November 2, 2025 • 74 mins
Oluwatobi Aluko photo by Praise Samuel
Show Notes:
2:50 Tobi Aluko–Nigerian copyright
5:30 Nigeria’s national AI strategy and creative sector
7:25 Emily Gould – query on tech sector’s strength in Nigeria
9:40 debate over UK’s copyright provision for machine-generated work
11:40 Johan Brandstedt - Sweden’s approach to copyright
13:40 Czech Republic case on AI and copyrightability
16:40 Aluko – prompting versus creating art that involves effort/process
19:00 US copyright for A Single Piece of American Cheese
20:30 Prof. Ahmed Elgammal - background of AI use in the arts
23:45 19th C. French case on copyright of photographs that show originality
24:45 US refusal to copyright AI outputs
26:35 Bartz v. Anthropic and Kadrey v. Meta
30:45 AICAN project
32:55 GAN-based art movement celebrated prior to prompting
34:25 artists survey on value from AI
37:15 backlash of Christies’ auction of Generative AI works
41:50 Elgammal – effort versus intentionality e.g., DuChamp/ready-made art
42:30 intentionality and agency missing with prompting and random output
51:40 AI's cultural impact
1:04:30 Beethoven X project
Please share your comments and/or questions at stephanie@warfareofartandlaw.com
Music by Toulme.
To hear more episodes, please visit Warfare of Art and Law podcast's website.
To leave questions or comments about this or other episodes of the podcast and/or for information about joining the 2ND Saturday discussion on art, culture and justice, please message me at stephanie@warfareofartandlaw.com.
Thanks so much for listening!
© Stephanie Drawdy [2025]
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
SPEAKER_02 (00:02):
I'm all for airing for good and not airing taking a
bunch of stalling creativity.
SPEAKER_04 (00:10):
Welcome to Warfare of Art and Law, the podcast that
focuses on how justice does ordoesn't play out when art and
law overlap.
Hi everyone, it's Stephanie.
And that was artist, lawyer, anddata protection specialist Toby
Aluko from a recent AI and IPpanel discussion.
(00:32):
What follows is a recording fromthat discussion in which Toby is
joined by Johan Branstead, an ITand publishing specialist and
artist rights advocate, andAhmed Alkama, an AI researcher
and creator, recordersprofessor, and director of
records art and AI, and thefounder of Playform AI, a
(00:56):
platform for artists tointegrate AI into their
practice.
These panelists are also joinedby Institute of Art and Laws
Assistant Director Emily Gouldfor a conversation in which we
look at the issues surroundingAI and IPA from a global
perspective, from Nigeria andthe UK to Sweden and the US,
(01:18):
considering both the concernsand the benefits of emerging
technology in the arts.
Welcome to Warfare, Heart andLaw and Second Saturday.
Thank you all so much for beinghere.
It is such a pleasure to bringtogether this panel of
individuals who have a deepunderstanding of the issues
(01:41):
around AI and IP.
I will give a brief introductionof the panel and then we will
jump in.
We have Toby Aluco, an artist,lawyer, and data protection
specialist.
Thank you, Toby, for being here.
SPEAKER_02 (01:57):
It's a pleasure to be here.
SPEAKER_04 (01:59):
Johan Brandstead, IT and publishing specialist and
artist rights advocate.
Thank you, Johan, for beinghere.
SPEAKER_03 (02:08):
Thank you.
Pleasure to be here.
SPEAKER_04 (02:10):
Ahmed Algemal, an AI researcher and creator, uh,
Rutgers professor and directorof Rutgers Art and AI Lab, and
the founder of Playform AI, aplatform for artists.
Thank you so much, Ahmed, forbeing here.
SPEAKER_00 (02:29):
Thank you for inviting me.
SPEAKER_04 (02:31):
Together with this panel, we will launch in.
And Toby, perhaps you couldbegin.
And earlier in 2025, there was anew AI initiative in Nigeria.
So I don't know if you want togive us a bit of background on
that and then tell us what yourthoughts are on the issues and
bonuses that you see coming outof that for artists.
SPEAKER_02 (02:53):
Right.
So Nigerian law on AI.
So in Nigeria, our copyright acthasn't been amended.
The way we see copyright inNigeria is that it has to be the
work has to be created by ahuman being.
That's the way to get copyright,or the work has to be created
for a corporate body.
(03:13):
That's like a company.
That's how you get copyrightprotected.
And also in Nigeria, to getcopyright protected, once the
work is created, it doesn't haveto be registered.
Once it's originally created,it's you own the copyright
already.
That's the way it is in Nigerianlaw.
But the Nigerian Copyright Acthas some um, will I say
(03:33):
loopholes or the way it'sdrafted?
It says that there's a part inthe Copyright Act that's section
two, section two of the NigerianCopyright Act says that for a
work to get copyrighted, therehas to be some effort that has
been expended on making the workto give it an original
character.
So I was looking at this fromthe AI perspective, because I
(03:55):
know in America, um to get AI,AI is not, AI can't own
copyright.
But I think there are some caseswhereby where there's a mixture
of human effort and AI effort,you could argue if the human
effort is more than the AI, ifAI effort, you could argue that
such work can be copyrighted.
(04:15):
So in Nigeria, currently, itsays that if there's some effort
that has been shown into a workto give it original character,
such work can be copyrighted.
So this makes me think that ifwe use AI to create a work and
you also put some human touchesto it, would the work pass as
copyrighted?
(04:36):
Would it be copyrighted?
And well, there have not beenlitigations on that yet to
determine what the cost wouldsay if you create something with
the help of AI.
But my view to that is that AI,AI's work at the moment, can
pass as copyrighted if you canthrow some effort as a human
(04:58):
being, human touches into it tocreate it.
So that kind of creates um, likeI say, a loophole for AI work to
be peripherated in Nigeria.
And um at the moment, eventhough there's no litigation to
see what the court says on it, Ithink that as it stands, AI work
(05:18):
could pass if you put some humantouches into it.
That's the way I say it.
Then also there was like anational AI strategy that was
put forward by the government.
And some of the things I read onit was that with every situation
whereby some works or some jobsthat that could be done by human
beings would be taken over by AIin the future, is it going to
(05:40):
affect like the economy from theperspective of job losses?
So these are like um issues thatare still new to the country and
which we're still watching tosee the way the whole landscape
evolves.
I think that's my perspective onit.
SPEAKER_04 (05:56):
Thank you so much.
It's very interesting politicalissues that are swirling around
how AI initiatives are goingforward.
For Nigeria, what I'd read wasthat they do have this desire to
be a leader of the Africancountries in its AI initiative.
And so I was curious, uh,because they're saying they're
(06:18):
getting all these differentperspectives from the different
sectors.
What do you what do you think?
SPEAKER_02 (06:24):
So in Nigeria, the creative sector is one of the
leading sectors in the economy.
Um, because Nigerian is knownfor, in Africa, for movies, um,
music, and art.
So the way the law stands, uh,because the act hasn't been
amended to take into cognizanceum what AI can do.
(06:45):
So my view to that is that um Ithink um maybe litigations have
to come forward so that thecourts can say whether could
interpret that aspect of thecopyright law, which says that
you just need to show someeffort of originality for work
to have copyright abilities tobe copyrighted.
(07:05):
So I feel that um my view on itis that we need to wait to see
if um cases will come forwardand sections of the copyright
act could be interpreted, givenreference to the current phase
we are in with AI.
SPEAKER_01 (07:21):
Just sort of thinking politically and in
terms of the different voicesthat um are you know are
involved in this debate, whetherthe tech industry is strong in
Nigeria and whether there are AIdevelopers who are maybe
lobbying for more sort ofliberal rules which would allow
them to have freer use ofexisting copyright works,
(07:45):
because I know certainly in theUK that has been a very sort of
powerful debate, and I don'tthink there's resolution yet
between those two sides, whichto some extent are are sort of
pitted against each other.
I mean it's obviously morecomplex than that, but um that
has certainly been uh sort ofthe essence of the debate in the
(08:05):
UK so far, sort of between thetech side, the developer side,
and the creative industry side.
So I was just wondering, Toby,with the saying that the you
know the creative um sectorvoice is is really strong in
Nigeria, whether there is youknow an equivalence strength of
voice from the tech side.
SPEAKER_02 (08:25):
Yeah, in the tech sector, we haven't really had
like uh would I say companieslike OpenAI or in Nigeria
currently, we have like thereare tech companies that use AI
to create um to for theirservices.
And but in Nigeria, um there'sbeen a push for, would I say,
(08:47):
from the part of like personaldata, protection of personal
data, there's been the push forstronger laws and the way you
use personal data in businesses,the way you use personal data in
in the works that creatives orwould I say technological
companies do.
So we don't really have thesame, last I checked, it's not
(09:09):
like at the same level with theUK when it comes to tech
companies pushing for umstronger, or would I say lenient
um routes to create theirtechnology and for lesser um
lesser regulations from the fromthe government.
So at the moment, we have like anational AI strategy in Nigeria,
and I think that's where last Ichecked, that's where it is
(09:33):
currently.
There haven't been issues ofregulators against tech
companies from the side of AI.
SPEAKER_04 (09:40):
Just a wrinkle I would bring up too, since we're
discussing the UK as well, theissues that the UK is seeing
with extending copyright tomachine-generated works and
whether or not that might comeup if indeed other countries
like Nigeria start to do that.
The idea of how you defineauthorship and originality and
(10:05):
how much of a disconnect theremight be if you're gonna
describe uh extending copyrightto an original work, but then is
the original work created solelyby a human in your in your
interpretation of your copyrightact versus extending that to
machine-generated works?
(10:25):
I don't know if anyone wants tojump in on just that point.
SPEAKER_01 (10:28):
Yeah, yeah, I think I think just from the UK
perspective, um, I I don't knowif that was what you were
alluding to, Stephanie, butthere is um quite an unusual
provision whereby there is aspecific copyright for
computer-generated works in theUK Act currently, although that
is very much up for debate.
(10:49):
So the consultation that themost recent consultation that
the UK government published umsuggested that that you know
that debate be opened again.
And is that a right that is usedvery much?
And is it appropriate?
Because obviously, you know, ourour copyright act dates back to
1988, well before AI wasanything like at the state of
(11:12):
development it is now, and so Ithink um it's possible that that
right might go altogether, um,which would put the UK in a much
more comparative position withmost other countries.
I think it's quite unusual tohave that provision.
Uh, I mean, lots of countrieswhich derive their laws very
closely from UK law don't havethat necessarily.
(11:33):
So um, yeah, it's very much anopen debate, I think.
SPEAKER_03 (11:38):
Yeah, I think in uh in Sweden we have a similar uh
principles to Nigeria.
It's you you own it what youmake by default, um no
registration needed, and andthat's kind of where that's the
first major clash, right?
If you need permission first,then we're in a totally
different space.
And then when it comes toauthorship, I believe it's uh
(11:59):
UK, India, and Ukraine, andpossibly a few others that have
these special cases where youcan actually delegate most of
the work to the computer andhave it made for you and still
claim copyright for it, which isvery odd.
Uh it's odd with the basicprinciple of what creates value,
right?
It's the human authorship that'sthe taxable income and the the
(12:22):
value add.
Uh, whereas you don't want toplace incentives around
automation, usually, right?
You that's not what we want tocreate incentives for because it
doesn't build a taxable uhincome base to to basic data on,
more or less.
So I I I think it's superinteresting to to compare these
(12:43):
because uh people usually don'treflect a lot, and then Sweden
is in a lot of cultural ways anda lot of other ways a colony of
the US.
Popular culture is inundated andalso the tech companies, there's
a very close links between theUS and and Swedish Sweden.
But but some of these issuesreally brings to light the
(13:05):
differences at the veryfoundational level.
We're talking about theconstitutions of the states
being built from the ground upvery differently.
Um but I think one thing to wemight throw in here because
we're in the same situation asin Nigeria, that it hasn't
really been tested locallywhether it's what's
(13:27):
copyrightable.
But we do have a thousandregistered hybrid works in the
US and some principles emergingfrom there, which sort of align
with, I believe it was in uhCzech Republic we had the first
European uh court and thedecision on copyrightability,
and they pretty much went withthe principle of well, you
(13:49):
didn't make it, so you can'tclaim it.
What are we even discussinghere, right?
Um, but I think one test that'sworth throwing in here is one
that I came across a lot.
So I have a background in in uhcomics and licensing,
international licensing, uh,from Japan to Sweden, from
Sweden to Germany, from US toSweden.
(14:11):
Um and uh those are very oftenjoint authorship works.
You have a writer, you have anartist, sometimes you have a
bigger team.
Um, and there's the principle ofstarting medium, which I think
is a very good one to bring inuh to get the distinction
between prompting versus otherways of hybrid process to make
(14:33):
an image, right?
And the the principle is verysimple: you can have joint
authorship and shared copyrightsin a children's book that has
pictures and words, but there'salso separate, so the writer
can't claim the pictures and theillustrator can't claim the
words.
So that's that's quite that'sthe test, right?
And that's basically correspondsto prompting, right?
(14:55):
If you started writing a text,you can't claim an image, uh,
and that's really being kind ofokay.
So, how much control over animage can we have via text, even
to begin with?
Uh, and to me it's very clearbecause I've I'm a cartoonist
originally and I've writtencomic scripts and I've also
(15:17):
illustrated comic scripts, whichmeans to me, well, it's obvious
that you write something andthen you get an interpretation,
and this is entirely at oddswith how it's being presented by
the major AI companies.
They want you to feel that youmade what you prompted, right?
And that's how everything isworded: the marketing, the
(15:40):
services.
It all says you created this,and even if you ask ChatGPT
about copyrightability, it willtell 700 million people that
yes, what you just promptedwithout doing anything else,
it's yours to claim.
Uh, so we have an issue therewith common sense from the
(16:00):
perspective of anyone workingwithin the creative industry and
the sort of mental models thatthe tech industry wants to
impose, which is well, this isjust a tool, it learned like
people, and you made what youasked for.
You made what you bought.
That's basically the picture,and it's a very incoherent
picture to anyone who worksinside the field, but to all
(16:22):
amateurs, it makes perfectsense.
So just thought I'd have youropinions on this because this is
something that uh really bugs meand I fight about every day.
Would love to hear hear youryour thoughts.
SPEAKER_02 (16:37):
Yeah, for me, I feel um when it comes to creating
like art, because I create artmyself, I do paintings and I
think prompting and creating artis totally different.
They're totally differentthings.
Like prompting, you don't itdoesn't take the effort it takes
is just to imagine something andit comes into fruition.
You imagine something and youtype it, and it is different
(17:00):
from learning how to create umart, which is usually the I
think the beauty in creating artis the process of creating it,
what you go through, thinkingabout it, putting it on canvas
or on your iPod, because I dodigital art on your iPod.
It still takes efforts tocreate.
So prompting and creating to meare two totally different
(17:21):
things.
And I feel I I I tilt towardsthe um position that those
shouldn't be copyrighted, thatprompting should be shouldn't be
copyrighted because it's it'sgoing to lead to dilution.
That's how I say dilution ofworks.
There'll be like proliferationof what you can't, you can't
really tell whether it's createdby a human or it's created by
(17:43):
machine.
So the law dragging its feet onwhether machines should have
copyright.
I think it's fair.
I feel it should follow thatway, should continue that way
till we totally figure out whatwe're gonna do in the end,
whether machines are gonna havecopyright in the end or it will
forever be humans.
And ITO was the human expert.
(18:04):
But the loophole, like lookingat Nigerian law, which says that
some efforts, since that hasn'tbeen tested, doesn't mean that
if you start with um uh with youprompt an image and you paint
over it, would you havecopyright for that?
Like you prompt something andyou paint halfway through.
That's that's something I thinkof.
Should you should you earncopyright for machine-assisted
(18:28):
work or should it all just behuman?
SPEAKER_03 (18:30):
Yeah, the hybrid processes are like the where
it's really at right now, wherewe had this uh, I think it was
called American Polone orsomething like that.
Uh was a work of art, if youwant, uh legally a work of art.
Uh it was basically promptedfill in the blanks.
(18:52):
You uh they they sketched outthe simple drawing and then had
uh generative AI fill in theblanks, and they got uh what
some US legal experts call athin copyright or a slim
copright, I forget which.
That says, well, the human-addedparts uh are for short
cooperative, and there's reallyno way to set of separating
(19:13):
them.
So it's kind of a proof ofconcept of minimum viable effort
to claim uh an output uh by wayof and and here's the
interesting thing is I thinkthey you're actually in the
starting medium.
So the principle of startingmedium, if we kind of okay, so
it starts with an image, a blankcanvas, but it's really minimum
(19:34):
viable effort.
Um yes, it kind of nominallyyou've composed the work, you've
decided where things go in animage, but from there it's five
minutes of work in total forsomething that looks like it
could have taken a month, and isprobably based on something that
took months or weeks or at leastdays.
(19:56):
Thoughts on that one?
Have you seen it?
Uh in another one?
SPEAKER_02 (20:00):
I haven't seen the baloney.
SPEAKER_03 (20:02):
I haven't seen the I forget if that's the exact
title.
SPEAKER_04 (20:06):
The American piece of Swiss cheese or American a
single piece of American cheese?
SPEAKER_03 (20:12):
Yes, yes, that's it.
Bologna, that's my that's it'sall baloney.
Oh I have a single cheese.
A single piece of Americancheese.
Yes.
I missed that.
Yeah.
SPEAKER_04 (20:28):
I would ask Ahmed, what are your thoughts on all of
this that we've been discussingfrom your perspective?
SPEAKER_00 (20:34):
Yeah, um uh first of all, I mean, um I would like to
remind everybody um and thelistener that uh the use of AI
in making art is not new.
It's have been around for as oldas AI itself, more than 50 years
uh now, where artists have beensome artists have been
pioneering exploring AI in theirown work.
(20:55):
AI itself has evolved a lot, andwhat we see right now in terms
of prompting uh is just thelast, the latest wave of AI,
which creates a lot of problems.
Uh but just if you go like fiveyears uh ago, before that wave
started in 2022, uh many artshave been using AI uh um around
(21:16):
2020.
Um uh we introduced Playform,for example, as a platform uh at
that time, and many arts havebeen using uh AI at that time,
before prompting, and it was avery different process because
you um can train your own AIfrom scratch based on your own
images that capture your ownwork and style and create art, a
(21:36):
very uh creative process.
Uh and then uh came prompting,and and all these issues came
out with prompting.
Um, fundamentally, because um,when you prompt um and you
create an image in a second, andyou think obviously that uh you
created this, uh, that's youhave the impression that you
create this because you writethis amazing prompt.
(21:58):
But I mean, suppose somebodyelse comes with exactly the same
prompt.
Um, so you have two people comewith the same prompt, and and uh
you're gonna have uh probably uhmost probably gonna have very
different images.
But why uh the reason why isbasically because um in the back
uh end of the AI, there's arandom number generator that
(22:20):
just gives a coin and uh usingthe same prompt you end up with
different things.
So the the the agency here isreally um uh missing in terms of
how the word translates to an afinal image is a black box in
the middle.
That that black box has beentrained on trillions of images
and involves a lot of randomnessengineering the image.
(22:42):
So the agency between the inputand output is totally uh missed
out.
Um so that's actually the reasonwhy um the copyright uh office,
uh particularly in the OS, haverejected several cases of uh uh
giving copyrights for uh AIgenerated uh artworks.
And um, however, this uh I don'tthink this will stand for
(23:05):
forever.
I I it all uh I believe that ifsomebody cab was a case where um
they could use AI to create uhnot necessarily prompt, but but
the process really involvedtheir own work and their own
control, the copyright officersmight accept that.
I haven't seen any cases likethat yet.
But the thing is also that uhthat um uh there are many things
(23:29):
in my head here, I want to getstraight.
Um uh that there's a in the 19thcentury, uh the history of
photography, there was a verysimilar scenario where um courts
in the beginning uh rejected togive copyright to photography.
And there was a famous case uharound 1861, 62 in France, where
(23:52):
um famous photographer Mayer anduh Bierson wanted to get
copyright for their work and thecourt rejected in the beginning.
And then after a lot of debates,they finally gave copyright to
photography.
And the main the fundamentalreason was uh um that the artist
has to show that their own umtheir own uh identity is
(24:12):
reflected in the photographythat they are making.
Uh what's different between metaking the same camera and and
having the same shots?
Why gonna be different?
If I if you can show me thatthere is something about my
artist's identity that reflectsin the art, then copyright can
can be granted.
And that's what's missing now inAI.
(24:33):
Uh, how to show that your art'sidentity is there when you are
using AI.
And I guess that depends on whatkind of AI you are using.
And um the uh when you look atnow uh the the courts and and a
copyright office, their standingin this issue is totally
confusing.
Because, in one hand, um uhthere have been lawsuits against
(24:54):
um companies making AI um uh artplatforms uh claiming that the
their work is infringement oncopyrights, and the courts have
been rejected in all these casesso far, um, uh basically saying
that copyright uh generationgeneration is transformative
enough and is not uh uhcopyright infringement, which
(25:16):
would imply that the generationis somehow original.
But at the same time, thecopyright office have rejected
any attempt to copyright thissupposedly original work uh
because you are not infringing.
So that's kind of uh kind of uhcontradicting uh situation at
this point.
And and most recently, um fromJune this year, uh there have
(25:38):
been cases where um uh um uh thecourt said that basically uh
training AI on copyrightedmaterial is a fair use.
Um and that basically makes uhall this company training AI
using other people's uh materialum not liable for copyright
infringement, it's fair use.
(26:00):
Um that's also a very uh oddsituation for artists.
I think now that situation isvery uh uh muddy and and against
artists in all cases.
You cannot use AI to getanything copyrighted.
If anybody else uses your AI umto generate uh to easy AI to
generate things that look likeyour work, you are not
protected.
And if companies use your AI totrain, uh that's fair use.
(26:22):
So it's a very bad situation forarts at this point.
SPEAKER_03 (26:25):
Yeah, and if we look at the Bards versus anthropic,
which is the one you referenced,I think is super interesting
that two days later in the nextdistrict court over, uh it
basically got scolded by anotherjudge saying, Well, your
reasoning is entirely flawed.
You uh and the entire premise ofthe case was what I like to call
(26:48):
anthroxic, uh meaning toxicallyanthropomorphic.
Uh so it's a portmanteau of likeyou bring in this human, you try
to humanize the system and makepeople think of it as a human in
order to establish a mentalmodel where the machine use is
comparable to the human use.
So there I believe what Bart'sjudge in that case is what he
(27:12):
said was well, it's just liketeaching a class of children to
write, and that's it, that's agood, right?
Um, and that that's not an issuecompetition-wise.
Uh and that's entire reasoningwas demolished by the next judge
to over.
But since the plaintiff camewith the anthropic and served
the anthroxic framing from theget-go and said it learns like
(27:35):
people, then all the judge cansay is well, it learns like
people.
They didn't do that in the nextcase over, but there they failed
to uh demonstrate market harm.
And I watched the session, itwas really painful to see
because the judge was basicallybegging them to bring some
evidence of having sufferedconcrete harm from the
(27:58):
plaintiffs in the room.
And I say, I understand theissue, it's about the next
Taylor Swift, it's about buddingtalents.
How do we make incentives forthem?
It's about market obliteration.
Um, that I believe that's theterm he used.
Um, and I would love to give youright, but you're not presenting
(28:21):
the arguments, and I can only goby your arguments.
So some commenters said, well,he basically served the
blueprint for how to argueagainst fair use in the next
ruling, but that won't happenthis year.
Uh it might happen next year.
Um, and he still ended upsaying, well, it's still
(28:41):
transformative fair use based onthese arguments, right?
Based on the facts in this case.
Um so I think we're we're kindof inching closer to the core
issue of this mad frame ofcomparing products to people and
making assumptions on moralityand ethicality and legality
(29:02):
based on products are likepeople.
Uh but it it might be years yet,and we're the markets is being
obliterated in real time.
So artists are basically beingbled out as this plays out in
court.
That's my view.
It's it's bleak.
Sorry.
SPEAKER_04 (29:23):
Thank you.
That second case, uh Kadre, Iwould just point to, I believe
Toby had raised earlier thisissue about market dilution,
which all came up with JudgeShabria's opinion in the Cadre
case.
And as Johan put out there, yes,it is like a blueprint going
forward for plaintiffs tohopefully successfully argue
against transformative fair use.
(29:45):
But yeah, this is definitely aconcerning space that we're in,
at least in the states, inregards to the way the cases are
unfolding.
Going back to you, Ahmed, Iwonder for all of the different
projects that you've been doing.
Doing, knowing that this is thiskind of situation where you can
are not able to uh receivecopyright protection.
(30:08):
I wonder what your thoughts areabout whether or not there is an
issue of creative motivation orlack thereof.
SPEAKER_00 (30:16):
Yeah, um frankly, I mean, when I I did most of this
project, um um I didn't havethat issue in my mind at all.
I didn't, I mean, as a as aresearch scientist or as uh an
artist also um approaching thisproject as a creator, uh I
really didn't think about that.
I I'm not doing that to getcopyright, and I'm uh I'm not
(30:37):
getting copyright that does notprevent me from doing them.
Uh that's fundamental.
Here's uh also a veryinteresting uh project that
relates to this.
Uh back in um 2017, um, we havedeveloped an AI algorithm that
uh looked at um a history ofWestern art and tried to
generate images that uh learnthe rules of art, but at the
(31:01):
same time uh don't uh follow uhany any art movements, try to
generate its own style,basically.
And that was called uh AI canICANN.
Um and and uh it made a lot ofuh media attention at the time,
back in 2017, that was waybefore uh the current generation
of AI art making.
(31:21):
And the headline at that timewas basically finally, things
like finally AI can generatesomething that we like, the
future of art is in danger,things like that.
Really a lot of debate about uhuh AI stepping into the creative
domain and can create somethinguh creative and novel um um and
and not never seen before.
Um uh one thing interesting hereis that um we applied for a
(31:46):
patent for the algorithm, andthe patent was accepted.
So we officially have uh um uhan algorithm that's patentable,
the S patented that create artby itself.
In the same time, copyrightoffice that's often uh issue the
same patents have been rejectedany copyright um for AI
generation from artists.
(32:07):
So that that uh for me it's abizarre situation because uh you
kind of patented an algorithm togenerate art, but in the same
time, you we didn't never try toclaim copyright for the
generation, but it would befunny if we uh asked uh
copyright uh office to um for acopyright for the generation and
how how that would be seen giventhat the algorithm itself is
(32:29):
patentable.
Um so it's it's uh the all theseissues came later.
I said as I said before, uh Imean there was not much debate
about that until really the theBrompt-based AI came around in
late 2022.
Um before that, uh I mean therehave been five years of uh at
(32:50):
least of AI generated art uhmovement called the GAN-based uh
art movement, where there wasactually AI art was have been
celebrated.
Uh there have been AI artexhibitions in museums, in in
art fairs, in galleries, and uhit was very, very healthy, and
(33:11):
and people really liked thatkind of art, um, AI art at that
time.
And I said that there are ahistory of that go back 50 years
uh older than this.
All this debate comes reallywith the prompt-based AI for the
two reasons I mentioned earlier,the fact that uh the
prompt-based AI really comeswith trillions of images of uh
(33:32):
uh that are copyrighted, bakedin, and using to train the
system, and that changedeverything.
Once this came around, we startseeing basically backlash
against AI, and we start seeingAI is being banned from artist
community and and uh uh lawsuitsstarted happening.
Um uh, and really that I have anarticle at the time uh in I
(33:54):
remember in Artnet News orsomething uh that really said
that basically the AI art era isover back in 2022, because
really the the main amazingthings that happened in the five
years prior to 2022 was reallyan art movement uh that where
artists use AI in amazing ways.
Once 2020 came around andprompt-based came around, it's
(34:17):
no longer art.
It's just basically makingimages using prompts.
Uh uh to make it art, to toqualify as art is a totally
different thing.
Um so um, and also I have beenworking with many artists in
that era um until now who havebeen using AI in amazing ways.
And and we we have a paper wherewe actually um we surveyed many
(34:39):
artists who are using AI inaround 2020, again, very early,
and ask them um uh uh why useAI, what do you find as a value
when using AI?
And basically there are twothings uh artists keep repeating
uh in terms of what they theysay.
First, um the fact that uh AIgive them um um um creative um
(35:02):
uh ideas, the fact thatbasically um you can plug in
your images or ideas and and cangive you something that you
never thought uh before about.
Because AI basically doesn'thave the same constraints as us
when looking at images.
And that again was before theprompt-based uh scenario.
Uh, because uh once we addprompt to the problem, now you
(35:24):
kind of render the image withthe lens of the language.
You are constrained only to whatcan be described by language.
Um uh and that's a limitationactually.
Before the prompt-based, uh youdon't have that limitation, you
can re-render the based thepixels based on your own images
without any um uh uh limit ofthe language.
(35:46):
Imagine you are working onabstract uh uh uh art and and
and you cannot describe whatyou're doing in with any words,
and and AI now is limiting toyou to do that.
Uh that's so that the firstthing is that this that the fact
that artists find AI really givethem creative inspiration.
And the second value was um uhthat uh artists always say that
AI gave them the creativevolume.
(36:08):
Um the fact that AI can helpthem in creating lots of and
lots of uh assets and materialthat they can use in their own
project or or they can use intowhatever they are making,
whether they are making video ormaking a mosaic or making uh
whatever project, the fact thatum AI can work as an uh uh
studio assistant for you uh thatyou can uh you can use for that
(36:32):
verbos.
Um so again, I mean, whenartists use AI at that time,
they didn't think much about uhthe copyright issues because
mostly they were using AI basedon their own images, not on
somebody's else's images.
And secondly, they are using AIas part of their own process.
It was not really uh somethingalien to them, it's part of the
(36:54):
process.
I'm using AI as any other toolthat I'm using.
And again, that whole thingschanged where you are just using
prompts and your whole processis just input a text and getting
an image, and and and and uh nowit's really uh problematic in in
many ways.
SPEAKER_01 (37:10):
I think um the there's a really good sort of
example or illustration of howthings change very much with
prompts, but so generative AItools.
Um so there was a there was aChristie's sale of AI works.
(37:31):
Um, I can't remember exactlywhen it was a fair while ago,
but after sort of 2022.
Right, okay.
So you know the one I'mreferring to, and there was a
real sort of backlash within theartistic community, and artists
got together and asked for thesale to be cancelled, but then
there was kind of a backlash tothe backlash by a lot of the
(37:52):
artists exhibiting because theysaid exactly as you were saying,
Ahmed, we are using AI as thiscreative tool, and we see it as
a way of you know progressingour work, moving moving uh the
artistic canon forwards, if youlike, uh within the digital art
sphere.
And so I think that I think thatdoes illustrate really sort of
(38:13):
clearly the the sort ofdifferent um approaches and the
way you know there isn't kind ofa one-size fits-all when we're
thinking, I think, aboutcopyright.
In we can't sort of generalize,I suppose, I don't think,
really, when we talking abouttalk about AI generated art
because it can be so verydifferent, you know, it AI might
(38:36):
be just a simple tool forsomebody using a um a generative
AI algorithm versus you knowartists who are working as
artists all day, every day, andusing a very different kind of
tool that they have created byusing often their own work and
you know by sort of curatingthat tool in the first place.
SPEAKER_03 (38:59):
Absolutely.
Yes, and I think the backlashwas about kind of this
piracy-based uh image modelsbeing fronted and invading the
space very obviously from theget-go.
And I I really think they didthemselves a disservice by
hosting this and and muddyingthe waters by mixing up people
(39:19):
like Rafi Canadol, who has donethis custom software that makes
custom living images, and it'svery much bespoke and the uh
artistic process, versus someonewho prompted something in stable
diffusion and then stitched somepictures together.
Um, and I guess I mean the youcan still copyright the work
(39:41):
that's based based on stolenpaint or stolen images, um, as
long as you add value to it.
But I think the red flag for redsheet for a lot of people was
those specific image modelsbeing used.
Um so yeah, there was aninteresting discussion after.
And I I I I'm totally on onboard with protesting against it
(40:05):
because it we need to debate andget to the point to the core of
what are the real issues here.
And I actually debated a bitwith one of the guys who had uh
an artist who makespseudophotography, calls it
post-photography, uh, where heuses these kind of models to
(40:27):
make something that's veryclearly photorealistic, but also
very clearly not photos, becausethings break apart at a closer
look, and he kind of uses thatconsciously.
Uh, whereas the issue is takingthe shortcut and never mind the
details, and just getting thework done in the smallest time
possible.
(40:47):
So he's putting himself in apretty tricky spot by using
these things, but I think theposition is defensible because
he really uses the tools and thecontent and the process in a
very conscious way to specificends uh creatively.
So yeah.
SPEAKER_02 (41:07):
Yeah, for me it's uh I'm also on the view of about
the process.
Like art is not just about thefinal image, it's about the
process and how you make it.
So, like um Ahmed said,prompting prompting is actually
a big issue because I don'tagree that uh putting images,
but what if you do a bunch ofprompting and you gather your
(41:30):
prompts as like a mosaic?
Sort of is could that pass outsome effort to create art if you
um clip different images you'veprompted together and as a
finished work to createsomething?
That could pass out some effort,or I don't know what do you
think about that?
SPEAKER_00 (41:48):
I don't think I mean the effort should be the the
metric here.
I mean, uh we are a hundredyears now after uh Marcel
Deschamps have his his fountainuh as an art, basically the
unary thing that he did as asculpture, and and that opened
the the whole era of ready-madeart, basically, when you can use
an object and and declare anart.
(42:08):
And he had a point basicallythat artist can can make
anything and call it art.
That's how his whole point.
And so the to come now 100 yearslater and and still talk about
the effort and the time spent asa as a metric to qualify
something to be an art or not, Idon't think that's right.
I I think uh the main issue hereis really um um uh
(42:31):
intentionality, obviously, isone important thing.
Um, and definitely Marcel Schamhad the intentionality of making
art this way, uh to make amessage.
Uh, when you're prompting, uhthe intentionality is broken.
Uh part of it, yes, you you youwrite a prompt with intention of
something, but you actually geton the other end is something
totally up to the uh torandomness in the process.
(42:53):
And and uh you might like it ornot, you might start tweaking
the prompt to try to control itto what you want.
But most of the time you end upwith something that you like.
So your role is more like acurator.
Uh, you start with with someidea, you got something, it's
not really what I like, but Ilike what I get.
Uh, and and here it is.
I mean, uh, so that's what'smissing in the process, the fact
(43:15):
that uh the agency is is notcomplete.
Um uh part of it is random, partof it is is uh is mashing up
millions of images uh based onlanguage to generate something.
Um but it's not about the effortbecause it can actually take you
a lot of effort to create someimage with a prompt.
It can a bit of trial and erroruh to to try to really control
(43:36):
it because we are very hard tocontrol.
So the effort is there, the theagency is not there, the
intentionality is what'smissing, or or or not 100%.
SPEAKER_03 (43:44):
Yeah.
One of the examples I like to goback to is uh I mean, generative
art has a prehistory to digitalart, right?
There were spiral graphs andthere were paint spinners and
people experimenting with thatin the in the 60s to figure out
some of these things, right?
Um, and the principle ofstarting medium is a is a good
(44:07):
one, but one that came out ofthat whole line of
experimentation where they didthings like okay, so here's a
paint spinner, we invite thevisitors to pour in the paints
and then uh press the button orspin the wheel, and we get an
original picture and we put iton the wall.
And then you get the same corequestion that you have today:
who made the picture?
(44:28):
Is it the guy who made themachine and patterned the
machine?
Is it the user or is it themachine?
Right?
Who is the real author here?
Uh and the principle ofindependent origination came out
of that, if I'm correctlyinformed.
And that says, well, if a laymanlooks at the pictures on the
wall and he says, Well, I can'ttell apart that these are being
(44:51):
done by different people, thenthere's no basis for a claim of
human authorship because there'sno originality in it.
Uh, and and it's kind of twolayers to it.
One is can you tell it apart atall?
And then the next layer is canyou tell apart that specific
people have made this?
(45:12):
Because if you can't, it doesn'texpress the personality of the
author, right?
Um, so it's one is yellow andone is green.
That's not enough personality,even if one person really likes
green, right?
Um, so I think that's a goodone.
And also if you look at thosethree, like user, machine, and
uh and the machine maker, uhthere's a really good paper, I
(45:34):
forget the title then, but thatreally went to the bottom of
from a societal perspective, ifwe look at copyright as an
incentive machine, an incentiveframework, if we place the
incentives around the user'seffort, the machine effort, or
the the machine maker, likelet's just look at what happens
(45:59):
if we give output rights to theguy who made the machine, to the
machine itself, or the user.
And the conclusion out of allthis was well, it's not about
who, it's about we don't want toincentivize automation, we don't
want to incentivize volume ofempty work.
(46:19):
We want to actually incentivizethe process, as you mentioned,
Toby.
That's where the value iscreated.
Um, and also as a matter ofscale, which they it's a few
years back and they kind ofpredicted where we are now, is
that if you incentivize emptyautomation, then you get
(46:40):
system-wrecking volumes.
So it doesn't really matter whogets to claim all of all of
these outputs, if there's anincentive to create more and
more and more, you get too much.
And market dilution is onething, but there's also a
democracy aspect and giving upon truth altogether and telling
(47:03):
what's real.
I mean, it's a it's it's a muchlarger issue than just the
incentives around art.
It becomes can we even navigatereality if we're surrounded by
fake pictures?
So yeah.
Um, and I to your point there umabout the patterns for the
(47:24):
machine, that to me sounds a lotlike oh, I forget now, but uh
the gates of paradise is thename of a work that's entirely
machine generated.
Um, so there was no human in theloop at all.
It's a randomizer that producesuh generative works.
And the guy uh I think it'sStephen Thaler, right?
(47:47):
Is on the mission around theworld to get authorship rights
for machines.
So that's a I see either he's alunatic or he's kind of pressure
testing the legal systems aroundthe world.
Either way, that's how it endsup working.
It's it's how it does the systemhold.
Um and I believe they assignedauthorship rights, I believe
(48:10):
maybe in South Africa.
Um, and I forget now because hedid a patent machine and he did
an art machine, uh, and he'strying to register the outputs
of both of them around theworld, right?
Um, and I believe he actuallymanaged to get one of his
inventions or artworks patentedor copyrighted, I forget which
(48:31):
one.
Uh, but they ended up notdesigning the copyrights to the
machine that he asked for, butto him, right?
So the author of the machine gotuh patents to the outputs, which
is an anomaly, as I understandit.
Um but yeah, uh so but I I guessone of the uh courts when they
(48:53):
assessed uh his attempts toregister the gates to paradise,
the randomized works without ahuman in the loop.
It was uh well, we alreadyincentivized you, we already
provide protection.
You're the patent owner of themachine that you made.
That's your incentive, right?
(49:13):
You don't need additionalincentives in claiming also the
outputs of the machine for thereasons discussed just now.
So I think that's how how thatstory ended up for them.
But as I understand it, this guygoes around and tries to
escalate.
And you have Jason Allen alsodoing his space theater opera.
(49:33):
Uh, one of my favorite kind ofone-liners from one of the court
rulings when he tried tocopyright this mid-journey
image, which looks like ageneric mid-journey version 3
image.
Um, is like, well, you're not anartist because you found a
pretty stick in the woods.
It doesn't matter that yousearched for 600 days, you still
(49:55):
didn't make the stick, right?
Um, and that's it's a bit mean,but uh there's a core of truth
to it, right?
And and I believe that one wasokay.
One part was he did all theseiterations until he found
something he liked.
And the next part was he didsome minor edits, and I believe
they assessed these edits to notadd any value, uh basically,
(50:19):
it's marginal adjustments, andthey say, Well, that's not
enough.
You didn't add any creativity inthat step, you just tweaked it a
little bit in the edges, so nocopyright for you.
That's I believe so.
Those are some of my favoritecases to track Steven Tyler and
Jason Allen.
And um, and they teach us a lotabout like what's the underlying
(50:44):
principles here uh that we'refinding out as we go on.
SPEAKER_04 (50:48):
Also, that whole layer of the training questions
that go into mainstream models,like the mid-journey models.
What are the infringing aspectsof those outputs?
Like, that's a whole otherconversation that's really not
even being factored in manytimes when they're having that
can we copyright an outputquestion.
(51:11):
That's what I love about modelslike Ahmed, like what you have
been talking about, where youcan train your own model.
And that to me, it really doesspeak to an ethical approach to
this.
Well, unless there are anyquestions about that, I also was
just gonna dovetail into thisquestion that Ahmed had raised
as a point about culturalstagnation, which I think
(51:34):
everything that we've beentalking about kind of smacks of
questions about where is ourculture going.
So, anybody want to jump in onthat?
SPEAKER_00 (51:41):
Yeah, let me start.
Um, um uh yeah, um I think uhsince um 2022, um Chat GPT and
and uh all prompt-based imagegeneration came around, um,
definitely the line betweenwhat's real and what's fake
become very blurry.
Uh so since then we have seenlots and lots of um images in
(52:05):
social media that are fake.
Um every day now it's uh beinggenerated.
Very good quality, obviously.
It's very hard to tell videosnow, real videos from fake
videos.
Uh right after Shad GBT camearound, there was a nice story,
funny story that um Amazon um uhthere was many Amazon books, uh
bookstores uh online just makingfake books and selling well,
(52:28):
actually, these fake books.
Um uh so uh really the the thethere are two big big problems
with that, obviously.
The problem of um that the linebetween what's real and what's
fake is totally blurred at thispoint and gonna be more blurred
to uh to in the future to thepoint that we will not really
know what's real and what's fakeanymore uh in any medium.
And the other thing is that allthis fake stuff is gonna be uh
(52:54):
um by scale, gonna be themajority of content on the
internet and and and and andeverywhere.
And that means that the nextwave of AI models will be
generated with more of AIgenerated images and videos and
text than human-generated ones.
Uh, because one of the problemswith AI now is that the lack of
(53:16):
data already digested all theinternet, so they were looking
for more synthetic data, more AIgenerated uh data, you're
welcoming that to train andskill their model.
But then we have a problem thatnow we are uh going into culture
with technicians because uh themajority of content there will
be AI generated.
Um, and um what that do to ourculture, that's really um a big
(53:40):
problem.
SPEAKER_03 (53:41):
Yeah, the if I may, there have been this discussed
to one of my pet peeves.
That that's again like what whenwe see the mainstream media
reporting on Sora, for instance.
Uh there are a couple offramings.
Uh one is can machines be heldliable?
Like, is the bot to blame, isthe AI to blame?
(54:04):
That's one of the framings,right?
People are really strugglingwith this, and the other is
users are using this tool to doX, Y, and Z.
And to me, that's the othercrazy frame.
If you have the maker and themachine and the user, right?
One is about blaming themachine, one is about blaming
(54:24):
the user.
But the user can't make this,and the machine can't make this
unless there's a serviceprovider providing this live
service, actually producing thework, and nobody points in that
direction.
Uh, and that's it's baffling andhorrifying, frankly, to me, that
nobody says, why aren't thesemedia production companies being
(54:46):
held accountable for what theyproduce?
That's absent in a lot of thediscussion.
And I I just don't understandit.
And the the best conspiratorialthinking is they have been
really good at PR, atestablishing some frames of
thinking around uh liable humanmachines and about users getting
(55:08):
tools that make that they use tomake things, uh, and avoiding
the obvious thing that is techcompanies moving into media
production.
That's what they're doing,right?
Except without the input costsand the output liabilities.
(55:28):
So they manage to kind of havethis old mental frame of these
toolmakers or these neutraldistributors, despite now being
producers.
And so this discussion becomesone of the DMCA, like, okay, so
the platforms can't be heldaccountable, right, for what
(55:50):
flows through the platforms, andwe've already seen what happens
if it sucks all the money out ofmedia, the traditional media,
and everybody else out of theirpersonal data, and you have an
ad monopoly and a socialdistribution, media distribution
monopoly, right?
Um, and now they're moving intoto even compete directly against
(56:14):
legacy media by producing theirown media based on theirs
without paying for it.
Uh, and we're still stuck inthinking about them as tools and
as distribution platforms wherereally they are production
houses now, uh, fully automatedproduction houses.
Um, yeah, uh, so I I think justplacing those the product
(56:38):
liability and the serviceprovider responsibility in the
right place is where we need tostart if we're going to get
anywhere with this.
SPEAKER_02 (56:46):
Yeah, I think my view on the my view on the
culture of stagnation is that itcan also take incentive away
from because when there is likeproliferation of AI
creator-generated work all overthe internet, if you're like a
graphic designer, it's it takesjobs away from graphic designers
because now you could just useuh an AI app to create graphics
(57:12):
and you no longer need to paygraphic designers.
And it kind of takes awayincentives from uh skills like
skills like graphic design, likebecause if you could uh just use
uh AI to create your to brush upyour images, you don't need to
pay photographers to take imagephotos of you.
You can use AI to uh brush upphotos to make it clearer and
(57:35):
fine.
Now you could use uh these appsto create designs and uh it kind
of kills that's what I say.
The incentive for people tolearn some of these skills that
we've been uh doing for years,that's how I say.
SPEAKER_03 (57:49):
Yeah, absolutely.
Uh and to me again it it's uh itgoes back to the framing.
Uh, if we think that an AI useris doing the same work as the
one who produces the originalwork that's being used as a
training material, or if wethink of the machine as the
actor, then we we miss theaspect of well, no, these are
(58:12):
the service providers, the nowmedia production companies
competing directly against theartists and graphic designers
and writers and translators andall of those, right?
Uh but they're dumping themarket, they're not selling
these services at nearly thesame price level.
So when OpenAI produces acartoon image, uh it does it in
(58:38):
seconds, right?
Fractions of a second, and itcompetes directly against the
work of whoever made theoriginal drawing who needs hours
to finish the same thing, andthey sell it for cents rather
than selling it for the well,maybe a hundred bucks or
whatever they would need tocharge to to make a living off
(58:59):
of making cartoons, right?
Um and one part of it isautomation.
That's fine.
I mean, it it does the workfaster, but the other is
sourcing, right?
Uh the cartoonist needsreference material, they need
production assets, they needtools, they need all of these
business inputs to produce theimage.
(59:22):
And this trillion dollar companyhas the competitive advantage of
not paying for any of it.
They get tax breaks for the datacenters and they they scrape the
internet to get the productionassets for free.
So it's really an even unevencompetition situation.
And to me, it's actually the bigpicture issue is more antitrust
(59:46):
and comp fair competition lawthan it's about copyright.
Because again, they don't owntheir output, but that doesn't
really matter.
You don't need a copyright tosell anything, you can still.
Sell bespoke images made in aninstant, right?
In a lot of business context, itdoesn't really matter if
(01:00:08):
somebody else has the same work,right?
You you would go to a stockagency or you would license
pre-made work anyway.
You don't need the exclusivity.
And you're not really using thepicture as your core competitive
thing.
It's maybe maybe used formarketing or as a web page
(01:00:28):
filler or whatever.
It doesn't really matter ifsomeone prompts the same headset
girl for your support page,right?
So yeah.
So to me, it's it's uhdefinitely a case of Microsoft
now competing directly againstthe Dartists they took from
Microsoft and OpenAI, not theirusers, Microsoft directly.
SPEAKER_00 (01:00:50):
Except that um that um Microsoft and OpenAI does not
at this point uh uh makeentertainment and production
their their main business modelor or thinking this way AI and
AI focus on AGI and Microsoft asa software company.
Uh however, we have to rememberuh how Netflix started.
(01:01:13):
Uh when Netflix started, we werebasically mailing CDs in the
mail and we're mainlyrecommendation system for
recommending you the next movieto watch.
And we evolved into the majorproduction company now of shows
and killing all other studios.
Um, so that's really um um howthings went um uh for Netflix.
(01:01:37):
Uh can we imagine major AIcompanies uh going into uh into
the direction of mediaproduction and abandoning the
idea of AGI?
I don't see it yet.
Probably there are somecompanies who will come up uh uh
and focus on the production inparticular.
It might be again Netflix or orthe Netflixes of the world where
(01:01:58):
basically they are wellpositioned to use AI uh in
production in the next level.
But um the problem was thatmajor AI companies have other
missions they're trying to focuson, thinking that these are
bigger business models toachieve, including automation of
the world, taking the jobs ofother people and tools and
(01:02:18):
robotics and things like that.
Uh so um they are not reallyfocusing on uh that
entertainment production angle.
SPEAKER_03 (01:02:27):
This is so interesting to me because I
agree that that's how theyposition themselves and how they
push things forward.
But when you look today at whatthey actually do, if I have 20
bucks a month to spend onpictures, I can spend it on
shutter stock, I can hire afairly rookie illustrator, I can
(01:02:48):
go to OpenAI.
So even though they're aiming tobe a tech and presenting as a
tech company in practice, theyare competing already today.
Uh, and we see them also movinginto media.
OpenAI is funding an animatedfeature field now.
Google is starting their ownproduction house.
So what we're seeing is thatthey're kind of shedding the
(01:03:12):
pretense of being a techcompany, they're spinning off
this purely media-centeredplace, right?
Where automation and everyoneelse's property as free business
inputs is the starting point.
And if you have that as yourhead start, then you can get
pretty far and make some prettybad movies until you get the
(01:03:33):
hang of the rest.
If you look at Altman and whathe's doing, he says, Well, I'm
making a religion, I'm making acult, and I'm incorporating it
as a company.
It's all about getting people tothink more about the future than
they hear now.
Uh, what might this be nextyear, right?
What might might it be in twoyears from now?
(01:03:54):
Rather than looking at, well,what are you doing right now?
You've already wrecked basicallyacademic research, um news
media, and you know uh you'redumping the market for
illustration and translationservices, but it's always about
looking into the future andseeing what's next and
forgetting about this pettysquabbles about the here and
(01:04:16):
now.
So it's fascinating to me.
SPEAKER_04 (01:04:21):
Thank you.
One upside I see of our culturegoing forward using AI, I would
just uh bring back to aBeethoven project that I learned
about through Ahmed.
SPEAKER_00 (01:04:38):
I haven't heard about this.
It's it was quite famous inEurope, not in the US,
unfortunately.
Uh so uh Beethoven uh uh madenine symphonies and uh he left
some sketches for a 10thsymphony, very rough sketches
for a 10th symphony.
And back in 2020, 2019,actually, uh Deutsche Telekom um
commissioned me and a group ofuh musicologists and a music
(01:05:01):
historian to try to uh come upwith a version of Beethoven 10th
symphony based on the sketchesand using AI.
I was the AI lead in thatproject.
Uh that was a celebration ofBeethoven 250th anniversary.
Uh COVID came obviously anddelayed everything, but back in
2021, we made the concert um uhuh in Bonn uh played by uh
(01:05:23):
Beethoven uh Symphony Orchestra,where we finished two movements
uh of uh of uh the symphonyusing uh AI.
It was really uh binding projectuh using the same kind of
technology that came late to beknown as ChatGBT, basically the
same language models, uh, tolearn about uh Beethoven style
(01:05:43):
of composition andorchestration, and now feed
these sketches and try to seewhat kind of composition the AI
can come up with.
And and it was really a goodproject to show the potential of
AI.
AI was a building block in thatproject, but didn't cancel the
human.
I mean, there was a humanmusicologist, composers, they
(01:06:04):
are historian, all the teamworked together with the AI to
come up with something, and itwas a creative process where you
use AI to come up with somecontinuation, for example, and
then the composer would look atthis and and approve it or not,
or come up with other ideas,give it to the AI again.
It was really a very goodproject to show how AI can be
(01:06:25):
used as a tool in the artistictoolkit of uh an artist or an
artist team to do somethingrather than just uh uh having a
black box that you prompt andand you get a final answer uh
that that uh really isproblematic.
So um that the the completion ofBeethoven X is available on
(01:06:47):
YouTube.
You can uh find it and listen toit and enjoy it.
Um uh it's it's it I mean I findit inspiring and in terms of uh
um listening to a beast thatBeethoven never wrote, but still
based on his own notes.
It's really interesting.
You can tell it for yourself,search for it.
Thank you.
SPEAKER_03 (01:07:08):
That's super cool, and I mean to not be the David
Honor here being the uh publicdiffusion is an interesting
project, right?
Which starts from the samepremise.
We we'll starting with fromcleared public domain images and
making an image model out ofthat.
That's still problematic forartists working in traditional
(01:07:29):
styles today.
If they manage to produce asgood input, we still have an
issue with automation, but wedon't have an IP issue, right?
Um, and it's uh I see a lot ofpotentially similar use cases
where you can expand on ourshared history.
But again, you need to sort ofhave output watermarking or
(01:07:51):
something like that, so you oror we will just confuse
ourselves by inventing a lot offake history, uh fake art
history.
But I think it's a it's aworthwhile project and it's
interesting, and there are manylikes it.
Uh there was another one inJapan where it took the the
grand old master of of manga,Osama Tezuka, um, and worked
(01:08:14):
with his estate to digitize hisentire archive and see if they
could have it come up with newcomic episodes in his style.
But similar to what youmentioned, the estate had
approved beforehand, and his oldassistants uh and his old
publishers were all involvedcreatively in the project to
(01:08:36):
make sure that they theyactually completed it with a
human touch and in a good way.
I don't think it wascommercially successful, I
haven't seen it, so I don't knowif it was creatively successful,
but it shows a good example ofhow to do it the right way.
Everyone approving and everyonebeing contributes in a creative
(01:08:57):
human way to keep a traditionalive.
It's about building on what camebefore, rather than
cannibalizing not onlyeverything that came before, but
everyone today as well.
SPEAKER_04 (01:09:09):
So yeah, and being transparent about the source
that there was that AI elementto it, then no one is misled.
SPEAKER_03 (01:09:17):
Absolutely.
Yeah, the three C synatry,credit consent, compensation,
that transparency.
SPEAKER_04 (01:09:24):
Yeah, yeah.
Toby, what would you think?
I I'm just curious because I Iam curious to uh use an AI model
built on my work, and and sowould you or have you done
anything like that?
What are your thoughts on that?
SPEAKER_02 (01:09:40):
So I think that's good.
I've because I also writefiction sometimes.
I play with stories, and sosometimes sometimes I feed my um
work into like a temporary chatGPT, the temporary part, and
hear suggestions on tweaking orviews.
Like if I can get a humaneditor, I could use AI trained
(01:10:04):
on my own work to give me viewsor feedback on what I could do
better.
And sometimes the suggestionscould be good and they make
sense sometimes.
Like when I write fiction, I useAI to just give me feedback.
And I think that's like apositive way to use it, to give
you feedback and um what youcould tweak.
(01:10:25):
And yeah, I do I feel that'sgood, but um not like the way
Johan said, not in thecannibalistic nature of AI
taking away our creative agentsto create stuff, but if we could
use it to assist and to um showus other ways or other
perspectives, I think that couldbe good.
SPEAKER_04 (01:10:47):
I would love it if each of you offered a closing
thought.
So please.
SPEAKER_03 (01:10:51):
It was really interesting to hear your your
perspectives and uh hope we cankeep the discussion going.
Thanks for the lot of uh goodexamples that came up that I
wasn't aware about and dig intothose.
So I look forward to trying outplay for.
It sounds like a lot of fun.
SPEAKER_02 (01:11:08):
AI is fed with uh what creatives already created.
That's what you used to trainthe machine.
So if you're taking awayopportunities for creatives to
make money, I think so.
In the future, would AI betrained on what AI is already
produced, or if like creativesare not like having
opportunities due to like thecannibalistic nature of the way
(01:11:33):
of prompting and the rest.
So I just feel that if we couldreach like a middle ground
between original work andgenerated work, I think that
would be better for the future.
SPEAKER_00 (01:11:46):
And definitely coming to uncharted territories
as we go and and uh lots ofquestions being raised every
day, and lots of new data pointscoming, and uh we have to keep
this discussion all going on anduh and uh continue.
SPEAKER_04 (01:12:03):
Emily, any thoughts?
SPEAKER_01 (01:12:06):
Oh, I think the main takeaway is that uh as Ahmed
said, it's more questions thananswers at the moment, but I
think I guess on a positivenote, uh and it's a bit trite,
but I think things will kind ofwork themselves out in the end,
just like you know, newtechnologies like photography,
and certainly in the kind of inthe narrow sort of IP world, I
(01:12:30):
think things will work out, andthere'll be kind of licensing
schemes, and um in time thingswill work out.
That said, you know, AI moregenerally, um there are so many
kind of questions, and I thinkthe answers to those questions
will very much sort of shape theworld we live in and we boss on
(01:12:54):
to future generations.
SPEAKER_04 (01:12:58):
There will be links in the show notes to learn more.
If you want to be able to bemuch pretty much time, this is
the thank you so much forlistening.
I'm all for airing for good and not airing taking a
bunch of stalling creativity.
SPEAKER_04 (00:10):
Welcome to Warfare of Art and Law, the podcast that
focuses on how justice does ordoesn't play out when art and
law overlap.
Hi everyone, it's Stephanie.
And that was artist, lawyer, anddata protection specialist Toby
Aluko from a recent AI and IPpanel discussion.
(00:32):
What follows is a recording fromthat discussion in which Toby is
joined by Johan Branstead, an ITand publishing specialist and
artist rights advocate, andAhmed Alkama, an AI researcher
and creator, recordersprofessor, and director of
records art and AI, and thefounder of Playform AI, a
(00:56):
platform for artists tointegrate AI into their
practice.
These panelists are also joinedby Institute of Art and Laws
Assistant Director Emily Gouldfor a conversation in which we
look at the issues surroundingAI and IPA from a global
perspective, from Nigeria andthe UK to Sweden and the US,
(01:18):
considering both the concernsand the benefits of emerging
technology in the arts.
Welcome to Warfare, Heart andLaw and Second Saturday.
Thank you all so much for beinghere.
It is such a pleasure to bringtogether this panel of
individuals who have a deepunderstanding of the issues
(01:41):
around AI and IP.
I will give a brief introductionof the panel and then we will
jump in.
We have Toby Aluco, an artist,lawyer, and data protection
specialist.
Thank you, Toby, for being here.
SPEAKER_02 (01:57):
It's a pleasure to be here.
SPEAKER_04 (01:59):
Johan Brandstead, IT and publishing specialist and
artist rights advocate.
Thank you, Johan, for beinghere.
SPEAKER_03 (02:08):
Thank you.
Pleasure to be here.
SPEAKER_04 (02:10):
Ahmed Algemal, an AI researcher and creator, uh,
Rutgers professor and directorof Rutgers Art and AI Lab, and
the founder of Playform AI, aplatform for artists.
Thank you so much, Ahmed, forbeing here.
SPEAKER_00 (02:29):
Thank you for inviting me.
SPEAKER_04 (02:31):
Together with this panel, we will launch in.
And Toby, perhaps you couldbegin.
And earlier in 2025, there was anew AI initiative in Nigeria.
So I don't know if you want togive us a bit of background on
that and then tell us what yourthoughts are on the issues and
bonuses that you see coming outof that for artists.
SPEAKER_02 (02:53):
Right.
So Nigerian law on AI.
So in Nigeria, our copyright acthasn't been amended.
The way we see copyright inNigeria is that it has to be the
work has to be created by ahuman being.
That's the way to get copyright,or the work has to be created
for a corporate body.
(03:13):
That's like a company.
That's how you get copyrightprotected.
And also in Nigeria, to getcopyright protected, once the
work is created, it doesn't haveto be registered.
Once it's originally created,it's you own the copyright
already.
That's the way it is in Nigerianlaw.
But the Nigerian Copyright Acthas some um, will I say
(03:33):
loopholes or the way it'sdrafted?
It says that there's a part inthe Copyright Act that's section
two, section two of the NigerianCopyright Act says that for a
work to get copyrighted, therehas to be some effort that has
been expended on making the workto give it an original
character.
So I was looking at this fromthe AI perspective, because I
(03:55):
know in America, um to get AI,AI is not, AI can't own
copyright.
But I think there are some caseswhereby where there's a mixture
of human effort and AI effort,you could argue if the human
effort is more than the AI, ifAI effort, you could argue that
such work can be copyrighted.
(04:15):
So in Nigeria, currently, itsays that if there's some effort
that has been shown into a workto give it original character,
such work can be copyrighted.
So this makes me think that ifwe use AI to create a work and
you also put some human touchesto it, would the work pass as
copyrighted?
(04:36):
Would it be copyrighted?
And well, there have not beenlitigations on that yet to
determine what the cost wouldsay if you create something with
the help of AI.
But my view to that is that AI,AI's work at the moment, can
pass as copyrighted if you canthrow some effort as a human
(04:58):
being, human touches into it tocreate it.
So that kind of creates um, likeI say, a loophole for AI work to
be peripherated in Nigeria.
And um at the moment, eventhough there's no litigation to
see what the court says on it, Ithink that as it stands, AI work
(05:18):
could pass if you put some humantouches into it.
That's the way I say it.
Then also there was like anational AI strategy that was
put forward by the government.
And some of the things I read onit was that with every situation
whereby some works or some jobsthat that could be done by human
beings would be taken over by AIin the future, is it going to
(05:40):
affect like the economy from theperspective of job losses?
So these are like um issues thatare still new to the country and
which we're still watching tosee the way the whole landscape
evolves.
I think that's my perspective onit.
SPEAKER_04 (05:56):
Thank you so much.
It's very interesting politicalissues that are swirling around
how AI initiatives are goingforward.
For Nigeria, what I'd read wasthat they do have this desire to
be a leader of the Africancountries in its AI initiative.
And so I was curious, uh,because they're saying they're
(06:18):
getting all these differentperspectives from the different
sectors.
What do you what do you think?
SPEAKER_02 (06:24):
So in Nigeria, the creative sector is one of the
leading sectors in the economy.
Um, because Nigerian is knownfor, in Africa, for movies, um,
music, and art.
So the way the law stands, uh,because the act hasn't been
amended to take into cognizanceum what AI can do.
(06:45):
So my view to that is that um Ithink um maybe litigations have
to come forward so that thecourts can say whether could
interpret that aspect of thecopyright law, which says that
you just need to show someeffort of originality for work
to have copyright abilities tobe copyrighted.
(07:05):
So I feel that um my view on itis that we need to wait to see
if um cases will come forwardand sections of the copyright
act could be interpreted, givenreference to the current phase
we are in with AI.
SPEAKER_01 (07:21):
Just sort of thinking politically and in
terms of the different voicesthat um are you know are
involved in this debate, whetherthe tech industry is strong in
Nigeria and whether there are AIdevelopers who are maybe
lobbying for more sort ofliberal rules which would allow
them to have freer use ofexisting copyright works,
(07:45):
because I know certainly in theUK that has been a very sort of
powerful debate, and I don'tthink there's resolution yet
between those two sides, whichto some extent are are sort of
pitted against each other.
I mean it's obviously morecomplex than that, but um that
has certainly been uh sort ofthe essence of the debate in the
(08:05):
UK so far, sort of between thetech side, the developer side,
and the creative industry side.
So I was just wondering, Toby,with the saying that the you
know the creative um sectorvoice is is really strong in
Nigeria, whether there is youknow an equivalence strength of
voice from the tech side.
SPEAKER_02 (08:25):
Yeah, in the tech sector, we haven't really had
like uh would I say companieslike OpenAI or in Nigeria
currently, we have like thereare tech companies that use AI
to create um to for theirservices.
And but in Nigeria, um there'sbeen a push for, would I say,
(08:47):
from the part of like personaldata, protection of personal
data, there's been the push forstronger laws and the way you
use personal data in businesses,the way you use personal data in
in the works that creatives orwould I say technological
companies do.
So we don't really have thesame, last I checked, it's not
(09:09):
like at the same level with theUK when it comes to tech
companies pushing for umstronger, or would I say lenient
um routes to create theirtechnology and for lesser um
lesser regulations from the fromthe government.
So at the moment, we have like anational AI strategy in Nigeria,
and I think that's where last Ichecked, that's where it is
(09:33):
currently.
There haven't been issues ofregulators against tech
companies from the side of AI.
SPEAKER_04 (09:40):
Just a wrinkle I would bring up too, since we're
discussing the UK as well, theissues that the UK is seeing
with extending copyright tomachine-generated works and
whether or not that might comeup if indeed other countries
like Nigeria start to do that.
The idea of how you defineauthorship and originality and
(10:05):
how much of a disconnect theremight be if you're gonna
describe uh extending copyrightto an original work, but then is
the original work created solelyby a human in your in your
interpretation of your copyrightact versus extending that to
machine-generated works?
(10:25):
I don't know if anyone wants tojump in on just that point.
SPEAKER_01 (10:28):
Yeah, yeah, I think I think just from the UK
perspective, um, I I don't knowif that was what you were
alluding to, Stephanie, butthere is um quite an unusual
provision whereby there is aspecific copyright for
computer-generated works in theUK Act currently, although that
is very much up for debate.
(10:49):
So the consultation that themost recent consultation that
the UK government published umsuggested that that you know
that debate be opened again.
And is that a right that is usedvery much?
And is it appropriate?
Because obviously, you know, ourour copyright act dates back to
1988, well before AI wasanything like at the state of
(11:12):
development it is now, and so Ithink um it's possible that that
right might go altogether, um,which would put the UK in a much
more comparative position withmost other countries.
I think it's quite unusual tohave that provision.
Uh, I mean, lots of countrieswhich derive their laws very
closely from UK law don't havethat necessarily.
(11:33):
So um, yeah, it's very much anopen debate, I think.
SPEAKER_03 (11:38):
Yeah, I think in uh in Sweden we have a similar uh
principles to Nigeria.
It's you you own it what youmake by default, um no
registration needed, and andthat's kind of where that's the
first major clash, right?
If you need permission first,then we're in a totally
different space.
And then when it comes toauthorship, I believe it's uh
(11:59):
UK, India, and Ukraine, andpossibly a few others that have
these special cases where youcan actually delegate most of
the work to the computer andhave it made for you and still
claim copyright for it, which isvery odd.
Uh it's odd with the basicprinciple of what creates value,
right?
It's the human authorship that'sthe taxable income and the the
(12:22):
value add.
Uh, whereas you don't want toplace incentives around
automation, usually, right?
You that's not what we want tocreate incentives for because it
doesn't build a taxable uhincome base to to basic data on,
more or less.
So I I I think it's superinteresting to to compare these
(12:43):
because uh people usually don'treflect a lot, and then Sweden
is in a lot of cultural ways anda lot of other ways a colony of
the US.
Popular culture is inundated andalso the tech companies, there's
a very close links between theUS and and Swedish Sweden.
But but some of these issuesreally brings to light the
(13:05):
differences at the veryfoundational level.
We're talking about theconstitutions of the states
being built from the ground upvery differently.
Um but I think one thing to wemight throw in here because
we're in the same situation asin Nigeria, that it hasn't
really been tested locallywhether it's what's
(13:27):
copyrightable.
But we do have a thousandregistered hybrid works in the
US and some principles emergingfrom there, which sort of align
with, I believe it was in uhCzech Republic we had the first
European uh court and thedecision on copyrightability,
and they pretty much went withthe principle of well, you
(13:49):
didn't make it, so you can'tclaim it.
What are we even discussinghere, right?
Um, but I think one test that'sworth throwing in here is one
that I came across a lot.
So I have a background in in uhcomics and licensing,
international licensing, uh,from Japan to Sweden, from
Sweden to Germany, from US toSweden.
(14:11):
Um and uh those are very oftenjoint authorship works.
You have a writer, you have anartist, sometimes you have a
bigger team.
Um, and there's the principle ofstarting medium, which I think
is a very good one to bring inuh to get the distinction
between prompting versus otherways of hybrid process to make
(14:33):
an image, right?
And the the principle is verysimple: you can have joint
authorship and shared copyrightsin a children's book that has
pictures and words, but there'salso separate, so the writer
can't claim the pictures and theillustrator can't claim the
words.
So that's that's quite that'sthe test, right?
And that's basically correspondsto prompting, right?
(14:55):
If you started writing a text,you can't claim an image, uh,
and that's really being kind ofokay.
So, how much control over animage can we have via text, even
to begin with?
Uh, and to me it's very clearbecause I've I'm a cartoonist
originally and I've writtencomic scripts and I've also
(15:17):
illustrated comic scripts, whichmeans to me, well, it's obvious
that you write something andthen you get an interpretation,
and this is entirely at oddswith how it's being presented by
the major AI companies.
They want you to feel that youmade what you prompted, right?
And that's how everything isworded: the marketing, the
(15:40):
services.
It all says you created this,and even if you ask ChatGPT
about copyrightability, it willtell 700 million people that
yes, what you just promptedwithout doing anything else,
it's yours to claim.
Uh, so we have an issue therewith common sense from the
(16:00):
perspective of anyone workingwithin the creative industry and
the sort of mental models thatthe tech industry wants to
impose, which is well, this isjust a tool, it learned like
people, and you made what youasked for.
You made what you bought.
That's basically the picture,and it's a very incoherent
picture to anyone who worksinside the field, but to all
(16:22):
amateurs, it makes perfectsense.
So just thought I'd have youropinions on this because this is
something that uh really bugs meand I fight about every day.
Would love to hear hear youryour thoughts.
SPEAKER_02 (16:37):
Yeah, for me, I feel um when it comes to creating
like art, because I create artmyself, I do paintings and I
think prompting and creating artis totally different.
They're totally differentthings.
Like prompting, you don't itdoesn't take the effort it takes
is just to imagine something andit comes into fruition.
You imagine something and youtype it, and it is different
(17:00):
from learning how to create umart, which is usually the I
think the beauty in creating artis the process of creating it,
what you go through, thinkingabout it, putting it on canvas
or on your iPod, because I dodigital art on your iPod.
It still takes efforts tocreate.
So prompting and creating to meare two totally different
(17:21):
things.
And I feel I I I tilt towardsthe um position that those
shouldn't be copyrighted, thatprompting should be shouldn't be
copyrighted because it's it'sgoing to lead to dilution.
That's how I say dilution ofworks.
There'll be like proliferationof what you can't, you can't
really tell whether it's createdby a human or it's created by
(17:43):
machine.
So the law dragging its feet onwhether machines should have
copyright.
I think it's fair.
I feel it should follow thatway, should continue that way
till we totally figure out whatwe're gonna do in the end,
whether machines are gonna havecopyright in the end or it will
forever be humans.
And ITO was the human expert.
(18:04):
But the loophole, like lookingat Nigerian law, which says that
some efforts, since that hasn'tbeen tested, doesn't mean that
if you start with um uh with youprompt an image and you paint
over it, would you havecopyright for that?
Like you prompt something andyou paint halfway through.
That's that's something I thinkof.
Should you should you earncopyright for machine-assisted
(18:28):
work or should it all just behuman?
SPEAKER_03 (18:30):
Yeah, the hybrid processes are like the where
it's really at right now, wherewe had this uh, I think it was
called American Polone orsomething like that.
Uh was a work of art, if youwant, uh legally a work of art.
Uh it was basically promptedfill in the blanks.
(18:52):
You uh they they sketched outthe simple drawing and then had
uh generative AI fill in theblanks, and they got uh what
some US legal experts call athin copyright or a slim
copright, I forget which.
That says, well, the human-addedparts uh are for short
cooperative, and there's reallyno way to set of separating
(19:13):
them.
So it's kind of a proof ofconcept of minimum viable effort
to claim uh an output uh by wayof and and here's the
interesting thing is I thinkthey you're actually in the
starting medium.
So the principle of startingmedium, if we kind of okay, so
it starts with an image, a blankcanvas, but it's really minimum
(19:34):
viable effort.
Um yes, it kind of nominallyyou've composed the work, you've
decided where things go in animage, but from there it's five
minutes of work in total forsomething that looks like it
could have taken a month, and isprobably based on something that
took months or weeks or at leastdays.
(19:56):
Thoughts on that one?
Have you seen it?
Uh in another one?
SPEAKER_02 (20:00):
I haven't seen the baloney.
SPEAKER_03 (20:02):
I haven't seen the I forget if that's the exact
title.
SPEAKER_04 (20:06):
The American piece of Swiss cheese or American a
single piece of American cheese?
SPEAKER_03 (20:12):
Yes, yes, that's it.
Bologna, that's my that's it'sall baloney.
Oh I have a single cheese.
A single piece of Americancheese.
Yes.
I missed that.
Yeah.
SPEAKER_04 (20:28):
I would ask Ahmed, what are your thoughts on all of
this that we've been discussingfrom your perspective?
SPEAKER_00 (20:34):
Yeah, um uh first of all, I mean, um I would like to
remind everybody um and thelistener that uh the use of AI
in making art is not new.
It's have been around for as oldas AI itself, more than 50 years
uh now, where artists have beensome artists have been
pioneering exploring AI in theirown work.
(20:55):
AI itself has evolved a lot, andwhat we see right now in terms
of prompting uh is just thelast, the latest wave of AI,
which creates a lot of problems.
Uh but just if you go like fiveyears uh ago, before that wave
started in 2022, uh many artshave been using AI uh um around
(21:16):
2020.
Um uh we introduced Playform,for example, as a platform uh at
that time, and many arts havebeen using uh AI at that time,
before prompting, and it was avery different process because
you um can train your own AIfrom scratch based on your own
images that capture your ownwork and style and create art, a
(21:36):
very uh creative process.
Uh and then uh came prompting,and and all these issues came
out with prompting.
Um, fundamentally, because um,when you prompt um and you
create an image in a second, andyou think obviously that uh you
created this, uh, that's youhave the impression that you
create this because you writethis amazing prompt.
(21:58):
But I mean, suppose somebodyelse comes with exactly the same
prompt.
Um, so you have two people comewith the same prompt, and and uh
you're gonna have uh probably uhmost probably gonna have very
different images.
But why uh the reason why isbasically because um in the back
uh end of the AI, there's arandom number generator that
(22:20):
just gives a coin and uh usingthe same prompt you end up with
different things.
So the the the agency here isreally um uh missing in terms of
how the word translates to an afinal image is a black box in
the middle.
That that black box has beentrained on trillions of images
and involves a lot of randomnessengineering the image.
(22:42):
So the agency between the inputand output is totally uh missed
out.
Um so that's actually the reasonwhy um the copyright uh office,
uh particularly in the OS, haverejected several cases of uh uh
giving copyrights for uh AIgenerated uh artworks.
And um, however, this uh I don'tthink this will stand for
(23:05):
forever.
I I it all uh I believe that ifsomebody cab was a case where um
they could use AI to create uhnot necessarily prompt, but but
the process really involvedtheir own work and their own
control, the copyright officersmight accept that.
I haven't seen any cases likethat yet.
But the thing is also that uhthat um uh there are many things
(23:29):
in my head here, I want to getstraight.
Um uh that there's a in the 19thcentury, uh the history of
photography, there was a verysimilar scenario where um courts
in the beginning uh rejected togive copyright to photography.
And there was a famous case uharound 1861, 62 in France, where
(23:52):
um famous photographer Mayer anduh Bierson wanted to get
copyright for their work and thecourt rejected in the beginning.
And then after a lot of debates,they finally gave copyright to
photography.
And the main the fundamentalreason was uh um that the artist
has to show that their own umtheir own uh identity is
(24:12):
reflected in the photographythat they are making.
Uh what's different between metaking the same camera and and
having the same shots?
Why gonna be different?
If I if you can show me thatthere is something about my
artist's identity that reflectsin the art, then copyright can
can be granted.
And that's what's missing now inAI.
(24:33):
Uh, how to show that your art'sidentity is there when you are
using AI.
And I guess that depends on whatkind of AI you are using.
And um the uh when you look atnow uh the the courts and and a
copyright office, their standingin this issue is totally
confusing.
Because, in one hand, um uhthere have been lawsuits against
(24:54):
um companies making AI um uh artplatforms uh claiming that the
their work is infringement oncopyrights, and the courts have
been rejected in all these casesso far, um, uh basically saying
that copyright uh generationgeneration is transformative
enough and is not uh uhcopyright infringement, which
(25:16):
would imply that the generationis somehow original.
But at the same time, thecopyright office have rejected
any attempt to copyright thissupposedly original work uh
because you are not infringing.
So that's kind of uh kind of uhcontradicting uh situation at
this point.
And and most recently, um fromJune this year, uh there have
(25:38):
been cases where um uh um uh thecourt said that basically uh
training AI on copyrightedmaterial is a fair use.
Um and that basically makes uhall this company training AI
using other people's uh materialum not liable for copyright
infringement, it's fair use.
(26:00):
Um that's also a very uh oddsituation for artists.
I think now that situation isvery uh uh muddy and and against
artists in all cases.
You cannot use AI to getanything copyrighted.
If anybody else uses your AI umto generate uh to easy AI to
generate things that look likeyour work, you are not
protected.
And if companies use your AI totrain, uh that's fair use.
(26:22):
So it's a very bad situation forarts at this point.
SPEAKER_03 (26:25):
Yeah, and if we look at the Bards versus anthropic,
which is the one you referenced,I think is super interesting
that two days later in the nextdistrict court over, uh it
basically got scolded by anotherjudge saying, Well, your
reasoning is entirely flawed.
You uh and the entire premise ofthe case was what I like to call
(26:48):
anthroxic, uh meaning toxicallyanthropomorphic.
Uh so it's a portmanteau of likeyou bring in this human, you try
to humanize the system and makepeople think of it as a human in
order to establish a mentalmodel where the machine use is
comparable to the human use.
So there I believe what Bart'sjudge in that case is what he
(27:12):
said was well, it's just liketeaching a class of children to
write, and that's it, that's agood, right?
Um, and that that's not an issuecompetition-wise.
Uh and that's entire reasoningwas demolished by the next judge
to over.
But since the plaintiff camewith the anthropic and served
the anthroxic framing from theget-go and said it learns like
(27:35):
people, then all the judge cansay is well, it learns like
people.
They didn't do that in the nextcase over, but there they failed
to uh demonstrate market harm.
And I watched the session, itwas really painful to see
because the judge was basicallybegging them to bring some
evidence of having sufferedconcrete harm from the
(27:58):
plaintiffs in the room.
And I say, I understand theissue, it's about the next
Taylor Swift, it's about buddingtalents.
How do we make incentives forthem?
It's about market obliteration.
Um, that I believe that's theterm he used.
Um, and I would love to give youright, but you're not presenting
(28:21):
the arguments, and I can only goby your arguments.
So some commenters said, well,he basically served the
blueprint for how to argueagainst fair use in the next
ruling, but that won't happenthis year.
Uh it might happen next year.
Um, and he still ended upsaying, well, it's still
(28:41):
transformative fair use based onthese arguments, right?
Based on the facts in this case.
Um so I think we're we're kindof inching closer to the core
issue of this mad frame ofcomparing products to people and
making assumptions on moralityand ethicality and legality
(29:02):
based on products are likepeople.
Uh but it it might be years yet,and we're the markets is being
obliterated in real time.
So artists are basically beingbled out as this plays out in
court.
That's my view.
It's it's bleak.
Sorry.
SPEAKER_04 (29:23):
Thank you.
That second case, uh Kadre, Iwould just point to, I believe
Toby had raised earlier thisissue about market dilution,
which all came up with JudgeShabria's opinion in the Cadre
case.
And as Johan put out there, yes,it is like a blueprint going
forward for plaintiffs tohopefully successfully argue
against transformative fair use.
(29:45):
But yeah, this is definitely aconcerning space that we're in,
at least in the states, inregards to the way the cases are
unfolding.
Going back to you, Ahmed, Iwonder for all of the different
projects that you've been doing.
Doing, knowing that this is thiskind of situation where you can
are not able to uh receivecopyright protection.
(30:08):
I wonder what your thoughts areabout whether or not there is an
issue of creative motivation orlack thereof.
SPEAKER_00 (30:16):
Yeah, um frankly, I mean, when I I did most of this
project, um um I didn't havethat issue in my mind at all.
I didn't, I mean, as a as aresearch scientist or as uh an
artist also um approaching thisproject as a creator, uh I
really didn't think about that.
I I'm not doing that to getcopyright, and I'm uh I'm not
(30:37):
getting copyright that does notprevent me from doing them.
Uh that's fundamental.
Here's uh also a veryinteresting uh project that
relates to this.
Uh back in um 2017, um, we havedeveloped an AI algorithm that
uh looked at um a history ofWestern art and tried to
generate images that uh learnthe rules of art, but at the
(31:01):
same time uh don't uh follow uhany any art movements, try to
generate its own style,basically.
And that was called uh AI canICANN.
Um and and uh it made a lot ofuh media attention at the time,
back in 2017, that was waybefore uh the current generation
of AI art making.
(31:21):
And the headline at that timewas basically finally, things
like finally AI can generatesomething that we like, the
future of art is in danger,things like that.
Really a lot of debate about uhuh AI stepping into the creative
domain and can create somethinguh creative and novel um um and
and not never seen before.
Um uh one thing interesting hereis that um we applied for a
(31:46):
patent for the algorithm, andthe patent was accepted.
So we officially have uh um uhan algorithm that's patentable,
the S patented that create artby itself.
In the same time, copyrightoffice that's often uh issue the
same patents have been rejectedany copyright um for AI
generation from artists.
(32:07):
So that that uh for me it's abizarre situation because uh you
kind of patented an algorithm togenerate art, but in the same
time, you we didn't never try toclaim copyright for the
generation, but it would befunny if we uh asked uh
copyright uh office to um for acopyright for the generation and
how how that would be seen giventhat the algorithm itself is
(32:29):
patentable.
Um so it's it's uh the all theseissues came later.
I said as I said before, uh Imean there was not much debate
about that until really the theBrompt-based AI came around in
late 2022.
Um before that, uh I mean therehave been five years of uh at
(32:50):
least of AI generated art uhmovement called the GAN-based uh
art movement, where there wasactually AI art was have been
celebrated.
Uh there have been AI artexhibitions in museums, in in
art fairs, in galleries, and uhit was very, very healthy, and
(33:11):
and people really liked thatkind of art, um, AI art at that
time.
And I said that there are ahistory of that go back 50 years
uh older than this.
All this debate comes reallywith the prompt-based AI for the
two reasons I mentioned earlier,the fact that uh the
prompt-based AI really comeswith trillions of images of uh
(33:32):
uh that are copyrighted, bakedin, and using to train the
system, and that changedeverything.
Once this came around, we startseeing basically backlash
against AI, and we start seeingAI is being banned from artist
community and and uh uh lawsuitsstarted happening.
Um uh, and really that I have anarticle at the time uh in I
(33:54):
remember in Artnet News orsomething uh that really said
that basically the AI art era isover back in 2022, because
really the the main amazingthings that happened in the five
years prior to 2022 was reallyan art movement uh that where
artists use AI in amazing ways.
Once 2020 came around andprompt-based came around, it's
(34:17):
no longer art.
It's just basically makingimages using prompts.
Uh uh to make it art, to toqualify as art is a totally
different thing.
Um so um, and also I have beenworking with many artists in
that era um until now who havebeen using AI in amazing ways.
And and we we have a paper wherewe actually um we surveyed many
(34:39):
artists who are using AI inaround 2020, again, very early,
and ask them um uh uh why useAI, what do you find as a value
when using AI?
And basically there are twothings uh artists keep repeating
uh in terms of what they theysay.
First, um the fact that uh AIgive them um um um creative um
(35:02):
uh ideas, the fact thatbasically um you can plug in
your images or ideas and and cangive you something that you
never thought uh before about.
Because AI basically doesn'thave the same constraints as us
when looking at images.
And that again was before theprompt-based uh scenario.
Uh, because uh once we addprompt to the problem, now you
(35:24):
kind of render the image withthe lens of the language.
You are constrained only to whatcan be described by language.
Um uh and that's a limitationactually.
Before the prompt-based, uh youdon't have that limitation, you
can re-render the based thepixels based on your own images
without any um uh uh limit ofthe language.
(35:46):
Imagine you are working onabstract uh uh uh art and and
and you cannot describe whatyou're doing in with any words,
and and AI now is limiting toyou to do that.
Uh that's so that the firstthing is that this that the fact
that artists find AI really givethem creative inspiration.
And the second value was um uhthat uh artists always say that
AI gave them the creativevolume.
(36:08):
Um the fact that AI can helpthem in creating lots of and
lots of uh assets and materialthat they can use in their own
project or or they can use intowhatever they are making,
whether they are making video ormaking a mosaic or making uh
whatever project, the fact thatum AI can work as an uh uh
studio assistant for you uh thatyou can uh you can use for that
(36:32):
verbos.
Um so again, I mean, whenartists use AI at that time,
they didn't think much about uhthe copyright issues because
mostly they were using AI basedon their own images, not on
somebody's else's images.
And secondly, they are using AIas part of their own process.
It was not really uh somethingalien to them, it's part of the
(36:54):
process.
I'm using AI as any other toolthat I'm using.
And again, that whole thingschanged where you are just using
prompts and your whole processis just input a text and getting
an image, and and and and uh nowit's really uh problematic in in
many ways.
SPEAKER_01 (37:10):
I think um the there's a really good sort of
example or illustration of howthings change very much with
prompts, but so generative AItools.
Um so there was a there was aChristie's sale of AI works.
(37:31):
Um, I can't remember exactlywhen it was a fair while ago,
but after sort of 2022.
Right, okay.
So you know the one I'mreferring to, and there was a
real sort of backlash within theartistic community, and artists
got together and asked for thesale to be cancelled, but then
there was kind of a backlash tothe backlash by a lot of the
(37:52):
artists exhibiting because theysaid exactly as you were saying,
Ahmed, we are using AI as thiscreative tool, and we see it as
a way of you know progressingour work, moving moving uh the
artistic canon forwards, if youlike, uh within the digital art
sphere.
And so I think that I think thatdoes illustrate really sort of
(38:13):
clearly the the sort ofdifferent um approaches and the
way you know there isn't kind ofa one-size fits-all when we're
thinking, I think, aboutcopyright.
In we can't sort of generalize,I suppose, I don't think,
really, when we talking abouttalk about AI generated art
because it can be so verydifferent, you know, it AI might
(38:36):
be just a simple tool forsomebody using a um a generative
AI algorithm versus you knowartists who are working as
artists all day, every day, andusing a very different kind of
tool that they have created byusing often their own work and
you know by sort of curatingthat tool in the first place.
SPEAKER_03 (38:59):
Absolutely.
Yes, and I think the backlashwas about kind of this
piracy-based uh image modelsbeing fronted and invading the
space very obviously from theget-go.
And I I really think they didthemselves a disservice by
hosting this and and muddyingthe waters by mixing up people
(39:19):
like Rafi Canadol, who has donethis custom software that makes
custom living images, and it'svery much bespoke and the uh
artistic process, versus someonewho prompted something in stable
diffusion and then stitched somepictures together.
Um, and I guess I mean the youcan still copyright the work
(39:41):
that's based based on stolenpaint or stolen images, um, as
long as you add value to it.
But I think the red flag for redsheet for a lot of people was
those specific image modelsbeing used.
Um so yeah, there was aninteresting discussion after.
And I I I I'm totally on onboard with protesting against it
(40:05):
because it we need to debate andget to the point to the core of
what are the real issues here.
And I actually debated a bitwith one of the guys who had uh
an artist who makespseudophotography, calls it
post-photography, uh, where heuses these kind of models to
(40:27):
make something that's veryclearly photorealistic, but also
very clearly not photos, becausethings break apart at a closer
look, and he kind of uses thatconsciously.
Uh, whereas the issue is takingthe shortcut and never mind the
details, and just getting thework done in the smallest time
possible.
(40:47):
So he's putting himself in apretty tricky spot by using
these things, but I think theposition is defensible because
he really uses the tools and thecontent and the process in a
very conscious way to specificends uh creatively.
So yeah.
SPEAKER_02 (41:07):
Yeah, for me it's uh I'm also on the view of about
the process.
Like art is not just about thefinal image, it's about the
process and how you make it.
So, like um Ahmed said,prompting prompting is actually
a big issue because I don'tagree that uh putting images,
but what if you do a bunch ofprompting and you gather your
(41:30):
prompts as like a mosaic?
Sort of is could that pass outsome effort to create art if you
um clip different images you'veprompted together and as a
finished work to createsomething?
That could pass out some effort,or I don't know what do you
think about that?
SPEAKER_00 (41:48):
I don't think I mean the effort should be the the
metric here.
I mean, uh we are a hundredyears now after uh Marcel
Deschamps have his his fountainuh as an art, basically the
unary thing that he did as asculpture, and and that opened
the the whole era of ready-madeart, basically, when you can use
an object and and declare anart.
(42:08):
And he had a point basicallythat artist can can make
anything and call it art.
That's how his whole point.
And so the to come now 100 yearslater and and still talk about
the effort and the time spent asa as a metric to qualify
something to be an art or not, Idon't think that's right.
I I think uh the main issue hereis really um um uh
(42:31):
intentionality, obviously, isone important thing.
Um, and definitely Marcel Schamhad the intentionality of making
art this way, uh to make amessage.
Uh, when you're prompting, uhthe intentionality is broken.
Uh part of it, yes, you you youwrite a prompt with intention of
something, but you actually geton the other end is something
totally up to the uh torandomness in the process.
(42:53):
And and uh you might like it ornot, you might start tweaking
the prompt to try to control itto what you want.
But most of the time you end upwith something that you like.
So your role is more like acurator.
Uh, you start with with someidea, you got something, it's
not really what I like, but Ilike what I get.
Uh, and and here it is.
I mean, uh, so that's what'smissing in the process, the fact
(43:15):
that uh the agency is is notcomplete.
Um uh part of it is random, partof it is is uh is mashing up
millions of images uh based onlanguage to generate something.
Um but it's not about the effortbecause it can actually take you
a lot of effort to create someimage with a prompt.
It can a bit of trial and erroruh to to try to really control
(43:36):
it because we are very hard tocontrol.
So the effort is there, the theagency is not there, the
intentionality is what'smissing, or or or not 100%.
SPEAKER_03 (43:44):
Yeah.
One of the examples I like to goback to is uh I mean, generative
art has a prehistory to digitalart, right?
There were spiral graphs andthere were paint spinners and
people experimenting with thatin the in the 60s to figure out
some of these things, right?
Um, and the principle ofstarting medium is a is a good
(44:07):
one, but one that came out ofthat whole line of
experimentation where they didthings like okay, so here's a
paint spinner, we invite thevisitors to pour in the paints
and then uh press the button orspin the wheel, and we get an
original picture and we put iton the wall.
And then you get the same corequestion that you have today:
who made the picture?
(44:28):
Is it the guy who made themachine and patterned the
machine?
Is it the user or is it themachine?
Right?
Who is the real author here?
Uh and the principle ofindependent origination came out
of that, if I'm correctlyinformed.
And that says, well, if a laymanlooks at the pictures on the
wall and he says, Well, I can'ttell apart that these are being
(44:51):
done by different people, thenthere's no basis for a claim of
human authorship because there'sno originality in it.
Uh, and and it's kind of twolayers to it.
One is can you tell it apart atall?
And then the next layer is canyou tell apart that specific
people have made this?
(45:12):
Because if you can't, it doesn'texpress the personality of the
author, right?
Um, so it's one is yellow andone is green.
That's not enough personality,even if one person really likes
green, right?
Um, so I think that's a goodone.
And also if you look at thosethree, like user, machine, and
uh and the machine maker, uhthere's a really good paper, I
(45:34):
forget the title then, but thatreally went to the bottom of
from a societal perspective, ifwe look at copyright as an
incentive machine, an incentiveframework, if we place the
incentives around the user'seffort, the machine effort, or
the the machine maker, likelet's just look at what happens
(45:59):
if we give output rights to theguy who made the machine, to the
machine itself, or the user.
And the conclusion out of allthis was well, it's not about
who, it's about we don't want toincentivize automation, we don't
want to incentivize volume ofempty work.
(46:19):
We want to actually incentivizethe process, as you mentioned,
Toby.
That's where the value iscreated.
Um, and also as a matter ofscale, which they it's a few
years back and they kind ofpredicted where we are now, is
that if you incentivize emptyautomation, then you get
(46:40):
system-wrecking volumes.
So it doesn't really matter whogets to claim all of all of
these outputs, if there's anincentive to create more and
more and more, you get too much.
And market dilution is onething, but there's also a
democracy aspect and giving upon truth altogether and telling
(47:03):
what's real.
I mean, it's a it's it's a muchlarger issue than just the
incentives around art.
It becomes can we even navigatereality if we're surrounded by
fake pictures?
So yeah.
Um, and I to your point there umabout the patterns for the
(47:24):
machine, that to me sounds a lotlike oh, I forget now, but uh
the gates of paradise is thename of a work that's entirely
machine generated.
Um, so there was no human in theloop at all.
It's a randomizer that producesuh generative works.
And the guy uh I think it'sStephen Thaler, right?
(47:47):
Is on the mission around theworld to get authorship rights
for machines.
So that's a I see either he's alunatic or he's kind of pressure
testing the legal systems aroundthe world.
Either way, that's how it endsup working.
It's it's how it does the systemhold.
Um and I believe they assignedauthorship rights, I believe
(48:10):
maybe in South Africa.
Um, and I forget now because hedid a patent machine and he did
an art machine, uh, and he'strying to register the outputs
of both of them around theworld, right?
Um, and I believe he actuallymanaged to get one of his
inventions or artworks patentedor copyrighted, I forget which
(48:31):
one.
Uh, but they ended up notdesigning the copyrights to the
machine that he asked for, butto him, right?
So the author of the machine gotuh patents to the outputs, which
is an anomaly, as I understandit.
Um but yeah, uh so but I I guessone of the uh courts when they
(48:53):
assessed uh his attempts toregister the gates to paradise,
the randomized works without ahuman in the loop.
It was uh well, we alreadyincentivized you, we already
provide protection.
You're the patent owner of themachine that you made.
That's your incentive, right?
(49:13):
You don't need additionalincentives in claiming also the
outputs of the machine for thereasons discussed just now.
So I think that's how how thatstory ended up for them.
But as I understand it, this guygoes around and tries to
escalate.
And you have Jason Allen alsodoing his space theater opera.
(49:33):
Uh, one of my favorite kind ofone-liners from one of the court
rulings when he tried tocopyright this mid-journey
image, which looks like ageneric mid-journey version 3
image.
Um, is like, well, you're not anartist because you found a
pretty stick in the woods.
It doesn't matter that yousearched for 600 days, you still
(49:55):
didn't make the stick, right?
Um, and that's it's a bit mean,but uh there's a core of truth
to it, right?
And and I believe that one wasokay.
One part was he did all theseiterations until he found
something he liked.
And the next part was he didsome minor edits, and I believe
they assessed these edits to notadd any value, uh basically,
(50:19):
it's marginal adjustments, andthey say, Well, that's not
enough.
You didn't add any creativity inthat step, you just tweaked it a
little bit in the edges, so nocopyright for you.
That's I believe so.
Those are some of my favoritecases to track Steven Tyler and
Jason Allen.
And um, and they teach us a lotabout like what's the underlying
(50:44):
principles here uh that we'refinding out as we go on.
SPEAKER_04 (50:48):
Also, that whole layer of the training questions
that go into mainstream models,like the mid-journey models.
What are the infringing aspectsof those outputs?
Like, that's a whole otherconversation that's really not
even being factored in manytimes when they're having that
can we copyright an outputquestion.
(51:11):
That's what I love about modelslike Ahmed, like what you have
been talking about, where youcan train your own model.
And that to me, it really doesspeak to an ethical approach to
this.
Well, unless there are anyquestions about that, I also was
just gonna dovetail into thisquestion that Ahmed had raised
as a point about culturalstagnation, which I think
(51:34):
everything that we've beentalking about kind of smacks of
questions about where is ourculture going.
So, anybody want to jump in onthat?
SPEAKER_00 (51:41):
Yeah, let me start.
Um, um uh yeah, um I think uhsince um 2022, um Chat GPT and
and uh all prompt-based imagegeneration came around, um,
definitely the line betweenwhat's real and what's fake
become very blurry.
Uh so since then we have seenlots and lots of um images in
(52:05):
social media that are fake.
Um every day now it's uh beinggenerated.
Very good quality, obviously.
It's very hard to tell videosnow, real videos from fake
videos.
Uh right after Shad GBT camearound, there was a nice story,
funny story that um Amazon um uhthere was many Amazon books, uh
bookstores uh online just makingfake books and selling well,
(52:28):
actually, these fake books.
Um uh so uh really the the thethere are two big big problems
with that, obviously.
The problem of um that the linebetween what's real and what's
fake is totally blurred at thispoint and gonna be more blurred
to uh to in the future to thepoint that we will not really
know what's real and what's fakeanymore uh in any medium.
And the other thing is that allthis fake stuff is gonna be uh
(52:54):
um by scale, gonna be themajority of content on the
internet and and and and andeverywhere.
And that means that the nextwave of AI models will be
generated with more of AIgenerated images and videos and
text than human-generated ones.
Uh, because one of the problemswith AI now is that the lack of
(53:16):
data already digested all theinternet, so they were looking
for more synthetic data, more AIgenerated uh data, you're
welcoming that to train andskill their model.
But then we have a problem thatnow we are uh going into culture
with technicians because uh themajority of content there will
be AI generated.
Um, and um what that do to ourculture, that's really um a big
(53:40):
problem.
SPEAKER_03 (53:41):
Yeah, the if I may, there have been this discussed
to one of my pet peeves.
That that's again like what whenwe see the mainstream media
reporting on Sora, for instance.
Uh there are a couple offramings.
Uh one is can machines be heldliable?
Like, is the bot to blame, isthe AI to blame?
(54:04):
That's one of the framings,right?
People are really strugglingwith this, and the other is
users are using this tool to doX, Y, and Z.
And to me, that's the othercrazy frame.
If you have the maker and themachine and the user, right?
One is about blaming themachine, one is about blaming
(54:24):
the user.
But the user can't make this,and the machine can't make this
unless there's a serviceprovider providing this live
service, actually producing thework, and nobody points in that
direction.
Uh, and that's it's baffling andhorrifying, frankly, to me, that
nobody says, why aren't thesemedia production companies being
(54:46):
held accountable for what theyproduce?
That's absent in a lot of thediscussion.
And I I just don't understandit.
And the the best conspiratorialthinking is they have been
really good at PR, atestablishing some frames of
thinking around uh liable humanmachines and about users getting
(55:08):
tools that make that they use tomake things, uh, and avoiding
the obvious thing that is techcompanies moving into media
production.
That's what they're doing,right?
Except without the input costsand the output liabilities.
(55:28):
So they manage to kind of havethis old mental frame of these
toolmakers or these neutraldistributors, despite now being
producers.
And so this discussion becomesone of the DMCA, like, okay, so
the platforms can't be heldaccountable, right, for what
(55:50):
flows through the platforms, andwe've already seen what happens
if it sucks all the money out ofmedia, the traditional media,
and everybody else out of theirpersonal data, and you have an
ad monopoly and a socialdistribution, media distribution
monopoly, right?
Um, and now they're moving intoto even compete directly against
(56:14):
legacy media by producing theirown media based on theirs
without paying for it.
Uh, and we're still stuck inthinking about them as tools and
as distribution platforms wherereally they are production
houses now, uh, fully automatedproduction houses.
Um, yeah, uh, so I I think justplacing those the product
(56:38):
liability and the serviceprovider responsibility in the
right place is where we need tostart if we're going to get
anywhere with this.
SPEAKER_02 (56:46):
Yeah, I think my view on the my view on the
culture of stagnation is that itcan also take incentive away
from because when there is likeproliferation of AI
creator-generated work all overthe internet, if you're like a
graphic designer, it's it takesjobs away from graphic designers
because now you could just useuh an AI app to create graphics
(57:12):
and you no longer need to paygraphic designers.
And it kind of takes awayincentives from uh skills like
skills like graphic design, likebecause if you could uh just use
uh AI to create your to brush upyour images, you don't need to
pay photographers to take imagephotos of you.
You can use AI to uh brush upphotos to make it clearer and
(57:35):
fine.
Now you could use uh these appsto create designs and uh it kind
of kills that's what I say.
The incentive for people tolearn some of these skills that
we've been uh doing for years,that's how I say.
SPEAKER_03 (57:49):
Yeah, absolutely.
Uh and to me again it it's uh itgoes back to the framing.
Uh, if we think that an AI useris doing the same work as the
one who produces the originalwork that's being used as a
training material, or if wethink of the machine as the
actor, then we we miss theaspect of well, no, these are
(58:12):
the service providers, the nowmedia production companies
competing directly against theartists and graphic designers
and writers and translators andall of those, right?
Uh but they're dumping themarket, they're not selling
these services at nearly thesame price level.
So when OpenAI produces acartoon image, uh it does it in
(58:38):
seconds, right?
Fractions of a second, and itcompetes directly against the
work of whoever made theoriginal drawing who needs hours
to finish the same thing, andthey sell it for cents rather
than selling it for the well,maybe a hundred bucks or
whatever they would need tocharge to to make a living off
(58:59):
of making cartoons, right?
Um and one part of it isautomation.
That's fine.
I mean, it it does the workfaster, but the other is
sourcing, right?
Uh the cartoonist needsreference material, they need
production assets, they needtools, they need all of these
business inputs to produce theimage.
(59:22):
And this trillion dollar companyhas the competitive advantage of
not paying for any of it.
They get tax breaks for the datacenters and they they scrape the
internet to get the productionassets for free.
So it's really an even unevencompetition situation.
And to me, it's actually the bigpicture issue is more antitrust
(59:46):
and comp fair competition lawthan it's about copyright.
Because again, they don't owntheir output, but that doesn't
really matter.
You don't need a copyright tosell anything, you can still.
Sell bespoke images made in aninstant, right?
In a lot of business context, itdoesn't really matter if
(01:00:08):
somebody else has the same work,right?
You you would go to a stockagency or you would license
pre-made work anyway.
You don't need the exclusivity.
And you're not really using thepicture as your core competitive
thing.
It's maybe maybe used formarketing or as a web page
(01:00:28):
filler or whatever.
It doesn't really matter ifsomeone prompts the same headset
girl for your support page,right?
So yeah.
So to me, it's it's uhdefinitely a case of Microsoft
now competing directly againstthe Dartists they took from
Microsoft and OpenAI, not theirusers, Microsoft directly.
SPEAKER_00 (01:00:50):
Except that um that um Microsoft and OpenAI does not
at this point uh uh makeentertainment and production
their their main business modelor or thinking this way AI and
AI focus on AGI and Microsoft asa software company.
Uh however, we have to rememberuh how Netflix started.
(01:01:13):
Uh when Netflix started, we werebasically mailing CDs in the
mail and we're mainlyrecommendation system for
recommending you the next movieto watch.
And we evolved into the majorproduction company now of shows
and killing all other studios.
Um, so that's really um um howthings went um uh for Netflix.
(01:01:37):
Uh can we imagine major AIcompanies uh going into uh into
the direction of mediaproduction and abandoning the
idea of AGI?
I don't see it yet.
Probably there are somecompanies who will come up uh uh
and focus on the production inparticular.
It might be again Netflix or orthe Netflixes of the world where
(01:01:58):
basically they are wellpositioned to use AI uh in
production in the next level.
But um the problem was thatmajor AI companies have other
missions they're trying to focuson, thinking that these are
bigger business models toachieve, including automation of
the world, taking the jobs ofother people and tools and
(01:02:18):
robotics and things like that.
Uh so um they are not reallyfocusing on uh that
entertainment production angle.
SPEAKER_03 (01:02:27):
This is so interesting to me because I
agree that that's how theyposition themselves and how they
push things forward.
But when you look today at whatthey actually do, if I have 20
bucks a month to spend onpictures, I can spend it on
shutter stock, I can hire afairly rookie illustrator, I can
(01:02:48):
go to OpenAI.
So even though they're aiming tobe a tech and presenting as a
tech company in practice, theyare competing already today.
Uh, and we see them also movinginto media.
OpenAI is funding an animatedfeature field now.
Google is starting their ownproduction house.
So what we're seeing is thatthey're kind of shedding the
(01:03:12):
pretense of being a techcompany, they're spinning off
this purely media-centeredplace, right?
Where automation and everyoneelse's property as free business
inputs is the starting point.
And if you have that as yourhead start, then you can get
pretty far and make some prettybad movies until you get the
(01:03:33):
hang of the rest.
If you look at Altman and whathe's doing, he says, Well, I'm
making a religion, I'm making acult, and I'm incorporating it
as a company.
It's all about getting people tothink more about the future than
they hear now.
Uh, what might this be nextyear, right?
What might might it be in twoyears from now?
(01:03:54):
Rather than looking at, well,what are you doing right now?
You've already wrecked basicallyacademic research, um news
media, and you know uh you'redumping the market for
illustration and translationservices, but it's always about
looking into the future andseeing what's next and
forgetting about this pettysquabbles about the here and
(01:04:16):
now.
So it's fascinating to me.
SPEAKER_04 (01:04:21):
Thank you.
One upside I see of our culturegoing forward using AI, I would
just uh bring back to aBeethoven project that I learned
about through Ahmed.
SPEAKER_00 (01:04:38):
I haven't heard about this.
It's it was quite famous inEurope, not in the US,
unfortunately.
Uh so uh Beethoven uh uh madenine symphonies and uh he left
some sketches for a 10thsymphony, very rough sketches
for a 10th symphony.
And back in 2020, 2019,actually, uh Deutsche Telekom um
commissioned me and a group ofuh musicologists and a music
(01:05:01):
historian to try to uh come upwith a version of Beethoven 10th
symphony based on the sketchesand using AI.
I was the AI lead in thatproject.
Uh that was a celebration ofBeethoven 250th anniversary.
Uh COVID came obviously anddelayed everything, but back in
2021, we made the concert um uhuh in Bonn uh played by uh
(01:05:23):
Beethoven uh Symphony Orchestra,where we finished two movements
uh of uh of uh the symphonyusing uh AI.
It was really uh binding projectuh using the same kind of
technology that came late to beknown as ChatGBT, basically the
same language models, uh, tolearn about uh Beethoven style
(01:05:43):
of composition andorchestration, and now feed
these sketches and try to seewhat kind of composition the AI
can come up with.
And and it was really a goodproject to show the potential of
AI.
AI was a building block in thatproject, but didn't cancel the
human.
I mean, there was a humanmusicologist, composers, they
(01:06:04):
are historian, all the teamworked together with the AI to
come up with something, and itwas a creative process where you
use AI to come up with somecontinuation, for example, and
then the composer would look atthis and and approve it or not,
or come up with other ideas,give it to the AI again.
It was really a very goodproject to show how AI can be
(01:06:25):
used as a tool in the artistictoolkit of uh an artist or an
artist team to do somethingrather than just uh uh having a
black box that you prompt andand you get a final answer uh
that that uh really isproblematic.
So um that the the completion ofBeethoven X is available on
(01:06:47):
YouTube.
You can uh find it and listen toit and enjoy it.
Um uh it's it's it I mean I findit inspiring and in terms of uh
um listening to a beast thatBeethoven never wrote, but still
based on his own notes.
It's really interesting.
You can tell it for yourself,search for it.
Thank you.
SPEAKER_03 (01:07:08):
That's super cool, and I mean to not be the David
Honor here being the uh publicdiffusion is an interesting
project, right?
Which starts from the samepremise.
We we'll starting with fromcleared public domain images and
making an image model out ofthat.
That's still problematic forartists working in traditional
(01:07:29):
styles today.
If they manage to produce asgood input, we still have an
issue with automation, but wedon't have an IP issue, right?
Um, and it's uh I see a lot ofpotentially similar use cases
where you can expand on ourshared history.
But again, you need to sort ofhave output watermarking or
(01:07:51):
something like that, so you oror we will just confuse
ourselves by inventing a lot offake history, uh fake art
history.
But I think it's a it's aworthwhile project and it's
interesting, and there are manylikes it.
Uh there was another one inJapan where it took the the
grand old master of of manga,Osama Tezuka, um, and worked
(01:08:14):
with his estate to digitize hisentire archive and see if they
could have it come up with newcomic episodes in his style.
But similar to what youmentioned, the estate had
approved beforehand, and his oldassistants uh and his old
publishers were all involvedcreatively in the project to
(01:08:36):
make sure that they theyactually completed it with a
human touch and in a good way.
I don't think it wascommercially successful, I
haven't seen it, so I don't knowif it was creatively successful,
but it shows a good example ofhow to do it the right way.
Everyone approving and everyonebeing contributes in a creative
(01:08:57):
human way to keep a traditionalive.
It's about building on what camebefore, rather than
cannibalizing not onlyeverything that came before, but
everyone today as well.
SPEAKER_04 (01:09:09):
So yeah, and being transparent about the source
that there was that AI elementto it, then no one is misled.
SPEAKER_03 (01:09:17):
Absolutely.
Yeah, the three C synatry,credit consent, compensation,
that transparency.
SPEAKER_04 (01:09:24):
Yeah, yeah.
Toby, what would you think?
I I'm just curious because I Iam curious to uh use an AI model
built on my work, and and sowould you or have you done
anything like that?
What are your thoughts on that?
SPEAKER_02 (01:09:40):
So I think that's good.
I've because I also writefiction sometimes.
I play with stories, and sosometimes sometimes I feed my um
work into like a temporary chatGPT, the temporary part, and
hear suggestions on tweaking orviews.
Like if I can get a humaneditor, I could use AI trained
(01:10:04):
on my own work to give me viewsor feedback on what I could do
better.
And sometimes the suggestionscould be good and they make
sense sometimes.
Like when I write fiction, I useAI to just give me feedback.
And I think that's like apositive way to use it, to give
you feedback and um what youcould tweak.
(01:10:25):
And yeah, I do I feel that'sgood, but um not like the way
Johan said, not in thecannibalistic nature of AI
taking away our creative agentsto create stuff, but if we could
use it to assist and to um showus other ways or other
perspectives, I think that couldbe good.
SPEAKER_04 (01:10:47):
I would love it if each of you offered a closing
thought.
So please.
SPEAKER_03 (01:10:51):
It was really interesting to hear your your
perspectives and uh hope we cankeep the discussion going.
Thanks for the lot of uh goodexamples that came up that I
wasn't aware about and dig intothose.
So I look forward to trying outplay for.
It sounds like a lot of fun.
SPEAKER_02 (01:11:08):
AI is fed with uh what creatives already created.
That's what you used to trainthe machine.
So if you're taking awayopportunities for creatives to
make money, I think so.
In the future, would AI betrained on what AI is already
produced, or if like creativesare not like having
opportunities due to like thecannibalistic nature of the way
(01:11:33):
of prompting and the rest.
So I just feel that if we couldreach like a middle ground
between original work andgenerated work, I think that
would be better for the future.
SPEAKER_00 (01:11:46):
And definitely coming to uncharted territories
as we go and and uh lots ofquestions being raised every
day, and lots of new data pointscoming, and uh we have to keep
this discussion all going on anduh and uh continue.
SPEAKER_04 (01:12:03):
Emily, any thoughts?
SPEAKER_01 (01:12:06):
Oh, I think the main takeaway is that uh as Ahmed
said, it's more questions thananswers at the moment, but I
think I guess on a positivenote, uh and it's a bit trite,
but I think things will kind ofwork themselves out in the end,
just like you know, newtechnologies like photography,
and certainly in the kind of inthe narrow sort of IP world, I
(01:12:30):
think things will work out, andthere'll be kind of licensing
schemes, and um in time thingswill work out.
That said, you know, AI moregenerally, um there are so many
kind of questions, and I thinkthe answers to those questions
will very much sort of shape theworld we live in and we boss on
(01:12:54):
to future generations.
SPEAKER_04 (01:12:58):
There will be links in the show notes to learn more.
If you want to be able to bemuch pretty much time, this is
the thank you so much forlistening.
Popular Podcasts
Las Culturistas with Matt Rogers and Bowen Yang
Ding dong! Join your culture consultants, Matt Rogers and Bowen Yang, on an unforgettable journey into the beating heart of CULTURE. Alongside sizzling special guests, they GET INTO the hottest pop-culture moments of the day and the formative cultural experiences that turned them into Culturistas. Produced by the Big Money Players Network and iHeartRadio.
The Joe Rogan Experience
The official podcast of comedian Joe Rogan.
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.