US
28:50
and the UK and many developed economies
28:52
around the globe. And we think the
28:56
economic impacts of undermining that contribution
28:59
to GDP can't be net beneficial
29:01
to society from
29:03
an economic standpoint, if you ultimately just
29:05
allow them to be sucked dry. I
29:08
use the example, everybody's watched
29:10
the movies, The Matrix and
29:12
ultimately what were humans in The Matrix? The humans
29:15
in The Matrix were the batteries. They
29:17
were used to create the power to fuel
29:19
the AI. Well, right now, I feel like
29:21
the creatives are already being consumed as batteries.
29:24
And the media is already being a battery. We are already
29:27
not being valued. We're
29:29
basically just being used in
29:32
order to create one of
29:34
the core pillars that is necessary to create
29:36
generative AI. It's
29:38
just being stripped and taken. And
29:41
I don't wanna see a world where there
29:43
aren't more creatives. I don't wanna see a world where
29:46
creatives aren't compensated for the work that they do. And
29:49
I don't wanna see a world that doesn't have creativity in
29:51
it. We'll
29:54
be back in a minute. Support
30:04
for On with Kara Swisher comes from NetSuite.
30:07
Your business cannot succeed if costs spiral
30:09
out of control, but there's so much
30:12
more required if you want to grow.
30:14
Tracking data, handling HR responsibilities, and mastering
30:16
accounting are just a few tasks on
30:18
every Entrepreneur's to-do list. But what
30:20
if you had a platform that could deal with all of them at once?
30:23
NetSuite is a top-rated financial system
30:25
bringing accounting, financial management, inventory HR
30:27
into one platform and one source
30:29
of truth. With NetSuite, you can
30:31
reduce IT costs because NetSuite lives
30:33
in the cloud with no hardware
30:36
required, accessible from anywhere. You
30:38
can cut costs of maintaining multiple
30:40
systems because you've got one unified
30:42
business management suite. Plus, NetSuite can
30:44
help you improve efficiency by bringing
30:46
all your major business processes into
30:48
one platform, slashing manual tasks and
30:50
errors. Over 37,000 companies
30:53
have already made the move to NetSuite.
30:55
By popular demand, NetSuite has extended
30:57
its one-of-a-kind flexible financing program for
30:59
a few more weeks. Head
31:02
to netsuite.com/on. netsuite.com/on.
31:07
That's netsuite.com/on.
31:14
Support for this show comes from Delete Me. Do
31:17
you ever wonder how much of your personal information
31:19
is out there on the internet? A lot. Chances
31:22
are a lot more than you think, too.
31:24
Your name, contact info, social security number, they
31:26
can all be compiled by data brokers and
31:28
sold online. It's happening right now. It might
31:30
be why you keep getting those insane spam
31:32
calls about your student loans, even if you
31:34
paid them off 20 years ago. And
31:36
worse. You might want to try
31:38
Delete Me to fix that. Delete Me is
31:40
a subscription service that removes your personal information
31:42
from the largest people search databases on the
31:44
web and in the process helps prevent potential
31:46
ID theft, doxxing and phishing scams. I have
31:48
really enjoyed Delete Me largely because I'm a
31:51
pretty good person about tech things and I
31:53
have a pretty good sense of privacy and
31:55
pushing all the correct buttons. But I was
31:57
very surprised by how much data was. out
31:59
there about me. Take control of your data
32:01
and keep your private life private by signing
32:03
up for Delete Me. Now at a special
32:05
discount to our listeners today, get 20% off
32:07
your Delete Me plan when you
32:10
go to joindeleteme.com/Cara and use the
32:12
promo code Cara at checkout. The
32:14
only way to get 20% off
32:16
is to go to joindeleteme.com/Cara, enter
32:19
the code Cara at checkout. That's
32:22
joindeleteme.com/Cara, code Cara.
32:29
This is advertiser content from
32:31
PBS. Jason
32:39
Scott practically grew up at the Pike Place Fish
32:41
Market in Seattle. His mom
32:43
was a fishmonger, and now he's one too. He
32:46
sees firsthand the toll that irresponsible fishing can
32:48
take on our oceans. I grew up here.
32:51
I'm only 51. Are you sure? And there's stuff
32:53
that we can't even get locally anymore because it's
32:55
been overfished. That's not fair. A
32:58
common misconception is that wild caught is good and
33:00
farmed is bad. But Jason says
33:03
it's not that simple. I think as a
33:05
consumer you shouldn't accept if the person selling
33:07
you, you know, your fish or meat or
33:09
whatever it is, doesn't know where it comes
33:11
from or how it's caught or those things. And,
33:13
you know, we're moving a lot of fish. We
33:16
need ways where everybody can eat it and
33:18
we're not depleting the oceans. Whether
33:20
the fish you eat is sustainably farmed or wild
33:22
caught, knowing the difference is the first step.
33:25
In PBS's new three-part docuseries, Hope in
33:27
the Water, scientists, celebrities and
33:30
regular citizens come together to explore
33:32
unique solutions and challenges facing our
33:34
oceans and how we preserve the
33:36
food it provides. In order
33:39
for you to keep enjoying the riches of the
33:41
ocean, you're going to have to also be a
33:43
friend of the ocean. Join
33:45
chef Jose Andres, Martha Stewart, Shailene
33:47
Woodley and Baratunde Thurston as they
33:49
deep dive into the powerful blue
33:51
food technologies that can not only
33:54
feed us, but help save
33:56
our threatened seas and fresh waterways. Stream
33:58
Hope in the Water now. on the PBS app.
34:03
So I want to zoom in a bit
34:05
to talk more broadly about the issue of
34:07
copyright intellectual property and standards. You've called for
34:09
industry standards around AI. What
34:11
do you think those standards should include?
34:13
Watermarking, copyright protections. Obviously the more confusing
34:15
it is, the better it is for
34:17
these tech companies who, this is me
34:19
saying this, could give a fuck about
34:21
creatives. They do not care.
34:23
They do not care. They never have. Well,
34:26
let me start. I mean, there are four pillars that we
34:28
think are important with respect to
34:30
visual generative AI. One of
34:33
those is transparency of training data so that
34:35
you know what's being trained and
34:37
where they're training. And that's not only important
34:39
to companies like Getty Images and creatives that
34:41
are creating works around the globe, but that's
34:43
important as a parent or someone
34:46
with children. If
34:48
I post this image to this social
34:50
media platform, is it being used to
34:52
train outputs in these models? So
34:55
that's one, permission of copyrighted work. So if
34:57
you are gonna train on copyrighted data, you
34:59
need to have permission rights in
35:02
order to do so. We want
35:04
these models to identify
35:06
their outputs in a way that
35:08
is persistent. Now that's likely to
35:10
involve watermarking. It might involve some
35:13
other standards like hashing to the
35:15
cloud, but the
35:17
technology is still in development, but we want
35:19
that to be at the model level.
35:22
So the model actually does the output and
35:24
it's persistent. And then
35:26
we want to make sure that model
35:28
providers are accountable for
35:31
the models they put out into the world. In essence,
35:33
there's no section 230, which kind
35:35
of protected platforms back in
35:37
the original Digital
35:39
Millennium Copyright Act, which
35:41
basically gave them indemnity
35:44
for many claims for content that was posted
35:46
on their platform. We wanna see a world
35:49
where model providers have some level of accountability
35:51
aren't given a government exemption in
35:54
indemnity. We wanna see people
35:56
be accountable. Who would should set up
35:58
the standards in your opinion, Paul, too? industry leaders,
36:00
there's business advocacy groups like C2PA, which
36:03
is the Coalition for Content, Provenance, and
36:05
Authenticity, which includes Google and OpenAI, and
36:07
your competitors, Adobe and Shutterstock. And who
36:10
should be responsible for enforcing them? Is
36:12
it a combination of courts and regulations?
36:14
I'll come back to C2PA, but I
36:17
think it's going to be a combination
36:19
of regulators, of legislators, and
36:22
industry, which is typically how all standards
36:24
evolve over time. I think like happened
36:26
in the EU AI Act, there's a
36:28
general requirement, but the definition of how
36:31
that requirement actually manifests itself is still
36:33
yet to be put
36:36
forward. And that's going to take collaboration
36:38
from the technology industries and
36:40
from creative industries in order
36:42
to get to something that works.
36:45
I think the C2PA, to be very
36:47
specific, it
36:50
is a foundation, but there's still some
36:52
flaws. It is relying upon the end
36:54
user of these platforms to identify the
36:56
generative output versus the model itself. That's
36:58
what happened with YouTube initially, as you
37:01
recall. Exactly. So if you say, if
37:03
I'm going to use a generative model and I'm a bad actor,
37:06
I can say it's authentic and
37:08
label it as such. Well, that's a
37:10
point of failure in the C2PA model
37:12
as it exists today. It's already been
37:14
exploited and we want to see that
37:16
close again. So that's why we want to see these models provide
37:19
the output at the model
37:21
level versus at the user level. It
37:24
also is one that they're trying to
37:26
shift the cost of
37:28
implementing C2PA to the individual versus
37:31
the model provider. There are
37:33
going to be costs if you're producing outputs and
37:36
then you have to store the output
37:38
in the cloud, the original and kind of hash it
37:40
and so that can be referenced and referenced back to,
37:43
I don't think that's a reality where individuals
37:46
are going to be making that investment. So you're
37:48
bearing all the costs. I believe the model providers
37:50
should bear that cost. So do I. We
37:53
are a model provider in partnership with
37:55
Nvidia and Pixar and others and
37:58
I believe that we should create standards.
38:00
standards that are immutable that store with
38:02
it. And I think C2PA is a
38:04
foundation, but I think it's
38:07
one that we need to kind of continue
38:09
to evolve it in partnership with the technology
38:12
industry to get it to something that truly
38:14
meets the challenges versus is a nice kind
38:16
of cover when you get
38:18
called to Congress or Parliament
38:21
or into the EU. Sure.
38:24
That you can just kind of throw it out is that we've got
38:26
a solution. We really want to prevent deep
38:28
fakes. We really want people to
38:31
have fluency in terms of the imagery
38:33
that they're looking at. And
38:35
we think C2PA can evolve to that and
38:37
we're going to work hard in
38:39
partnership with the members of C2PA to try to get
38:42
it there. And ultimately what I
38:44
don't want to see though, you go
38:46
back to some of the original mistakes of digital media
38:48
and such, is I don't want to see it be
38:50
a big cost shift from
38:52
technology industries into the media
38:54
industry where what I
38:56
mean by that is I'm hearing some,
38:58
well, we're never going to be able
39:01
to identify generative content. No, it's very
39:03
hard. Yeah, it's very hard. It's a
39:05
hard problem. That's their excuse. We can
39:07
invent these tools, but we can't invent
39:09
a way to identify. So what we
39:11
want the media industry to do is
39:13
we want them to identify all of
39:15
their content is authentic. So they have
39:17
to buy all new cameras. They have
39:19
to create entirely new workflows. They have
39:21
to fundamentally change how
39:23
they bring content onto their websites.
39:27
That investment, we want the
39:29
media to make. And I don't think that's the
39:31
right solution. I'm not saying
39:33
that Getty Images won't invest in
39:36
technologies in order to better
39:38
identify track and identify provenance
39:41
of content. But I don't think kind
39:43
of saying, well, the solution for the tools that
39:46
were created that create all this generative content and
39:48
are producing the problem that D fakes, the
39:51
creators of those tools don't have any accountability
39:53
for creating solutions to identify. Absolutely. I think
39:55
we have to have a balanced solution. Absolutely.
39:58
It does play into the hands. that actor
40:00
is for sure. That's right. There's a really
40:02
interesting book that I read a
40:04
couple of years back, Power in Progress. And
40:07
it's an interesting study where I think we've
40:09
always been adapted. I'm a pro technologist. Me
40:11
too. I started my career out in the Bay Area. But
40:15
we always assume that just technology for technology's
40:17
sake will be beneficial to society. That's what
40:20
we've been fed and that's
40:22
what we have been taught. And
40:24
their study really looks at
40:26
technology over a thousand years.
40:28
And whether it is innately
40:30
beneficial to society or whether
40:32
society needs to put some
40:35
level of rules around
40:38
it in order to make it
40:40
net beneficial. And you go
40:42
back to industrial revolution and basically
40:45
the first 100 years of the industrial
40:47
revolution were not beneficial to society
40:50
as a whole. They benefited a
40:52
small number of individuals but largely
40:54
gave society nothing more than disease.
41:00
And it took things like regulations
41:02
on limiting work week, regulations
41:06
on child labor. It
41:08
took certain organizations like labor
41:10
unions, et cetera, to
41:13
ultimately get that technology
41:15
leap forward to be something that was
41:17
broadly beneficial to society. And I think
41:19
that is something that we need to
41:21
think a little bit more about is
41:23
not how do we stop AI. Not
41:26
how to, I'm not trying to put this technology
41:28
in a box and kill it. I
41:31
think it could be net beneficial to society. I
41:33
think we need to be thoughtful about
41:36
how we bring it into society. And
41:38
I don't think racing it out of the
41:40
box, absent
41:43
that thought is necessarily gonna lead
41:45
us to something that is net
41:47
beneficial. Yeah, you're gonna lose your
41:49
libertarian card from the boys of Silicon Valley.
41:51
Just so you know, you just lost it.
41:56
We'll be back in a minute. Have
42:08
a question or need how-to advice? Just
42:11
ask Meta AI. Whether you
42:13
want to design a marathon training program that
42:16
will get you race-ready for the fall or
42:18
you're curious what planets are visible in tonight's
42:20
sky, Meta AI has the
42:22
answers. Perhaps you
42:24
want to learn how to plant basil when
42:26
your garden only gets indirect sunlight. Meta
42:29
AI is your intelligent assistant. It
42:32
can even answer your follow-up questions, like
42:34
what you can make with your basil and other ingredients
42:36
you have in your fridge. Meta
42:39
AI can also summarize your class notes,
42:41
visualize your ideas, and so much more.
42:44
The best part is you can find Meta AI right
42:46
in the apps you already have. Instagram,
42:49
WhatsApp, Facebook, and Messenger.
42:52
Just tag Meta AI in your chat or
42:54
tap the icon in the search bar to
42:57
get started. Place the
42:59
most advanced AI at your
43:01
fingertips. Expand
43:03
your world with Meta AI. This
43:08
episode is brought to you by Shopify. This episode is brought
43:10
to you by Shopify. Forget
43:12
the frustration of picking commerce platforms
43:14
when you switch your business to
43:16
Shopify, the global commerce platform that
43:19
supercharges your selling wherever you sell.
43:21
With Shopify, you'll harness the same
43:23
intuitive features, trusted apps, and powerful
43:25
analytics used by the world's leading
43:27
brands. Sign up today for your
43:29
$1 per month trial
43:32
period at shopify.com/tech, all
43:34
lowercase. That's shopify.com/tech. That's shopify.com
43:36
slash tech. Speaking of deepfakes before we
43:38
go, I do want to talk about
43:40
the election and political images. Getty has
43:42
put editorial notes on doctored images like
43:44
the ones of Princess of Wales, Kate
43:46
Milton that were sent from the Royal
43:49
Palace. You said earlier you don't use
43:51
editorial content for your AI database, but
43:53
your competitors like Adobe and Shutterstock have
43:55
been called out for having AI-generated editorial
43:57
images of news events in their databases
43:59
like the... were in Gaza, which were then
44:01
used by media companies and others. And as
44:03
we said before, their databases are also being
44:05
licensed to train some of the biggest AI
44:08
models. So, you know, I think it's
44:10
the expression in software is garbage
44:12
in garbage out. How
44:14
worried should we be about synthetic
44:16
content, basically fake or altered images
44:18
impacting the outcome of elections, including
44:20
ours in November, and
44:23
those images then circulating back into the system
44:25
like a virus? I think we
44:27
should be very worried. I think it
44:29
ultimately, it goes back to some of the
44:32
things that why we're really focused
44:34
in on some of the, you know,
44:36
the premises of the regulatory elements I
44:38
talked about. We want to see a
44:40
situation where we can
44:43
identify this content. We can
44:45
give society as a whole fluency
44:47
in fact, right? If you don't have fluency
44:50
in fact, you don't fact have one of
44:52
the pillars of democracy. So
44:54
I think it is something
44:56
that we should be very worried about. I
44:59
think the pace of AI
45:01
is moving so fast. The ability
45:03
for it to permeate society across
45:05
social networks and those algorithms,
45:07
I think is something that we need to be
45:09
highly concerned about. And I think we
45:12
need industries and governments
45:14
and regulatory bodies to come together
45:16
in order to mitigate
45:19
that risk overall. I'm
45:22
hopeful that we can get there. It's
45:24
a tough problem and it's going to take a lot
45:26
of people putting energy against it
45:28
to solve for it. What can news
45:30
or should news or issues do about
45:32
this specific kind of visual misinformation and
45:34
what responsibly do the agencies have? Obviously
45:36
you are not doing that, but your
45:38
competitors are. Well, I think ourselves,
45:40
I think the AP, I think Reuters,
45:42
I think other news agencies around the
45:44
world, I think we have
45:47
to continue to make sure that our services
45:50
and our content is incredibly
45:52
credible and ultimately
45:54
lives up
45:58
to a standard that is beyond the And
46:00
I think then we need to work
46:02
with the technology industry and again, regulatory
46:04
bodies and legislative bodies on solutions that
46:07
address the generative side of things. Because there
46:09
are bad actors in this world and there
46:12
are models that allow people
46:14
to create not only misinformation
46:16
with respect to elections, but
46:19
some very other societally harmful
46:22
outcomes like deep
46:24
fake porn that
46:26
we need to address. And
46:28
these tools are readily available. What's your
46:30
nightmare and how quickly can you react
46:32
when the Kate Middleton things took a
46:34
few days? It's not the biggest deal
46:36
in the world, but it was still
46:39
problematic. Like that took a while. So
46:42
how do you, it's
46:45
expensive to do this to really constantly
46:47
be picking up trash that everybody's throwing
46:49
down. It's very important that we spend
46:52
a ton of time, first
46:54
off, vetting all of our sources of content. It's
46:56
not something we just take any piece of content
46:58
and put it on our website and then provide
47:00
it out. There is a
47:02
tremendous amount of vetting that goes into the
47:04
sources of content that we have. So
47:07
like the NBC News is or the Bloomberg's
47:09
of the world or the BBC's
47:11
of the world, but all the way down to
47:14
the individuals that we have covering conflicts in Gaza
47:17
or in Ukraine and making
47:20
sure that we know that
47:22
the provenance of their imagery is authentic.
47:24
We know the standard of their journalism is high.
47:28
That takes a tremendous amount of
47:30
effort, time, and investment in order
47:32
to do. And we
47:35
need to make sure that we don't in
47:37
any way, shape, or form reduce that investment.
47:39
We need to increase that investment in the
47:41
face of generative AI. We need to tell
47:43
that story. We need
47:45
to make sure that our customers have,
47:49
who are the media outlets like yourself, know
47:51
that that is something that we're doing.
47:55
Sure. So what's your
47:58
nightmare scenario? Give me a nightmare.
48:00
The primary scenario is that we
48:02
ultimately continue to undermine the public
48:04
trust in fact. And
48:08
we've seen that over
48:11
recent history. It's
48:14
been enabled by some technology
48:16
and platforms. We
48:18
need to make sure that we
48:21
don't restrict the
48:23
public square in terms of debate
48:25
and ideas and different
48:29
opinions, but at the same time that
48:31
we don't feed into an undermining of
48:33
what is real and what is authentic
48:36
and what is fact. And
48:38
those two things to me sometimes get conflated, and
48:41
I think that's a false conflation. I
48:44
think they can both stand as true. We
48:46
can have debate. We can have
48:48
different points of view, but
48:51
we can't have different facts and we can't have
48:53
different truths. And
48:55
ultimately, I think that's the long haul. It's
48:58
not a particular event. It's
49:02
a continual undermining
49:04
of the foundation of that truth. Well, it is
49:06
certainly easier to be a bomb thrower than a
49:08
builder. One of the issues is the fall off
49:10
in economy for all media,
49:12
not just your company, but everybody.
49:15
The costs go up, the revenues go down.
49:17
In the case of media, they suck up
49:20
all the advertising. You see declines in all
49:22
the major media who are fact-based, I would
49:24
say. Getty's value dropped
49:26
by two-thirds since you went public. Now, I
49:28
know SPACs have had all kinds of problems.
49:31
That's a trend in the SPAC area, which is how you
49:34
went public. But even with
49:36
all your AI moves, your shares are down.
49:38
Do you think it's because you're taking a more
49:41
cautious approach than your competitors? And
49:43
what turns it around? I want you to
49:45
speak for all of media, but you're running
49:47
a company here, and you're trying
49:49
to do it a little safer and
49:51
more factual. And that's not as profitable
49:53
as other ways. How
49:55
does that change from an economic point of
49:57
view? And I think all
50:00
of media. has this challenge going forward?
50:02
Well, I can't speak to specific drivers
50:04
of stock price or not. What
50:07
I can tell you is that we aren't taking
50:09
a more conservative view or
50:11
cautious view. We are taking a long-term
50:13
view. We are taking a
50:16
view that we are going to spend money and
50:18
invest to bring services to the market that we
50:20
think are helpful to our customers. We're
50:22
going to invest in lawsuits that we
50:25
think underpin the value of our content
50:27
and information that we have. We
50:31
are going to continue to invest
50:33
in bringing amazing content and services to
50:35
our customers. We think over time that
50:38
is something that will be to
50:40
the benefit of all stakeholders and
50:43
Getty Images, our photographers and videographers
50:45
around the world, our partners around
50:48
the world, our employees and our
50:50
shareholders. That's what we are
50:52
focused in on doing. I
50:55
think there is a backdrop within
50:57
recent trading of the entertainment strikes
50:59
and the impacts of those against
51:01
our business, some of the massive
51:03
impacts against large advertising and some
51:05
of the end-use cases where
51:07
our content gets utilized. Your
51:10
customers are also suffering too, yes. It
51:13
hasn't been a great environment for media companies
51:16
in the recent 18 months. I
51:18
think the end of the average... I would say 10
51:20
years, but okay. But more
51:22
recently, it's been more acute. You're seeing
51:24
streaming services cut
51:26
back. You're seeing ad revenues
51:29
down double digits. So
51:31
there are some challenges out there. But you aren't
51:33
seeing that for the tech companies. No, you're not.
51:35
You're not. Again, that goes where
51:38
I think we need to have some rules to
51:40
the road as we approach generative AI that we
51:42
already learned some of the things that maybe we
51:44
didn't do 20 years ago or 25 years ago. But
51:49
our business is about a long term.
51:51
The images has been around almost 30
51:53
years. And we've focused
51:55
in on making sure that the foundations of
51:59
copyright and... intellectual property remains strong throughout. Yeah.
52:01
And so that's what we'll continue to do
52:03
going forward. The hits keep on coming and
52:05
I don't mean good hits. Then
52:08
I have a final question for you. Are they
52:10
ever going to get fingers right? What
52:13
is the fucking problem? Well, look, I think if
52:15
you use our model, and I know that you're
52:17
pretty sure I had some time with it and
52:19
hopefully you'll get some time with it as well.
52:21
I think you'll see that we largely do. We've
52:24
trained on some things to try to make sure
52:26
that those outputs are of high quality. The
52:28
reality is it does matter. You
52:31
said kind of crap in, crap out.
52:33
Well, I would argue the exact opposite.
52:35
So quality in means quality out. And
52:38
so we've addressed those issues with the
52:40
model. Those aren't going to
52:42
be the hard things. Fingers in the
52:44
eyes. Fingers, eyes, hair, hair lines, smiles
52:47
and teeth. I think
52:49
we've largely solved those items. But
52:51
what we're always going to struggle
52:53
with is solving for the blank
52:55
page issue that customers have. And how
52:57
do you really get to authenticity? I think one of
52:59
the most interesting things to me, and I know we're
53:02
short on time, is how generative
53:04
AI comes directly in
53:06
conflict with another massive trend of the last
53:08
10 years plus, which is
53:10
authenticity. And how
53:13
does authenticity, how
53:15
am I portrayed? How am I
53:17
viewed? Am I positively represented? How
53:20
am I brought to bear? And what do I
53:22
see in the content that's presented to me? Body
53:26
representation, gender representation. How
53:29
did those things come? And I think that
53:32
is still a world that that trend
53:34
is not going away. And I think
53:36
it's a world that Getty Images helps
53:38
our customers address. Although I'm
53:40
still going round after round with Amazon about
53:42
my fake books and the fake Kara Swishers
53:44
who look corny. They just look corny. Strange
53:47
Kara Swishers are all over Amazon. Go
53:50
check it out. Anyway, I really
53:52
appreciate it. This is a really important discussion. People
53:54
understand what's happening to companies like yours. Well, I
53:56
appreciate you making time and thank you again for
53:58
having me. Thanks so much. Craig. On
54:05
with Kara Swisher is produced by
54:07
Christian Castro Roussel, Kateri Yocum, Joly
54:09
Myers and Megan Bernie. Special thanks
54:11
to Kate Gallagher, Andrea Lopez-Druzado and
54:13
Kate Furby. Our engineers are
54:15
Rick Kwon and Fernando Arudo. And our
54:17
music is by Trackademics. If you're already
54:19
following the show, you're authentic.
54:22
If not, be careful that you're not
54:24
being used as a battery. Go wherever
54:26
you listen to podcasts, search for On
54:28
with Kara Swisher and hit follow. Thanks
54:30
for listening to On with Kara Swisher
54:32
from New York Magazine, the Vox Media
54:34
Podcast Network and us. We'll be back
54:36
on Thursday with more. Have
54:44
a question or need how to advice? Just
54:47
ask Meta AI. Whether you
54:49
want to design a marathon training program,
54:51
or you're curious what planets are visible
54:53
in tonight's sky, Meta AI
54:55
has the answers. It can
54:58
also summarize your class notes, visualize your
55:00
ideas, and so much more. It's the
55:02
most advanced AI at your fingertips.
55:06
Expand your world with Meta
55:08
AI. Now on
55:10
Instagram, WhatsApp, Facebook and Messenger.
55:18
If you've been enjoying this, here's just
55:20
one more thing before you go from
55:22
New York Magazine. I'm Corey Sica. Recently,
55:25
our writer Rebecca Tracer noticed something. Republican
55:27
and right wing women have been flourishing
55:29
and prospering in the last year. From
55:31
Marjorie Taylor Greene to Kristi Noem. They're
55:33
tough, they're chaotic, and they tend to
55:36
have really great teeth. They're also swirling
55:38
in the orbit of Donald Trump as
55:40
he seeks to seize the country with
55:42
an iron fist this fall. Rebecca wondered,
55:45
is this empowerment or are
55:47
they just Trump's handmaidens? I
55:49
brought her in to explain to all of us what's going on here.
55:52
Hello, Corey. I'm gonna start
55:54
asking really boring questions. Here's one.
55:56
Which of the many exciting recent
55:59
dustups incited to want to talk
56:01
to and write about right-wing women
56:03
politicians. The first time I floated
56:05
a version of this was after Katie
56:07
Britz post-State of the Union. But I
56:09
can't say that at that point I
56:12
thought like I want to do a
56:14
whole scope of Republican women. I was
56:16
just really into Katie Britz because it
56:18
was very old school in certain ways
56:20
like the kitchen, like it
56:22
was very white suburban, middle
56:25
class, mommy presentation. But it
56:27
was also like, gothic
56:30
horror. You know, there's blood
56:32
of the Patriots right here in
56:34
my kitchen with an apple. But it wasn't
56:36
enough. I wasn't going to write a whole
56:39
piece about Katie Britz. It might have been
56:41
the infomercial that South Dakota Governor Kristi Noem
56:43
cut for the dental work she'd had done
56:45
where I was like, what is happening with
56:48
the Republican women? Right.
56:51
Like, so Kristi Noem did sort
56:53
of a physical self-redivation to make
56:55
herself either more palatable or more
56:57
powerful. I'm not sure. When
56:59
she began, she had a very no-nonsense,
57:02
boxy, Pelosi-esque, Hillary sometimes haircut, that like
57:04
sort of choppy haircut. And then in
57:06
recent years, since she's become a little
57:08
bit of a right-wing star, one of
57:11
the things she's done is really changed
57:13
her look. Now, Donald Trump is very
57:15
open about how he feels about women,
57:17
how he evaluates women. And Noem has
57:20
clearly remade herself into somebody
57:22
who looks like somebody Donald Trump has expressed
57:24
physical appreciation for. So part of my question
57:26
in this piece is, what does political power
57:28
mean if you conform to those kinds of
57:30
aesthetic standards, but then in some way that
57:32
winds up diminishing the respect that the people
57:34
who set those standards have for you? Well,
57:36
your point is a great one that Trump
57:38
hangs over a lot of this. Both soliciting
57:42
him for a big job at the same
57:44
time as they know he has standards. But
57:47
also at the same time, Trump's big innovation
57:49
was like performance is power and they're enacting
57:51
their own narratives. They've all become Trumpy in
57:53
their own weird way. Yeah. And I have to
57:56
tell you that it's very frustrating for me because I read about politics
57:58
and I hate the thing where everything is about I
58:00
always want to make it not about Trump. But writing
58:02
about these women really challenged that conviction
58:05
in me because it is clear that
58:07
at least for some of them, so
58:09
many of the new behaviors they're enacting
58:11
are in response to
58:13
Trump, are about the single demand
58:16
in the Republican Party right now,
58:18
which is showing him loyalty, fealty to
58:20
this guy. Like, all these
58:22
people, Valentina Gomez, Laura Loomer, they're enjoying
58:24
the fruits of choice in career
58:26
and motherhood. Like, does this
58:28
mean feminism won? Ah. Well,
58:32
this is what's so dystopian and scary
58:34
about their project, is that they're all
58:36
doing these things which are really fascinating,
58:38
right? Marjorie Taylor Greene's lifting weights
58:40
in a video and not behaving
58:43
classically demure and all of this
58:45
sense of empowerment is absolutely
58:48
what feminism gave to women. OK, so great.
58:51
Here is its success. But also, the
58:55
party and the ideology that
58:57
these women are using these
58:59
feminist gains to promote is
59:02
openly dedicated to the rolling back
59:04
of those feminist gains. That's
59:07
Rebecca Tracer. You can read her
59:09
work on Republican Women and more
59:12
in your home in our glorious
59:14
print magazine and at nymag.com/lineup.
Podchaser is the ultimate destination for podcast data, search, and discovery. Learn More