Full transcript
0:01good afternoon
0:02everyone and and welcome to the
0:05manhattanville sessions of the first
0:07Columbia AI Summit I'm garu dangar I'm
0:10the director of the data Science
0:12Institute and a professor at Columbia
0:13engineering the data Science Institute
0:16is the operational home of Columbia Ai
0:18and a hub for interdisciplinary research
0:20in Ai and data science at this institute
0:23we don't just ask what Ai and data
0:25science can do but rather emphasize what
0:27they should do our state Mission data
0:30for good is about ensuring that Ai and
0:33data science are developed and applied
0:34in ways that serve Humanity AI is
0:38advancing rapidly shaping Industries
0:41institutions indeed everyday life itself
0:44today's discussion the first one on AI
0:46in creativity and the second one on ai's
0:49role in leadership and the future of
0:51work reflect its B
0:53impact Colombia is uniquely positioned
0:56to bring together experts from across
0:58disciplines from law policy journalism
1:01to engineering from The Sciences to the
1:03Arts to ask the right questions and
1:05ensure AI reflects human values and
1:08serves the public good indeed few
1:11universities in the world have
1:13Colombia's breadth and depth of
1:15expertise the data Science Institute and
1:17Colombia AI are able to leverage these
1:20deep inter interdisciplinary strengths
1:23differentiating us from similar efforts
1:25in other Universities at the data
1:28Science Institute we lead into this
1:29disciplinary work through research
1:31through events like this Summit through
1:33Partnerships that bridge Academia
1:35industry in the public sector Colombia
1:38AI is another hub for
1:40engagement the website ai. columbia.edu
1:43let me repeat it ai. columbia.edu a very
1:47simple URL is your entry point for AI
1:50research news events at Colombia if
1:53you're looking to get involved please
1:55start there everything AI related is
1:58going to be reflected on that
2:00website I also encourage you to connect
2:03with us on data science day on April 2nd
2:06that's data science institute's Flagship
2:08event showcasing Cutting Edge Ai and
2:10data science research and its industrial
2:12impact many themes of today's Summit
2:16will continue in those
2:18discussions the conversations that are
2:20happening here today here in Manhattan V
2:22across Colombia's campuses reflect the
2:25strength of Columbia AI the ideas and
2:28Partnerships that we hope emerg from
2:30these discussions will have the
2:32potential to shape how AI is adopted in
2:35the creative Fields how it influences
2:38leadership and how we navigate ai's
2:41growing role in the workforce these are
2:43just few of the critical areas that we
2:45are examining in the summit
2:46today I encourage you to engage connect
2:50think boldly about the role we
2:51Colombians can play in shaping ai's
2:54future and for those of you who can't be
2:56in the room there's the live stream and
2:58most of these events will be in the key
3:00takeaways from these will be posted on
3:01ai. columbia.edu and data science.
3:04columbia.edu so follow us there thank
3:07you for being part of the summit and I
3:09look forward to the discussions ahead
3:11and now I would like to introduce
3:13Professor Naim moan a professor in the
3:16department of Visual Arts at the school
3:17of the arts who will be introducing this
3:19session thank
3:25you confuse
3:27you thank you gud uh thank you everyone
3:30for being here good
3:32afternoon um just a few introductory
3:34remarks before I hand over to my
3:36colleague Laura Kuran to moderate and
3:38run the rest of the event as some of my
3:41colleagues at the School of Arts have
3:43commented the future impact of AI on
3:45Creative culture is hazy in the short
3:48term we expect massive dislocations in
3:50the valued compensated work of creatives
3:53in all spaces the long term that is one
3:56of the things our panelists want to
3:58think through with you
4:00our students entered this period via
4:02science fiction Paths of cyborg
4:04imaginary and sensient decision- making
4:07encapsulated in Philip K Dick's novel do
4:10Android's Dream of Electric
4:12Sheep 90 years ago Walter Benyamin
4:15presented the work of art in the age of
4:17mechanical reproduction his idea of an
4:20aura that was present in handmade
4:23painting and absent from early
4:25mechanical photography this was
4:281935 while providing a caution Benjamin
4:31also had Great Hopes for photography as
4:34an emancipatory toolkit that could see
4:36things with
4:37Clarity last year in the School of Arts
4:40campus lecture series machine Visions we
4:43encountered hesitation among industry
4:45leaders to come and speak to our
4:47students as the question of plundering
4:50visual artists work for machine learning
4:52training sets had become a key concern
4:54in the
4:55media considering the challenging
4:57questions our students ask visiting
4:59speakers I can presume some technology
5:02leaders do not have an answer about the
5:06why during the pandemic I started an
5:08exercise in my class where I would ask
5:11students what they remembered about the
5:13morning of
5:149/11 then I would show magazine covers
5:17from 2001 followed by covers of
5:20September
5:212020 eventually I had to put aside this
5:24exercise Because unless you had ma and
5:26PhD students in your class the reply was
5:29frequently I was not born
5:31yet an exercise in faded memory Works
5:35differently when it's a received memory
5:37from others and now we live in a world
5:39of constantly contested memories and we
5:42are working within an AI tool set that
5:44can invent photorealistic Tableau out of
5:47synaptic
5:49calculations imagine the illustrator who
5:52in 1953 was handed volume one of JRR
5:56tolkien's Lord of the Rings with an
5:58instruction to dream up what Mordor and
6:01saon look like the ultimate Act of
6:04invention out of whole cloth with no
6:06precedent because fantasy novels don't
6:09have to make serialized sequence sense
6:11AI engines that have vacuumed up fantasy
6:13novel covers can quickly generate a new
6:16genre book cover look at the artist
6:19class action lawsuit against mid Journey
6:22fantasy illustrators were among the
6:24first
6:25plaintiffs many artists now Express a
6:28desire to slow image consumption down we
6:31spend our time with existing images in
6:33the archive using AI tools to give them
6:36new meanings rather than giving into the
6:38pressure to always make new
6:41images we find our students to be a
6:44healthy mixture of curious and skeptical
6:47they are not attached to any imperative
6:49to always be Eternal
6:52Sunshine and now in the spirit of many
6:55parallel universes at the same time we
6:57turn to our six panelists who will each
7:00present Concepts and case studies from
7:02their fields on the question of AI and
7:04the creative impulse and I turn over to
7:06my colleague Laura Kuran director of
7:09computational design and practice at The
7:11Graduate School of Architecture planning
7:14and preservation thank
7:16[Applause]
7:23you thanks so much Naim for setting up
7:26the panel so well um I'm going to do
7:30something very small at the beginning
7:31and just tell you who's coming up to
7:34talk so our first panelist is Dennis
7:37Tenon he's an associate professor of
7:40English and comparative literature and
7:43he also co-directs the center for
7:45comparative media he's a rare English
7:47Professor whose code runs on millions of
7:50personal computers worldwide and he's
7:52published two books plain text the PO
7:55Poetics of computation and literary
7:57theory for robots
8:01[Applause]
8:07thank you
8:09Laura and then how we doing okay oh
8:13looks like they're all number one that's
8:14great so um uh uh I'd like to use the
8:18time to talk about Ai and creativity and
8:21specifically what the English Department
8:23and more broadly the humanities how
8:25these fields are being transformed by
8:27artificial intelligence now the first
8:29first and perhaps most obvious place to
8:31start is to say that uh artificial
8:34intelligence has a history beyond the
8:37last decade you'll be surprised to know
8:39and so there are a number of courses and
8:42a number of interesting research
8:43projects coming out uh of our discipline
8:47that deal with the history of artificial
8:49intelligence I'll just tell you since
8:51time is short I'll tell you a quick
8:53anecdote so for example uh Mark of
8:55chains uh which are which is a
8:57statistical algorithm that is lies at
9:00the heart of some of the technology
9:03producing AI the original uh re the
9:06original paper written by marov was the
9:08analysis of pushkin's uh poetry and so
9:11that lies squarely in the field of um
9:13literary scholarship and then as my
9:15research shows some of the first
9:17generative AI the kind that you can talk
9:20to and the kind of that can generate
9:22stories was built on the work of Roman
9:25yakobson out of Harvard and Levy Strauss
9:28who was an anthropologist and other
9:30researchers who were interested in the
9:32structure of fairy tales and generally
9:35kind of plot and narrative and other uh
9:39themes that are key that are Central to
9:41to our discipline of of literary
9:43analysis so there's much to do on the
9:45historical front um I teach a course
9:48called um called um uh literature in the
9:52age of artificial intelligence I've
9:55taught it for many years now it's a you
9:57know there's always a weight list uh and
9:59it's a course where we take some of
10:01these historical machines and some of
10:04these sort of paper and ink Bots and
10:06also early you know uh story generators
10:10that were designed for Mainframe
10:12mainframe computers and we recreate them
10:14in class to both think about the history
10:17and the future of possibilities of
10:19writing with AI so that's the historical
10:21part the second uh the second thing I
10:24want to cover is that um it is clear
10:28that the way we write today is changing
10:31and and uh we need to prepare our
10:33students to write uh in in the industry
10:36now whether they're going uh to uh into
10:39a legal profession or medical profession
10:41or the many creative Industries or the
10:43publishing industry uh that um that the
10:48uh the practices that are observed in
10:50the real world are changing rapidly and
10:52also we are in this moment where it's
10:54not clear how those changes will exactly
10:57take shape and so I feel
10:59responsibility of studying those changes
11:02more systematically and responding more
11:05quickly to the needs of creative
11:07Industries and the need to other
11:09Industries where knowledge creation and
11:12writing are at the very center of what
11:14they do uh to this end uh to this end um
11:21we have partnered I have partnered or
11:23the department has partnered with the
11:24Writing Center we have a very Lively
11:27Writing Center that's part of the
11:28English Department they are tasked to uh
11:31uh teach our students bringing their
11:34level of writing to to kind of a
11:38Colombia standard but but it is also
11:40becoming clear to us that thinking about
11:43how writing with AI how that should look
11:46like and what it means to write with an
11:48an with AI in a way that answers to our
11:52values and to our standards of
11:54excellence um we have a pro pilot
11:56program that was uh sponsored by Amy
11:59office by our EVP uh and so together
12:02with the Writing Center there'll be a
12:04class uh that that um kind of it's part
12:08a theoretical Class part a philosophical
12:11class but part also a practicum on what
12:13it means to write for AI and that class
12:16will be offered as soon as next semester
12:19in the fall um finally I want to say
12:23that um um AI lowers the barrier of
12:27computation to computational analysis
12:29and I think uh uh scholars in all Fields
12:32will relate to this whereas a decade ago
12:35it would take you several years to get
12:37up to speed to doing computational work
12:39now whether you're doing physics or
12:41computational chemistry or computational
12:44biology or computational social science
12:46or indeed computational literary
12:48analysis that training period was quite
12:51intense and difficult the low the
12:53barriers to entry have been lowered and
12:56that means that more of our graduate
12:58students are able to participate in
13:02doing computational uh research in our
13:05field to this end uh uh we have
13:08partnered with uh skku which is uh one
13:12of the oldest universities in Soul in
13:14Korea Toronto and Harvard and we're
13:17launching a pilot summer school this
13:19summer actually in Soul where we'll be
13:22inviting our graduate students to skill
13:24up and to and to kind of an entry point
13:28of computation in literary analysis in
13:31history and in social sciences now a
13:34little
13:36bit my own work also falls somewhere
13:39between those three kind of those three
13:42uh bullet points my most recent book is
13:45literary theory for robots which is uh
13:47the historical part uh as I've mentioned
13:50um and also these days and probably my
13:53next projects uh in the last year or so
13:57we have uh uh there is uh been renewed
14:00interest in the The Narrative
14:02intelligence lab which is a lab I
14:05co-direct with several colleagues uh
14:07including colleagues from Harvard
14:10University and we are interested in
14:12broadly the impact of artificial
14:15intelligence and other sort of uh other
14:19uh mechanical or automated incursions
14:23let's let's call them into our uh
14:25discursive media sphere uh and we uh
14:30look forward to many more interesting
14:32projects hopefully research projects
14:34coming out of this lab thank
14:37[Applause]
14:44you okay um next up is Katherine
14:48Griffiths who has recently joined
14:51Colombia as an assistant professor um at
14:55at Gap in computational design practices
14:58she's a a designer and researcher who's
15:01work who works at the intersection of
15:03critical AI studies social documentary
15:05and algorithmic visualization she makes
15:08visible the relationships among the
15:10ethics of machine learning Labor
15:12Relations algorithmic governance and the
15:15future of
15:18[Applause]
15:26work hi thank you for the opportunity
15:29teach to share some of my work with you
15:31um do
15:34I okay so my work I would describe it as
15:37sitting at the intersection of an
15:39emerging field called critical AI
15:41studies and computational design where I
15:44focus on the politics of machine
15:46learning in society and the development
15:48of custom software to visualize probe
15:52and raise questions about those issues
15:54so today I'd like to share um some
15:57excerpts of several projects with you
15:59and some of the concepts behind their
16:03development so critical code studies
16:06informs um my approach to thinking about
16:08Ai and its entanglements and I think of
16:12software as an El sorry I think of
16:14source code as a neglected but
16:17omnipresent cultural text that should be
16:21we should think of it as being available
16:23to for humanistic and creative
16:26interpretation and so that means taking
16:28code out of it's the idea of it being
16:32purely technical um and only available
16:36for computer readability and computer
16:39execution and trying to bring it into
16:41legibility and meaning for broader
16:44audiences um so for me that process
16:46extends into Art and Design and through
16:50this approach I try to understand the
16:52ways that
16:54algorithms um and today that frequent
16:56frequently means machine learning
16:58algorithms
16:59um shape sociopolitical Dynamics um
17:02especially as they're deployed in more
17:04socially sensitive
17:07domains so visualizing algorithms is a
17:10kind of design tactic that's a core part
17:13of my practice which for me is
17:16spatializing computational structure and
17:19process in real
17:23time so what we see here is a um
17:26interactive software piece developed in
17:28the unity game engine and it visualizes
17:32a simple machine learning algorithm a
17:34generative decision treat that's used to
17:36make
17:36classifications that become decisions
17:39about who has access in this case to
17:42social housing and who has access to
17:45disability benefits so very much
17:47decision making um that is of social
17:51consequences and in making this work I
17:53was interested in opening up a
17:55predictive system and understanding
17:58first of all spatially where decisions
18:00are made so you can watch as individual
18:04data points which are the colored
18:05circles which are
18:07people um being passed by this oops
18:11being passed by this internal
18:13Network so a question for me is can this
18:16approach to software support a a sort of
18:18re-engagement with decision making in
18:21prediction systems where we're perhaps
18:23at risk of losing our connection to
18:26it and I'm interested in
18:30um these terms like the problem of
18:32interpretability and explainability
18:34Solutions in machine learning and a
18:37desire to reframe them as not just
18:39technical issues but also entangled
18:42sociopolitical dilemas and therefore
18:45there as much um a question for the Arts
18:48and Humanities to Grapple with and
18:50contribute to as they are for computer
18:52scientists to reconcile and
18:56solve so for me it's about bringing code
18:59from the inside of an algorithm into
19:01legibility and using a visual language
19:04to make that
19:09accessible and then slow computation or
19:12maybe slow AI now is another tactic that
19:15I work
19:19with for me this is purposefully slowing
19:22down computational process to meet the
19:24human scale of perception and
19:27understanding
19:28the nature of computation is a process
19:31that happens beyond the human scale of
19:33perception it's hyperfast it's
19:36obfuscated for most it's technical it's
19:40abstract so this quality of
19:43slowness um especially in this new world
19:46of massive interpretable impenetrable
19:49models that we're living through I'm
19:51thinking about how knowledge is said to
19:53now emerge through an AI system and how
19:56how can we find a way back to that
19:58relationship to knowledge
20:04generation so the these pieces were
20:06shown in um an exhibition at the center
20:08pompo in Paris um called neurons
20:12simulated
20:14intelligences and as an exhibition piece
20:16I'm interested in how we can render
20:18algorithms more
20:21sensorial how to engage the human gaze
20:24in algorithmic process in
20:27predictions um in the same way that
20:29might one might spend time in front of a
20:32painting in a museum engaging that
20:36particular type of time a kind of
20:38contemplative
20:44time and then reflexive
20:47software um for me this entails
20:49recognizing an algorithm as constructing
20:51an argument from it it has a Viewpoint
20:55and it's trying to achieve not just an
20:56extrinsic task but potential an
20:58ideological operation so then the the
21:01visualization of algorithms and
21:03interface design software development
21:07these all become a medium for exploring
21:09this type of soci technical
21:12investigation so in contrast to
21:14traditional software development that
21:16produces tools to perform an extrinsic
21:20task reflexive software development is
21:22critically reflecting on its own uh code
21:25and functions and processes
21:29um it's what Yanni lisses might call
21:31interfaces that cause
21:35friction and then another um creative
21:38component of my practice um is counter
21:41data sets which means building
21:43alternative data sets to probe some of
21:45the issues that I'm interested
21:49in so working with counter data is a
21:53design propositional practice for how we
21:55can think about counter arguments in
21:57software design also building on the
21:59work of MIM Ona's project of missing
22:04data so it includes what I term the
22:06unmodeled and how we can think about
22:08particular human values or
22:11perspectives um that could be encoded
22:13into algorithms but in fact are
22:17missing and then finally um this piece
22:21uh also comes from a counter counter
22:24data set approach um and it be it's a
22:27it's part of a much bigger project
22:29um in which I filmed my mother in her
22:31own home and then reenacted her domestic
22:34tasks to create an alternative motion
22:37data set as an algorithmic rendition of
22:40this
22:42labor and I'm interested in the way that
22:443D digital asset libraries um and data
22:47sets as a whole often embody this notion
22:51of agnosticism the idea that they can be
22:53taken from one place and used anywhere
22:55else and I wanted to build a more
22:57politicized creative library of motions
23:00and activities in this case pointing to
23:03a gendered um labor so it's also a data
23:07set that comes from part performance or
23:08reenactment of someone's life
23:13um
23:14and the final
23:16slide um is just trying to think about
23:20the near future of the home as a space
23:22that produces these data sets that could
23:24be used to train domestic robots and
23:26other learning algorithms that is set to
23:29enter intimate spaces and Care settings
23:32so thinking about the privacy of the
23:34home as a domain targeted by kind of the
23:36looming vectorization of the body and
23:39intimate
23:41space I'll leave you there thank
23:49you great okay next up is David Benjamin
23:54who is um founding principal of the
23:57living it's a a architecture small
24:00architecture practice and an associate
24:02professor at Gap as
24:06[Applause]
24:13well oops sorry about
24:16that uh so uh thank you Laura Naim I'm
24:20excited to be part of this important
24:24conversation let's see does that work
24:27yes uh
24:28so first I'd like to acknowledge the
24:30incredible team that I work
24:33with for the past 15 years we've been
24:36aiming to combine uh some of the
24:39benefits uh of human creativity with
24:41some of the benefits of computational
24:44analysis and in addition we've been
24:46trying to test out some different AI
24:49approaches in real world physical
24:51projects so this is Twin mirror for the
24:54sole bional of architecture in 2017
25:01and here
25:05oops uh and here we combined two machine
25:08learning models based on two different
25:11sets of data so visitors approached and
25:14saw themselves reflected back through
25:16these two versions of reality neither
25:19version of reality was neutral or
25:20objective both of them were biased and
25:22by seeing them together we could start
25:24to develop a kind of critical Instinct
25:26for how the technology works in addition
25:28to the gallery component we also showed
25:31this project on the streets of the city
25:33in
25:35so and so with this as background I'd
25:39like to quickly describe four different
25:41approaches uh that we've been exploring
25:44uh with that combine Ai and creativity
25:47number one generative design for
25:49lightweight
25:51structures uh a few years ago uh Airbus
25:55the big airplane manufacturer developed
25:57a lightweight concept plane for the year
26:012050 but they also wanted to start
26:03testing this concept with a real world
26:07demonstrator and the demonstrator that
26:10they selected was a partition for
26:12today's A320 planes so the partition
26:16shown here in red uh looks simple but
26:18it's actually very difficult to design a
26:20partition that is both strong and
26:23lightweight so we used a computational
26:26method called generative design to
26:27explore a wide design space in search of
26:31options that were both High performing
26:33and
26:34unexpected in other words we created a
26:37custom software workflow to generate
26:39evaluate and evolve literally tens of
26:42thousands of design options including
26:44some that were very unusual and many
26:47that were Beyond typical rules of
26:49thumb and then we used techniques like
26:52principle component analysis to explore
26:54tradeoffs and Home in on some of the
26:56best possibilities
27:00we also used 3D printing and a new alloy
27:03to test the partition out in the
27:05physical
27:06world this is the largest metal 3D
27:09printed airplane part which consists of
27:1360,000 microl latus
27:15bars and here's the 24g static test of
27:20the fulls scale component which we
27:22successfully passed and the partition is
27:24now being certified to fly in today's
27:26commercial airplanes
27:29the new Partition is 45% lighter than
27:32the traditional component and if it's
27:33installed in all of the A320 planes
27:35flying today it will reduce carbon
27:37emissions by about 1 million tons per
27:40year two generative design for complex
27:43buildings so here we used computation in
27:45the design of an airplane Factory
27:48instead of two goals for the airplane
27:50part we had 10 different goals for the
27:51factory including operational Financial
27:54environmental and social goals and for
27:57example here's the way we calculated the
27:59production efficiency of a factory
28:01layout measuring the flow of materials
28:03and people between logistic spaces
28:05assembly spaces and paint
28:08spaces and here's a self-organizing map
28:10that compresses some of the
28:1110-dimensional data that we had into a
28:13two-dimensional graph and here it's
28:16important for me to note that the
28:17computer and the math offer us a good
28:20set of designs but they don't produce a
28:22single design the final design requires
28:25human stakeholders to exercise judgment
28:27and make make collaborative
28:29decisions three machine learning for
28:31circular materials so this is a new
28:34building that we designed for Princeton
28:36University uh for things like
28:38experiments with robotics and sensors
28:41and everywhere that computers meet the
28:43physical world and become so-called
28:46embodied
28:49computation and to reduce carbon in this
28:51building as one of our kind of
28:53sustainability ideas we decided to make
28:55the facade out of repurposed scaffold
28:58boards uh from New York City
29:00construction and these boards would
29:02otherwise end up in a landfill so this
29:03was a kind of idea of uh using a
29:07sustainable material we basically
29:09developed the hypothesis that this kind
29:11of weathered wood with these micro
29:13Contours you see here might offer some
29:15thermal performance benefits uh in terms
29:18of trapping small pockets of air and
29:21also some hydr performance benefits
29:24through shedding
29:25water so we knew that with a material
29:28offer these kind of performance benefits
29:29but we also knew that zones of the
29:31facade had different performance
29:32requirements as this solar analysis
29:34shows and we knew that we had thousands
29:36of boards with different properties to
29:39arrange on the facade so the question
29:40became which board with which properties
29:42should we put in each
29:45position the areas of wood with the
29:47narrowest grain tended to provide the
29:49best performance and the areas around
29:51the knots in Wood tended to have the
29:54narrow grain we wanted so we set out to
29:56identify and treat the knots in in each
29:58of our boards first we photographed all
30:01the boards and then we used a little bit
30:02of human training asking people to look
30:04at photos and click not with the knen or
30:07not with an
30:09n and after some initial training the
30:12algorithm actually got pretty good at
30:14seeing where the knots were even in
30:16photos that had never been seen or
30:18classified by a
30:20human here's the algorithm homing in on
30:23the knots and it was fascinating that
30:24the human had simply stated whether each
30:26photo had a knot but the human didn't
30:29Circle it or describe it and then the
30:30computer was able to look at a new photo
30:33and not only figure out whether the
30:34photo had a knot but also find the
30:35location and the size of the
30:37knot so after using machine learning to
30:40identify the knots we created a custom
30:42robotic sand blasting machine to treat
30:44the knots and produce the kind of micro
30:46Contours we wanted and overall this
30:48allowed us to explore kind of a new
30:50approach to materials practically it
30:53allowed us to reuse materials and reduce
30:55the carbon footprint of the building
30:58and conceptually it allowed us to
31:00celebrate variation rather than reducing
31:02it and to overturn this kind of lowest
31:05common denominator approach to
31:08materials the facade itself became an
31:10example of embodied computation and in
31:12the end the building suggests a new
31:14hybrid design approach that combines
31:16human intelligence and machine
31:20intelligence so for the last one uh
31:23large language models for site layout so
31:25this is a software project called tile
31:27GP and a building project called The
31:29Phoenix which is a 300 unit affordable
31:31housing complex and here we started with
31:34the prototype for so-called
31:36example-based design where designer
31:38provides a catalog of parts and enters
31:40some examples of how those parts should
31:41fit together and then a software system
31:44that we created learns the desired
31:45patterns and creates new designs with
31:47the same properties as the example
31:49designs and then we developed the
31:51Prototype into a no code software
31:54workflow where you can enter natural
31:56language requests in including requests
31:58for formal features or performance
32:00features and then you receive layouts
32:02that correspond to your
32:03request so the result is a design for a
32:06specific plot of land in Oakland and
32:07also a tool that can be reused to make
32:10sustainable decisions on other projects
32:12in the future so here's a vacant lot
32:15that will soon become this Net Zero
32:17affordable housing complex thank you
32:20[Applause]
32:30um so next up is Francisco
32:34Ramirez who is um pursuing a master's
32:38degree here in the v in the visual arts
32:40with a concentration in
32:42photography um and transformation is a
32:44constant subject within his work so I
32:47hope he's going to show some of that
32:50[Applause]
33:00thank you for the introduction hi I'm
33:03Francisco Javier Ramirez and I am an MFA
33:07first year MFA Visual Arts student in
33:09the photo concentration like Laura said
33:12my practice explore transformation but
33:14focuses on queerness and immigration
33:16through my personal experience and new
33:18ideas of the photographic image in my
33:21recent work I've been thinking about
33:24Utopias myths and the American dream
33:27focusing on my father's migration from
33:30Mexico and how it repeats with my own
33:34story my father's family converted to
33:36Mormonism when he was a child and he
33:38migrated to the US in 1990 when my
33:41mother was pregnant with me these are
33:44passages for which I do not have any
33:46images therefore I was interested in
33:49ai's capabilities to create an archive
33:51of those images that don't exist for
33:54instance here in this photo
33:58um there's a man who could resemble me
34:01and be my
34:04father
34:05and
34:08um yeah it could be an image from him in
34:12his first years in the US a family
34:13portrait that previously only existed in
34:16my
34:18mind I think I maybe I can I use this
34:21one yeah this one's a little short for
34:23me
34:25um through my work I am often
34:27questioning religion and what it
34:30represents in American life I believe
34:32Mormonism was important in my father's
34:34story and his aspirations for white
34:36American Suburbia for this project an
34:39arrow band right above the Horizon I
34:41wanted to collapse the real and the
34:43fiction I used AI to generate new old
34:46archive and blended it with the real
34:48archive I found images that included the
34:50historicity that we carry combined with
34:52prompts based on my personal
34:56history I was conscious of the cliches
34:59and stereotypes that AI could easily
35:01reproduce and I was looking for
35:02contradictions finding artifice in
35:04reality and reality in artifice
35:07collapsing the idealized images from
35:09media and Ai and the
35:11real reality of History it is reflected
35:15in the Scrolls that we'll see in a
35:16minute um where images from various
35:20sources were flattened for multiple
35:21layers of meaning into a single one
35:23through a series of processes first I
35:27printed an Archive of man-made and
35:28machine-made images in different sizes
35:31but similar materials then I manipulate
35:34images by painting and sexing
35:36them before arranging them into a
35:39composition like I was doing there on
35:41the table to be scanned with a handheld
35:43device the lower resolution handheld
35:45scanner the saturates images and causes
35:47a loss of detail introducing
35:49imperfection where there was only
35:52certainty creating a blend of realities
35:55that took away the sheen of the AI
35:57images in made it more
36:00believable when I generated a Mormon
36:02temple it was always an
36:05idealized white building with this all
36:09inspiring mountainous Landscapes and
36:12Cascades flowing and if I asked it to
36:15generate a Mormon missionary it was an
36:16attractive white young man every time
36:19technology is simplifying things into a
36:21single possibility that cannot
36:23understand the contradiction of The
36:24Human Experience so when generative AI
36:27think of a Mexican man in Utah it still
36:30represents the cliches the Hat the
36:33pickup
36:36truck it can more easily render a
36:39believable American Suburban street
36:41because its points of references are
36:43those that have already been shown many
36:50times when I was selecting the images
36:52from the three archives I compiled I was
36:54not fully interested in the reality in
36:56which we exist but rather in reflecting
36:58my own thoughts of what I understand it
37:01to be and what I thought it to be when I
37:03was a
37:05child using the
37:07error of the mechanical device like the
37:10scanner I can control the artificiality
37:13and resemble the flow in which we exist
37:15and as the title of this series suggests
37:18an arrow bound right above the Horizon
37:21is some
37:24Mirage I consider this works to be impr
37:27process but I like how they start to
37:29create an
37:30atmosphere I still think that for the
37:34audience it needs a point of access so
37:38in that sense there are other projects
37:39that I've created where I do Straight
37:41photography and incorporate certain
37:43elements of AI that are more
37:49legible this like this years is David
37:52cruising at Prospect Park and well this
37:54is just a photo
38:01okay the project is a sequence of photos
38:03where I am using an app called sne to
38:05cruising Prospect Park this room may not
38:08know but sniffies is a wellknown
38:10map-based cruising app so I am there
38:13using the
38:14app
38:16um and cruising is an act of seeking out
38:19sexual encounters historically and
38:20semi-public Spaces by queer
38:23people as I'm
38:25cruising I feel a presence and as I turn
38:30around I
38:32see David
38:34wasnik David was a gay artist writer and
38:36political activist from New York who
38:38died of AIDS in the early
38:401990s and when I was making this series
38:42I was thinking of query life all the
38:45people who died in the eights crisis and
38:47how even now as queer people we carry
38:49that stigma so when I am bringing David
38:52back from the dead through
38:53AI I am asking myself what would it mean
38:56for David to to still be alive to be
38:59older and to still be doing the things
39:02that he would do like
39:05cruising that's the queer Utopia that
39:07Jose Stan MOS talked about where past
39:10and future exist in the present and we
39:11have hope but also examining how queer
39:14culture has been commodified by
39:15technology to the point where we now
39:17have apps for cruising and while I'm
39:19cruising I find David and what do I do I
39:21take a photo thank you
39:37um so next is Mark Hansen who joined the
39:41Columbia journalism School in 2012 from
39:44a career a long career in statistics and
39:48he runs the Brown Institute for media
39:51innovation
39:53[Applause]
40:01thank you Laura thank you night no
40:04pictures um I'm very happy to be here
40:07today um and my contribution will focus
40:09on creativity in
40:12journalism uh my background as Laura
40:14mentioned is in statistics and I teach
40:16computational journalism in the J School
40:18here in my classes we explore the way
40:21new technologies like AI might impact
40:24the practice of Journalism and the flip
40:27how the needs of Journalism its unique
40:29methods of investigation and its
40:31underlying values and ethics how these
40:33might shape the priorities for new
40:35research in
40:38AI what just happened
40:45there
40:51um well this will be an
40:55adventure um uh journalists uh both
40:59report on as well as with AI um are
41:02called to serve the public good we are
41:04called to serve the public good but do
41:06so in a way that PRI prizes verification
41:09these are the starting points for
41:10creativity in the profession across the
41:13country AI is finding its way into to
41:15sourcing and information gathering into
41:18structuring and analyzing data and
41:20eventually shaping how we present um our
41:23work through new ways of reaching
41:26audiences um
41:28so I'll briefly uh talk about the state
41:30of play in journalism um the
41:33relationship between AI companies and
41:35journalistic Outlets is complicated um
41:38but generative AI has found its way into
41:40newsrooms small and
41:42large uh this is the Washington posts um
41:47uh uh uh ask the post uh AI uh
41:51application that distills answers to
41:53news questions from published reports um
41:56also from the the post
41:59um we have uh uh and elsewhere AI is
42:03creating automated summaries of reader
42:05comments um and their
42:07articles this is why I said this would
42:09be a little bit of an adventure um uh at
42:12the city a local publication serving New
42:15York City uh they've developed an AI tip
42:18bot that makes it easier for you to
42:19submit uh tips about uh landlord
42:22violations or issues with local
42:23government by dynamically asking you for
42:26as many details as possible about the
42:29event um under the banner of AI uh
42:33newsrooms use models to sort through um
42:36vast quantities of of
42:39information uh uh uh including uh text
42:44uh images and videos creating structure
42:46and spotting patterns this has fueled
42:48incredible new reporting like this piece
42:51from the times um that poured through
42:54satellite imagery of Gaza to spot where
42:562,000lb bomb had been dropped you might
42:59be surprised at how sophisticated these
43:01kinds of projects had become um in an
43:04early move uh one Newsroom even trained
43:06in llm dubbed Bloom Bloomberg
43:09GPT um although uh recent advances in
43:12the big commercial models have outpaced
43:14the ADV uh advantage that this model
43:16once had and of course any discussion of
43:21um of
43:23creativity um uh uh in and they a should
43:27address the importance of Journalism as
43:30content content used in training models
43:33um with local newsrooms uh across the
43:36country um blocking AI Bots with um
43:39entries in their robots.txt files and
43:42others making
43:44deals I'm not sure how these got so out
43:46of order but anyway um others making
43:48deals um with AI companies local news is
43:52staking out unique territory its
43:54creativity dealing in people places and
43:57situations that might not appear in
43:59other available
44:00content uh to give an example of the
44:03twin roles of reporting on as well as
44:06with AI um let's consider the Washington
44:09Post again um after Michael Brown was
44:12shot in 2014 the post began collecting
44:14data about um the extent of police
44:17shootings across the United States um
44:20coling data from uh uh news reports law
44:24enforcement websites social media and
44:27creating structured databases from those
44:29unstructured sources this kind of manual
44:32uh this kind of task was manual in 2015
44:35and is now automatable in 2025 uh with a
44:38balance between public good and
44:40verification found at the center uh full
44:43disclosure the post is shutting down
44:45this site but looking for modernization
44:47in a new home so if anyone here is
44:50looking for um a project to a worthy
44:52project to keep alive this would be it
44:55um once regular ized uh the facts of
44:58each event are exported into a table um
45:01which can be read into an llm in this
45:03case chat GPT um and analyzed um was uh
45:08what's the incident count per year here
45:10we have it as a table or as a line plot
45:13importantly by generating
45:16code um uh via an llm in the context of
45:19data analysis we have a secondary
45:22artifact that a data desk or a colleague
45:24might be able to check before
45:26publication appealing to AI um or to
45:29various analysis agents we largely shift
45:32the prized form of creativity in data
45:35journalism away from the ability to turn
45:37questions into code and instead value
45:40the ability to ask good questions in the
45:42first place and this is what the J
45:44School curriculum is all about curiosity
45:47and asking questions but even in this
45:49simple context we see the reporting on
45:52versus reporting with roles in my
45:55computational journalism class
45:57um uh last year for example we looked at
46:00the household pulse survey uh from the
46:03Census Bureau it was the bureau's first
46:04product to ask about sexual orientation
46:07and gender identity um and somewhat
46:09remarkably it's still up on the website
46:13um uh uh this was part of um a larger
46:17theme in my class uh last year about
46:19queer Tech establishing a beat about how
46:21technology impacts the lives of lgbtq
46:24people and how perhaps their lives have
46:26inspired ired um the development of new
46:29technologies it was an introduction to
46:31data and computation that was Heavy with
46:32GPT assistance um in an early move um we
46:36looked at um uh uh the ages of
46:39respondents to the household pulse
46:41survey to compare it against the US
46:43population GPT gave us this um this
46:46lovely uh histogram um fine um I then
46:50found from the the the Census Bureau
46:54that there were these other amazing uh
46:56lovely plots called pyramid plots um
46:59that compared and um uh compared
47:02population across age and Sexes um Light
47:04Blue for males dark blue for females um
47:07we asked GPT to to make us one from our
47:09data and it gave us uh well this is
47:12asking GPT and it gave us
47:15this um uh what's the difference here um
47:19uh male and female pink and blue uh my
47:22point in bringing this up is uh simple
47:25is is is my point in bringing up this
47:28simple example is that um the biases in
47:30training data slip into even what we
47:32might think of as very dry technical
47:34work journalism's duty to report on and
47:37with AI um uh is perhaps something we
47:41should take on broadly as we use these
47:43models and platforms especially when it
47:45comes to our creative practices thank
47:47you
47:49[Applause]
47:57um so we have our last speaker and the
48:00fifth out of 17 schools across Colombia
48:03represented on our panel so Sheena
48:06enangar is the inaugural St Lee
48:09professor of business in the management
48:11division at the Columbia business school
48:13and a world expert on choice and
48:21decisionmaking good afternoon
48:24everybody now I'm blind so you need to a
48:26little better than that good afternoon
48:30everybody perfect thank you um need to
48:33make sure you're all awake so um my name
48:36is Sheena Angar and I've been at the
48:38Columbia business school for the last 25
48:41years and uh um as a moderator just
48:44announced I've been studying Choice my
48:46entire career and the big question that
48:49I study is how do we get the most from
48:52choice and and I've I've written a
48:55couple of books on it and about 10 years
48:58ago I started to ask the question that
49:02more than just being able to find the
49:04best choice and a whole array of lots
49:07and lots of choices how do we create
49:09meaningful choices after all we live in
49:12a world where we're often confronted by
49:14meaningless trivial ucer choices how do
49:17we create those meaningful choices and
49:20that led me to a path to create a
49:23methodology that's based on
49:25neurocognitive
49:27economics um Sciences um I began to
49:31teach that methodology in the last 10
49:33years and more recently I've even
49:35enhanced it now through the use of
49:37generative AI so let's start with the
49:40question that underpins this
49:43methodology where do big Ideas come from
49:46who gets them and how do we get them and
49:50so in the interest of the topic of our
49:52panel I figured I would start with art
49:55so
49:57I'm going to ask you a question and I
49:58want you to just call out your answers
50:00who do you think made this piece of art
50:02just call
50:04out ah she's a smart
50:07one okay who do you think indeed it is
50:11Picasso 1901 who made this piece of
50:15art absolutely notice the transformation
50:18in the style and certainly historically
50:22whenever we think of great artists great
50:24thinkers of any kind we're we convince
50:27ourselves that these big Ideas come to
50:29us almost like magic you know maybe you
50:32were given some Divine Revelations when
50:34you climbed the mountain maybe an apple
50:37just happened to fall upon your head
50:38when you were sitting and then you got
50:41an Insight that showed you some great
50:44insight about the world well we now know
50:47that it wasn't magic and in fact we know
50:50through a lot of recent advances in
50:52science about how our mind actually
50:54creates ideas and in large part ma uh ML
50:58and AI have actually have taken
51:01advantage of that knowledge so how did
51:04Picasso get this
51:07transformation what was a
51:09technology that came into being in the
51:1219th century that completely disrupted
51:16art guesses yes did you say the yes it
51:21yes
51:22indeed it was photography right because
51:25back then we would make port portraits
51:27and suddenly you had this thing that
51:30could take pictures and capture people
51:32in real life more precisely more quickly
51:36more cheaply from today painting is dead
51:40cried one critic yet another critic
51:42cried it's not long before art will be
51:45entirely supplanted by the camera but
51:47the only thing I have to do is change
51:49the word camera with a generative Ai and
51:51I think we're perfectly comfortable in
51:54modern times so
51:57Picasso like all the other realists is
52:00searching for a technique you know
52:03because we all know that art didn't die
52:05they just found new ways to see and so
52:09in
52:111906 Gerard Stein invites him to her
52:14Salon in Paris and there he beholds this
52:17famous painting by matis the joy of Life
52:21take a look at this
52:23painting now matis was very much
52:25influenced by his mentor say take a look
52:29at this five
52:31bathers and Picasso asks to meet matis
52:35and they meet in a
52:36cafe and while heading to the cafe matis
52:40happens to pick up an African sculpture
52:43that had just come in from one of the
52:44French Colony so he brings that along to
52:47his meeting it is recorded by mati's
52:50secretary that throughout their meeting
52:52Picasso held that African sculpture and
52:55he kept fingering its long its ridges
52:57its angles and that very night he
52:59started to paint de moisel de
53:03aen take a look at
53:06this so a contemporary of Picasso TS
53:10Elliot once quipped immature poets
53:12imitate mature poets steal so was
53:17Picasso a mere thief or was he like any
53:21great innovator artist science scientist
53:25entrepreneur someone who takes ideas
53:29drawn from distant and diverse sources
53:32and combines them in a new and
53:35meaningful
53:37way Sher in the same era the great
53:41Economist said that all Innovation is
53:43nothing more and nothing less than a new
53:46combination of old ideas and certainly
53:49the work of Eric kendel the neuro the
53:51Nobel prizewinning neuroscientist showed
53:54us how in our brains we simply take lots
53:58of different memories bits of
53:59information and keep combining and
54:01recombining them until we might stumble
54:05stumble upon a big
54:07idea so a number of years ago I began to
54:11think maybe we could actually take this
54:14knowledge and create something
54:16prescriptive a and give people a type of
54:19toolkit that would enable them to create
54:21their best ideas their best solutions to
54:24whatever complex problem that they were
54:26attending to either in their personal or
54:29in their professional lives and I
54:30created a course around it and the
54:32course is called think bigger and
54:34essentially Drew on what we had already
54:37learned from Neuroscience about how the
54:39brain works from cognitive science about
54:41all the things that get in our way and
54:43it essentially applied the theory of
54:46shumer so these are the six steps choose
54:49a problem break it
54:51down compare wants search in and out of
54:54the box um
54:57a choice map and the third eye
54:59essentially what it is is you define the
55:03problem you break it down and you go out
55:06into the world and you search how has
55:09anyone solved this problem in different
55:12places and you take those pieces that
55:15you get from dis different from diverse
55:18places and you combine them until you
55:21create a unique
55:23solution now I used to teach this the
55:26hard way the long way and would have
55:29students spend weeks searching but just
55:33this past July we rolled out a new
55:36generative AI app that literally does
55:39those six steps and we found that in a
55:425-day boot camp you could actually get
55:45students to now create entrepreneurial
55:48ideas which they can actually get funded
55:50for these are just some of the ideas
55:53that just came out in the last few
55:55months
55:56like a way to help people detect if this
56:00food they're about to eat in a
56:01restaurant um might affect their
56:04allergies um how you might diagnose if
56:07your kid has a learning disability to
56:09name just a few and I'm going to end by
56:12showing you the QR code in case any of
56:14you want to test it out so with that I
56:18want to encourage everybody to believe
56:21that yes we can all think bigger thank
56:24you very much
56:44please can you hear me yeah okay um
56:49so thank you all this was a really
56:52amazing um panel and as I predicted um
56:56yeah I should have I should have written
56:57an algorithm for it but I there would
56:59have been no purpose for it um not one
57:03person over here told us that AI was
57:06going to replace us as humans in making
57:10art not not they it was furthest away
57:13from that and in fact what everybody did
57:16was um tell us how creative they are in
57:20using AI um in alternate ways than it
57:24might have been in you know for for
57:26which it might have been you know
57:28invented or created or stumbled upon Etc
57:32you know so Dennis is talking about you
57:34know not not using AI to to write but to
57:40figure out how to use it in a in a more
57:42creative way let's call it creative
57:44right in a in a way that is actually um
57:47a good thing to to teach our students
57:49and Catherine was um you know obviously
57:53taking apart AI but really interpret it
57:56and explaining it to us in a in a
57:59creative way that allowed us to enter
58:02into the algorithm in a creative way and
58:05David um brings qualitative as well as
58:08quantitative um uh what inputs right
58:12into into a design problem really
58:15massively expanding the dimensions um
58:18through which we create a design space
58:20for architecture um Francisco you know
58:23beautiful in terms of how to recover
58:26memory um through AI which is not
58:28something I had ever thought about I
58:31think that is such a creative use of
58:33generative um imagery and Mark you know
58:36reporting on AI I think has some
58:39similarity with the way Katherine um is
58:42in interpreting and the accountability
58:44that we all have to bring to Ai and
58:47Sheena I would associate um you with a
58:51kind of design thinking uh process but
58:53in a more in in um but specifically for
58:57art and that the the ways in which um uh
59:02design thinking actually comes out of a
59:05computational design space and M you
59:08know method makes everything a little
59:10bit more methodical than artists would
59:13um so I don't know if we're allowed to
59:15actually have a discussion but I would
59:17like to sort of ask you know everyone I
59:19don't yeah we all know I think that's a
59:23good summary I think of the of
59:26everybody's of everybody's talk and I
59:28really am you know hugely optimistic
59:31that no one even even thought through
59:34the idea that you know AI is ever going
59:37to completely take over human creation
59:41of art or
59:43creativity thank you okay and I've also
59:46been instructed to tell everyone to stay
59:48in their in their seats so that the
59:51transition to the next panel goes e e
59:54more easily thanks so much
1:00:01than