A.I.: Savior or Destroyer? | PhotoVogue Festival 2023: What Makes Us Human? Image in the Age of A.I.

Can you guys okay, it s working right?

Yeah, you can hear me, okay.

Hi, I am Charlie Engman.

Thank you for that nice introduction.

So yeah, thank you for coming

to my very click bait titled talk.

I think there s a lot of kind of doomerism

and utopianism surrounding what s going on with AI

and so I kind of just wanted to highlight that.

But spoiler alert, it s neither of those things.

I kind of take a Buddhist approach to this,

which is, it exists, it s there, what is it doing,

what can we make it do?

Questions about going forward?

So yeah, some qualifications

before I start, I guess I m the last

kind of solo speaker in this whole panel,

which is both wonderful but also a bit problematic for me

because I prepared this long thing

where a lot of the topics have already been covered

in great detail and very well.

So I ll just skip around

and if I kind of bounce around in my logic

and something doesn t make sense to you,

please just come at me in the Q and A

and I can fill in the logic gaps where applicable.

But yeah, yeah, I think it s important to sort of address

and what I was trying to do with this title

is also kind of point

towards this idea that the kind of the meaning

of any technology is defined by how we the humans use it

and more importantly, kind of what incentive structures

are embedded within it

and then how those are reinforced

through kind of increased use and shared experience of it.

And if it s not already clear,

I m gonna focus primarily on image generation

through artificial intelligence.

And one way I m gonna start

is first I m gonna show you an overview

of the kind of things

that I have made with image generation,

but I m also gonna start with kind of trying

to explain how image generation works,

because I think that s often like a misunderstanding

of the technology is often what leads to I think,

some misconceptions around problems and issues with it.

And then I m also gonna talk

about how it informs my practice

and you know, all those other things.

I also wanted to kind of address this image,

which I think is already gonna hint at a lot of things

or a lot of central issues around this,

which this is, I literally just put into Midjourney,

which is, as you all know now,

a very popular AI image generator model.

I just put in the prompt AI Savior Destroyer

Photo Vogue Festival.

And this is the first one that came up.

So we re already seeing a lot

about the conventions of looking

that are sort of embedded into the models.

So obviously AI savior destroyer,

there s AI is represented by robots.

That s the sort of popular imagining of what an AI is.

Saviors are destroyers is this sort

of battlefield aesthetic, Photo Vogue Festival

I guess gives you a Ukrainian

or Eastern European looking white woman

of a very specific stature.

And I just thought that was interesting

that that s sort of the like preloaded kind of creature

that comes up in that.

So first I m just gonna show you a little overview

of my work and I had someone who s a little bit

more technically proficient than I am.

I just gave them the first like 3000 pictures I made

that I thought were interesting from my explorations and AI.

And I had an algorithm sort of organize them chronologically

and with themes that were sort of preloaded in.

And I set it to music that was made by Grimes AI.

So Grimes, if you all know, she s a musician

who has done an open source AI of her voice.

So the music that I set it to is also using AI tools.

So I m just gonna play this to foreground

so you know kind of where my aesthetics are coming from.

[upbeat music]

So everything there was.

[people clapping]

Yeah, so everything there including the music

and the structure of the video

and everything was kind of made

through different AI techniques,

which I think is very interesting.

But more so that s just to give you

an aesthetic foregrounding of who I am

and what I kind of look at.

Yeah, so what is AI art and why does it look like that?

Basically, I ll get into that.

So another kind of caveat is I am a lay person,

I m not a computer engineer, so I only understand

how it works through absorbing a lot of podcasts

and videos and articles and whatever.

So I m just gonna distill down

my lay person s understanding.

If someone here has a more technical information

and wants to sort of elaborate on what I m saying

or contest some of the explanations that I ve kind of,

or analogies I m making to explain this,

please challenge me afterwards.

This is a really great YouTube video by the way.

AI Art Explained by Vox.

Vox has actually a lot

of really good videos if you re trying

to understand just the technical

and sort of like mechanical processes around it.

I highly encourage that.

So yeah, so the steps of generating, so if we have a prompt,

a banana inside of snow globe from 1960, we get this output.

How do we get there?

And the general steps are you have training data,

you have deep learning, there s a latent space,

then there s a generation process

and then you get an output.

So I m gonna just kind of briefly scan through

what those are

and that will help inform

kind of the conversations going forward.

So first obviously is training data, which is basically,

what are the materials given to the model?

And for our purposes, a model is a program.

So training data is a set of examples

or input information used to teach a machine learning model,

providing the foundation for the model to learn patterns

and make predictions or generate outputs.

So for example, you can give it a million pictures of hands

and it will kind of learn what a hand is.

And I put hands here very intentionally

because they re very contentious part of AI.

A lot of people like to point to hands as an example

of how kind of inhuman and weird AI is.

And that s because hands are sort of notoriously

mechanically complicated.

They can do a lot of things

and they can be visually represented in a lot of ways.

And also humans have a very intense bias

towards the physical body.

So how we look at the physical body is really important.

How the physical body is represented

and manifested is something

that we humans have a very attuned

and very kind of sensitive relationship to.

Whereas AI has a very different kind of set of biases

and we ll get to biases of course.

Yeah, and you know also humans kind of,

we have sort of analog and schematic connections to things.

We take things kind of in as whole pictures

and our data includes experiential, emotional, spatial data.

And so certain things matter more to us than others,

but to a model kind of all pixels matter,

all pixels are equally important.

And so yeah, it s humans

that weigh different components of an image differently.

And so a bot s natural bias

is kind of more towards probability

and a computational defined form of accuracy.

So generative image, AI images,

they actually have a foundation in language,

which is often why we talk about prompts and text prompts.

And already by 2015, AI has developed

to generate captions in natural language from images.

So that s what image to text is.

So this came from a variety of sources,

including kind of alt text that people would do

for search engine optimization, for example,

people would label things to get them higher up on Google

and that was helping the original early bots

kind of figure out what they are.

And also disability access, trying to explain for people

with visual impairments what an image was.

Those all kind of went into doing the image to text.

And also these annoying captures that we all had to do

was kind of helping them understand like the word car,

how it s associated to pictures.

And kind of as a fun aside, there s some models

that have been trained on very specific archives.

So a lot of, for example, a lot of photos

have been labeled with text as like author unknown.

So there s a lot of models where if you type in unknown

as a general prompt,

you ll get a sort of like nostalgic

historical output that s based

on that kind of association with that word,

which I think is interesting.

And I just like these, these are some of the early,

very early kind of successes when it came to trying

to make a text image, which is sort

of reverse process of that.

And so you can kind of already see

that it s sort of getting at what it is.

But these are only 32 by 32 pixels at the beginning.

Yeah, so basically the training data,

let s say it s like the whole internet s worth of images,

which is what we like to talk about.

So let s say, okay, so you give

a whole internet s worth of images and text to a model.

Deep learning is what the model does

with all that training data.

So I think there s a common misperception which comes up

in like a lot of the copyright questions

for example, is that, there s not like individual images

that are stored, that are used

in piecemeal fashion to create images.

The models are actually sort of breaking things down

into vectors and variables.

So generally speaking, I think the models

kind of look at things

as kind of a collection of pixel values.

So like this for example, showing all the RGB values

for this little point on the pepper.

And so after kind of showing it a million examples

and having it iterate many a times,

we ll basically learn how to identify

and anticipate patterns in these picture arrangements

as they correspond to image descriptions.

So then through that we ll establish variables

with which to accurately describe various inputs.

So for example, like a pepper has a lot

of different variables, like it has color,

so it s generally a certain number of colors,

it has a certain kind of shape,

it has a certain kind of texture.

There s a lot of different variables

that you can kind of map.

So if you think about it as like an access

where color is one axis and then you have shape

and then you have shininess,

there s all sorts of different accesses.

And then basically inputting the text pepper

will cause a machine to sort of access a point

in that multi-dimensional space.

So that s how it kind of creates the latent space.

And this is where it gets really kind of complicated

and hard for even for me to understand

because it sort of surpasses human comprehension.

Cause we can kind of think in max four dimensions,

We have three dimensional space

and maybe time is the fourth dimension,

but these models are using over 500 different variables.

So using the sort of human approach,

I find it very helpful to kind of visualize

this sort of latent space

where actually all of this kind of thinking,

let s say of the bots

and the models are happening,

it s happening in this latent space.

And I like to think of that kind of as topography with peaks

and valleys or maybe as like interlocking webs

with strong and weak points of connection.

So the sort of density or like the density in the web

or sort of the peaks, the high points in the topography,

those represent kind of high

what we call high resource areas.

So that s things that have been

kind of weighted strongly in the training data.

For example, we ve given models a lot of face images

because we humans care about faces.

And then later, because we decided

we also cared about hands,

a lot of the models were then given a lot

of hands to reference.

So that s a very high resource area.

Also, English is a very high resource language.

That s something that s very commonly,

a lot of the languages are very biased

towards English naturally.

And then there s the sort of esoteric empty areas

that show a sort of the low resource parts

are sort of the impoverished parts of the data.

Things that are less well known to the bot or to the models.

And the more well known kinda the more embedded

something gets into the visual landscape.

So I kind of like kind of visualizing the sort of shape

of a model s imagination in this way,

although that s not really accurate

because we can t really fully understand how that works.

And I m gonna kind of skip through diffusion,

but this is kind of an arbitrary point,

but this is the last step that kind of gets you

into a model you can kinda look up diffusion later.

Yeah, and then you got an output,

which is an image in this case.

So this is the first AI image that I made

where I like many people, I started using AI tools

simply because they had become kind of accessible

and popularly available.

So I was playing around

just putting little text prompts and seeing whatever.

And I think this was something like,

two teen boys share their first kiss on a couch

that is also a horse in a suburban living room.

Something like this, this is the text point.

And this is just me being fun

with my friends and being silly.

And when this image came out I kind of thought like,

this actually has a lot of artistic,

creative, photographic value to me.

I find this image strangely compelling.

And also it s sort of confusing to me in a way too,

there s already a lot going in there,

like why did they make them this young?

That s not what I would imagine a teen to be.

Why is this this sort of common understanding

of what a teen looks like?

Why are they white?

I didn t specify their race.

Why does the hand look weird?

Why does the right hand pillow have no definition

while everything else does?

Why is the spatially reason incorrect?

But then it s still kind of in this weird uncanny realm

of acceptability, which also I think is very interesting

because the Overton window of acceptability

is kind of shifting already

as culture has kind of absorbed

and processed the aesthetics of AI.

And I was also very interested in how quickly I was able

to sort of move through this kind of wow factor.

And I think culture has the same kind of ability

to absorb things and create

kind of expectations out of them.

So parenthetically, I also just got really interested

in how AI s generated horses

because horse appears to be a very kind of dense

or high resource point in the latent space

of many models, but they re also kind of

as mechanically complicated as hands.

So I found the way that they generated them

and kind of the way that they had them interact

with with space was really interesting.

This is a, yeah, this is a work I made for Gucci

that was exhibited at Miaki here in Milan.

That was kind of an exponent of my obsession with horses.

Yeah, and I was also interested really specifically

in how it depicted intimacy

and how that was kind of relayed onto the human body.

So what were the conventional metrics

used to describe intimacy

and where did the nuance get lost

in the algorithm, if at all?

And kind of also around that,

what expectations are we bringing to depictions

of bodies and depictions of intimacy?

So these are some very early,

and I can t believe I m showing this to a public audience

cause I made this when I was in university.

This is me like with a self-timer

in my boyfriend at the time s House.

So these are like very early photographic works

that I had started to make

right as I was getting into photography as a medium at all.

And they re based on my sort of interest in dance

and performance, which I was very involved

with before I got into visual medias.

Yeah, and I think, I m showing these

because I think they kind of help explain why I personally

as a kind of creative person sort of took

to AI generative tools like a duck to water.

So yeah, and I think these help illustrate

kind of some of the questions

that are always at the forefront

of my artistic practice.

Questions like how do my images relate to the images

that proceed and surround them?

What expectations are we bringing to images

and how are these expectations established and reinforced?

What purpose do these expectations serve

and how can we interact with these expectations?

Yeah, so I m kind of interested in looking at familiarity

as it relates to vulnerability.

So why does something become familiar?

Why are we invested in protecting

this feeling of familiarity?

So these were kind of earliest sketches of me

trying to think like, okay, my body is the most thing

I have most familiarity with.

How can I sort of destabilize that idea of familiarity

with myself and with an audience?

Yeah, so here s some more examples of that.

And yeah, and then I think I m probably best well known

for the work I ve been doing with my mom.

I ve been photographing my mom for over a decade.

I, in 2020 released a book called Mom

that encapsulates a lot of that work.

And part of that work was about

kind of taking the singular subject of my mother

and treating her in a range of different affectations

as a means of revealing the sort of mechanics

and emotional qualities

that were embedded in those affectations.

So for example, if you take an iPhone photo,

how does that kind of affect a subject

versus if you kind of give it very expensive lighting

versus if you use kind of, yeah,

various techniques applied to a singular subject

kind of reveals what those techniques

are actually doing and how they impact.

And I feel like AI has a very similar way of like adducing

and revealing conventions

of looking and conventions of making,

and this is kind of what my current,

my studio wall currently looks like, which is me

sort of trying to make my own

kind of latent space around my work

and trying to figure out what are the sort

of like high resource areas

and low resource areas in my practice

and trying to fill in those gaps

and make those kind of connections.

So that s again, just to kind of bring my own work

back as a grounding.

So yeah, let me get into the common critiques of AI

and I realize I m already going way over,

so I m gonna zip through a lot of these.

Obviously, one is data bias,

which was already covered in amazing talk yesterday

by Mutale Nkonde.

So I m kind of gonna skip through this if we have questions

that we want to get into about data bias,

I have a lot of opinions so you can ask me,

but yeah,

and part of that is representational harm

and I think, I just find it very interesting here

that especially because Midjourney,

which generated all of these,

it s a proprietary software that was trained

kind of mostly on DeviantArt, which is predominated

by certain forms of illustration like comic books

and video game art.

And so kind of what does it mean when you bring

that into a more general use,

those kind of aesthetics become embedded

in kind of visual culture in a very interesting way.

This I just, yeah.

So another common critique is it can be used

for propaganda misinformation, which I think

is kind of one of the most salient critiques of AI

and is a real and present danger.

I m just gonna play this deep fake cause I love it.

All right everyone, so it s Friday night,

I m getting ready to go out, I m feeling cut,

but before I go out I ve gotta pregame

and the only way I pregame

is with and ice cold Bud Light.

So good, happy Friday.

So that s Joe Biden feeling cunt on a Friday evening,

which just boggled my mind the first time I saw it.

Yeah, so, and I think what s interesting here

is that these really reflect the kind of biases

that are already present in media consumers.

So something s perceived

as true if it corroborates your worldview

and is fake, if it contradicts it,

basically haters will say it s AI.

And this I think is part of a broader general trend

of distrust in legacy media and institutions.

So yeah, and I think

we can get into this in the talk, the panel discussion later

or in talk backs, but I think this requires,

the solution is both about regulation

and about increased education around data literacy.

And there s a lot of other systemic solutions

that I think are involved in that.

Another really important critique

that I actually think it s left out

a lot is the environment.

AI takes a lot of electricity

and it s pretty damaging to the environment.

And there s no equivocation I have about that.

That is just the fact.

And I think it s important to kind of remember

that that s part of the conversation.

And again, I think this involves regulation,

et cetera, as a solution.

And probably the most popular critique I hear

is that AI is sort of the death of creativity.

And I think the fact that you re all here

is that you re probably on the other side of the argument

where you re at least thinking

that creativity can lurk around AI.

So I don t think I need to get into this too interestingly,

but I do think that there s a lot

of interesting things that are brought up in this critique.

So yeah, the critique is that AI is uncreative

or creatively lazy

and will ultimately lead to the death of creativity.

And first to that I would like to say

that if creativity can be killed,

I think we should kill it posthaste so that we

can kind of move on past it as a marker of value.

If creativity is something that is this

kind of fragile, like let s find something

that has a little bit more robustness

to invest our energy into.

But yeah, I think this view of the kind of death

of creativity through AI is based on a number

of assumptions about what creativity is and how it works.

So one is around ideas of craft and labor.

So I had a New Yorker profile written about my AI work

and the clickbait kind of headline

or subline around that profile

was me saying I can make 300 pictures a day,

which obviously got the comment section really inflamed.

People really were in my DMs

and my inbox about me saying

that I can make 300 pictures a day

and how this kind of cheapens,

this is part of the cheapening of creativity.

And so I think within that

there s the kind of pervasive idea that labor

should be visible in a creative output

and that the only kind of meaningful labor

and time is what goes into the actual construction

of a work of art.

So yeah, the value of something is basically apportioned on

based on how hard it is to make,

which is kind of the classic,

my kid could make that, argument,

which is kind of a gross flattening of taste.

It sort of makes a claim to what is valuable

or tasteful kind of, I don t like it

and so it s not valuable, basically is the argument.

And I think it s also a willful ignorance

of context where how a thing is received,

where the visual goes, context matters

in how it s perceived creatively.

I think it also presumes that the labor

of the artist is in hierarchy over other forms of labor,

which I personally take a lot of issue with.

Labor is present across all aspects of life

and all goods and services.

You know, for me as a practicing artist,

my entire life is a form of labor

and service to my artwork, if we wanna put it that way.

All of my experiences, all the research I have,

all of my kind of knowledge

around whatever is a kind of labor

that goes into any type of creative output that I have.

And so like for example, this is something

that I often find also is missing.

Like this is Kenyan labor that is present in AI software

that is routinely invisibilized.

So there s questions of supply chain

that I think are being kind of left out

or purposefully sort of avoided in those kind of critiques.

Yeah, so what this was basically was that

like real human Kenyans were employed

to get the sort of filters in AR.

So they were to manage manually label toxic imagery

in order to make the labels sort of consumer ready

so that it wouldn t always,

obviously if you train a model

on the entire Internet s worth of images,

you ll get a lot of porn, you ll get a lot of violence

and you needed a human presence to sort of train that out

and to kind of weight the models so against that.

And they used real human people who were getting traumatized

and underpaid to do that work.

So it s important to kind of think

about when we re talking about labor

as it revolves around creativity,

what kind of labor are we invisible in that conversation?

And I think that s kind of related to this idea

of the kind of myth of the individual genius,

the individual creative genius.

And this is sort of again, this idea,

there s an idea that the model is the artist

and the inputter is just sort of,

the prompter is kind of irrelevant,

which I think is a polemic of creativity

there s already been answered by the ready made

and other art movements.

And you know, also like Jeff Koons,

he s not actually creating any of his work physically,

but we still sort of give him the credence of the artist.

Yeah. And I think this response is also intention

with the argument that yeah, the AI models of the artists.

Yeah, and I think there is an argument

or there s an argument

that at some point error, randomness,

happenstance stops coming under the purview

of the artist in a meaningful way.

Which is an interesting question.

So like is the randomness

of the paint splashes in a Jackson Pollock for example,

when we think about that, does Jackson Pollock

have a creative claim of ownership over the randomness

of his paint splatters?

I think that s kind of an analogy.

I would encourage people to read this book.

I really got a lot out of this book

Shanzhai: Deconstruction in Chinese by Byung-Chul Han

who s a German Korean philosopher.

And so Shanzhais are basically like Chinese fakes.

They re things that sort of like make puns

and kind of take like we have a Harry Potter

and the porcelain doll.

It s sort of reiterating

the kind of known quantity of Harry Potter.

Yeah.

So their creativity which cannot be denied,

is determined not by the discontinuity

and suddenness of new creation that completely breaks

with the old but with a playful enjoyment

and modifying, varying, combining and transforming the old.

So I think that s also important to kind of,

this sort of helps clarify

what my personal view about creativity is,

is that it s something that we ve kind of done collectively

and we re all kind of constantly doing and it is a process

and conversely under a sort of capitalist ideology,

every utterance is viewed as the property

of the utter and non-physical ideas

and images which are infinitely reputable.

They can be viewed as private property, which I think is,

this leads up to, well this is also,

I just quickly threw this in here

because I saw this show by Camille Henrot

at the Fondazione ICA Milano,

it s still up, I encourage you to see the show

and I really like this little quote that she put here.

She certainly never founded anything.

She had loyalty to sameness,

good artists reproduce the resemblance of old work

but not the very same work.

So I think that s kind of an important framing

to keep in mind.

So yeah, obviously the most important critique

or most loudly opted critique

is the copyright that AI is dealing.

Yeah, and yeah, first I think

some of the explanation I made of how the techniques work

I think already helps kind of unpack some of this.

But there s already been a lot

of conversations around copyright.

I have a lot to say about it, you can ask me about it,

but I m gonna kind of just skip through those things.

But one thing that s kind of interesting here is,

so I ve been trying to train my own models on my mom

because I ve made thousands of pictures of her.

And so I have a lot of material

that I could kind of create my own model.

And I think this is really interesting that it s sort of,

that brings up more interesting questions to me

around consent and around what does it mean

to have ownership over someone s likeness.

Yeah, those are kind of interesting things.

And then we also have,

copyright I think is also a very flimsy mechanism

for sort of understanding and kind of policing creativity

and also, there s a lot of problems of access,

who has the ability to litigate, for example.

So, and I think we have a lot

of like social mechanisms like shame for example.

So cultural preparation is an example

where shame has been developed as a cultural mechanism

to sort of rear orient or direct attention

and other resources that have utter unfairly been usurped

where copyright probably usually fails copyright law.

Yeah, I realize I m like way over time.

So I m just gonna kinda skip,

basically the problem is capitalism,

which we can talk about it.

But yeah, I think a very legitimate concern

is that a lot of people s data,

a lot of the training data was harvested in service

to basically three mega corporations

and the idea is that they re kind of the sole

financial beneficiaries of the use of that data.

Yeah, and so then the argument

is that everyone whose data was being trained,

they should be kind of paid for that use of data.

And again, I think that goes back to this idea

of like whose labor is being offered payment,

who deserves the sort of recompense for what labor?

And I think if you take that kind of logic

of copyright infringement to its conclusion,

everyone should be paid

because we re all participating in society.

And so I think there s kind of like two extremes of that.

It s like either I m the boss or we have socialism.

Those are the kind of two things

because I think there s a legitimate jealousy there

that the people kind of higher up

in the food chain, like these big tech companies,

are getting some kind of value out of this.

And you know, either the artists want to be a beneficiary

of that exploitation process

or the artists take issue

with the entire exploitation process in general.

And I m more sympathetic to that latter.

And so I kind of believe that the energy

that s put around copyright and like whose labor

is being exploited, where, that s a really good energy

and really good conversation we should have

that s being a little bit mistargeted I think

in the conversation around AI.

We should be kind of looking at sort of social systems.

So yeah, so another critique, AI will take our jobs

as artists and I think it s more important

to kind of think generally and systematically

about how artists are valued.

How are we paying for that labor

and the labor that sort of surrounds the work of artists.

Yeah, because of course jobs will be lost

and new jobs will be created,

as with any technology.

And it s important to kinda remember that it s bosses

who will get rid of jobs, not machines.

But yeah, we can get into that and talk back.

So I can pick up some speed.

But this is basically, this is an amazing video

and actually Daniel Felstead and Jenn Leung

have made a series of really great videos

for Disc Magazine that I would encourage you to watch

cause they re really entertaining

and they think they say a lot of good things.

So this basically sums up my position.

AI is going to take all of our jobs and render us useless.

And I for one am stoked, I hate jobs.

[Speaker] I hate jobs.

I had a job once

and everyone there talked in weird voices,

AI is gonna 86 all of that.

But we ll still need money.

That is why I m asking the government to step up

and make sure we re breaded.

We are proposing a small payment plan

or a small PP of 10G month for every citizen

so we can party and look hot and enjoy our free time.

Yeah, so AI to UBI, let s get into it.

We can talk about that.

But yeah, okay,

so probably actually the more important thing,

and I m way over time so it s sad

that I m rushing this part, but basically this is me

trying to figure out how to manage capitalism

and creativity as it relates to AI within all of this.

And obviously I think it s important to foreground

that I had kind of the privilege

of a preexisting photography career,

which was built on top of a lot of different accesses

of my own privilege as a person.

And so I personally have benefited directly from AI.

Here I am talking to you guys about it.

Yeah, but also AI kind of already exists

in a lot of different tools and interfaces

with things that we re already benefiting

and suffering from.

Yeah, so this is some of the work that I made

that s sort of a hybrid process

of kind of traditional photography, AI

and a lot of other different things that I made

for Vogue China, which felt very relevant.

It was actually brought up in the previous talk already

with Margaret Zhang who s the editor there.

Yeah, and I think there s something

very important around this idea

of kind of physical reality and touch.

And so an issue in integrating AI into the commercial world

is that there s sort of an element

of randomness that s built into the process

and when you kind of prompt something, it goes

through this sort of computational randomness

in order to get out a result,

which is why you can t reliably get the same image

over and over again even if you put in the same prompt.

And so this obviously has an impact

where I did this campaign for Acne Studios

and they wanted to show the bags in a very literal way,

this is a product they were trying to highlight.

AI doesn t have an ability yet to do that.

So there s already an interesting sort of refraction

and sort of need, callback

to the physical world in this process

that I think is important to kind of note

that we re not there yet with AI basically.

And I don t know if this was already brought up,

but there s this very kind of contentious use,

Levi s was using this for E-comm.

This has been talked about in previous talks,

so I ll skip through that.

But yeah, I think like something really important

to think about is a sort of detachment

from the physical world that s sort of built in

to the use of technologies for visualizing.

And so what does it mean

when we sort of leave the physical world?

I think that s a good question

to sort of meditate on around AI.

And then quickly, I also work as the art director

for the fashion brand based in New York, Collina Strada.

And so part of the process we did

for developing the last runway show that we had

was we were using AI tools to develop and iterate

and kind of concept the process.

And so this is something that we did

where we input the entire archive of the Collina Strada

plus lots of different textual ideas

of things we were thinking,

different kind of references or whatever.

And we kind of iterated and iterated and iterated

until we got certain kind of models.

So you know, very similar to what we see outside.

And then we actually had to figure out

how to kind of physically instantiate these things.

So here s an example of something that we were able

to sort of conceive of in AI

and then this is the sort of

physical output that we ended up.

And we ve also done this kind of in a visual way.

We do this with making prints, for example.

So this is a print that I made

kind of in a very traditional way with Photoshop and camera.

And then we took some references

from some of the dead stock materials that we use.

We can blend them, mix them around,

them with AI and we get something like this.

Ultimately, I think though kind of the main issue

with AI is its relationship to scale

and this idea of productivity

and kind of overabundance,

which I think is both its strength

and its weakness is how it kind of relates to scale.

I m grossly skipping through this so I can just end

and we can get to the next conversation,

which is gonna be more interesting.

But yeah, I think in the commercial world

I ve already come up a lot

against this sort of issue of productivity

as it relates to AI.

There s this idea of, oh, you can make 300 pictures a day,

that means we should make 300 more pictures a day.

Rather than kind of using that as a way

to maybe create more leisure space or kind of sit back

and kind of give more space to creativity, for example.

So I m gonna leave you with another clip

from that video I showed you earlier

because it s just more entertaining

than I am and this kind of encapsulates everything

and then we can move

right into the next conversation.

Technology always acts

all to cast our desires and fears.

And not to get all psychoanalytic on you baby girl,

but don t you get the impression that in both the utopian

and Duma rhetoric, AI is, you know, being used

as a displacement for the batshit crazy horrors show

that is contemporary capitalism.

But instead of treating AI as salvation

or apocalyptic, what if we like understood it

as an abstracted mirror of our present

in all its deranged, fucked up wonder?

Period.

[person cheering]

[people clapping]