Full transcript
0:14the AI for good Global Summit aims to
0:16bring together industry leaders to try
0:18and harness AI to further the un's
0:20development goals in areas like Health
0:23climate and sustainability how do we
0:25govern Technologies if we don't yet know
0:28their full potential they mean purpose
0:30of this Summit is to explore how the AI
0:33Solutions can help sustainable
0:34development AI will come to influence
0:37almost every aspect of our societies and
0:42economies for good I would say the most
0:45pressing Topic at the Summit is going to
0:48be how do you handle deep fakes
0:52misinformation the good news is that uh
0:55the future of undetectable deep fakes
0:57can be avoided through the development
1:00an implementation of robust detection
1:02Technologies and international
1:11[Music]
1:13standards you know what I've really seen
1:16is the fact that we can really affect
1:17amazing change but we have to do this
1:19together and we have to do it in
1:20collaboration with partners with
1:22organizations with institutions and
1:24that's what's making me so excited about
1:26what AI for good is all about
1:33onethird of humanity remains completely
1:37offline excluded from the AI Revolution
1:42[Music]
1:57[Music]
1:58[Applause]
2:01[Music]
2:03artificial intelligence is changing our
2:05world and our lives and it can
2:07turbocharge sustainable development we
2:10are the AI generation and this is our
2:14moment and it's our responsibility to
2:17write the next chapter in the great
2:20story of humanity and Technology
2:28[Applause]
2:30[Music]
2:35welcome to AI for good the leading
2:38action oriented Global and inclusive
2:40United Nations platform on AI organized
2:43by itu in partnership with 40 un sister
2:47organizations and co-convened with
2:49Switzerland the goal of a for good is to
2:52identify practical applications of AI to
2:55advance the United Nations sustainable
2:57development goals and scale those
2:59solutions for Global impact in today's
3:02session we're counting on you to use the
3:04live video wol feature to ask questions
3:06and post comments to help create an
3:08engaging discussion we encourage you to
3:11stay until the end to chat connect ask
3:15questions and network with our
3:17distinguished panelists and worldclass
3:19AI experts in the neural network it is
3:22now time to kick off the session and
3:24welcome our first Speaker the floor is
3:27yours
3:38hello everyone I warmly welcome you to
3:40our webinar today on AI and sustainable
3:44urbanization for good on behalf of the
3:47organizers the so Institute lemburg
3:51Institute of social economic research
3:54and the Korea Planning
3:56Association I'm Kim from Luxemburg
3:59Institute Ute of social economic
4:01research I'll be sharing today's
4:04session As Cities face increasingly
4:07complex challenges AI emerges as a
4:11powerful tool for enhancing urban
4:14planning and decision making today's
4:17webinar brings together distinguished
4:20experts who will explore various
4:22dimensions of AI and its role in
4:26creating sustainable Urban futures
4:29before we begin I'd like to invite Dr
4:32junong Tre from the soul Institute to
4:36outline our workshops objectives and
4:39future
4:40directions Dr Jun CH please proceed with
4:44your opening
4:46remarks thank you for
4:49introducing uh Dr Kim um
4:52I'm I'm Jun and I'm working at the S
4:56Institute as a director of
4:59collaborative
5:01research uh last one years we are
5:05organizing this event with with the not
5:09just planning but the mobility and the
5:12environment and every aspect of abound
5:15planning and and Orban is built
5:18environment issues so um so this um
5:23webinar is
5:25organized uh with the three partner two
5:29partner organization s Institute and
5:31Luxemburg Institute of social economic
5:33research and Korea Planning Association
5:37we hope to continue this kind of the
5:41technology andan Oran issue and the um
5:46the we continue to uh dealing with the
5:49issues in the next other event also
5:51thank you
6:00okay so I'd like to introduce today's
6:02program our program today reflects the
6:05multifaceted nature of AI in Urban
6:08Development bringing together experts
6:11from research institutes universities
6:14and Industry across the world we will
6:17explore how AI is transforming cities
6:21through four key
6:22perspectives strategic planning for mega
6:25cities optimization of urban Mobility
6:28systems environmental resilience through
6:31satellite technology and the critical
6:34Frameworks needed for responsible AI
6:37governance these interconnected themes
6:40will help us understand both the
6:43Practical and applications and policy
6:46implications of AI in creating sustain
6:49sustainable Urban Futures our discussion
6:53will be enhanced by perspectives from
6:56experts in spal geospacial Analytics and
7:00Urban Design helping us exploring the
7:03Practical implications of these
7:07developments and I'd like to happy to
7:09introduce uh our first Speaker Dr
7:14junong uh from the soul Institute as a
7:18director and a research fellow at the
7:20center for collaborative research he
7:23brings extensive experience in Smart
7:26City Planning and egovernment initiative
7:29tips his work focuses on AI and XR based
7:34decision support systems and he has
7:37contributed to numerous national
7:40projects on special policy and Urban
7:43Development Dr CH will share insights on
7:47digital uh technology for Mega City
7:49Planning with a particular focus on
7:52artificial intelligence in planning Dr
7:55CH please proceed with your presentation
8:10yes yes thank you for
8:12introducing my topic my presentation is
8:16about AI artificial intelligence for
8:18Mega City
8:20Planning my presentation is uh five sub
8:24chapters from planning issue to digital
8:27technology in the AI era
8:30the S Institute before before starting
8:32the presentation I would like to
8:34introduce my research institution the S
8:38Institute established in
8:411922 and fully funded by S Metropolitan
8:44government and six research division one
8:47AI lab and one
8:49Center and three hund researchers and uh
8:54the budget is about 37 million
9:00the plan issue in the metropolitan area
9:03is uh is a kind of the expansion of
9:08Orban area built built uh environment
9:11over administrative boundary in case of
9:15soul soul is surrounded by kungi and
9:19inan and in consist Soul metropolitan
9:23area met mega mega mega City region and
9:26it expanded from um um very uh rapidly
9:30from the early
9:321960s and from 1980s and 2010 is most of
9:38the uh uh area is urbanized and the as
9:43of
9:442019 it is U committing from the very
9:49long distance so long distance time and
9:53uh long travel distance so uh in this
9:57metropan area there are many issues
10:00increasing increasing Orban problem
10:03over inside not just inside but the
10:06between in between the administrative
10:10boundaries such as Regional
10:12Transportation conflict of Waste
10:15Management installment deregulating the
10:19green belt for housing provision
10:22Regional m transportation service common
10:25use of water resources and uh
10:30n issues that need to to to deal with
10:35this kind of issues we have a complex
10:38and integrated decision
10:44making so so metropan area we uh deal
10:48with this issues uh we have some set
10:52setting some vision and objectives such
10:54as competitiveness industrial space
10:58citizen needs for culture and climate
11:01change and it also some kind of
11:05integrated and complex decision making
11:13process planning and
11:15planners naturally basically needs a
11:18wise investment of limited public
11:20resources and management of the
11:22complexity of human settlements
11:25combining quantitative and qualitative
11:28information for optimal decision
11:31making so planning through rational
11:34models processes and participation
11:38strategies evaluation of planning tools
11:40from statistical techniques to map based
11:44suitability analysis traffic models and
11:47digital
11:49twins the but the role of AI and
11:52planning decision making that is as a
11:56automated decision making tool AI
11:58operator as an artificial intelligence
12:01capable of making decisions without
12:05expit human
12:07command and it it predict Behavior
12:11outcomes based on patterns identified by
12:14from accurate data and
12:17rules but the interpretation and
12:20response to unintended outcomes and
12:22unexpected errors so there there's some
12:27good outcomes and we need to very
12:29cautious of outcomes when applying or
12:33using AI in planning
12:37processes there are many element of uh
12:40AI element in about planning and design
12:44from machine learning de learning Neal
12:46Network to real time emotion
12:50analytics s control gaming real time
12:53Universal trans translation to Virtual
12:57personal assistance so
12:59the the reference uh depicted this
13:04diagram there is more than 100 component
13:07of artificial intelligence that
13:09currently impacting planning and aan
13:12design
13:14practices so we we we have to need to
13:17consider the implication of AI ucg in
13:20aan
13:21planning there are synergies and
13:24similarities between Ai and planning the
13:27essense of planners is creativity and
13:30human centeredness and AI cannot replace
13:34the metaphors of expression and style
13:38that plannings bring AI should be used
13:41as augmented
13:43intelligence but the rapid development
13:45of generative AI technology is exceeding
13:48previous expectations so impl
13:51implication of AI in planning that is uh
13:57the technical accuracy of models and
14:00limitations in available data the need
14:04for literacy in automating qualitative
14:07aspect into numbers and entities within
14:10software code discovery of best
14:13practices collaboration with engineers
14:16and AI Literacy for
14:21planners so the use of AI for planners
14:26is a somewhere between creativity and
14:29Humanity you may you may heard about the
14:32Deep blin chess the
14:351997 uh exported system by IBM that that
14:40is compete against the human chess
14:42champion Gary KAS
14:45Paro but listen to Alpo in
14:49go the developed by De mind in 2016
14:53artificial intelligence developed by the
14:56mind which competed against human go
14:59Champion
15:00Lor the the creativity we think the
15:05creativity is comes from the the 37th
15:09move played by De mind in the in it
15:11second match against RoR
15:15someone told this kind this is a a kind
15:18of creativity but this is creativity
15:21than the computer or AI is a it is a
15:25triumph over human creativity
15:29so the technology in terms of Technology
15:32it developed from Big Data plus AI that
15:36enhancing analytical capability that
15:39expanding the Target and scope of
15:41analysis to AI plus robotics that
15:44focused on automation replacing human
15:47physical and emotional labor and
15:50generative AI this is a decisionmaking
15:53support and reduce times and cost of
15:56information search and acquisition but
15:59rapidly developing so we need some
16:01creativity to control the control and uh
16:06this kind of technology in the planning
16:10field so the problem solving with
16:14complex and integrated thinking for
16:17planning for Mega City Planning that
16:20means if necessary data such as
16:23population land use demand facility
16:26demand and Market policies are
16:28sufficient efficiently studied would it
16:30be possible to support the establishment
16:34of Orban
16:35planning some export very uh some exper
16:39in the Korean export he talked about the
16:42L ability left to
16:46humans uh that is the attitude l l
16:51capability of a human that means the
16:53knowy being replaced or removed
16:56eliminated by eliminated by generative
16:59AI technology replaced by AI powered
17:05robots so we think about the planning
17:08process decision making process of
17:10planning and design so as to we use AI
17:14adaptive use of AI in as a planning tool
17:19and we so we also consider ined planning
17:22and design workflow so the step is from
17:25The Briefing example to completion
17:30or Pro from pully digitalized planning
17:33and Oran design service workflow to high
17:36adaptation to high design support AI
17:39augmentation
17:41to design process that means we consider
17:44planning press ened or
17:47adapted with or inside AI tools for
17:55planning there are some strengths and
17:58weaknesses
17:59of AI as a planning tool that is
18:02technology and in terms of technology
18:04and planning such as spread ISO
18:07statistical packages and G already
18:10enhance the efficiency of planning tasks
18:13by
18:14automating but uh need to need for
18:18evaluation to assess the relative
18:20strengths and weaknesses of AI tools
18:23when applying this already uh the the
18:28changing this um technology into
18:32AI such as such as chat
18:36GPT and also we have need to consider
18:39time and wicked problems the potential
18:41of AI is recognized but its development
18:44is limited by a lack of data on about
18:47spaces and
18:49processes and also other cutting of
18:52issues such as data issues we need to
18:55deal with the special phenomenon so uh
18:58we uh we are developing or planning or
19:02planning process inside not just 2D
19:05space but also 3D space but there is
19:08limitation and OPP also opportunity to
19:12opportunities for the use of 2D and 3D
19:15spatial information in application
19:19field so uh as AI as a planning tool the
19:24go of planning is to strive for the
19:26common good while AI can address address
19:29problems that humans cannot Sol solve
19:32alone so planners reel to ethical
19:34guideline in their responsibility
19:36whereas AI LS on on ethical
19:40framework we we count on values and
19:43ethical decision making the topics
19:47include the ethical handling of data
19:49moral decisions or ethical
19:52algorithms and also we con consider
19:55inclusiveness and transparency when a
19:58makes decisions based on data sets
20:01people not represented in the data may
20:04be excluded incomplete data sets can
20:08lead to inequality and algorithm Biers
20:12can exacerbate
20:14disparities um as a users of AI planners
20:18need to understand the sources of data
20:20who is represented in the data and what
20:23is represent and absent in the black box
20:26aiy is a key for planet
20:29to use AI fairly and com
20:33purpose in the AI in the AI era ital
20:37technology and the planning process we
20:40also um use geospatial data from the
20:44200000 uh in Soul in the OB planning
20:48committee also 2022 we use 3D uh
20:52Dimension but the the planning process
20:56itself is still a same but we more
20:59detailed data set but need to consider
21:02more planning process into this virtual
21:05space so in inside virtual space we need
21:08to interactions between humans NPC
21:11non-player character is uh the main main
21:16kind of NPC is interact then we can make
21:19a virtual environment very similar or
21:22can be a norm for the human behavior and
21:26it it can be used as a guideline for the
21:29other decision making
21:32process so in the virtual space digital
21:36space we uh make or try or test the many
21:42policy intermissions and policy uh
21:45simulation cannot be done in the
21:48physical
21:51world in the near future the rapid
21:55development of digital T technology it
21:59isut realistic
22:03and
22:07very what is what is impacting
22:10development and how people or
22:15C this kind
22:18of already
22:20have plan that is 20 year long home plan
22:27inside the
22:30[Music]
22:33making
22:36mother
22:39[Music]
22:49and
22:52[Music]
22:56B be done in this watch world that is
22:59the twin of the physical
23:03world so it is the end of my
23:05presentation thank
23:13you thank you for your presentation and
23:17your
23:18insights um now I I'm pleased to
23:22introduce our second speaker of the day
23:25Dr yanan Sin Dr Sin is a
23:29assistant professor at the Department of
23:31Transport and planning at T Del in the
23:35Netherlands she also serves as the
23:37co-director of diamond lab which stands
23:41for digitalization and AI for Mobility
23:44Network
23:45Dynamics Dr sin's research is
23:48particularly relevant to our discussion
23:51today as she focuses on promoting
23:54responsible AI transformation in
23:57transportation systems her expertise
24:00spend several crucial areas including
24:03interpretable machine learning special
24:06caer inference and Mobility based
24:09anomaly detection before joining Tu de
24:13she was a senior assistant and lecturer
24:17at eth JK today she will present on
24:22unboxing AI to assist assist
24:25Transportation decision making where she
24:28will dis discuss both opportunities and
24:31challenges of implementing AI in Real
24:34World Transportation settings Dr Sin we
24:38look forward to your presentation please
24:41begin thank you very much for the
24:50introduction so today uh my talk is on
24:53AI for sustainable Urban Mobility
24:59AI offers many opportunities for
25:01promoting sustainability in the
25:03transportation field and it has been
25:06utilized in many areas of the
25:09transportation system for example in
25:12traffic management and control AI can
25:15provide us realtime traffic prediction
25:18to enable Dynamic Traffic Control in
25:22public transport
25:23optimization um AI has also been
25:25utilized to predict the demand in other
25:29ones and help in allocating resources
25:32and scheduling of those public
25:34transportation
25:36system and for electric and autonomous
25:39vehicles uh we all know that autonomous
25:41vehicles are built on AI Technologies so
25:45as you can see AI is almost part of our
25:48daily operation of various
25:51Transportation Systems
25:54nowadays as we witness the development
25:57of a I in the transportation field there
26:00is a strong necessity to improve the
26:03interpretability of robustness uh of AI
26:08this is because Transportation
26:09applications many of them are high stake
26:13applications so it's important for us to
26:15make sure that when those applications
26:17are deployed in practice they are
26:20reliable and also when we use AI to
26:25support decision making it's important
26:27for us to offer the end users with uh
26:31explanation that we um about the a
26:34models that we use in the decision-
26:36making
26:38process the need for enhancing the
26:41interpretability and robustness of a
26:43models also aligns with regulations on
26:46AI for example the eu's right to an
26:49explanation rule or the ethics
26:52guidelines for trustworthy
26:55AI to demonstrate the benefit pH of
26:58developing interpretable and robust um
27:01AI in the transportation field I will
27:04use the traffic forecasting application
27:07as a
27:08example traffic forecasting is uh
27:11crucial in helping uh reduce congestion
27:15and CO2
27:17emissions and nowadays AI method has
27:20achieved the state-ofthe-art performance
27:22in traffic forecasting so we're
27:24interested to know uh what knowledge has
27:28as the AI model learned from traffic
27:30forecasting and whether we could utilize
27:32those knowledge to make the AI models
27:35more efficient and
27:38reliable so next I will share a couple
27:41of studies we investigated on this topic
27:44the first study focuses on image based
27:47traffic forecasting with unet
27:50models this case study is inspired by
27:52the AY traffic forecast competition so
27:56in the three years of traffic forecast
27:58competition the unet architecture has
28:01achieved the best prediction
28:04performance um outperforming other
28:06models that have temporal uh that that
28:10are considered to be better at capturing
28:12the tempor
28:14dependencies um so we are interested to
28:17know why the U unet model has such a
28:21superior
28:23performance to answer that question we
28:26applied different types of ging based
28:28feature attribution explainable AI
28:31method to uh help us look into those
28:34blackbox AI
28:36models so attribution can be understand
28:39as a contribution of pixels uh in the
28:43input feature to a particular prediction
28:46in the
28:49output by analyzing the contribution of
28:52the input data from both the temporal
28:55Dimension and spatial Dimension we found
28:57out that um when're looking at the
29:00temporal contribution of the input uh
29:03sequences of a historical traffic
29:06data contrary to our hypothesis that
29:10having a longer historical input data
29:12would help in improving the prediction
29:16performance it turns out only the latest
29:20historical State um help the most in the
29:24prediction so when considering um 12
29:28input temp steps in total the previous
29:31first 11 temp steps historical state in
29:35only contribute in lowering the
29:38prediction error uh in this case mean
29:41squared error by
29:440.6% this is uh this partially explains
29:48why a unit architecture could outperform
29:52other models that have stronger ability
29:54to learn temporal dependence because um
29:57the tempor dependence especially
29:59long-term temper dependence is not a key
30:03contributor to traffic forecasting in
30:05this
30:06case when looking at the spatial
30:09attribution we also found out that for a
30:12specific Target
30:14prediction the neuron Network only
30:16utilize uh part of the spatial
30:19dependencies um that correlate with the
30:22target uh traffic flow or speed
30:25prediction it does not utilize the
30:27entire um scope of uh pixels or input
30:31data that have correlation with the
30:35target this Insight also shows that um
30:39we sometimes could reduce the scope of
30:42input data to make the model uh more
30:45efficient and also produce good
30:47prediction
30:50performance in the second case study I
30:53want to um show you another perspective
30:56on utilizing explainable AI method for
31:00understanding traffic forecasting um
31:03models so on the second key study is
31:07more is about counterfactual
31:10explanations the question we want to
31:12answer in this case is what input
31:15feature particularly contextual features
31:17contribute most to the traffic
31:20forecasting and what features with
31:22minimal change can alter the prediction
31:25so this could help us investigate the
31:28sensitivity of the model against the
31:31perb of input
31:34features and to answer this question we
31:36utilize the counterfactual explanations
31:38the logic behind counterfactual
31:40explanations is um you have a original
31:43input features and you fit them into a
31:45blackbox model you have a the original
31:49uh prediction outcome and now you define
31:52a alternative prediction that you would
31:54like the model to achieve for example in
31:57increasing the traffic speed during peak
32:00hours and once we set a Target
32:03prediction we go back to the input space
32:06and change the input features generate
32:08counter factual features to um achieve
32:11that alternative
32:13prediction so this allows us to identify
32:17the features that with minimal change
32:19that can help us alter to prediction to
32:22a desired outcome um why we utilize
32:25counterfactual explanations because it
32:27can help us identify the importance of
32:29contextual features and also it helps us
32:32identify the vulnerability of machine
32:35learning models against adversarial
32:37attacks um particularly perturbations in
32:41the input
32:44feature so I would like to share with
32:46you one interesting uh result from our
32:49study when looking at um the
32:54counterfactual predictions across
32:56different special environment
32:59particularly we studied um a segment in
33:02the Suburban Road a segment in the urban
33:05road and a segment in Highway um we
33:09realize that by altering the static
33:12contextual features for example speed
33:15limit number of length and number of
33:18point of Interest those features can
33:21help us um alter the traffic speed to
33:26the predefined traffic speed uh in this
33:29case 56 kilm per
33:31hour however it does not have much
33:35influence on altering the traffic speed
33:37in highway so this shows us um although
33:41also contrary to our private belief that
33:44contexual features in general
33:45contributes to um improving traffic
33:48forecasting accuracy um in Highway the
33:52static contal features do not help
33:55much but in contrary uh those features
33:59will have a uh considerable impact for
34:03improving or altering traffic speed in
34:05suburban and urban
34:10roads next I want to talk about
34:12robustness aspect of AI so in the third
34:16case study is um the question is
34:19motivated by the need to make AI models
34:24more reliable when deployed in practice
34:27most of the uh predictive models are
34:30trained in using observational data so
34:34and evaluated using
34:36accuracy once we deploy those models in
34:40practice because Mobility patterns
34:42continuously change so it's very common
34:45for us to observe data distribution
34:48shapes in this case the model that used
34:50to work in uh in training may not work
34:55anymore so
34:58we propose that beyond the accuracy
35:01evaluation we also should evaluate the
35:04robustness of those models uh
35:07particularly against distribution shift
35:10before deploying them in
35:12practice and to achieve this we proposed
35:15a robust benchmarking framework this
35:18framework consists of four uh components
35:22first we start with a mechanistic
35:24Mobility simulator um and bu
35:28Cal graphs um and Cal models based on
35:31those mechanistic Mobility
35:33simulator with Cal graphs we could
35:36conduct a cal interventions to generate
35:39controlled Interventional data with uh
35:42no distribution shifts and those
35:45interational data could help us evaluate
35:48the robustness of different blackbox
35:50machine learning
35:52models we also uh integrated this uh
35:57robust benchmarking framework with open
35:59Digital TN
36:01platform um which helps to provide a
36:06userfriendly interface um particularly
36:08to empower domain experts and policy
36:12makers to utilize this um robust things
36:15framework this open Digital Tain
36:17platform can also support traceability
36:21reusability inspection and quiring of
36:24specific functions and considering some
36:27of the mobility datas are sensitive data
36:30through this integration it can also
36:32support uh access control for sensitive
36:37data so with this I conclude my talk
36:41thank
36:50you thank you for the great talk
36:53[Music]
36:55from uh and we have next speaker Dr
36:59Frankin van
37:01vinsent
37:03um Dr Frankin who leads uh he leads AI
37:08solution development at Veil a
37:11Luxembourg based company specialized in
37:14environmental
37:16analytics with his expertise in data
37:19science and remote sensing he focuses on
37:22leveraging satellite data and AI to
37:24create practical Environmental Solutions
37:28Dr Van Vincent was a PhD in light biolog
37:32biological sciences and had spent the
37:36past six years applying deep learning to
37:39Earth observation data helping cities
37:42make informed decisions about
37:45Environmental Management today he will
37:48present on harnessing satellite data and
37:51AI for urban resilience showing us how
37:54this technology can help cities address
37:57climate challenges and other Urban
38:00issues Dr Van Vincent please begin your
38:12presentation I think we are
38:22muted is that
38:25better now is better hi Frankin yeah and
38:29we could see your presentation in full
38:31screen mode earlier but now we can see
38:33it with the notes so if you could go
38:35back to the
38:39[Music]
38:48other yes sorry for that no that's okay
38:52we can still see all the slides it works
38:55works like this no we can still see all
38:58the
39:03slides on the left hand
39:09side can you go to slideshow perhaps and
39:12PR and click
39:17Start or to the bottom right hand corner
39:20for the presentation
39:25mode yes the if I if it's okay with you
39:28I will do it like this that
39:31good okay yeah I mean yeah yeah yeah if
39:34you could perhaps increase the size of
39:35your screen so that we can see it a bit
39:37better thank
39:42you all
39:43right so sorry that thank you very much
39:46for the um nice
39:50introduction
39:51and uh what do you see do you see my
39:54notes as well or not
39:57and just want to make clear do you see
39:59only my PowerPoints or do you also see
40:01the word doc we see we don't see the
40:03notes we just see the slides all of the
40:05slides on the left hand side with the
40:07slide that you're presenting as
40:09well
40:11[Music]
40:12yeah um all
40:15right again thanks for that introduction
40:18um as you said I'm the head of
40:20development at R Luxemburg based startup
40:24and uh we indeed leverage AI in big
40:26Earth data mainly open source satellite
40:29data to create maps and indicators um
40:33that drive environmental analytics and
40:36that help our clients with their
40:37environment
40:38management so our clients are mainly
40:42municipalities and other local and
40:44National governments and we also have
40:46some private sector
40:48clients and in my role as Aid I'm
40:51overseeing this development of the aori
40:53the
40:54company uh that's mainly computer vision
40:56model work a lot of conclution neural
40:59networks and I've been doing this in the
41:01past few years for R but also before for
41:05different start Earth that and with that
41:08experiment experience um in the next 10
41:11minutes I would be happy to share a bit
41:14about how I think that uh cities can and
41:17then should use satellite data for
41:19promoting Urban resilience especially in
41:22the face of clate
41:24change so let me start by
41:27um speaking a little bit about satellite
41:30data so the previous speaker already
41:33showed some examples of using spatial
41:35data and indeed satellite um for Earth
41:39observation has been used
41:41since around uh the mid of the previous
41:45Century so since
41:471950s uh in the beginning mainly for
41:49meteorological and Atmospheric
41:51monitoring but since the' 70s the NASA
41:54launched a lens set program for land
41:56monitoring that includes Urban
41:58Landscapes uh Europe was a bit behind
42:01they launched the Copernicus program
42:04together with the European space agency
42:07and the
42:08eiss um and the Sentinel missions
42:11starting the
42:132010 but uh they now have a very very
42:16good openly and freely available
42:19earthstation program and the fact that
42:21it's openly available including for
42:23commercial use really make a change
42:26because well of course it made Earth
42:29station wiely available and it was an
42:31enormous boost for this Downstream
42:34sector
42:36um and in the recent decades we also
42:39seen an enormous launch of commercial
42:43satellites this is because of the
42:45technology U mainly the cube sets are
42:48the smaller sensors small satellites but
42:50also because of the cheaper Alles made
42:53it affordable for commerciales launch
42:55and actually now in the decade the
42:58number of commercial satellites
43:00outnumber the public satellites and I
43:04believe that planet alone has about 300
43:07satellites in arm so with that there is
43:10a global coverage um and there's also
43:14very high revisits very high frequency
43:17of this coverage and it means also that
43:20there is enormous volume and velocity of
43:23data the new data is uh into the
43:28hundreds of terabytes per day uh and and
43:32that of course makes the total amount of
43:34historical data available into the py
43:37skill which also means that this is a
43:39huge opportunity for machine learning
43:41and indeed now the largest challeng is
43:44no longer IM availability but it's more
43:48the ground to thing or the labels for
43:50both training and validation of this
43:52remote sing
43:53models and apart from uh this big volume
43:57and velocity there's also a big variety
44:00of different types of sensors so there's
44:02the optical
44:03imagery which consist uh the offering of
44:06lens set and the central two program and
44:09this is not just the the optical lights
44:13that we can see the visible spectrum
44:15which is the red green and blue like in
44:18your camera and also how we perceive
44:20images but this extends into the
44:22infrared and the termal
44:24wavelength um of the electromagnetic
44:26spectrum and we have now multispectral
44:28even hyperspectral data available with
44:31hundreds of Bens within the Spectrum
44:33available and then there's also active
44:35remote sensing that un like the optical
44:37sensors that use reflectance of the
44:39visible light the sun they use uh
44:43instead radio waves and that has the
44:45advantage that they can look through
44:47clouds uh during and also during the
44:51night time so this richness in different
44:55types of data and fact that this data is
44:57very scalable and also to a large extent
45:01quite unbiased provides a very good case
45:04study for monitoring Urban l
45:08u
45:11Landscapes so also this High revisits Ur
45:15that can have consistent monitoring and
45:17check changes over time quite precise um
45:21which allows urban planners for example
45:23to monitor the base of uh Urban SC of
45:28development over timey high grow zones
45:32Etc
45:33s capture radar which can capture also
45:37true clouds has been widely used in
45:42flood assessments flood Assessments in
45:44cities um specifically for PL floods
45:48where flouds would hinder the other
45:52remote sensing means and uh we can also
45:56for example monitor sealed surfaces
45:59which then can be used as an input for
46:00flop
46:02models disaster Rel that light is also
46:05used to understand what are the
46:07vulnerable zones within cities or even
46:10mapping available
46:12Roots um to the the vable
46:17zones um and one of our main products at
46:21Rio that's another case study that is
46:24our green monitor where we monitor for
46:27urban green spaces and for example the
46:30growth and health of
46:32vegetation which helps in is in urban
46:35green
46:36management so before this and still the
46:40case for example in Paris uh AR they are
46:44having an iPad in the hand going to all
46:46of the different city trees taking notes
46:48on the health of and the growth of each
46:51individual
46:52tree and now we can give the same amount
46:55of data
46:57on a very very frequent basis and set
47:00doing door to door in this Cas Tre to
47:02Tre cus we which normally only happen
47:06once a year if already once a year we
47:08can do this at M scale and and multi
47:12times a year and then of course also the
47:15global cence elim biases uh that are
47:19associated with groundbased
47:21sensors and allows to compare different
47:25municipalities across the GL for
47:28example uh we can also monitor things
47:31like green roofs existing or potential
47:33for green roofs
47:36um we can relate this to air quality for
47:40example and then of course also um we
47:43can look into the urban heat island
47:46effect so we all know the urban heat is
47:49effect effect that course cities are
47:52warmer than the surrounding rural areas
47:56and this has different all this is the
47:58fact that cities Asel
48:02concrete buildings they absorb water and
48:05reflect heat and uh they have a high
48:07ternal storage capacity so reflect
48:11is throughout the day and throughout end
48:15of the night then there also the fact
48:18that within cities there is less
48:20vegetation that has a natural cooling
48:23effect from fading and transpiration and
48:26then there's some androgenic heat
48:28sources uh like air conditioning but
48:31also buildings industrial
48:33facilities vles even that contribute to
48:36warming
48:38cities uh and like I said Urban is
48:41particularly problem at night because of
48:44these cement asals absorb heat and then
48:48slowly release their heat when the Sun
48:51goes down this has that to um seriously
48:56health related issues um such as
49:00repository difficulties extion but even
49:03in severe cases
49:05death um and in fact in a recent study
49:09uh in Europe uh it was shown that
49:13202022 I think um 61,000 people in
49:18Europe um death were H by heat related
49:25issues there's there's also a socio
49:27economic aspect to this story because
49:29forign neighborhoods are usually more
49:32exposed to um heat
49:35exposure um because lack of air
49:38conditioning would also often because
49:40there is less Green in these
49:43neighborhoods and
49:45um uh which the climate change we are
49:49expecting to have increased frequency
49:51and also intensity of heat whs so this
49:53is a major concern to cities um and one
49:58obvious aspect that they're thinking
50:00about as a solution is Greening so for
50:03example Paris is planning to plant trees
50:06and half 50% canop be cover by the end
50:09of 2030 and then satellite data can
50:12definitely help in assessing where to
50:15plant so monitor this
50:19vegetation so at R we provide different
50:24layers maps of urban green spaces is uh
50:28is individual trees Ming individual
50:30trees but also going into not only
50:32public parks but also private Gardens
50:34and green WS this is data that's not
50:36always easy to obtain for municipalities
50:39and then we can track as such the growth
50:41and the health of this
50:43vegetation um so that we can also
50:47correlate to temperature we can make
50:51heat Maps where we can capture why is a
50:55day every day the day and night
50:57temperature of at the 10 meter
50:59resolution so that means you can
51:01identify neighborhoods that are even
51:04streets that are more at
51:07risk and then related to the Green in
51:10that neighborhood so indeed from a study
51:13it shows that in cities the hotos are us
51:17having less than 6% vegetation cover and
51:19then the cool spots has over 70%
51:22vegetation
51:24cover so city of ual can use this
51:27information uh to identify the areas
51:30that are most at risk and then they can
51:32prioritize the planning of trees uh
51:35other green infrastructure in these
51:37areas to help cool down the urban
51:40environment uh optimize also building
51:43materials and designing the city to
51:45reflect heat uhing the urban layouts you
51:49can think about anding Ting
51:53corridors uh to a certain extend also
51:57the historical perspective is important
52:01in in this and is a key advantage of
52:03satellite data because it allows us to
52:05look back in time uh and and That's
52:09essential for projects focus on
52:11long-term analysis for example for
52:13indentifying Trend and then urban
52:15planners can analyze historical data to
52:19trace development of urban heat Islands
52:21over the years and Implement more
52:24effective strategies
52:26um and and even investigate climate
52:29change
52:31scenarios and and compare that to
52:33current
52:36conditions um Dr Van vinsen could you
52:40wrap up in two minutes yes okay so I
52:44just want to wrap up showing a few more
52:48examples of the maps that we uh
52:51create and speak maybe also a little
52:56about about the downside so the downside
52:59we mainly use open source uh satellite
53:03data and downside there is usually the
53:05resolution uh in general withe sensing
53:09is a resolution issue
53:11potentially um so we try to tackle this
53:16by things like super resolution un
53:18mixing at the pixel level and of course
53:22you can combine it with nc2 sensors so
53:26just to wrap up I think there's a lot of
53:28opportunities there's a lot of new
53:31technologies coming on like hos spectral
53:33data that allows us to even better un
53:36miix at cix level different materials
53:39like different types of gr materials or
53:41even different types of so I think uh
53:44I'm very excited about these coming
53:48opportunities and I guess with that I'd
53:50like to thank you very much for for this
53:53opportunity to
53:54here thank you
53:58thank you for your
54:11presentation okay uh our next speaker is
54:14Dr Sin koseki who hosts the UNESCO chair
54:18in urban landscape at the University of
54:21Montreal he will present the LA of
54:23cities in responsible AI governance his
54:27presentation it has been recorded So I
54:31let's turn to his presentation Ki I'm an
54:33assistant professor at the University of
54:35Montreal and I'm a holder of the UNESCO
54:37chair in urban landscape I'm also a
54:39member of Mila the Quebec Institute for
54:42AI and uh my presentation today
54:45unfortunately I cannot be there with you
54:46I'm currently sleeping or on a plane so
54:50thank you very much EK for having me for
54:52inviting me to this event um I will talk
54:55about some of the work that we've been
54:56doing at the UNESCO chair and at Mila
54:58for the past few years uh focusing on uh
55:02what are the risks of AI in cities and
55:05how we can harness and uh address those
55:08risks with better governance but also
55:10with a more inclusive development of the
55:13algorithms so um basically this is
55:17something that we've been working on
55:18since 2017 uh before I actually ad
55:21joined the chair I was a researcher PhD
55:24student and then postdoc and different
55:26universities in Switzerland and also in
55:29different uh countries in Europe and in
55:31Asia my name is shin koseki I'm an
55:33assistant professor at the University of
55:35Montreal and I'm a holder work that
55:36we've been doing at the unco chair and
55:38at Mila for the past few years uh
55:41focusing on uh what are the risks of AI
55:44in cities and how we can uh countries in
55:47Europe and basically this is something
55:49that we've been working on since 2017 uh
55:52before I actually joined a chair I was a
55:55researcher PG student and then postto in
55:58different universities in Switzerland
56:00and also in different uh countries in
56:02Europe and in Asia and overall
56:06um
56:08oh hi enyong sorry you stopped sharing
56:11but the video was running
56:13smoothly ah
56:18okay sorry about
56:23this perect
56:30cor are the impact of AI in cities um
56:34and how to better govern those impacts
56:37as well as how to um you know create
56:40algorithms that might be more
56:42responsible more inclusive then that
56:44might hold less biases towards uh the
56:48people that are you know uh the most
56:51vulnerable to this new
56:53technology the objective of the
56:55presentation today today will be to look
56:57at how we can apprehend the deployment
56:59of AI in cities through responsible
57:01governance and development and I want to
57:04emphasize on this word apprehend which
57:07can mean both understand fear and
57:11control the structure of the
57:13presentation is very simple I will talk
57:15about what is AI and how we address it
57:18how we govern it and how we develop it
57:21so you know the first thing to know is
57:24why do we need to do this work as you
57:25know AI is a very disruptive technology
57:28a lot of people have been raising red
57:29flags about its potential impacts on
57:32humankind and the environment uh there's
57:35been a call for a um moratorium to stop
57:39the development of certain AI that come
57:42that came both from the industry but
57:44also from researchers and from uh
57:47different uh different um influential
57:49people around the world including uh or
57:52dear elen
57:53musk uh AI is first a social project and
57:57my first relationship to AI was this
57:59character played by Al schwazer in the
58:02movie Terminator so most of you might
58:04have seen this movie of course back then
58:07I didn't realize uh the character who
58:09was a robot um was trying to portray
58:13this kind of idealic um vision of what
58:16AI could become uh in uh the near future
58:19but also the most uh you know non idic
58:22version of what it can become uh AI is
58:25also an expensive technology it's been
58:27around for quite a few years now the
58:29concept itself has evolved uh from its
58:32first use in the 1950s all the way to
58:35today with uh different kind of like um
58:38Milestones such as when AI first won
58:41against the Grand Master red chest or
58:43the launch of chat GPT and other
58:45generative AI
58:47algorithms AI can also be understood as
58:50a multi-layered concept so basically how
58:52we look at AI can be you know taken
58:55taken from the perspective of the data
58:57that we are using to train the algorithm
58:59but also the algorithms themselves how
59:02they are being embedded in what we call
59:04artificial artificial intelligence
59:06systems and those systems then included
59:09in applications that we use on a daily
59:11basis through a series of utilization
59:13that we make uh
59:17ourself AI is also an industry of
59:20knowledge uh with a lot of major actors
59:23that have been around for quite a few
59:25years that have kind of like
59:26progressively or radically changed their
59:30business model to fit this new
59:32technology such as Google MAA Amazon uh
59:36Microsoft and so on also some new
59:38players such as open Ai and uh no
59:42research and academic actors such as the
59:44tring institute in London Mila in
59:46Montreal the F in Brussels the MIT cell
59:50in Boston and also a series of
59:52governmental or parnal agencies that
59:55both kind of like look at what are the
59:57impacts of AI Society but also how to
59:59promote it and how to um facilitate its
1:00:04use by both the people and by the
1:00:07industry and finally AI can be
1:00:09understood as a life cycle and this is
1:00:11how we actually address it so basically
1:00:12looking at how the technology itself is
1:00:14being developed into different faces
1:00:17starting with framing for example then
1:00:19design implementation deployment and
1:00:21maintenance and how every um Step of
1:00:25this process kind of introduces new new
1:00:28biases and new risk to the technology
1:00:30and how we can address those biases and
1:00:33those
1:00:34risks as I said one of the major issues
1:00:37with AI is the inent biases if we try to
1:00:40generate images of toys in the US versus
1:00:42to in Iraq using generative AI today
1:00:46what we see is that uh those models are
1:00:49aligned with the biases that are
1:00:51inherent to the data and that have might
1:00:53have been introduced later on in the uh
1:00:56life cycle process where uh you know we
1:01:00have these kind of cultural deformation
1:01:03associating countries with different uh
1:01:06social and political
1:01:08contexts AI is also Insidious and
1:01:11omnipresent it's basically everywhere
1:01:13and it's getting you know in more places
1:01:17uh it's in the military it's in the uh
1:01:20use of prime or Netflix it's on our
1:01:22smart smartphone it's in our research of
1:01:25course uh uh in the transportation that
1:01:27we use and even in toys and uh you know
1:01:30language
1:01:32[Music]
1:01:34correctors and one of the big issue with
1:01:36this is that AI is you know deeply
1:01:38unregulated so basically um it is one of
1:01:42the most unregulated technology today uh
1:01:45with only one legal framework being uh
1:01:48in place right now which is the EU AI
1:01:50act which still has to be kind of like
1:01:53you know fully implemented otherwise
1:01:55what we see that countries are really
1:01:57reluctant in engaging in regulating AI
1:02:00the most recent example was of course
1:02:02the uh veto by Governor G newon of
1:02:06California on the California AI act
1:02:09which basically killed the project just
1:02:12after it was uh voted by the uh
1:02:17representatives and this is a problem or
1:02:20these are problems that we've been
1:02:21trying to address in a series of tracks
1:02:23at the applied marchine learning days
1:02:25since 18 uh incl you know working with
1:02:28for example EK on this uh and um you
1:02:31know we've we've tried to gather a bunch
1:02:33of different people from different
1:02:35perspective and different expertise um
1:02:37researchers
1:02:39thinkers politicians AI makers uh and so
1:02:44on to kind of understand what are the
1:02:46actual implication of AI for cities and
1:02:49basically so by the way the next uh
1:02:52edition of this is in February of next
1:02:55year and we hope hope for those who live
1:02:56in Switzerland or can come to loan uh
1:02:59that you'll be
1:03:01there and um I'm going to stain the call
1:03:05yes uh back in 2021 so Mila was
1:03:08approached by un Habitat to create this
1:03:11framework which is called uh AI in
1:03:13cities risk applications and governance
1:03:15which is actually derived from the
1:03:17series of tracks that we've been
1:03:19organizing um um you know un habitat is
1:03:23the UN Agency for cities and human
1:03:25settlement
1:03:26and they basically oversee issues
1:03:28relating to Human Rights sanitation
1:03:32Transportation uh equ diversity in
1:03:34cities and human settlements across the
1:03:36world and so we came up with this kind
1:03:39of like very um accessible simple
1:03:42framework that kind of address the major
1:03:45issues on using AI within andb cities
1:03:50nowaday uh across the world right and uh
1:03:53this kind of booklet divides into three
1:03:55Parts one on applications one on risk
1:03:58framework and one on the strategies that
1:04:00cities can put in place to kind of
1:04:02reduce the risks introduced by the
1:04:05technology so the on the application
1:04:07part we kind of took a very generic um
1:04:12sectorial division of cities uh looking
1:04:15at energy Mobility Public Safety water
1:04:17and waste management urban planning and
1:04:19City governance and what we did is that
1:04:21we tried to for each sector uh
1:04:24illustrate series of ongoing or you know
1:04:27upcoming applications and um also
1:04:31provide some context especially for
1:04:33those application that we deem the most
1:04:35problematic such as for example uh
1:04:37predictive
1:04:39policing then the risk framework as I
1:04:41mentioned earlier try to identify what
1:04:44are the risks that every step of the a
1:04:46life cycle kind of like you know holds
1:04:49in it holds into and how we can address
1:04:52uh those risks through different
1:04:54practices and by asking ourselves
1:04:56different
1:04:57questions and finally the urban AI
1:05:00strategy part actually exposes a series
1:05:04of Concepts and principles and ways of
1:05:07doing that cities could and should
1:05:09embrace in order to reduce the risk of
1:05:11AI on their population and on through
1:05:13the environment such as starting from
1:05:15the local context prioritizing capacity
1:05:18building and fostering cross sectoral
1:05:21collaboration and this is actually what
1:05:23we're doing also at the chair and at
1:05:25Mila when we try to develop our own AI
1:05:28algorithm for cities so we take um an
1:05:31approach that is both based on the AI
1:05:34and Ci's white paper but also on the
1:05:37Montreal declaration for responsible AI
1:05:39which was created here at the University
1:05:41of Montreal and we apply it by a series
1:05:44of um strategies such as again um you
1:05:47know part SP conception so working with
1:05:49the people with citizens to Define um
1:05:52you know the framing design
1:05:53implementation deployment and
1:05:55maintenance of algorithms by working in
1:05:58intersectoral manner so working across
1:06:01uh Computer Sciences but also with
1:06:02social sciences philosophy art and so on
1:06:06and by focusing on underprivileged
1:06:09individuals who are more at risks of uh
1:06:11dealing with issues uh related to
1:06:15Ai and this brings us to uh the last
1:06:18part of my presentation which is uh
1:06:21where I'm going to present very briefly
1:06:23two projects that we've been working on
1:06:25for the past last uh year or two the
1:06:27first one called evalene uh its goal was
1:06:30to measure the um quality of public
1:06:32space using a parts in local AI so you
1:06:35know back in 2022 we were you know right
1:06:38during or after the pandemic and what
1:06:41Urban researchers saw is that you know
1:06:44people who come from underprivileged
1:06:47groups or marginalized individuals uh
1:06:49especially those coming from
1:06:51underrepresented uh ethnic cultural and
1:06:53religious minorities also lgbtq people
1:06:56people living with disabilities and
1:06:58women had a much lower access to public
1:07:01space at a moment where it was the most
1:07:03so during the uh you know Global
1:07:05lockdown um and one of the reason for
1:07:08this is because cities are inherently
1:07:11structurally un you know unequally
1:07:13accessible to people and tend to favor
1:07:16uh dominant groups so to address this
1:07:20what we did is that we partnered up with
1:07:21a bunch of different local organization
1:07:23here in Montreal working with with
1:07:26underrepresented individuals and we're
1:07:28trying to create a series of algorithms
1:07:31that trying to that can understand what
1:07:33makes um public space good or livable or
1:07:38uh playful or uh beautiful from the
1:07:42perspective of those different groups
1:07:44and the idea here is that we I mean it's
1:07:46a project we still continue working on
1:07:48but the idea here is to eventually be
1:07:50able to map uh those qualities across
1:07:53very large um Urban territories such as
1:07:56Montreal Metropolitan region but also
1:07:59perhaps test it in other cultural
1:08:01contexts and see which spaces are the
1:08:04most let's say um problematic from the
1:08:08perspective of those underrepresented
1:08:11groups a second project that we've been
1:08:13working on here uh the development of
1:08:16participatory and local AI that is
1:08:18capable of aligning Jive AI images of
1:08:22public space for more inclusion so again
1:08:25what we we saw from our research is that
1:08:28a lot of generative AI tends to
1:08:30integrate biases social and cultural
1:08:32biases within its data that is then
1:08:34reproduced and Amplified in the images
1:08:38of public space it creates right uh this
1:08:40is a problem that has been long studied
1:08:43and what we do here is that we again we
1:08:45work we partner up with local
1:08:47organizations here in Montreal working
1:08:49with people from underrepresented
1:08:51communities and with them we try to
1:08:54create this series of algorithm
1:08:56alignment algorithm that can correct um
1:09:00existing uh open source or openly
1:09:04accessible um uh
1:09:06algorithms uh to create images in order
1:09:10to see if we can realign or redefine
1:09:14their
1:09:15output to fit those um people's
1:09:20aspiration a bit closer right so if you
1:09:23look at the bottom uh we have a series
1:09:25of images one is prior to fine tuning
1:09:27one is post fine tuning and a third one
1:09:29is post full fine-tuning and what
1:09:32happens here is that our algorithms
1:09:35integrates the uh feedback from those
1:09:38underrepresented groups u based on the
1:09:41uh series of Workshop that we've done
1:09:43with them and uh corrects or um you
1:09:47know steers the um production of images
1:09:51towards what those groups might have
1:09:54wanted or expected so it's in a sense
1:09:57creating creating this kind of like
1:09:58positive bias towards under represented
1:10:02groups and to conclude uh a few you know
1:10:06basic principles on how to create an AI
1:10:08and cities for good well first conceive
1:10:11as regies that are adapted to your
1:10:13context so work locally especially with
1:10:15local population local organizations
1:10:17start small and try to build things from
1:10:20scratch develop an inclusive network
1:10:23with citizens and interstructural actors
1:10:25especially bridging between let's say um
1:10:28engineering and social sciences but also
1:10:31with industry and uh you know um civil
1:10:35society and so on build capacities on a
1:10:38across Urban context so really try to
1:10:42promote a knowledge on AI uh again um in
1:10:47citizens in researchers in students in
1:10:50colleagues in uh City professionals and
1:10:53and politicians and finally deploy air
1:10:56systems created with marginalized group
1:10:58and this is something that we're
1:10:59becoming pretty good at so really
1:11:02working with people that come from
1:11:04underprivileged uh um context or people
1:11:08who belong to a marginalized uh group
1:11:11actually enhances the chance that the AI
1:11:13that we create does not represent a risk
1:11:16for those
1:11:17people so on this thank you very much I
1:11:20hope you enjoy this presentation uh
1:11:23unfortunately I'm not here but I'm
1:11:24always reachable by email so please feel
1:11:26free to uh write to me thank
1:11:33you thank you for his talk
1:11:37and so we had us multiple presentations
1:11:43for the topic and now we have a
1:11:46discussion
1:11:48session
1:11:51so uh we can explore the implications of
1:11:54today's presentation and dig deeper into
1:11:57the challenges and opportunities of AI
1:11:59in Urban Development I'm delighted to
1:12:02introduce our two disc curent who brings
1:12:06valuable perspectives to this
1:12:08conversation first we have kolina zba
1:12:12Kik research associate at lier who
1:12:16brings expertise in Remos sensing and AI
1:12:18applications in urban Environmental
1:12:21Studies her work on integration new
1:12:24technology for analyzing
1:12:26cityscape um changes and also Urban
1:12:30forest forest forestry mon monitoring
1:12:34offers valuable insights into practical
1:12:37imp implementation of
1:12:39AI and following her we'll hear from Dr
1:12:44Janu assistant professor at the
1:12:47University of Soul whose work Bridge
1:12:49datadriven urban planning with a value
1:12:53value based Urban Design her research on
1:12:56sustainable Urban processes with
1:12:59empirical methods in special network
1:13:02analysis will help us understand the
1:13:04human centered aspects of AI
1:13:07implementation in cities so Dr jba Kik
1:13:11please share your thoughts on today's
1:13:14presentation yes hello uh thank you for
1:13:17your today presentation this was really
1:13:20really insightful and I as as I am
1:13:23landscape architect and scientist
1:13:25working on housing and Greenery H I
1:13:28would like to add a brief maybe more
1:13:31General uh comments to the discussion if
1:13:34uh we are talking about the sustainable
1:13:36urbanization for good uh because uh I
1:13:40think it can mean the different things
1:13:43depending on the specific Urban
1:13:45context As Cities face different
1:13:49challenges of course based on this size
1:13:53geography um economic development and
1:13:57environment and uh
1:14:00other and while the main goal uh is to
1:14:04create the cities that are resilient and
1:14:08inclusive with the access to affordable
1:14:10housing for all and uh of course
1:14:13environmentally and climate friendly uh
1:14:16the way to achieve this depend on the
1:14:19unique uh Urban settings uh and as
1:14:23presented today efforts are being out in
1:14:25many fields including the remote sensing
1:14:28to provide the the specific data and and
1:14:32products which is great and very very
1:14:35helpful for the
1:14:37cities and what is new in the urban
1:14:39studies uh it is the integration of AI
1:14:43tools um and it still raises many
1:14:46concerns and and
1:14:48questions uh but I see uh that this
1:14:53progress and inde this Revolution we can
1:14:56say I think we can no longer stop and
1:15:00the real question now is maybe how uh do
1:15:05we guide it and which direction we
1:15:07choose to stare it and of course as long
1:15:11as we uh also mean Academia and the the
1:15:15private sector not just the political
1:15:18actors and political environment H
1:15:21because as you know the technology is
1:15:23one thing and the applications in
1:15:26everyday life is another and sustainable
1:15:30urbanization is only a smart part of the
1:15:34application on on these new
1:15:36technologies um but it allow us to see
1:15:39the possibilities and and risks that
1:15:43come with AI H I would like to share
1:15:47with you as specific housing related
1:15:50issue that has already happened in in
1:15:53the world and there was a case in the
1:15:56San Francisco where Property Owners use
1:16:01the software and AR algori to gain more
1:16:04money by increasing prices of the houses
1:16:07and plats H even if it meant that the
1:16:11some of the plats were vacant so of
1:16:14course the this use of technology is
1:16:17directly conflicting with the idea of
1:16:21sustainability uh and we need to be
1:16:23aware that AI can also be used to to
1:16:27damage the for example the property
1:16:30Market uh
1:16:31with in this case with negative impact
1:16:35uh on housing and and on
1:16:38people but the fact is that the
1:16:41potential of this technology is here uh
1:16:44we use it on a daily basis uh we
1:16:47creating new products that would be
1:16:49maybe labor intensive or maybe
1:16:52impossible with our AI
1:16:55and I believe that the awareness of the
1:16:59hazard will make us better prepared and
1:17:03uh being prepared uh will allow us to
1:17:07avoid these
1:17:08problems and this is my hope this is my
1:17:12real hope and
1:17:14U maybe if we have a time uh I would
1:17:17like to ask the one general question to
1:17:21to our pre presenters uh
1:17:25if we have a
1:17:29time yeah yeah I think we have a time
1:17:32and five minutes yeah okay so so maybe
1:17:35the quick more more general question
1:17:38because we see that the city is
1:17:40increasing become a a focal point to the
1:17:44technological innovation and the
1:17:46question arises uh where should Urban
1:17:50centers begin developing the and
1:17:53implementing this AI strategies uh is
1:17:57such more general question where you can
1:18:00refer to the specificat Thematic area of
1:18:03your interest or U
1:18:08yeah I I don't indicate who should
1:18:12answer but okay so uh any speakers wants
1:18:18to address this
1:18:23question or or kolina did you have any
1:18:26specific question to specific
1:18:31topic specific Maybe not maybe the
1:18:34another general
1:18:36question how maybe how and can and and
1:18:40should Academia and Industry and
1:18:44governments collaborate to to ensure
1:18:47that this this smart cities to REM
1:18:49remain human centered and sustainable
1:18:53driven maybe the the
1:18:56other other question thank you uh any
1:19:01speakers wants to address
1:19:05this I can uh maybe say a few things how
1:19:08we do that and uh
1:19:12feel what we see is that many of the
1:19:15municipalities that we work with um they
1:19:19do
1:19:21have they just starting with again so
1:19:24they do have to certain extent GIS
1:19:29department and they do usually want to
1:19:34have kind of based input there to that
1:19:37and work with it
1:19:39themselves uh but often or part goes
1:19:44with aning the information on Sat image
1:19:47in our case they ask that to be done
1:19:51right so we that to them it's hard for
1:19:54them so
1:19:55to do the themselves and even sometimes
1:19:57to understand that so in communicating
1:19:59with them spend some time to explain
1:20:03what are the limitations of the AI um
1:20:08and yeah I I think it will depend City
1:20:11by city we are working mainly with
1:20:13municipalities in Luxemburg um also with
1:20:17the larger municipalities in Luxenberg
1:20:19but those are not you know the cities
1:20:21that probably will have the capabilities
1:20:24themselves
1:20:25and in Luxemburg you have the good
1:20:26experience with the contact with the
1:20:29ministry or some Administration unit
1:20:31with this yeah yeah so we also work with
1:20:34the ministry um the ministry of
1:20:37environment for
1:20:39example living similar prods
1:20:43um I think for the moment it has been
1:20:46the same as with the municipality so the
1:20:49ministry definitely has a good GIS
1:20:52Department uh we deliver maps to then
1:20:55but again the I aspect is probably they
1:20:59working on it internally but they often
1:21:02ask us to do that yeah okay thanks
1:21:10Services okay thank you for sharing your
1:21:13thoughts and Dr yanin do you have any
1:21:17comment
1:21:18answers um yeah um I think to answer um
1:21:24har question um in short yes absolutely
1:21:28I think we should uh you know as
1:21:30researchers collaborate with the
1:21:32government agencies and also industry to
1:21:35make sure that the AI progress is um
1:21:39going forward in a secure safe reliable
1:21:43and inclusive um um path
1:21:47so because ultimately and the research
1:21:51we produce needs to have the support
1:21:54from policy makers and also um we want
1:21:58to you know ensure that
1:22:00it's consistent with the Practical
1:22:04applications or the development um from
1:22:08the
1:22:08industry so one example we are currently
1:22:12um working on or new project that we're
1:22:14working on at T is the AI Compass um
1:22:18which is to use AI technology for crowd
1:22:21management and for that um project um we
1:22:25have more than 20 collaborators that
1:22:27includes both government agencies and
1:22:30also uh industry Partners so the goal is
1:22:33really um closely couple the development
1:22:38in the research field with the
1:22:40application and
1:22:42deployment um in municipalities and um
1:22:46companies um I believe we can benefit
1:22:49both from this integration and we can
1:22:52better understand the real world problem
1:22:55and the and prioritize the most critical
1:22:57issues as researchers and also those uh
1:23:01municipalities and also um companies
1:23:04could benefit from our um um Advanced um
1:23:09outcome from research
1:23:11so so I think you know if um possible we
1:23:16should always advocate for a close
1:23:18collaboration between the three
1:23:22parties yeah okay great um so now can we
1:23:27move to Dr
1:23:30Janu could you share your thoughts on
1:23:33today's
1:23:34presentations yeah sure um yeah uh
1:23:38really uh I really thank to the present
1:23:42uh speakers today it's very insightful
1:23:44uh presentation so um I can say the
1:23:48three uh uh the points uh after
1:23:53listening all the this kind of uh
1:23:55presentations say uh first thing is that
1:23:59I think the AI technology uh can pursu
1:24:02of value of
1:24:04customization because AI technology can
1:24:07uh even analyze user emotion through
1:24:10text voice or facial expressions or um
1:24:15also providing appropriate responses uh
1:24:19throughout all this kind of data
1:24:22personal data let's say uh uh and um
1:24:26leveraging such this kind of AI uh
1:24:30capabilities I think it is uh possible
1:24:33that Urban environments and Designs can
1:24:37reflect the individual preferences and
1:24:41characteristics uh including uh unique
1:24:44uh personal identities and those kind of
1:24:49uh values can be achievable uh so I
1:24:54think AI
1:24:55technology uh can uh uh the contribute
1:25:00uh value of each persons uh the the
1:25:05preferences and
1:25:07characteristics and my second point is
1:25:10that uh AI technology also can pursue of
1:25:14more diversity uh as a core value for
1:25:19example uh the V varieties of mode of
1:25:23transportation can expand uh the the
1:25:28broader range of C consumers and um the
1:25:34audience and increase the overall
1:25:37satisfactions as well uh for example the
1:25:40emerging of autonomous vehicles uh or
1:25:43robot taxes uh might initially seem like
1:25:47a competition for tra traditional car
1:25:51consumers but in reality it provides a
1:25:54greater uh spectrum of choices and
1:25:57accommodating diverse needs and
1:26:01preferences uh of
1:26:04consumers uh so uh AI technology really
1:26:07help out uh to make a more diverse uh
1:26:11City or diversity can pursue the
1:26:15diversity uh as a core value um uh in
1:26:19the
1:26:22city and lastly I um can say that uh AI
1:26:28technology can be utilized as a
1:26:31negotiation device in urban
1:26:34planning um I think it's really helpful
1:26:37to create a more sustainable Equitable
1:26:40and inclusive and ethical Urban
1:26:43spaces uh as we know the public
1:26:47participation and Community engagement
1:26:50process are very essential and very
1:26:53important while we
1:26:55are U making a
1:26:59city so I think AI powered algorithms uh
1:27:04can generate uh numerous design
1:27:08Alternatives with uh in uh with um
1:27:12diverse characteristics and uh
1:27:15quantitative evaluation uh each um
1:27:19design alternative value so this
1:27:22capability can make the
1:27:24uh inclusion of broader range of
1:27:27stakeholders and audience uh in design
1:27:31process it is allowing their
1:27:33perspectives to be uh Incorporated in
1:27:37the planning uh through this Al
1:27:39algorithm analysis so I think this kind
1:27:43of approach enhance uh transparency and
1:27:47fosters collaborative uh design decision
1:27:51making uh and that lead uh to make a
1:27:54more inclusive and effective Urban
1:27:57Design results so those are the three
1:28:01key things what I uh thought about the
1:28:05um the AI technology related with the
1:28:08urban planning and
1:28:10design um and uh here is uh I have a one
1:28:15common or generation for all speakers so
1:28:20um I'm wondering about uh the which
1:28:25stage or process uh of urban planning or
1:28:29design um um this kind of AI technology
1:28:34can be most powerfully utilized and why
1:28:37because uh I think there are so many AI
1:28:41related research uh for uh Urban or for
1:28:47City but I'm still I I don't think like
1:28:52AI technology can serve everything
1:28:55not yet but uh I just wondering which uh
1:29:00state
1:29:01or process of urban planning uh uh can
1:29:06be mostly uh powerfully
1:29:09utilized uh yeah by AI
1:29:17technology thank you for your uh thank
1:29:20you for sharing your thoughts and also
1:29:22some f
1:29:24questions uh due to the time limits I
1:29:27think um yeah if you have a quick
1:29:31answers yeah then you can answer but
1:29:35otherwise uh I think we can conclude
1:29:37this
1:29:38Workshop um yeah so for the closing I'd
1:29:42like to ask Dr Jun
1:29:44Tre for remarks and then I can
1:29:49conclude uh thank you for the giving us
1:29:54and audience of valuable insight and
1:29:58presentations so I think the one uh
1:30:01answer for the professor 's question the
1:30:05the stagey they can be more valuable and
1:30:10insightful uh utilizable applic giving
1:30:14us some uh good impact I think the we we
1:30:19also think about the predictive thinking
1:30:21using the AI in the many uh indust
1:30:24domain we think about the user
1:30:26experience or user interactions using
1:30:29the like the chpt the user interaction
1:30:32interactions but the planning itself we
1:30:35are our thinking our creative thinking
1:30:38need some kind of some um the Basic
1:30:42Instinct about predictions so I think
1:30:45the the planning stage will be the very
1:30:48powerful stage for applying a uh
1:30:52applying stage it is my personal thing
1:30:55personal opinion and about the closing
1:30:59remarks uh we
1:31:01are organized this event with the many
1:31:05big support and the from the the Dr Kim
1:31:10from Liah and any thanks for um the the
1:31:15Dr Kim and the I think the we are as a
1:31:20very various kind of domain and
1:31:22expertise
1:31:24the in the in the future we need some
1:31:26kind of collaborative AI research to
1:31:29deal with the very complex Urban
1:31:32issues and we also s Institute and S
1:31:36Metropolitan government organizing some
1:31:39ongoing event uh for with the metropolis
1:31:43and also I hope to the the continue the
1:31:47webinar with the itu but anyway the the
1:31:50issue of the applying or utilizing the
1:31:53digital technology AI to planning issue
1:31:56is many thing to consider uh is the
1:31:59basic I think the basic research from
1:32:02the shin kosi is is very impressive out
1:32:07output outcomes yeah so in the future we
1:32:11have to gather together to to make some
1:32:14kind of collaborative AI research with
1:32:17together thank
1:32:19you thank you for your remarks and atic
1:32:23express our sincere thanks to our
1:32:25speakers for your insightful
1:32:29presentations and our discussions for
1:32:32discussion for the thoughtful
1:32:34contribution on the discussion um so we
1:32:38can understand better practical
1:32:40implications of our development and I
1:32:43believe today's presentation and
1:32:45discussions have highlighted both
1:32:46enormous potential and and the
1:32:48responsibility we have in implementing
1:32:51AI for Urban Development
1:32:54and I'd like to espcially thank to our
1:32:56organizing partners and itu AI for good
1:33:00team for making this Workshop possible
1:33:02and thank you all participants and
1:33:04audience who joined us today from around
1:33:07the world uh we look forward to
1:33:10continuing these important discussions
1:33:12and collaborations as we work together
1:33:15towards more sustainable and inclusive
1:33:17Urban Futures thank you all and have a
1:33:20wonderful
1:33:22day thank you by thank you thank you by
1:33:26thank
1:33:27you thank
1:33:35[Music]
1:33:38you thank you for participating in
1:33:40today's AI for good session we hope
1:33:43you've learned something new Innovative
1:33:45and engaging in today's event we now
1:33:48encourage you to continue the
1:33:49conversation on the live video wall in
1:33:51the neural network here you can ask
1:33:54questions like and comment share links
1:33:57complete the poll connect with
1:33:58interesting profiles or speak one-on-one
1:34:00using the chat and video function we
1:34:03invite you to explore the lobby try the
1:34:05smart matching quiz visit the virtual
1:34:07exhibits poster boards the ehhop and
1:34:11build your personalized AI for good
1:34:13program let's shape the future of AI for
1:34:16good the AI for good Global Summit aims
1:34:19to bring together industry leaders to
1:34:21try and harness AI to further the un's
1:34:23development goals in areas like Health
1:34:25climate and sustainability how do we
1:34:27govern Technologies if we don't yet know
1:34:30their full potential the main purpose of
1:34:32this Summit is to explore how the AI
1:34:35Solutions can help sustainable
1:34:37development AI will come to influence
1:34:40almost every aspect of our societies and
1:34:44economies AI for good I would say the
1:34:47most pressing Topic at the Summit is
1:34:50going to be how do you handle deep fakes
1:34:52misinformation
1:34:55all for good the good news is that the
1:34:58future of undetectable deep fakes can be
1:35:00avoided through the development and
1:35:02implementation of robust detection
1:35:05Technologies and international
1:35:13[Music]
1:35:16standards you know what I've really seen
1:35:18is the fact that we can really affect
1:35:20amazing change but we have to do this
1:35:22together and we have to do it in
1:35:23collaboration
1:35:24withn with organizations withs and
1:35:27that's what's making me so excited about
1:35:28what AI for good is all
1:35:30[Music]
1:35:35about onethird of humanity remains
1:35:38completely offline excluded from the AI
1:35:42Revolution
1:35:44[Music]
1:35:59[Music]
1:36:00[Applause]
1:36:03[Music]
1:36:06artificial intelligence is changing our
1:36:08world and our lives and it can
1:36:09turbocharge sustainable development we
1:36:12are the AI generation and this is our
1:36:16moment and it's our responsibility to
1:36:19write the next chapter in the great
1:36:23story of humanity and
1:36:26[Music]
1:36:30[Applause]
1:36:32[Music]
1:36:41[Music]
1:36:52Technology
1:36:53[Music]
1:37:22for e