Full Transcript

Outsourcing the Muse? AI Creativity vs. Human Creativity | Boris Eldagsen | TEDxESMTBerlin

19:361,076 summary words · ~5 min readEnglishBy TEDx TalksTranscribed Aug 26, 2026

Analyze another video with Pro30-day money-back guarantee

Summary

Generative AI surpasses human capacity in combinatorial and exploratory creativity, freeing artists to act as directorial conductors who focus entirely on conceptual, transformative meaning.

As AI models automate technical execution and reprompting, artists risk losing personal subjectivity unless they actively retain control over the initiation, iterative blending, and critical evaluation stages of creation.

Section summaries

The Ambiguity of AI Creativity Tests

optional

Boris Eldagsen opens by polling the audience on whether AI is more, equally, or less creative than humans, noting that he personally raised his hand for all three perspectives. He reviews recent academic and media studies, such as the University of Essex alternative-use toothbrush test and the Torrance Tests of Creative Thinking in Montana, which claim AI matches or beats human originality. Eldagsen points out a core conceptual flaw: public discourse uses terms like creativity, intelligence, and consciousness without concrete, agreed-upon definitions.

Standardized creativity benchmarks (such as alternative-use and Torrance tests) rank LLMs at or above human averages in fluency and originality.
Public and academic debates about AI capabilities remain confused due to undefined terms regarding consciousness, intelligence, and creativity.

Useful contextual setup regarding recent psychological tests, but the theoretical core begins in the next section.

Margaret Boden's Model and Consciousness

watch

Eldagsen introduces cognitive scientist Margaret Boden's tripartite theory of creativity: combinatorial, exploratory, and transformative. He illustrates combination with surrealist mashups and Midjourney prompts, exploration as exhaustive mapping within defined parameters, and transformation as paradigm shifts that break conceptual limits. He critiques a 2023 study by 19 scientists claiming AI exhibits consciousness, highlighting that their definition equates consciousness merely to information processing—a criterion satisfied by basic smartphones.

Margaret Boden divides creativity into combination (novel juxtapositions), exploration (navigating conceptual spaces), and transformation (rewriting the rules).
Many scientific assertions of machine consciousness rely on shallow definitions that reduce consciousness to bare information routing.

Provides the foundational theoretical framework used to understand the division of labor between AI and human thought.

Promptography and the Conductor Role

watch

Eldagsen breaks down the power dynamics of human-machine interaction in visual generation, contrasting passive users who get dragged along by model defaults with trained practitioners who act as film directors directing specialized AI assistants. Reflecting on 30 years of solo photography, he describes coining the term 'promptography' to capture his new role. Rather than acting as a solo instrumentalist expressing isolated subjectivity, the artist becomes a conductor synthesizing personal intent with the anonymous choir of collective cultural training data.

Unskilled users are led by algorithmic defaults, while domain experts use historical and technical vocabulary to direct the model.
AI art ('promptography') shifts the artist's role from a solo instrumentalist into a directorial conductor harmonizing collective training datasets.

Crucial philosophical reframing of artistic identity, authorship, and the mechanics of prompt-based creation.

Defending the Human Loop Against Full Automation

watch

The creative workflow consists of three essential phases: human ideation and prompt drafting, machine generation, and human aesthetic evaluation. Eldagsen warns against recent tools like ChatGPT-4 paired with DALL-E 3 that automatically rewrite user prompts and select outputs, thereby erasing human subjectivity from the beginning and end of the pipeline. To maintain genuine artistic control, he outlines a complex iterative pipeline involving image blending, multi-step prompting, in-painting, and out-painting.

Subjectivity resides exclusively at the initiation (ideation) and termination (curation/evaluation) phases of generative workflows.
Automatic prompt expansion and algorithmic output filtering threaten to eliminate authorial subjectivity entirely.

Delivers the central actionable insight regarding where human input must be defended against black-box automation.

Case Study: The 'Trauma Porn' Exhibition

optional

Eldagsen presents his multi-part art project 'Trauma Porn,' created with Bangladeshi artist Tanvir Tahed. Built on authentic historical photographs collected over 20 years from his father's WWII past and Nazi-era Germany, the work explores the psychology of the silent, complicit masses under totalitarianism. By blending authentic archival photographs with generative prompts in Midjourney, he staged complex installations featuring surreal cultural motifs—such as a Berlin meat dish morphing into a soldier's helmet and ruined urban landscapes—that would otherwise have required Hollywood-level production budgets.

Generative synthesis enables independent creators to realize complex historical and cinematic visions without multi-million-dollar production infrastructures.
Iterative blending transforms authentic historical artifacts into surreal conceptual commentary through human thematic direction.

Compelling visual and narrative case study demonstrating how Eldagsen applies his theoretical framework in practice.

The Chess Paradigm for Transformative Creativity

watch

Eldagsen addresses AI doomers who predict the demise of human creativity by drawing an analogy to competitive chess. When chess engines outmatched humans, the game flourished as players utilized machine calculation to push human play to higher strategic heights. Applying this to Boden's model, Eldagsen resolves his opening paradox: AI is equal and superior to humans at combinatorial and exploratory tasks, meaning humans can outsource those dimensions to AI and focus their cognitive energy on achieving unprecedented transformative breakthroughs.

AI superiority in combinatorial tasks mirrors chess engines, which augmented rather than destroyed human mastery.
Humans should deliberately outsource combination and exploration to AI while cultivating transformative creativity as our unique domain.

Provides the synthesizing conclusion and philosophical resolution to the initial question of human-AI artistic coexistence.

Key points

Boden's Three-Tier Creativity Framework — Creativity operates across three dimensions: combination (mashups of familiar concepts), exploration (mapping possibilities within an established conceptual space), and transformation (fundamentally altering rules to create entirely new paradigms). Generative models excel at combination and exploration, leaving transformation as the distinct domain of human intent.
Promptography: From Solo Instrumentalist to Dataset Conductor — Traditional photography functions like a solo musician expressing an individual perspective, whereas AI-generated art ('promptography') positions the creator as a conductor harmonizing an anonymous collective choir embedded within vast training data.
The Threat of Full-Loop Automation — The authentic creative loop requires human subjectivity at the start (ideation and prompt formulation) and the end (aesthetic evaluation and curation). Automating the input via LLM reprompting and the output via automated selection removes the human element entirely.
The Chess Paradigm for Human Augmentation — When chess computers surpassed grandmasters, the game did not die; instead, players trained against engines to elevate human competitive play to historic heights.

“In the creative workflow I can do this at two points... you need a human being to prompt first to have an idea what to prompt... then again you need a human to evaluate was it good enough.” — Boris Eldagsen

“I was like the solo singer the solo instrument it was my individual perspective... now that I'm working with AI and this is what I call ponography... the training data is an anonymous choir and my role is suddenly a conductor.” — Boris Eldagsen

AI-generated from the transcript. May contain errors.

0:06

[Applause]

0:11

hello

0:13

everybody I have some questions for you

0:15

and I would like you to respond by

0:17

raising your hand question number one do

0:20

you think AI is as creative as you I

0:24

think

0:25

yes that's like 15%

0:28

10 do you think think I is more creative

0:31

as

0:33

you that's a little bit

0:35

more last question do you think you are

0:38

more creative than an

0:40

AI

0:43

wow so I was raising my hand three times

0:47

yeah why am I crazy why am I doing so we

0:51

going to get the answer by the end of my

0:54

talk and we'll start uh to have a closer

0:57

look at the question if human creativity

1:01

is replaceable by

1:03

AI in last Autumn there have been many

1:06

studies uh being released telling us

1:09

that AI is as creative as we are this is

1:14

the daily mail and Daily Mail is telling

1:16

their readers that AI is even more

1:19

creative and we had a different uh study

1:22

here in burin and the University of

1:26

eics and the question was very simple

1:29

what can we do with a toothbrush if we

1:31

are not brushing our teeth and the

1:34

result was very convincing they said CH

1:37

GPT could come up with as many um

1:42

results that are as original as humans

1:46

uh there was a different study in

1:48

Montana doing the torrent test of

1:51

creative thinking and it was like 201700

1:54

students versus Jet

1:56

GPT and the result here as well there

1:59

was had like four things they were

2:01

looking at and in two of those ease and

2:05

originality CH GPT was a more the best

2:08

percentage of those

2:11

tested what does it mean and what is the

2:14

future of creativity in the question is

2:16

what is creativity what we are having in

2:19

the debate of AI is that many terms are

2:24

not defined creativity is not defined

2:27

intelligence is not defined

2:29

consciousness is lot defined if you look

2:32

at

2:34

Wikipedia the common demeanor is that

2:37

creativity is something that is new and

2:42

valuable and um is that on a global

2:47

scale is that on a

2:50

continent country a city a school a

2:55

person how do you scale those

2:58

things and there is one creativity

3:00

theory that I like the most right now

3:03

because it can help us to get a better

3:07

understanding of what is going on with

3:09

creativity and AI right now this is ret

3:12

and bden have you heard of her before

3:14

anyone in the room

3:18

no her theory is not not that new and

3:22

she's quite an old person right now but

3:24

she studied computer science AI

3:27

philosophy and psychology for quite some

3:30

years so what she proposes is that

3:34

creativity has three different parts the

3:37

first part is combination and that is

3:39

what we all know from M Journey For

3:43

example Batman tressed as Barby it's

3:46

classical mashups It's a combination of

3:49

things surrealistic artists did it

3:52

before they use teacups and put fur on

3:56

it that's a

3:57

combination the second part is

4:00

exploration and I'm standing here I'm

4:02

having a certain perspective on the room

4:05

um and a I would be able to have an

4:08

overall perspective it could look below

4:11

your chairs behind that screen below

4:14

what I'm standing on and take it into

4:17

account and then we have the third part

4:20

that is transformation whereas the first

4:22

and the second is a given content a

4:25

given structure transformation would be

4:29

jumping out of this getting onto a new

4:32

level is that possible for

4:35

AI that is the big question that more

4:39

people could talk

4:41

about there are some feedbacks that we

4:44

get on the third question is um

4:47

computers can create but because they

4:50

have no awareness they have no

4:52

consciousness they can't be

4:55

creative but that is only creating more

4:59

problems because the question what is

5:02

consciousness um cannot really be solved

5:06

yeah it's a philosophical question you

5:08

could go to

5:09

university one semester two semesters

5:12

and you will have many answers but not

5:15

one

5:17

answer there was a study in

5:21

August a group of 19 scientists that

5:24

came up with the result that there is

5:27

evidence AI is the developing

5:31

Consciousness and that was picked up by

5:34

the media like many of those

5:36

things but if you look closely you

5:39

realize that how they define

5:42

consciousness is very

5:44

basic it's processing

5:47

information my smartphone is processing

5:49

information my computer is processing

5:52

information that means Consciousness

5:54

here in its definition is not attached

5:58

to a living organism and that makes it

6:01

problematic

6:02

so my first

6:05

um advice is whenever you read studies

6:09

about AI Consciousness Ai and creativity

6:14

take a closer look if there is a

6:16

definition of these terms and if that

6:20

definition is something you would

6:23

subscribe

6:25

to I'm going on back to my practice to

6:29

work with AI for really two years now

6:33

and to tell you more about the creative

6:36

workflow and what I've learned from

6:41

it man and machine are working together

6:44

it's a

6:45

collaboration this is the usual uh

6:50

collaboration if somebody is using an AI

6:53

image generator the machine is dragging

6:57

you behind it knows

7:00

what to look for what to decide how to

7:02

combine

7:03

pixels and you get a good result you

7:06

don't need to know about image making

7:09

photography composition and those

7:11

things if you have a knowledge of image

7:14

making you are having the leete because

7:17

you know the keywords you know how to

7:19

F if you have a knowledge of AI

7:22

platforms and you can identify strength

7:24

and

7:25

weaknesses you can combine platforms you

7:29

can inv event new workflows and then you

7:31

are having a very strong Le then you are

7:34

the director in a movie and the AI is

7:38

your assistant your set design actor

7:41

script writer camera and

7:44

sound and what I learned in working with

7:48

AI for two

7:50

years is um my role as an artist has

7:56

changed I was doing photography for

8:00

30 years for 15 years I was going out at

8:03

night shooting here in the neighborhood

8:05

at Museum's Island and I was like the

8:09

solo singer the solo instrument it was

8:13

my individual

8:16

perspective and uh that was a voice of a

8:19

single

8:20

human now that I'm working with AI and

8:23

this is what I call ponography it's an

8:26

ey generated image I have quite

8:30

the training data is an anonymous choir

8:34

and my role is suddenly a conductor I

8:36

need to make sense out of that training

8:39

data and I need to connect my

8:42

subjectivity to that Anonymous data that

8:46

I don't even know what it consists

8:49

of and that is a very fascinating

8:52

process and in the creative workflow I

8:55

can do this at two points

9:00

I'm going to explain

9:02

it most people think the creative

9:05

workflow is just generating images but

9:08

you need a human being to prompt first

9:12

to have an idea what to prompt to Define

9:14

what he would like to get then you have

9:17

the machine generating images and then

9:20

again you need a human to evaluate was

9:23

it good enough is something missing how

9:26

do I change the promp to get a better

9:28

result

9:29

but that can take many many Loops to

9:33

improve the result and to continue and

9:35

to continue sometimes you're stuck and

9:39

then a new feature is going to save you

9:41

or you need to change the platform or

9:44

the

9:46

workflow what we are going to have in

9:48

the future and what's already happening

9:50

with J GPT

9:53

for uh combined with d

9:56

E3 is that the first first and the last

10:01

step are going to be more and more

10:04

outsourced to AI when I'm using uh jet

10:08

GPT 4 to generate

10:11

images I don't even know what happens

10:14

with my text how it is transformed into

10:18

reprompt by CAD GPT 4 I can look it up

10:23

but they do something

10:25

so my skill in manually prompting is not

10:31

needed anymore and I think same things

10:34

are going to happen with evaluation and

10:36

for certain jobs that might be useful it

10:39

saves

10:40

time uh for me as an artist it's not

10:43

useful because the first and the last

10:46

step is where I can connect with the

10:48

training data and get my subjectivity

10:51

into the

10:53

process I like to work with AI because

10:55

it is a knowed booster

10:59

and that means I'm working with my

11:01

experience with my knowledge and I

11:04

combine

11:05

things and the combination is of course

11:09

of man and

11:10

machine but it's also a combination of

11:13

certain tasks do I use AI for

11:17

inspiration to analyze data or to

11:22

produce something or to use it for post

11:24

production or a combination of these

11:27

things

11:30

what kind of features do I combine to

11:34

create

11:35

something and I often start with text

11:38

prompts the result I blend the result I

11:41

use an image prompt the result I go to

11:44

in painting and then I go to out

11:46

painting and at the end I have an image

11:49

my workflow has become quite complex

11:51

it's not done in 4 seconds but that

11:53

combination of features is opening up

11:56

new

11:57

possibilities and I give you an example

11:59

of the work I did

12:02

recently and um that is called trauma

12:05

porn it's rooted in my family history I

12:09

had a very old father he was born in 24

12:12

and he went to

12:14

war uh when he was

12:16

16 he sured and he never talked about it

12:19

he got flashbacks at the end of his

12:22

life and he died in 2000 i r 30 there um

12:28

after his death I found images from my

12:32

family from My Father's Side from Nazi

12:35

times and I started to connect more and

12:38

more of those images I did this for over

12:40

20 years and um I turned this material

12:45

into a work by using original authentic

12:50

photographs as material to create images

12:53

with AI and Trauma porn is about a state

12:58

becoming

13:00

uh totalitarian system giving up for war

13:05

then War and the aftermath of war and

13:08

it's not seen Through The Eyes of the

13:11

victim or the perpetrators it is someone

13:14

of the big silent Mass someone that

13:17

possibly supported it in the beginning

13:19

or was too young and then was just

13:23

tracked

13:24

along so um in the

13:27

work that that I've done with my artist

13:30

friend tat from Bangladesh we have

13:33

original photographs and they are all

13:35

faint these are the photographs with the

13:38

little brown

13:40

background and this is the world of the

13:45

time that happened in front of the lens

13:48

and all the other images is something I

13:50

created uh with my partner tamat so that

13:53

was part one part two is war and there's

13:57

a trigger warning because I have

13:59

original images from World War II if you

14:01

don't want to see them close your eyes

14:03

for 10 seconds I'm going to tell you

14:05

when it's over so this is part

14:09

two again you have real photographs with

14:13

the brown background and the other

14:15

pictures we created out of

14:17

it and part three you can open your eyes

14:20

again it's a void it's the aftermath

14:22

it's a certain emptiness it's things

14:24

that are hidden that are not talked

14:26

about and how did I use AI here here and

14:29

what's the connection with what I've

14:30

said

14:31

before I'm using images for my archive

14:35

that I collected I blend them in mid

14:37

Journey that means it's a fusion of

14:39

images that create something new and

14:42

that is something that AI can do very

14:44

well but then I need to decide how can I

14:47

take it further what do I do out of it

14:50

and again it's another step in

14:52

creativity to goow from this result to

14:55

this result where you say hey I'm

14:56

playing with the 1936 Olympics

14:59

aesthetic I'm going to have this berian

15:02

meat dish and it looks like a helmet of

15:05

a soldier that is all done by me because

15:08

the

15:09

AI produced

15:11

something that sparked off this idea in

15:17

myself and then I have this idea of of

15:19

meat meat dish and so on it's a symbol

15:22

that I would like to use again um

15:25

another step using original photographs

15:28

blending it into this and then not

15:31

stopping here then it's me the director

15:34

taking this image further and saying

15:36

here it is again it's the bone from the

15:38

mid dish it's a destroyed City and the

15:42

little people are like ants gring up

15:45

that bone to

15:47

nowhere and out of this it's me the

15:50

director creating those images and

15:52

choreographing them I created three

15:55

different installations one was in a

15:57

gallery here in November

15:59

and then those images become objects

16:03

become an installation become something

16:06

that people interact

16:08

with and uh this work was only material

16:13

for 20 years until AI enabled me to take

16:16

it further because otherwise I would

16:19

have not known how how to get there I

16:21

would have needed a team of uh Hollywood

16:26

actors production lines special effects

16:30

and here it was just me and a

16:34

computer and that is something that um

16:38

is a very positive result for me I like

16:41

to work with AI and we have two

16:43

different teams

16:45

here that think um human creativity is

16:50

not needed anymore it can be replaced

16:53

it's all going downhill from here and

16:55

that is the P Doom it's the Dooms team

16:59

that think so and then you have the

17:00

other team that thinks um why does it

17:04

need to go downhill can it be augmented

17:07

can it become something else can it

17:08

become stronger instead and I think so

17:11

it can and it can why through um

17:17

interacting with AI

17:20

creativity there's one positive example

17:22

it goes back to chess I played love of

17:25

Chess when I was a

17:26

teenager and chess computers became very

17:30

very strong years ago and we as humans

17:35

can't win again a chess computer anymore

17:38

did we stop playing chess no we use

17:42

computers for training and what we are

17:45

having now is that humans playing chess

17:47

against each other on a much higher

17:50

level than before the invention of a

17:52

chess computer so that could be a

17:55

positive model for

17:57

creativity that we can get to um while

18:02

having AI creativity and interacting

18:05

with it so how could it look like going

18:07

back to maret and Bowden

18:09

system of combination exploration

18:13

transformation

18:15

so all the things that um had been

18:21

studied explored last year with the

18:23

creativity tests had been combination

18:26

and exploration and we have seen mean

18:29

that AI is very good in doing so this is

18:32

why I said yes AI is as good as me and

18:36

yes AI is better than me because it just

18:39

works faster in combining things so that

18:43

would be a model for the future where we

18:45

say um we Outsource two types of

18:49

creativity to Ai and we concentrate on

18:52

the remaining part of creativity as

18:55

humans to get more transformative

18:58

creative uh results than

19:02

before and um I hope this could give you

19:06

an Impulse to think about in the future

19:08

and to take a closer look because so far

19:11

I think not that many people

19:15

um discuss this but we need to AI is

19:18

going to stay and we need to find the

19:21

best workflow between man and machine in

19:24

the future so thanks for listening

19:30

[Applause]

Continue with YouTLDR

Analyze another video with Pro

Process a new video, search every timestamp, compare sources, and keep the result in your library.

Get Pro — $12/month30-day money-back guarantee

More transcripts

Explore other videos transcribed with YouTLDR.