April 25, 2025 • 62 mins
Guest: Dr. Christina Shivers, School of Architecture at Georgia Tech.
First broadcast April 25 2025. Transcript at https://hdl.handle.net/1853/77549
Playlist here
"This is like a split single for Lost in the Stacks."
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
[MUSIC PLAYING]
CHRISTINA SHIVERS (00:01):
The act of drawing and design
really has significantconnections to music.
I mean, even the processof musical notation,
there are so many reallyfascinating connections
between musical notationand architectural notation.
They're both languages--graphic languages.
And in both cases,music, and architecture
have significantcommonalities because they
(00:24):
both are disciplinesthat are conducted
through the act of drawing.
CHARLIE BENNETT (00:52):
You are listening to WREK Atlanta,
and this is Lost in theStacks, the research library
rock and roll radio show.
I'm Charlie Bennett in thestudio with everybody--
Alex McGee, Marlee Givens,Fred Rascoe, and Cody Turner.
I can't say anybody's name.
FRED RASCOE (01:08):
Rascoe.
CHARLIE BENNETT (01:09):
Rascoe.
Each week on Lost in theStacks, we pick a theme
and then use it to create amix of music and library talk.
Whichever you are herefor, we hope you dig it.
ALEX MCGEE (01:18):
Our show today is called
How to Train Your Algorithm.
That's the titleof an exhibit, you
can visit in theGeorgia Tech Library
right now on the secondfloor of Crosland Tower.
MARLEE GIVENS (01:29):
It is a set of objects
made from recycled paper, meltedplastics, wood, Styrofoam,
burlap, and probably a fewother unexpected materials
displayed on stands made fromunpainted, unvarnished wood
planks.
FRED RASCOE (01:42):
And it's all about AI.
CHARLIE BENNETT (01:44):
Yeah, this show is like a split
single for Lost in the Stacks.
One side is AI, and the otherside is the material umph.
ALEX MCGEE (01:52):
How to Train Your Algorithm is an ongoing project
exploring AI withmaterial practices created
by Professor Christina Shivers,the 2023 Ventulett NEXT
fellow in the School ofArchitecture at Georgia Tech.
MARLEE GIVENS: Charlie spoke with Dr. (02:06):
undefined
Shivers in the library'sscholars event theater in front
of a live audience duringMedia Arts Day 2025..
And our interview segments todaycome from that conversation.
FRED RASCOE (02:17):
And our songs today are
about architecture anddesign, repetition,
and getting machines to workwell with humans and vice versa.
No matter how many experimentaltechnological tools
we have at our disposal,the product of design still
has to incorporate thehuman factor, right?
Doesn't it?
So let's start with humanfactor by Music for Pleasure
(02:40):
right here on Lostin the Stacks.
CHARLIE BENNETT: Be assertive, Fred. (02:43):
undefined
[MUSIC FOR PLEASURE, "HUMANFACTOR"]
That was Human Factorby Music for Pleasure.
Our show today is calledHow to Train Your Algorithm.
Our guest is Dr. ChristinaShivers, an academic, architect,
and musician whoseresearch broadly focuses
(03:04):
on the intersection ofenvironmental and economic
thought specifically withinthe design disciplines.
ALEX MCGEE (03:09):
Dr. Shivers discussed her work
in an interview recorded livein the Georgia Tech Library.
For our firstquestion, we asked,
was it a straight line fromyour work in urban landscapes
and reclamation tothis AI project?
CHRISTINA SHIVERS: I think there's (03:21):
undefined
a couple steps in between.
I mean, my background is indesign and making, so music
and design.
So I always sort of had asomewhat applied approach
to things, or to the world.
And then throughstudying for a PhD,
I had to start thinking aboutthings very differently.
So how do I write anddo historical research,
(03:43):
understand primary sourcesas almost my kind of material
that I'm working with moreso than the act of drawing.
It's very kind of--
not a learning process,but just a different way
of thinking or using mybrain a little differently.
And so in coming back intomore of a design setting,
as a professor and as theVentulett NEXT fellow here,
(04:05):
I began to startthinking about passion
I have for making drawings,for making models,
for really investigatingarchitectural and craft
questions, how do Ibegin to approach that
through the lens of all of whatI was doing with my dissertation
and with my research?
And I think there are many majorenvironmental questions facing
(04:28):
architects today, from whatsort of materials do we use?
how are the materials processed?
what kind of chemicalsare in them? how
are they coming frompoint A to point B? where
are they extracted from?
Those are just some questions.
I mean, there are otherquestions about the carbon
emissions buildings produced.
So beginning to approachmy project last year,
(04:52):
AI, as we all know, has reallyballooned in terms of popularity
in the last twoyears, three years.
I guess ChatGPT emergedin January of 2022.
CHARLIE BENNETT: Yeah, there seems (05:04):
undefined
to be a metaphoricalstarting gun.
But every time we talkabout it, there's always
been the phrase, well, thathad been happening before,
but it just hit the popularconsciousness somehow.
CHRISTINA SHIVERS (05:17):
Yeah.
And so in architecture,I think it's
been a bit of an open questionabout how one uses it.
For many people, it has amazingpotential to produce renderings.
Renderings are one ofthe things architects
kind of have to do to show theclient what the building, what
the project will look like.
(05:38):
And many architectswill approach renderings
as like a sort ofnecessary evil.
So with AI, you can justexport renderings or out--
[CHUCKLING]
CHARLIE BENNETT (05:47):
Are renderings the pitch PowerPoint?
CHRISTINA SHIVERS (05:52):
A rendering is more of like a--
you model your buildingin a 3D modeling software,
and then you use a renderingsoftware to basically put all
the materials, lighting,put it in its environment,
make it look very realistic.
Because a clientwill often not really
want to look at plans, andsections, and elevations.
(06:14):
They want to knowwhat it's going
to feel like in the building.
So AI has a lot of potential forrendering uses in architecture.
I was really interestedin what other questions
I posed, particularlybecause it has this--
it does have an immenseenvironmental toll.
But it also is, in many ways, akind of reflection of ourselves,
(06:34):
of our society, because it'strained with all of our data.
There are many controversiesaround the use and acquisition
of that data, but it'sbasically, in this way,
a kind of mirror of our society.
So I became interestedin trying to understand,
potential uses oflike an AI platform,
(06:56):
how that couldmaybe begin creating
unexpected stories orunexpected ways of understanding
our society.
And I was interestedin primarily
in understanding the kind ofenvironmental underside of that.
CHARLIE BENNETT (07:11):
So when you say AI is a reflection of society
or the culture atlarge, were you
interested in messing withhow society was reflected?
Were you interested inmessing with society
so that it wasreflected differently,
or just messingup the AI itself?
CHRISTINA SHIVERS (07:29):
So I'm interested in--
so AI has very particularways in which it functions.
It is trained to reproduceidentifiable objects.
And it has a really difficulttime thinking abstractly
or producing abstract images.
CHARLIE BENNETT (07:47):
Wait.
Let me interrupt you there.
Has a difficult timethinking, you said?
CHRISTINA SHIVERS (07:51):
Or I guess not thinking, but producing.
CHARLIE BENNETT (07:53):
But is that what you kind of--
the thinking ofthe AI is when it
tries to produce something outof what raw materials it has
or raw data it has?
CHRISTINA SHIVERS (08:03):
Yeah.
So it will revert to somethingthat looks like an object--
that table, something like that.
And I was really interestedin trying to say,
what happens if you try tocircumvent that process?
What would these AI algorithmsand softwares begin producing?
(08:26):
And so I was reallyinterested in developing
a series of what are called--what I called camouflage models.
So noise algorithm cansometimes kind of mess
with the imagerecognition parts of AI.
There were--
CHARLIE BENNETT (08:41):
I think we should unpack that real quick.
So a noise algorithm--what is that?
CHRISTINA SHIVERS: Yeah, so if you (08:46):
undefined
think about static oran old TV when it just
is static on the screen,that's noise, basically.
It's just visual nothingness.
It's the same with sound, solike white noise, brown noise,
all these different noises.
CHARLIE BENNETT (09:06):
It's wild that static on a TV
is not really a thinganymore, that there's
a moment when people didnot grow up with that.
But white noise or pinknoise is still totally
in people's consciousness.
CHRISTINA SHIVERS (09:17):
Yeah, I have a white noise machine
I use very frequently to sleep.
[CHUCKLES]
CHARLIE BENNETT (09:22):
So a noise algorithm
is something that generatesthat kind of static or disarray?
CHRISTINA SHIVERS: Yeah, something-- so (09:27):
undefined
because of the way that theimage producing AI systems
work, they basicscreen of noise,
and they slowly whittle it downinto something like an object.
CHARLIE BENNETT (09:42):
Oh, wow.
CHRISTINA SHIVERS (09:43):
Yeah.
And so if you interfere withthat process with noise,
it can throw it off.
So there were someartists fairly recently--
I don't remember how long ago,but in the last two or three
years--
who were basically printingstickers with a noise image
and putting it ontheir face so it
(10:04):
wouldn't be ableto recognize it.
I'm pretty sure the algorithmsand the companies that
produce these AI algorithmsare always updating.
So I'm sure they've kindof circumvented that a bit.
But I still was veryinterested in this process.
So I created a sort of seriesof abstract three-dimensional
(10:26):
models in a 3D software thathad different noise algorithms
applied to their materiality,to their surface.
CHARLIE BENNETT (10:36):
So if I can try to understand it
in a very simple metaphor--
I know you know thisphrase, but Michelangelo
said the way to sculpt somethingwas to look at the raw material
and then take away anythingthat wasn't the sculpture.
It sounds like you arethrowing rocks at a sculpture
while it's being sculpted.
CHRISTINA SHIVERS: Yeah, so basically, I (10:56):
undefined
created these things.
And I asked an AIprogram, what is this?
And in all the cases,it would come up
with really interesting answers.
There's a captiongenerating software
called Pallyy AI softwareis kind of in this process.
(11:17):
But I asked it, what isthis? or describe this?
And it would say somethingalong the lines of,
this is a blue and green orturquoise and green plastic bag.
I like to put myart supplies in it.
CHARLIE BENNETT (11:32):
Whoa.
Where did that come from?
That additional--
CHRISTINA SHIVERS (11:38):
So that image or caption generator
is for use for Instagramand for basically quickly
generating captionsthat, I guess,
you don't want to think of.
CHARLIE BENNETT (11:54):
I hate this so much.
I'm so upset.
I don't know if I cancontinue the interview.
MARLEE GIVENS (12:02):
This is Lost in the Stacks.
We'll be back with more from Dr.Christina Shivers about messing
with AI after a music set.
FRED RASCOE (12:09):
You can file this set under NA 2520.H73.
[MANOR, "ARCHITECTURE"]
[GENTLE GIANT, "DESIGN"]
ALEX MCGEE (12:26):
That was Design by the not-Welsh band Gentle giant.
CHARLIE BENNETT: You don't know that. (12:29):
undefined
[CHUCKLING]
ALEX MCGEE (12:32):
There's a debate.
CHARLIE BENNETT (12:33):
Just sounds like it was
recorded in a field near cows.
ALEX MCGEE (12:36):
And before that, Architecture by Manor.
Songs about, well,design and architecture.
[MUSIC PLAYING]
MARLEE GIVENS (12:48):
This is Lost in the Stacks,
and today's show is calledHow to Train Your Algorithm.
Dr. Christina Shiversfrom Georgia Tech's School
of Architecture hasbeen exploring AI
and materiality usingimages, noise algorithms,
and fabrication.
You're listening to excerptsof an interview with her,
recorded live in the libraryon Media Arts Day 2025.
CHARLIE BENNETT (13:11):
In the last segment,
we talked about howthis project works.
Let's review.
Dr. Shivers createsan image of an object
obscured with visual noise, whatshe calls the camouflage model,
and then asks an AI systemto identify the object.
That identification becomesthe prompt for an AI generated
(13:33):
image.
FRED RASCOE (13:34):
Then, the camouflage model and the AI
image are blendedinto a new image.
And that new hybrid imageis used as part of a prompt
to design anarchitectural pavilion.
ALEX MCGEE (13:46):
Which will eventually
be built using recycledand unexpected materials.
If you need more detail orwant to see these images,
the whole process is onchristinashivers.com--
Christina spelled with a C-H.
CHARLIE BENNETT: This AI system-- (13:59):
undefined
don't even have to name it--
you showed it this image, or itcreated this image from text?
CHRISTINA SHIVERS (14:06):
So I showed it the image on the left.
It said it was a blueand green plastic bag.
And then I took thesentence that it
gave me and had it produce theblue and green plastic bag.
Yeah, so it's like onestep of translation
from left to the next one.
CHARLIE BENNETT (14:26):
And how is it that it can make the plastic bag
from an image that, ifI can just describe it
very quickly foranyone listening,
it looks like a landscapewith some popsicle sticks?
I mean, I don't-- help me.
I don't see how it canthen say that that's a bag.
CHRISTINA SHIVERS (14:48):
That's why it's really fascinating.
That's what I thought reallybecame really fascinating about
this process is the objects thatit's-- the data that is trained
with it reverts to something.
And so something asubiquitous as a plastic bag
that we see everywhere thatwe use for everything, it's
like, OK, I guessthis is a plastic bag.
CHARLIE BENNETT (15:10):
So the AI was able to peel the noise off
of that image andget a plastic bag?
CHRISTINA SHIVERS (15:17):
It said it was a plastic bag, yeah,
through its computer visionimage recognition process.
CHARLIE BENNETT (15:24):
So if I go back to my silly metaphor
of the sculptor and throwingrocks at the sculpture,
do you not like the sculptor?
Are you tired ofwhat he's creating?
Are you messing with the AIin order to know it better,
or are you trying to figureout ways to disrupt it?
CHRISTINA SHIVERS: I'm trying to-- (15:41):
undefined
I think it's probably both.
I think it has potential.
With any tool, it's reallyinteresting to use it in a way
that it may not beintended to use.
Many, many artists andmusicians use things
quite differentlythan maybe the way
(16:03):
they were engineeredand stumble upon,
in the process, something new--
a way of using thatmedium that can reveal
something that is unexpected.
So I think that's part of it, somessing with the AI algorithm.
A, I am trying toget under the surface
to begin understandingwhat the data
(16:26):
that it's trained with startsto-- what it starts to reveal
about that dataand, in turn, how
it is this reflectionof our world around us.
So the plastic bagis so ubiquitous
that it looks at thatimage on the left
and it's like, Iguess this is a bag.
CHARLIE BENNETT (16:42):
Wow.
So there's the noisy image, andthen the rendering of a prompt--
that noisy image, andthen another landscape.
So what's the next image then?
CHRISTINA SHIVERS (16:55):
That is a combination of those two images.
So basically, justthose two images
were blended togetherto begin understanding--
starting to develop a processof how this can create
an architectural application.
CHARLIE BENNETT (17:12):
And did you do that with an AI tool--
CHRISTINA SHIVERS (17:15):
Yeah.
CHARLIE BENNETT (17:16):
--or is that--
OK.
It's just digesting,digesting, digesting.
CHRISTINA SHIVERS (17:20):
Yeah, it's a kind of recursive process.
And so that third stepis the bridging point
between the objectthat's been described,
but then kind of incorporatingwhat it originally was used
to create that description.
CHARLIE BENNETT: So again, I'm going (17:40):
undefined
to do for people listening.
So I see what looks likea landscape with popsicle
sticks stuck in it.
And what AI describethat as used as a prompt
produces a pretty realisticplastic bag photo.
I mean, I'm looking at itnow and I can tell, oh, yeah,
that's not real.
But I didn't until you told me.
(18:00):
And then the nextimage is actually 3D,
and it is another landscape,like maybe some rock strata with
plastic bag handles straight up.
Is that what you see?
CHRISTINA SHIVERS (18:12):
Yeah-- sort of also kind of melted plastic
almost in some way.
CHARLIE BENNETT (18:18):
Yeah, it's got that texture to it.
But I see rock in it too, likefrom a lava flow or something.
And then the purpleunderneath-- well,
let me not gettoo far into this.
So then you blend them.
And what does that tell you?
CHRISTINA SHIVERS: It basically is just (18:34):
undefined
a process of making those twothings that were created part
of--
not allowing the bag to be the--
because basically, in AIsoftwares that create images,
there's weights that canbe assigned to an image
(18:56):
that you put in.
CHARLIE BENNETT (18:57):
Oh, right.
How much should this beused in the end product?
CHRISTINA SHIVERS: Yeah, so trying (19:00):
undefined
to not allow the bag to befully dominant because I wanted
the original camouflage modelto also be part of this process.
CHARLIE BENNETT: I think we should (19:09):
undefined
get a t-shirt thatsays not allow the bag
to be fully dominant.
[CHUCKLING]
CHRISTINA SHIVERS: Yeah, so that was (19:15):
undefined
sort of a process step in thecreation of the final output
on the right, which is anarchitectural, what I call,
a sort of research pavilion.
It's something that couldbe built given some time.
(19:35):
In the exhibitionI'm making, it's
broken up into aseries of modules.
So I'm creating one moduleof what this would be.
If you wanted tobuild this in reality,
it would be one piece of it.
So showing how, basically,through this process, creating
(19:57):
an architecturalobject or pavilion that
displays a potentially new--
or somewhat new-- not everythingunder the sun is not--
everything's a recombination ofsomething else, especially when
you're working with AI.
But something thatcan become innovative
in architectural constructiontechniques and processes.
(20:21):
AI is really thisinteresting process
where it's always givingyou multiple variations
of something.
And so you have to workwith images in this way.
That's parametricalmost, because you
can change a number slider tochange the height of a box,
for instance.
CHARLIE BENNETT (20:40):
So I'm flashing on like Photoshop
and using color levels,like desaturating.
But this affectsan entire rendering
of an architectural idea?
CHRISTINA SHIVERS: Yeah, and that's (20:50):
undefined
what's reallyinteresting about AI
is that it isinherently parametric,
but through a kind of verydifferent way of thinking
about it.
Instead of thinkingabout numbers, sliders
or different ways ofmanually coding an interface,
(21:12):
you are working withtext descriptions,
and in this instance,a series of images that
are being combined, recreated.
Again, it hearkens back to thatanalog method of photocopying,
cutting up, recutting up.
CHARLIE BENNETT (21:30):
This seems like a very involved way
of doing a conceptual model.
ALEX MCGEE (21:34):
--to Lost in the Stacks.
And we'll hear more about theresearch pavilions and melted
plastic on the leftside of the hour.
[MUSIC PLAYING]
[MUSIC PLAYING]
(SINGING) One, two
(21:56):
One, two, three
- Hi, I'm JoelleDietrick, and I'm
an artist that makeswork about moving around.
You are listening to Lost inthe Stacks on WREK Atlanta.
(SINGING) When you were a child,you were touched by the muse
And she said you were on firefrom your head to your shoes
[PHONE RINGING]
CHARLIE BENNETT: Today's show is called (22:17):
undefined
How to Train YourAlgorithm, although it might
be how to trick your algorithm.
There's been new informationbrought into the studio.
Named after the AI andarchitecture research project
created by Dr. ChristinaShivers of the Georgia Tech
School of Architecture.
You're hearing excerptsfrom an interview
I recorded with Dr. Shivers.
And for our mid-showbreak, I wanted
(22:38):
to pull a minutefrom the interview
when she talked aboutAI and intentionality.
This is for you, Fred.
[MUSIC PLAYING]
CHRISTINA SHIVERS: What I also am (22:46):
undefined
trying to explorein this process
is this blending between whatagency me or the artist has
and what agency ispart of the AI program.
And I think in manydiscussions about AI and art,
(23:10):
many people are concernedabout losing artistic agency.
And I have manysuspicions about AI just--
but I think it'ssomething that's
necessary to engagewith because it's
being engaged with as we speak.
And so beginning tothink about it critically
(23:32):
and through an artisticand architectural lens
is really important.
But the process thatI'm working with,
and I think all processesin working with AI,
there is actuallya lot of agency
that the artist orthe user of AI has.
But I do think you have tothink about it intentionally
(23:52):
because it's very easyto say, make this great.
That's it.
Where in reality,AI, in essence,
just from a very basicstandpoint or a basic level
is a kind of form of curating.
It always gives you variationswhen you say, which one is best?
But then when you continuewith this process of curating,
(24:15):
but also combination blending,again, these parametric methods
that always are goingto create something
a little bit differentor drastically different,
but you're not quitesure what it is,
that is where youbegin to have agencies.
It's very process-based insteadof planning based, I think.
[MUSIC PLAYING]
(24:38):
CHARLIE BENNETT (24:40):
File this set under PA 4037.C25.
[BOBBY DAY, OVER AND OVER]
(SINGING) Doot, doot, doot, doot
Well, I
[PEEL DREAM MAGAZINE, "MACHINEREPEATING"]
(25:10):
That's Machine Repeating
by Peel Dream Magazine.
And we started with Overand Over by Bobby Day.
Those are songsabout the three R's--
Repetition, Repetition,and Repetition.
Or to put it anotherway, the three R's--
Repetition, Repetitionand Repetition.
Fred?
FRED RASCOE (25:31):
Repetition, repetition, repetition.
CHARLIE BENNETT (25:33):
That's right.
What was that thing aboutvegetative electron microscopy?
[MUSIC PLAYING]
MARLEE GIVENS (25:42):
This is Lost in the Stacks,
and our show today is called Howto Train Your Algorithm, which
is also the name, we'repretty sure, of a research
project by Dr. ChristinaShivers of Georgia Tech's
School of Architecture.
CHARLIE BENNETT (25:56):
What could the other name be?
Marlee, what do you think?
MARLEE GIVENS (25:58):
How to trick your algorithm?
CHARLIE BENNETT (26:01):
Alex was jumping for the mic on that one.
ALEX MCGEE (26:04):
How to do repetition,
repetition, repetition,your algorithm.
[CHUCKLING]
CHARLIE BENNETT: We're playing excerpts (26:07):
undefined
from an interview with Dr.Shivers that only happened
once recorded live inthe Georgia Tech Library
this year on Media Arts Day.
This last segment hasto begin with the answer
to the question, whathave you learned?
How is this project useful?
CHRISTINA SHIVERS: --interesting, (26:22):
undefined
because the whole process--
it's doing something a littlebit different each time.
And again, it's hard to fullyunderstand why it does that.
You do multiple iterations--or I did multiple iterations.
And I'm like, why is itdoing it this little way?
(26:44):
And then you tweak it, and youunderstand certain parts of it.
But it does haveparticular features
that you just don't knowwhat's going on underneath.
CHARLIE BENNETT (26:53):
This is the black box thing.
It goes in and comes out,and you don't exactly
know what happened.
CHRISTINA SHIVERS: Yeah, you have no idea (26:56):
undefined
what's exactly going on.
CHARLIE BENNETT (26:58):
And so then the final results
are, in some cases,less realistic
than the blended images.
But then you're callingthose research pavilions.
So why researchand why pavilion?
CHRISTINA SHIVERS (27:12):
Yeah, so each of them is--
basically, I gave it a parameterof being a 10 by 10 cube,
and something thathas to exist as kind
of an architectural object.
It has to stand up, basically.
CHARLIE BENNETT (27:28):
Was that literally the thing you gave?
You said it has tostand up, or did you
say it in a different way?
CHRISTINA SHIVERS (27:35):
I'm trying to remember my prompt.
It became veryregular, but it was
10 by 10 cube in isometricview because those
are each in isometric view--
a piece ofarchitecture, a pavilion
in the style of acertain architect.
It's interestingbecause you continually
(27:58):
give it very similardescriptions so you can get it--
CHARLIE BENNETT (28:01):
But slightly [INAUDIBLE].
CHRISTINA SHIVERS: --yeah, slightly different (28:03):
undefined
in the process.
That's where you kind ofcan begin exerting control,
at least I found in thisprocess a little more.
So it will alwaysgive you a variation
of something within theparameters that you want,
which is an architectural 10by 10 cube that has to be--
(28:23):
10 by 10 by 10 cube thathas to, again, stand up.
I didn't use theword "stand up."
CHARLIE BENNETT (28:30):
Pavilion is what kind of made that.
CHRISTINA SHIVERS (28:31):
Yeah.
CHARLIE BENNETT: So you said give me (28:32):
undefined
a pavilion to start with.
CHRISTINA SHIVERS: Yeah, basically, (28:35):
undefined
and the dimensions of it and howit would produce the drawing.
And using the features ofthese different materials
or different objectsthat were made--
so in each of thesepavilions, I said,
(28:57):
using the kind ofplastic combination bag--
is made out of that.
So the material isconnected to those images.
CHARLIE BENNETT (29:06):
And you can make these.
That's the other piece of it?
CHRISTINA SHIVERS (29:10):
So that's the larger research question,
is how AI begins todevelop or allow architects
to develop innovative uses ofmaterials in their construction
methods.
CHARLIE BENNETT (29:28):
The materials--
that's the most important part.
Because what I see withan untrained eye is,
oh, this is a sort of renderingof the feel of a building
that you'd go for.
But you're looking atwhat material it suggests.
CHRISTINA SHIVERS (29:44):
Yeah, so in each of these,
they have been modeled andrun through the process,
I guess-- the ringer,where I model all of them,
I do dimensioneddrawings, and begin
to create a process bywhich these are built.
(30:04):
And that's been done with thelarger question of beginning
to develop a research or a kindof thesis of what AI can present
for architects outside ofsimply creating renderings.
CHARLIE BENNETT (30:22):
And is the idea that these
could be built at like apeople-walking through scale?
CHRISTINA SHIVERS (30:29):
Yeah, so they're 10 by 10--
so they'd be 10 feet.
Each module would be 2-- youcould imagine it's cut every 2
feet in space--
every two feet this way.
So they're all made ofdifferent 2 by 2 foot
by 2 foot modules, solike bricks basically.
CHARLIE BENNETT (30:46):
And you want to fabricate those yourself?
CHRISTINA SHIVERS (30:49):
Yeah.
CHARLIE BENNETT (30:50):
How far along are you in the process of?
CHRISTINA SHIVERS: Pretty far along. (30:52):
undefined
So the first one is madewith recycled plastic bags,
plastic sheets, basically.
So you have to be very careful.
Obviously, you have toyour plastic appropriately.
You don't want to just inhaleany plastic that's melting.
(31:12):
This is where it becomesreally interesting.
Plastic bags are very easy tobe melted down and reformed
into flat sheets and recycled,basically, and reused.
I've been melting-- well,getting the process going.
I've done some experiments.
I'm still waiting fora piece of machinery
to arrive to get thefull fabrication going.
(31:35):
Some hiccups, but I've stillmade test runs of this.
But basically, melting plasticbags down into plastic sheets.
So you do that process.
You create a sheet.
You let it cool.
It takes a little bit of time.
But then you have abuilding material from there
to work with.
To make one of these modules,you create a frame underneath.
(31:59):
You reheat the sheetfor just a little bit,
so it becomes malleable.
And then drill it.
I haven't drilled anythingyet, but that's the intention
if you don't wantit to move around.
CHARLIE BENNETT (32:12):
Were you hoping to get to a place
where you could build something?
CHRISTINA SHIVERS (32:17):
Yes.
CHARLIE BENNETT (32:17):
And what does that do
for you to be able to buildthese research pavilions?
CHRISTINA SHIVERS (32:23):
It helps concretize AI research and also
present new means of obtainingsomething, quote, unquote,
"sustainable."
Because innovativematerials are possibly
one of the more promisingdirections with which architects
(32:46):
probably need to be workingto think about producing maybe
more sustainable orenvironmentally less harmful
buildings--
in many ways, somethinglike almost a resource.
Each of these is built out ofsome type of off-cast object.
(33:11):
But yeah, in each of these--
or basically, eachof these cases,
the idea is to buildthem in some capacity
to begin looking at theapplied methods, which
in architecture, it'sbuilding something
or understanding howto build something.
So using AI in a wayof concretizing it
(33:31):
as an applied thing more sothan just producing a rendering.
CHARLIE BENNETT (33:35):
This is Lost in the Stacks.
Today's show was How toTrain Your Algorithm.
It was inspired by anexhibit currently installed
in the Georgia TechLibrary on the second floor
of the Crossland Tower.
MARLEE GIVENS (33:45):
And our guest was Dr. Christina Shivers,
the creator of How to TrainYour [INAUDIBLE] Georgia Tech.
And you can learn more aboutthis project and other work
at christinashivers.com,and that's
Christina spelled with a C-H.
CODY TURNER (33:59):
You can file this set under QA 76.9.I58I35.
[TRICERATOPS, "MECHANICALFRIEND"]
[MANFRED MANN, "MACHINES"]
MARLEE GIVENS (34:22):
Machines by Manfred Mann.
And before that, MechanicalFriend by Triceratops.
Songs about humans tryingto get machines to do what
the humans want them to do.
[MUSIC PLAYING]
CHARLIE BENNETT (34:42):
Today's episode is called
How to Train Your Algorithm.
And to finish the AIand art discussion,
I'd like to askeverybody, what's
some aspect of creative AI Iyou would like to interrogate?
I'm waiting for thelongitudinal study
of a set of artistswho use AI to see
if the qualities of theirwork begin to become
(35:02):
similar to each other.
How about you, Marlee?
MARLEE GIVENS (35:06):
I'm not that deep today.
I need something todo interior design.
[LAUGHTER]
CHARLIE BENNETT (35:13):
You want to know if it works or not?
MARLEE GIVENS (35:15):
Well, I'm not really good at design,
and I'm worse with 3D spaces.
And so I just needsomething to--
if I give it a vibe, likecreate a room for me,
that would be great.
Fred?
FRED RASCOE (35:29):
So I think--
well, to start off with,I don't know a whole lot
about art criticism,academic study of art.
But I do know that no matterwhat process, or technique,
or genre that people rise upand say, ugh, that's not art,
(35:49):
it always, always, always turnsout to be art in later years.
So I'm interested to see wherethat tipping point comes in
with AI-generated creative work.
CODY TURNER (36:03):
Yeah, Fred, I'll answer the next.
I think AI is going to leadto this era of perfectionism,
like everything's perfect.
So I'm curious as to whatconscious imperfections are
going to creep back in onceeveryone gets used to everything
being AI generated.
What about you, Alex?
ALEX MCGEE (36:22):
Mine's kind of in a similar vein to Charlie's.
I'm curious to seehow much theft there
is of other people's work inAI, especially with artists.
I follow this cakedecorator, actually,
and she regularly calls out AIfor then creating similar things
(36:46):
that people are circulatingand passing off as her own,
and also creating outlandishexpectations for people actually
doing this work.
CHARLIE BENNETT (36:56):
That could be like a poetry degree--
the words we use to describetheft, and take, and use.
[CHUCKLING]
Let's just rollthe credits, man.
[MUSIC PLAYING]
MARLEE GIVENS: Lost in the Stacks (37:10):
undefined
is a collaboration between WREKAtlanta and the Georgia Tech
Library, written and producedby Alex McGee, Charlie Bennett,
Fred Rascoe, and Marlee Givens.
Legal counsel and a set ofanti-surveillance noise masks
were provided by the BurroughsIntellectual Property Law Group
in Atlanta, Georgia.
CHARLIE BENNETT (37:29):
Special thanks to Christina
for being on the show, toKirk Henderson and Connor
Lynch for inspiringthe episode, and also
for Connor running to seewhat the name of things were.
To Quincy Thomas forthe interview audio.
And thanks, as always, toeach and every one of you
for listening.
MARLEE GIVENS (37:44):
Our web page is library.gatech.e
du/lostinthestacks, where you'llfind our most recent episode,
a link to our podcastfeed, and a web form
if you want to getin touch with us.
CHARLIE BENNETT: Next week's show (37:56):
undefined
is our last GT library guidebookepisode for the semester.
We're checkingout the info desk.
FRED RASCOE (38:03):
It's time for our last song today.
Humans will never becompletely removed
from the architecturaldesign process.
CHARLIE BENNETT (38:10):
You hope.
FRED RASCOE (38:10):
Right?
But automated toolslike AI will inevitably
be incorporated into ourprocesses, into our designs,
into our world.
So let's close witha track about humans
coexisting with technologyand making the best of it.
This is Our Worldby The Individuals
right here on Lostin the Stacks.
And have a greatweekend, everybody.
(38:31):
[MUSIC PLAYING]
[MUSIC PLAYING]
CHRISTINA SHIVERS (00:01):
The act of drawing and design
really has significantconnections to music.
I mean, even the processof musical notation,
there are so many reallyfascinating connections
between musical notationand architectural notation.
They're both languages--graphic languages.
And in both cases,music, and architecture
have significantcommonalities because they
(00:24):
both are disciplinesthat are conducted
through the act of drawing.
CHARLIE BENNETT (00:52):
You are listening to WREK Atlanta,
and this is Lost in theStacks, the research library
rock and roll radio show.
I'm Charlie Bennett in thestudio with everybody--
Alex McGee, Marlee Givens,Fred Rascoe, and Cody Turner.
I can't say anybody's name.
FRED RASCOE (01:08):
Rascoe.
CHARLIE BENNETT (01:09):
Rascoe.
Each week on Lost in theStacks, we pick a theme
and then use it to create amix of music and library talk.
Whichever you are herefor, we hope you dig it.
ALEX MCGEE (01:18):
Our show today is called
How to Train Your Algorithm.
That's the titleof an exhibit, you
can visit in theGeorgia Tech Library
right now on the secondfloor of Crosland Tower.
MARLEE GIVENS (01:29):
It is a set of objects
made from recycled paper, meltedplastics, wood, Styrofoam,
burlap, and probably a fewother unexpected materials
displayed on stands made fromunpainted, unvarnished wood
planks.
FRED RASCOE (01:42):
And it's all about AI.
CHARLIE BENNETT (01:44):
Yeah, this show is like a split
single for Lost in the Stacks.
One side is AI, and the otherside is the material umph.
ALEX MCGEE (01:52):
How to Train Your Algorithm is an ongoing project
exploring AI withmaterial practices created
by Professor Christina Shivers,the 2023 Ventulett NEXT
fellow in the School ofArchitecture at Georgia Tech.
MARLEE GIVENS: Charlie spoke with Dr. (02:06):
undefined
Shivers in the library'sscholars event theater in front
of a live audience duringMedia Arts Day 2025..
And our interview segments todaycome from that conversation.
FRED RASCOE (02:17):
And our songs today are
about architecture anddesign, repetition,
and getting machines to workwell with humans and vice versa.
No matter how many experimentaltechnological tools
we have at our disposal,the product of design still
has to incorporate thehuman factor, right?
Doesn't it?
So let's start with humanfactor by Music for Pleasure
(02:40):
right here on Lostin the Stacks.
CHARLIE BENNETT: Be assertive, Fred. (02:43):
undefined
[MUSIC FOR PLEASURE, "HUMANFACTOR"]
That was Human Factorby Music for Pleasure.
Our show today is calledHow to Train Your Algorithm.
Our guest is Dr. ChristinaShivers, an academic, architect,
and musician whoseresearch broadly focuses
(03:04):
on the intersection ofenvironmental and economic
thought specifically withinthe design disciplines.
ALEX MCGEE (03:09):
Dr. Shivers discussed her work
in an interview recorded livein the Georgia Tech Library.
For our firstquestion, we asked,
was it a straight line fromyour work in urban landscapes
and reclamation tothis AI project?
CHRISTINA SHIVERS: I think there's (03:21):
undefined
a couple steps in between.
I mean, my background is indesign and making, so music
and design.
So I always sort of had asomewhat applied approach
to things, or to the world.
And then throughstudying for a PhD,
I had to start thinking aboutthings very differently.
So how do I write anddo historical research,
(03:43):
understand primary sourcesas almost my kind of material
that I'm working with moreso than the act of drawing.
It's very kind of--
not a learning process,but just a different way
of thinking or using mybrain a little differently.
And so in coming back intomore of a design setting,
as a professor and as theVentulett NEXT fellow here,
(04:05):
I began to startthinking about passion
I have for making drawings,for making models,
for really investigatingarchitectural and craft
questions, how do Ibegin to approach that
through the lens of all of whatI was doing with my dissertation
and with my research?
And I think there are many majorenvironmental questions facing
(04:28):
architects today, from whatsort of materials do we use?
how are the materials processed?
what kind of chemicalsare in them? how
are they coming frompoint A to point B? where
are they extracted from?
Those are just some questions.
I mean, there are otherquestions about the carbon
emissions buildings produced.
So beginning to approachmy project last year,
(04:52):
AI, as we all know, has reallyballooned in terms of popularity
in the last twoyears, three years.
I guess ChatGPT emergedin January of 2022.
CHARLIE BENNETT: Yeah, there seems (05:04):
undefined
to be a metaphoricalstarting gun.
But every time we talkabout it, there's always
been the phrase, well, thathad been happening before,
but it just hit the popularconsciousness somehow.
CHRISTINA SHIVERS (05:17):
Yeah.
And so in architecture,I think it's
been a bit of an open questionabout how one uses it.
For many people, it has amazingpotential to produce renderings.
Renderings are one ofthe things architects
kind of have to do to show theclient what the building, what
the project will look like.
(05:38):
And many architectswill approach renderings
as like a sort ofnecessary evil.
So with AI, you can justexport renderings or out--
[CHUCKLING]
CHARLIE BENNETT (05:47):
Are renderings the pitch PowerPoint?
CHRISTINA SHIVERS (05:52):
A rendering is more of like a--
you model your buildingin a 3D modeling software,
and then you use a renderingsoftware to basically put all
the materials, lighting,put it in its environment,
make it look very realistic.
Because a clientwill often not really
want to look at plans, andsections, and elevations.
(06:14):
They want to knowwhat it's going
to feel like in the building.
So AI has a lot of potential forrendering uses in architecture.
I was really interestedin what other questions
I posed, particularlybecause it has this--
it does have an immenseenvironmental toll.
But it also is, in many ways, akind of reflection of ourselves,
(06:34):
of our society, because it'strained with all of our data.
There are many controversiesaround the use and acquisition
of that data, but it'sbasically, in this way,
a kind of mirror of our society.
So I became interestedin trying to understand,
potential uses oflike an AI platform,
(06:56):
how that couldmaybe begin creating
unexpected stories orunexpected ways of understanding
our society.
And I was interestedin primarily
in understanding the kind ofenvironmental underside of that.
CHARLIE BENNETT (07:11):
So when you say AI is a reflection of society
or the culture atlarge, were you
interested in messing withhow society was reflected?
Were you interested inmessing with society
so that it wasreflected differently,
or just messingup the AI itself?
CHRISTINA SHIVERS (07:29):
So I'm interested in--
so AI has very particularways in which it functions.
It is trained to reproduceidentifiable objects.
And it has a really difficulttime thinking abstractly
or producing abstract images.
CHARLIE BENNETT (07:47):
Wait.
Let me interrupt you there.
Has a difficult timethinking, you said?
CHRISTINA SHIVERS (07:51):
Or I guess not thinking, but producing.
CHARLIE BENNETT (07:53):
But is that what you kind of--
the thinking ofthe AI is when it
tries to produce something outof what raw materials it has
or raw data it has?
CHRISTINA SHIVERS (08:03):
Yeah.
So it will revert to somethingthat looks like an object--
that table, something like that.
And I was really interestedin trying to say,
what happens if you try tocircumvent that process?
What would these AI algorithmsand softwares begin producing?
(08:26):
And so I was reallyinterested in developing
a series of what are called--what I called camouflage models.
So noise algorithm cansometimes kind of mess
with the imagerecognition parts of AI.
There were--
CHARLIE BENNETT (08:41):
I think we should unpack that real quick.
So a noise algorithm--what is that?
CHRISTINA SHIVERS: Yeah, so if you (08:46):
undefined
think about static oran old TV when it just
is static on the screen,that's noise, basically.
It's just visual nothingness.
It's the same with sound, solike white noise, brown noise,
all these different noises.
CHARLIE BENNETT (09:06):
It's wild that static on a TV
is not really a thinganymore, that there's
a moment when people didnot grow up with that.
But white noise or pinknoise is still totally
in people's consciousness.
CHRISTINA SHIVERS (09:17):
Yeah, I have a white noise machine
I use very frequently to sleep.
[CHUCKLES]
CHARLIE BENNETT (09:22):
So a noise algorithm
is something that generatesthat kind of static or disarray?
CHRISTINA SHIVERS: Yeah, something-- so (09:27):
undefined
because of the way that theimage producing AI systems
work, they basicscreen of noise,
and they slowly whittle it downinto something like an object.
CHARLIE BENNETT (09:42):
Oh, wow.
CHRISTINA SHIVERS (09:43):
Yeah.
And so if you interfere withthat process with noise,
it can throw it off.
So there were someartists fairly recently--
I don't remember how long ago,but in the last two or three
years--
who were basically printingstickers with a noise image
and putting it ontheir face so it
(10:04):
wouldn't be ableto recognize it.
I'm pretty sure the algorithmsand the companies that
produce these AI algorithmsare always updating.
So I'm sure they've kindof circumvented that a bit.
But I still was veryinterested in this process.
So I created a sort of seriesof abstract three-dimensional
(10:26):
models in a 3D software thathad different noise algorithms
applied to their materiality,to their surface.
CHARLIE BENNETT (10:36):
So if I can try to understand it
in a very simple metaphor--
I know you know thisphrase, but Michelangelo
said the way to sculpt somethingwas to look at the raw material
and then take away anythingthat wasn't the sculpture.
It sounds like you arethrowing rocks at a sculpture
while it's being sculpted.
CHRISTINA SHIVERS: Yeah, so basically, I (10:56):
undefined
created these things.
And I asked an AIprogram, what is this?
And in all the cases,it would come up
with really interesting answers.
There's a captiongenerating software
called Pallyy AI softwareis kind of in this process.
(11:17):
But I asked it, what isthis? or describe this?
And it would say somethingalong the lines of,
this is a blue and green orturquoise and green plastic bag.
I like to put myart supplies in it.
CHARLIE BENNETT (11:32):
Whoa.
Where did that come from?
That additional--
CHRISTINA SHIVERS (11:38):
So that image or caption generator
is for use for Instagramand for basically quickly
generating captionsthat, I guess,
you don't want to think of.
CHARLIE BENNETT (11:54):
I hate this so much.
I'm so upset.
I don't know if I cancontinue the interview.
MARLEE GIVENS (12:02):
This is Lost in the Stacks.
We'll be back with more from Dr.Christina Shivers about messing
with AI after a music set.
FRED RASCOE (12:09):
You can file this set under NA 2520.H73.
[MANOR, "ARCHITECTURE"]
[GENTLE GIANT, "DESIGN"]
ALEX MCGEE (12:26):
That was Design by the not-Welsh band Gentle giant.
CHARLIE BENNETT: You don't know that. (12:29):
undefined
[CHUCKLING]
ALEX MCGEE (12:32):
There's a debate.
CHARLIE BENNETT (12:33):
Just sounds like it was
recorded in a field near cows.
ALEX MCGEE (12:36):
And before that, Architecture by Manor.
Songs about, well,design and architecture.
[MUSIC PLAYING]
MARLEE GIVENS (12:48):
This is Lost in the Stacks,
and today's show is calledHow to Train Your Algorithm.
Dr. Christina Shiversfrom Georgia Tech's School
of Architecture hasbeen exploring AI
and materiality usingimages, noise algorithms,
and fabrication.
You're listening to excerptsof an interview with her,
recorded live in the libraryon Media Arts Day 2025.
CHARLIE BENNETT (13:11):
In the last segment,
we talked about howthis project works.
Let's review.
Dr. Shivers createsan image of an object
obscured with visual noise, whatshe calls the camouflage model,
and then asks an AI systemto identify the object.
That identification becomesthe prompt for an AI generated
(13:33):
image.
FRED RASCOE (13:34):
Then, the camouflage model and the AI
image are blendedinto a new image.
And that new hybrid imageis used as part of a prompt
to design anarchitectural pavilion.
ALEX MCGEE (13:46):
Which will eventually
be built using recycledand unexpected materials.
If you need more detail orwant to see these images,
the whole process is onchristinashivers.com--
Christina spelled with a C-H.
CHARLIE BENNETT: This AI system-- (13:59):
undefined
don't even have to name it--
you showed it this image, or itcreated this image from text?
CHRISTINA SHIVERS (14:06):
So I showed it the image on the left.
It said it was a blueand green plastic bag.
And then I took thesentence that it
gave me and had it produce theblue and green plastic bag.
Yeah, so it's like onestep of translation
from left to the next one.
CHARLIE BENNETT (14:26):
And how is it that it can make the plastic bag
from an image that, ifI can just describe it
very quickly foranyone listening,
it looks like a landscapewith some popsicle sticks?
I mean, I don't-- help me.
I don't see how it canthen say that that's a bag.
CHRISTINA SHIVERS (14:48):
That's why it's really fascinating.
That's what I thought reallybecame really fascinating about
this process is the objects thatit's-- the data that is trained
with it reverts to something.
And so something asubiquitous as a plastic bag
that we see everywhere thatwe use for everything, it's
like, OK, I guessthis is a plastic bag.
CHARLIE BENNETT (15:10):
So the AI was able to peel the noise off
of that image andget a plastic bag?
CHRISTINA SHIVERS (15:17):
It said it was a plastic bag, yeah,
through its computer visionimage recognition process.
CHARLIE BENNETT (15:24):
So if I go back to my silly metaphor
of the sculptor and throwingrocks at the sculpture,
do you not like the sculptor?
Are you tired ofwhat he's creating?
Are you messing with the AIin order to know it better,
or are you trying to figureout ways to disrupt it?
CHRISTINA SHIVERS: I'm trying to-- (15:41):
undefined
I think it's probably both.
I think it has potential.
With any tool, it's reallyinteresting to use it in a way
that it may not beintended to use.
Many, many artists andmusicians use things
quite differentlythan maybe the way
(16:03):
they were engineeredand stumble upon,
in the process, something new--
a way of using thatmedium that can reveal
something that is unexpected.
So I think that's part of it, somessing with the AI algorithm.
A, I am trying toget under the surface
to begin understandingwhat the data
(16:26):
that it's trained with startsto-- what it starts to reveal
about that dataand, in turn, how
it is this reflectionof our world around us.
So the plastic bagis so ubiquitous
that it looks at thatimage on the left
and it's like, Iguess this is a bag.
CHARLIE BENNETT (16:42):
Wow.
So there's the noisy image, andthen the rendering of a prompt--
that noisy image, andthen another landscape.
So what's the next image then?
CHRISTINA SHIVERS (16:55):
That is a combination of those two images.
So basically, justthose two images
were blended togetherto begin understanding--
starting to develop a processof how this can create
an architectural application.
CHARLIE BENNETT (17:12):
And did you do that with an AI tool--
CHRISTINA SHIVERS (17:15):
Yeah.
CHARLIE BENNETT (17:16):
--or is that--
OK.
It's just digesting,digesting, digesting.
CHRISTINA SHIVERS (17:20):
Yeah, it's a kind of recursive process.
And so that third stepis the bridging point
between the objectthat's been described,
but then kind of incorporatingwhat it originally was used
to create that description.
CHARLIE BENNETT: So again, I'm going (17:40):
undefined
to do for people listening.
So I see what looks likea landscape with popsicle
sticks stuck in it.
And what AI describethat as used as a prompt
produces a pretty realisticplastic bag photo.
I mean, I'm looking at itnow and I can tell, oh, yeah,
that's not real.
But I didn't until you told me.
(18:00):
And then the nextimage is actually 3D,
and it is another landscape,like maybe some rock strata with
plastic bag handles straight up.
Is that what you see?
CHRISTINA SHIVERS (18:12):
Yeah-- sort of also kind of melted plastic
almost in some way.
CHARLIE BENNETT (18:18):
Yeah, it's got that texture to it.
But I see rock in it too, likefrom a lava flow or something.
And then the purpleunderneath-- well,
let me not gettoo far into this.
So then you blend them.
And what does that tell you?
CHRISTINA SHIVERS: It basically is just (18:34):
undefined
a process of making those twothings that were created part
of--
not allowing the bag to be the--
because basically, in AIsoftwares that create images,
there's weights that canbe assigned to an image
(18:56):
that you put in.
CHARLIE BENNETT (18:57):
Oh, right.
How much should this beused in the end product?
CHRISTINA SHIVERS: Yeah, so trying (19:00):
undefined
to not allow the bag to befully dominant because I wanted
the original camouflage modelto also be part of this process.
CHARLIE BENNETT: I think we should (19:09):
undefined
get a t-shirt thatsays not allow the bag
to be fully dominant.
[CHUCKLING]
CHRISTINA SHIVERS: Yeah, so that was (19:15):
undefined
sort of a process step in thecreation of the final output
on the right, which is anarchitectural, what I call,
a sort of research pavilion.
It's something that couldbe built given some time.
(19:35):
In the exhibitionI'm making, it's
broken up into aseries of modules.
So I'm creating one moduleof what this would be.
If you wanted tobuild this in reality,
it would be one piece of it.
So showing how, basically,through this process, creating
(19:57):
an architecturalobject or pavilion that
displays a potentially new--
or somewhat new-- not everythingunder the sun is not--
everything's a recombination ofsomething else, especially when
you're working with AI.
But something thatcan become innovative
in architectural constructiontechniques and processes.
(20:21):
AI is really thisinteresting process
where it's always givingyou multiple variations
of something.
And so you have to workwith images in this way.
That's parametricalmost, because you
can change a number slider tochange the height of a box,
for instance.
CHARLIE BENNETT (20:40):
So I'm flashing on like Photoshop
and using color levels,like desaturating.
But this affectsan entire rendering
of an architectural idea?
CHRISTINA SHIVERS: Yeah, and that's (20:50):
undefined
what's reallyinteresting about AI
is that it isinherently parametric,
but through a kind of verydifferent way of thinking
about it.
Instead of thinkingabout numbers, sliders
or different ways ofmanually coding an interface,
(21:12):
you are working withtext descriptions,
and in this instance,a series of images that
are being combined, recreated.
Again, it hearkens back to thatanalog method of photocopying,
cutting up, recutting up.
CHARLIE BENNETT (21:30):
This seems like a very involved way
of doing a conceptual model.
ALEX MCGEE (21:34):
--to Lost in the Stacks.
And we'll hear more about theresearch pavilions and melted
plastic on the leftside of the hour.
[MUSIC PLAYING]
[MUSIC PLAYING]
(SINGING) One, two
(21:56):
One, two, three
- Hi, I'm JoelleDietrick, and I'm
an artist that makeswork about moving around.
You are listening to Lost inthe Stacks on WREK Atlanta.
(SINGING) When you were a child,you were touched by the muse
And she said you were on firefrom your head to your shoes
[PHONE RINGING]
CHARLIE BENNETT: Today's show is called (22:17):
undefined
How to Train YourAlgorithm, although it might
be how to trick your algorithm.
There's been new informationbrought into the studio.
Named after the AI andarchitecture research project
created by Dr. ChristinaShivers of the Georgia Tech
School of Architecture.
You're hearing excerptsfrom an interview
I recorded with Dr. Shivers.
And for our mid-showbreak, I wanted
(22:38):
to pull a minutefrom the interview
when she talked aboutAI and intentionality.
This is for you, Fred.
[MUSIC PLAYING]
CHRISTINA SHIVERS: What I also am (22:46):
undefined
trying to explorein this process
is this blending between whatagency me or the artist has
and what agency ispart of the AI program.
And I think in manydiscussions about AI and art,
(23:10):
many people are concernedabout losing artistic agency.
And I have manysuspicions about AI just--
but I think it'ssomething that's
necessary to engagewith because it's
being engaged with as we speak.
And so beginning tothink about it critically
(23:32):
and through an artisticand architectural lens
is really important.
But the process thatI'm working with,
and I think all processesin working with AI,
there is actuallya lot of agency
that the artist orthe user of AI has.
But I do think you have tothink about it intentionally
(23:52):
because it's very easyto say, make this great.
That's it.
Where in reality,AI, in essence,
just from a very basicstandpoint or a basic level
is a kind of form of curating.
It always gives you variationswhen you say, which one is best?
But then when you continuewith this process of curating,
(24:15):
but also combination blending,again, these parametric methods
that always are goingto create something
a little bit differentor drastically different,
but you're not quitesure what it is,
that is where youbegin to have agencies.
It's very process-based insteadof planning based, I think.
[MUSIC PLAYING]
(24:38):
CHARLIE BENNETT (24:40):
File this set under PA 4037.C25.
[BOBBY DAY, OVER AND OVER]
(SINGING) Doot, doot, doot, doot
Well, I
[PEEL DREAM MAGAZINE, "MACHINEREPEATING"]
(25:10):
That's Machine Repeating
by Peel Dream Magazine.
And we started with Overand Over by Bobby Day.
Those are songsabout the three R's--
Repetition, Repetition,and Repetition.
Or to put it anotherway, the three R's--
Repetition, Repetitionand Repetition.
Fred?
FRED RASCOE (25:31):
Repetition, repetition, repetition.
CHARLIE BENNETT (25:33):
That's right.
What was that thing aboutvegetative electron microscopy?
[MUSIC PLAYING]
MARLEE GIVENS (25:42):
This is Lost in the Stacks,
and our show today is called Howto Train Your Algorithm, which
is also the name, we'repretty sure, of a research
project by Dr. ChristinaShivers of Georgia Tech's
School of Architecture.
CHARLIE BENNETT (25:56):
What could the other name be?
Marlee, what do you think?
MARLEE GIVENS (25:58):
How to trick your algorithm?
CHARLIE BENNETT (26:01):
Alex was jumping for the mic on that one.
ALEX MCGEE (26:04):
How to do repetition,
repetition, repetition,your algorithm.
[CHUCKLING]
CHARLIE BENNETT: We're playing excerpts (26:07):
undefined
from an interview with Dr.Shivers that only happened
once recorded live inthe Georgia Tech Library
this year on Media Arts Day.
This last segment hasto begin with the answer
to the question, whathave you learned?
How is this project useful?
CHRISTINA SHIVERS: --interesting, (26:22):
undefined
because the whole process--
it's doing something a littlebit different each time.
And again, it's hard to fullyunderstand why it does that.
You do multiple iterations--or I did multiple iterations.
And I'm like, why is itdoing it this little way?
(26:44):
And then you tweak it, and youunderstand certain parts of it.
But it does haveparticular features
that you just don't knowwhat's going on underneath.
CHARLIE BENNETT (26:53):
This is the black box thing.
It goes in and comes out,and you don't exactly
know what happened.
CHRISTINA SHIVERS: Yeah, you have no idea (26:56):
undefined
what's exactly going on.
CHARLIE BENNETT (26:58):
And so then the final results
are, in some cases,less realistic
than the blended images.
But then you're callingthose research pavilions.
So why researchand why pavilion?
CHRISTINA SHIVERS (27:12):
Yeah, so each of them is--
basically, I gave it a parameterof being a 10 by 10 cube,
and something thathas to exist as kind
of an architectural object.
It has to stand up, basically.
CHARLIE BENNETT (27:28):
Was that literally the thing you gave?
You said it has tostand up, or did you
say it in a different way?
CHRISTINA SHIVERS (27:35):
I'm trying to remember my prompt.
It became veryregular, but it was
10 by 10 cube in isometricview because those
are each in isometric view--
a piece ofarchitecture, a pavilion
in the style of acertain architect.
It's interestingbecause you continually
(27:58):
give it very similardescriptions so you can get it--
CHARLIE BENNETT (28:01):
But slightly [INAUDIBLE].
CHRISTINA SHIVERS: --yeah, slightly different (28:03):
undefined
in the process.
That's where you kind ofcan begin exerting control,
at least I found in thisprocess a little more.
So it will alwaysgive you a variation
of something within theparameters that you want,
which is an architectural 10by 10 cube that has to be--
(28:23):
10 by 10 by 10 cube thathas to, again, stand up.
I didn't use theword "stand up."
CHARLIE BENNETT (28:30):
Pavilion is what kind of made that.
CHRISTINA SHIVERS (28:31):
Yeah.
CHARLIE BENNETT: So you said give me (28:32):
undefined
a pavilion to start with.
CHRISTINA SHIVERS: Yeah, basically, (28:35):
undefined
and the dimensions of it and howit would produce the drawing.
And using the features ofthese different materials
or different objectsthat were made--
so in each of thesepavilions, I said,
(28:57):
using the kind ofplastic combination bag--
is made out of that.
So the material isconnected to those images.
CHARLIE BENNETT (29:06):
And you can make these.
That's the other piece of it?
CHRISTINA SHIVERS (29:10):
So that's the larger research question,
is how AI begins todevelop or allow architects
to develop innovative uses ofmaterials in their construction
methods.
CHARLIE BENNETT (29:28):
The materials--
that's the most important part.
Because what I see withan untrained eye is,
oh, this is a sort of renderingof the feel of a building
that you'd go for.
But you're looking atwhat material it suggests.
CHRISTINA SHIVERS (29:44):
Yeah, so in each of these,
they have been modeled andrun through the process,
I guess-- the ringer,where I model all of them,
I do dimensioneddrawings, and begin
to create a process bywhich these are built.
(30:04):
And that's been done with thelarger question of beginning
to develop a research or a kindof thesis of what AI can present
for architects outside ofsimply creating renderings.
CHARLIE BENNETT (30:22):
And is the idea that these
could be built at like apeople-walking through scale?
CHRISTINA SHIVERS (30:29):
Yeah, so they're 10 by 10--
so they'd be 10 feet.
Each module would be 2-- youcould imagine it's cut every 2
feet in space--
every two feet this way.
So they're all made ofdifferent 2 by 2 foot
by 2 foot modules, solike bricks basically.
CHARLIE BENNETT (30:46):
And you want to fabricate those yourself?
CHRISTINA SHIVERS (30:49):
Yeah.
CHARLIE BENNETT (30:50):
How far along are you in the process of?
CHRISTINA SHIVERS: Pretty far along. (30:52):
undefined
So the first one is madewith recycled plastic bags,
plastic sheets, basically.
So you have to be very careful.
Obviously, you have toyour plastic appropriately.
You don't want to just inhaleany plastic that's melting.
(31:12):
This is where it becomesreally interesting.
Plastic bags are very easy tobe melted down and reformed
into flat sheets and recycled,basically, and reused.
I've been melting-- well,getting the process going.
I've done some experiments.
I'm still waiting fora piece of machinery
to arrive to get thefull fabrication going.
(31:35):
Some hiccups, but I've stillmade test runs of this.
But basically, melting plasticbags down into plastic sheets.
So you do that process.
You create a sheet.
You let it cool.
It takes a little bit of time.
But then you have abuilding material from there
to work with.
To make one of these modules,you create a frame underneath.
(31:59):
You reheat the sheetfor just a little bit,
so it becomes malleable.
And then drill it.
I haven't drilled anythingyet, but that's the intention
if you don't wantit to move around.
CHARLIE BENNETT (32:12):
Were you hoping to get to a place
where you could build something?
CHRISTINA SHIVERS (32:17):
Yes.
CHARLIE BENNETT (32:17):
And what does that do
for you to be able to buildthese research pavilions?
CHRISTINA SHIVERS (32:23):
It helps concretize AI research and also
present new means of obtainingsomething, quote, unquote,
"sustainable."
Because innovativematerials are possibly
one of the more promisingdirections with which architects
(32:46):
probably need to be workingto think about producing maybe
more sustainable orenvironmentally less harmful
buildings--
in many ways, somethinglike almost a resource.
Each of these is built out ofsome type of off-cast object.
(33:11):
But yeah, in each of these--
or basically, eachof these cases,
the idea is to buildthem in some capacity
to begin looking at theapplied methods, which
in architecture, it'sbuilding something
or understanding howto build something.
So using AI in a wayof concretizing it
(33:31):
as an applied thing more sothan just producing a rendering.
CHARLIE BENNETT (33:35):
This is Lost in the Stacks.
Today's show was How toTrain Your Algorithm.
It was inspired by anexhibit currently installed
in the Georgia TechLibrary on the second floor
of the Crossland Tower.
MARLEE GIVENS (33:45):
And our guest was Dr. Christina Shivers,
the creator of How to TrainYour [INAUDIBLE] Georgia Tech.
And you can learn more aboutthis project and other work
at christinashivers.com,and that's
Christina spelled with a C-H.
CODY TURNER (33:59):
You can file this set under QA 76.9.I58I35.
[TRICERATOPS, "MECHANICALFRIEND"]
[MANFRED MANN, "MACHINES"]
MARLEE GIVENS (34:22):
Machines by Manfred Mann.
And before that, MechanicalFriend by Triceratops.
Songs about humans tryingto get machines to do what
the humans want them to do.
[MUSIC PLAYING]
CHARLIE BENNETT (34:42):
Today's episode is called
How to Train Your Algorithm.
And to finish the AIand art discussion,
I'd like to askeverybody, what's
some aspect of creative AI Iyou would like to interrogate?
I'm waiting for thelongitudinal study
of a set of artistswho use AI to see
if the qualities of theirwork begin to become
(35:02):
similar to each other.
How about you, Marlee?
MARLEE GIVENS (35:06):
I'm not that deep today.
I need something todo interior design.
[LAUGHTER]
CHARLIE BENNETT (35:13):
You want to know if it works or not?
MARLEE GIVENS (35:15):
Well, I'm not really good at design,
and I'm worse with 3D spaces.
And so I just needsomething to--
if I give it a vibe, likecreate a room for me,
that would be great.
Fred?
FRED RASCOE (35:29):
So I think--
well, to start off with,I don't know a whole lot
about art criticism,academic study of art.
But I do know that no matterwhat process, or technique,
or genre that people rise upand say, ugh, that's not art,
(35:49):
it always, always, always turnsout to be art in later years.
So I'm interested to see wherethat tipping point comes in
with AI-generated creative work.
CODY TURNER (36:03):
Yeah, Fred, I'll answer the next.
I think AI is going to leadto this era of perfectionism,
like everything's perfect.
So I'm curious as to whatconscious imperfections are
going to creep back in onceeveryone gets used to everything
being AI generated.
What about you, Alex?
ALEX MCGEE (36:22):
Mine's kind of in a similar vein to Charlie's.
I'm curious to seehow much theft there
is of other people's work inAI, especially with artists.
I follow this cakedecorator, actually,
and she regularly calls out AIfor then creating similar things
(36:46):
that people are circulatingand passing off as her own,
and also creating outlandishexpectations for people actually
doing this work.
CHARLIE BENNETT (36:56):
That could be like a poetry degree--
the words we use to describetheft, and take, and use.
[CHUCKLING]
Let's just rollthe credits, man.
[MUSIC PLAYING]
MARLEE GIVENS: Lost in the Stacks (37:10):
undefined
is a collaboration between WREKAtlanta and the Georgia Tech
Library, written and producedby Alex McGee, Charlie Bennett,
Fred Rascoe, and Marlee Givens.
Legal counsel and a set ofanti-surveillance noise masks
were provided by the BurroughsIntellectual Property Law Group
in Atlanta, Georgia.
CHARLIE BENNETT (37:29):
Special thanks to Christina
for being on the show, toKirk Henderson and Connor
Lynch for inspiringthe episode, and also
for Connor running to seewhat the name of things were.
To Quincy Thomas forthe interview audio.
And thanks, as always, toeach and every one of you
for listening.
MARLEE GIVENS (37:44):
Our web page is library.gatech.e
du/lostinthestacks, where you'llfind our most recent episode,
a link to our podcastfeed, and a web form
if you want to getin touch with us.
CHARLIE BENNETT: Next week's show (37:56):
undefined
is our last GT library guidebookepisode for the semester.
We're checkingout the info desk.
FRED RASCOE (38:03):
It's time for our last song today.
Humans will never becompletely removed
from the architecturaldesign process.
CHARLIE BENNETT (38:10):
You hope.
FRED RASCOE (38:10):
Right?
But automated toolslike AI will inevitably
be incorporated into ourprocesses, into our designs,
into our world.
So let's close witha track about humans
coexisting with technologyand making the best of it.
This is Our Worldby The Individuals
right here on Lostin the Stacks.
And have a greatweekend, everybody.
(38:31):
[MUSIC PLAYING]
Popular Podcasts
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.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.