How can AI transform the future of smart cities to create a more efficient, sustainable, and inclusive urban environment for citizens? Antibes, a city on the French Riviera, is leading the way in its AI-driven innovation to enhance public services and operational efficiency. From automating budget alignment with sustainability goals to running on-premise AI for data privacy, Antibes is redefining digital transformation in the public sector. Tune in as experts explore how technology is driving ethical, citizen-focused innovation across government services. Featured experts Patrick Duverger, CIO and CTO, City of Antibes Michael Bradshaw, SVP & Global Practice Leader, Applications, Data, and AI, Kyndryl
How can AI transform the future of smart cities to create a more efficient, sustainable, and inclusive urban environment for citizens? Antibes, a city on the French Riviera, is leading the way in its AI-driven innovation to enhance public services and operational efficiency.
From automating budget alignment with sustainability goals to running on-premise AI for data privacy, Antibes is redefining digital transformation in the public sector. Tune in as experts explore how technology is driving ethical, citizen-focused innovation across government services.
Featured experts
Patrick Duverger, CIO and CTO, City of Antibes
Michael Bradshaw, SVP & Global Practice Leader, Applications, Data, and AI, Kyndryl
Tom Rourke 00:02
Hello, and welcome to The Progress Report. I'm Tom Rourke, Vice President for Design, Insights and Innovation here at Kyndryl, and your host for today's podcast. I'm delighted to be joined today by two highly experienced practitioners in the area of AI. Patrick Duverger, who is the CIO and CTO for the City of Antibes, and my colleague, Michael Bradshaw, who is SVP and Global Practice Leader for Applications, Data, AI here at Kyndryl, but also himself, a highly experienced CIO. Patrick, Michael, you're both very welcome to The Progress Report.
Michael Bradshaw 00:33
Thank you.
Patrick Duverger 00:33
Thank you.
Tom Rourke 00:34
So Patrick, maybe I could start with you, and perhaps you could just tell us a little bit about the City of Antibes for those of our listeners who aren't already aware, and perhaps a little bit about the mission of the city towards its citizens.
Patrick Duverger 00:45
So the City of Antibes is a mid-sized city located in the south of France, on the French Riviera. We try to promote two things: quality of life and innovation. Quality of life because there is a lot of tourism in Antibes. There is 80,000 inhabitants during winter and 250,000 inhabitants during summer. We try to promote a mix between quality of life and innovation.
Tom Rourke 01:16
I mean, if we think about the impact of IT, and we will eventually get to AI, but what does a smarter city mean for Antibes because, obviously, there are so many natural features that add to the quality of life. Where does IT contribute to the quality of life as a citizen of the city?
Patrick Duverger 01:32
IT mainly is an employment advantage for the citizen, okay? Because there are a lot of people who work in Sophia Antipolis and live in Antibes. So the first advantage is employment. Second is that when you manage a smart city, you try to make everything smooth, you try to provide services the way people can live and can get connected easily to their job. So it's a very big challenge to manage that.
Tom Rourke 02:17
Indeed, and you have recently undergone quite a number of initiatives using and applying AI to some of those processes. I mean, perhaps you may just talk about the motivation behind that, and what were you trying to achieve with that application?
Patrick Duverger 02:31
Artificial Intelligence is a very interesting topic, and we tried to leverage AI, and first we try to understand what it can do and what it cannot. Because currently, as a CIO, when you speak about AI, you speak about a conversational chat that is very general, and it's very difficult to control the data that are sent to this chat, and also to control the efficiency of the answers. So we try to think small, with the help of Kyndryl, and to implement our own AI on premise. It's a very big challenge to have your own AI model on premise. We tried to do natural language processing on top of requests for proposals, because we are managing in my department 140 RFPs, and it's very difficult to analyze all the documents. So we tried to instantiate NLP models in order to get insights from those documents to get the contact, the minimum, the maximum, the delays, the late penalties, and technical stuff, in order to, after that, start a blockchain transaction. So, it was a project we started two years ago. And the latest project, also using AI, is more recent, and we used AI to "green" the budget. Every cities have the obligation to analyze their budget regarding the Sustainable Development Goals. In input, you have a budget with 6000 lines, approximately. And in output, you need for each line to distribute them in the different SDG objectives, Sustainable Development Goals. So we try to automate it, thanks to AI, because large language models can very efficiently understand the semantics and the text. So, when you have one budget operation, there is a text describing it, and you compare it to the descriptions of each Sustainable Development Goal, and there is a similarity algorithm that will give you the percentage that fits this goal or this one or this one. So, this work is not done manually. It is automated thanks to AI. And it was a very big project, and we did not use a remote AI. We tried to have our own model implemented on our premises, in our data center. And this is a real challenge.
Tom Rourke 05:41
There really are so many things that I could pursue on this. And I think the first thing that strikes me is that point you made about, you know, the sort of noise and the buzz around conversational AI, ChatGPT, and these types of models, tends to blind us to the fact that there are actually many different forms and implementations of artificial intelligence, and also different contexts. Obviously, public sector being one. I mean, Michael, perhaps you might just comment on that in terms of, you know, reminding people that artificial intelligence there, we need to get beneath the publicity depth to understand there are a variety of models and approaches depending on the context. And maybe you could expand a little on that.
Michael Bradshaw 06:23
In the large language models, there was actually a lot of tuning that was done to really pare down the capabilities, because these large language models have all kinds of capability and the use case that Patrick was describing for the city, it was really to map to the Sustainable Development Goals and then align that to the budget process, and then come back with recommendations. So, the process of extraction of data, of understanding how to do the pruning, if you will, of that model, to then create a tiny, because typically you talk about large language models, small language models. Patrick likes to talk about the tiny language model that was implemented, because the use case for this was actually to run within their own infrastructure, to run within a browser, so and then to do that, I mean, obviously today, everybody's talking about AI, talking about 10s of 1000s of GPUs needed, and here we're talking about something that was done for the city that was running in somebody's browser, right? So again, there's a lot of hype around AI. There's a lot of terrific capability, but it's bringing that capability to bear to solve the actual problem for the entity, and the point being is bringing these services to bear for the citizenry and for those people that come to visit the city. You know, the budget being one aspect, and this is where the example is a force multiplier for the existing employees within the city government. So, one of the things the city is aware of is, "How am I more productive for the citizenry?" So, the ability to take the Sustainable Development Goals to implement this within the budget process, and not have to hire new people to do this, but to do it with the folks they had, and I imagine there's more conversations that Patrick's been having with the city CFO, in terms of what are other things that can be done to free up that manual activity that people were doing and actually get them to focus on more forward looking activity. So, more forward planning, how they can continue to evolve and develop services. You know, what I think is really interesting is how Patrick and the city have been very forward leading with technology and bringing it to its citizens. So, I don't know, Patrick, if you want to talk about maybe the use of blockchain and that more consumer type of interactions.
Patrick Duverger 08:52
We started the blockchain project in order to instantiate smart contracts. When you finish your RFP, you have a contract with the supplier, and the problem is not when everything goes okay. The problem happens when you have a late delivery, or something like this. And so 20% of those use cases take 80% of our time. So we try to automate transaction in the blockchain in order to be sure that everything can be understood by everyone, that the information can be read and seen very precisely in order to see when you have a problem, where it comes from in order to solve the 80% of time we spend in bad cases. So, we try to automate, thanks to IT, the case, the industrial cases, where we spend time stupidly, without any value added. For example, for winning the budget, as Michael said, either we hire new people, and it's not a very good idea in the administration nowadays, or we try to automate the thing. So, it would have taken two new people in the finance department and plenty of hours of meetings in order to set up this distribution in the SDGs for each budget line. So, AI is very useful in those industrial use cases. And like Michael said, the budget is the most critical piece of information of the organization, so you cannot send that to a remote AI. You need to operate, to do the treatment, on the desktop of the adjunct of the people in charge of the finance budget. So, that's why we try tiny models that can be loaded into memory of a normal desktop, and moreover, we succeeded thanks to pruning and quantization. It means reducing the model size to run the model inside the browser. So at the end, it's not the data that goes to the model. It's a model that goes to the data.
Michael Bradshaw 11:43
Back to your question, Tom. So, what I think is really incredible about this work with the city is the innovation and the importance of what Patrick just said, right? Bringing the model to the data, because there are a lot of different approaches that are being bandied about right now, and right now I think everybody's focused on the build out of infrastructure, all the GPUs, as I mentioned before. And the reality is, we can do a lot now with what entities already have. So for example, Patrick didn't have to go buy a lot of equipment in order for them to do this for the budget. And as you can see, that paradigm of tuning, pruning, making smaller, so that you can apply it to specific problems that we think is going to be very, very important for enterprises to really scale real kinds of challenges.
Tom Rourke 12:35
No, no, absolutely. And I think what struck me as you were talking, Patrick, is there is a sort of an arms race in terms of people throwing large numbers amounts of money at AI, and yet getting frustrated about the return. And what struck me more than anything else, as you spoke about this, is just how purposeful you have been as a city in thinking through the use cases you wanted to apply it to. And I think there's something in that combination of possibly not unlimited resources. You can't just throw money at the thing, but that requires you to be purposeful, but also then probably inspires another level of innovation. And I think it is an important point that organizations see that, my perception would be the key to success here is just how purposeful you are in choosing your use cases. A question I would have for you is, how aware are your citizens actually of this, of AI and the application in the city? Is it something that is highlighted and is seen as you know people are embracing? Or how does that play in terms of the public awareness of the programs?
Patrick Duverger 13:39
The innovation we do are currently very internal. It's in order to help the services that do services to citizen. So, it is internal into the city administration. But it will not stay like this, because digital transformation means that IT services will participate to the value added to the customer and to the citizen. So currently, we are working very hard in order to be more efficient internally, but it is straightforward that the consequences will be seen by the citizen. And I will take one example. Video production is a use case where you can add a lot of artificial intelligence in order to analyze the video in real time, in order to count the number of cars, etc. So, we rerouted the video stream to AI algorithms in order to detect some strange cases if you have an engine. If you have a truck staying too long time near school, you can imagine that it is not normal, and you send the police in order to see what the guys are doing with this truck near the school during one or two days. So, policemen cannot see 170 cameras every day, every hour. So, AI is a very helpful tool in order to analyze what must be seen at this time. So we have a front big screen, a big wall with a lot of screens, and the screens are highlighted thanks to AI. When AI detects something, abnormal, something wrong, something risky, then it appears on the screen, on the wall. So, this is a real use case where customers benefit directly from AI algorithms.
Tom Rourke 16:07
And Michael, you obviously work with quite a range of organizations, both public and private. I mean, what are the lessons that you feel we could all draw from Patrick and the City of Antibes experience that may apply, actually, to two completely different types of entities, but informed by by their experience.
Michael Bradshaw 16:26
What's as important, if not probably even more important, are the people that whether they are finance, whether it's HR, whether it's legal, whether it's operations, all those different folks need to be involved in this so that we understand, you know, what's required on them to do this. I mean, Patrick mentioned an example with the police, right? Understanding what their view of the potential challenges are, how to bring that technology to bear to solve those problems, and then what that means for them, how they deal with it, how they respond. So a lot of, for example, we've focused on a lot of our own research, going to customers, both through our Readiness Report that we released last fall, and then an AI-focused series of reports earlier this year really focused on that people readiness. Because the introduction of these capabilities, it also will change some of the roles of what folks are doing. So as I mentioned earlier, being able to have people that are unfortunately having to do a lot of manual activities today, just based on the systems and the processes as they're defined, and the more you're able to take that manual work away and let them focus on more value additive types of activities, whether that's strategy of where should we be thinking about how to invest for the next set of services based on feedback we're getting from the citizens, in the case of a city, or for another company that's a service or a product company, what's the next area of priority and focus for them in terms of bettering their product, bettering their cost structure, and how they're doing things? So again, I think that the broader set of things that we're getting aside from the technologies are, how are we addressing this with our employees? Are we being very open and transparent with our employees as we're looking at these capabilities? How are we prioritizing these problems so that we can actually scale what we do? There's a lot that's out there in terms of investment that folks have made in AI yet little return that they've seen. And then the other, I'll say, paradox that we have here is, and I'll use autonomous vehicles as a great example. I mean, we have traffic statistics today that tell us how many automobile accidents there are, how many fatalities there are with humans at the wheel. Yet, when we start talking about autonomous driving, somehow our tolerance for any defect goes to zero, right? So the expectation is autonomous driving should be perfect, whereas that's not even our expectation when humans are behind the wheel. So, I think one of the challenges is we have to eliminate our own, let's call it human bias, and we have to have definitions for what are acceptable levels of performance. There's a lot of these different aspects that are coming together that, frankly, I'll say companies haven't been that adept at. I mean, if you look at what we've done over the last three plus decades, with all the business transformation efforts that have been done, they've predominantly been focused on from a technical aspect. They've not really brought that organizational change management and governance structures to bear. And I think in the city's case, again, what Patrick and team have been able to do, partnering with the stakeholders and working this together. This wasn't just Patrick and us and others going off and working on this, from a technology standpoint, it was integrated with the business team that needed to execute this. So, that's the other piece that I think is critically important here, is making sure, even more so now than in the past, is making sure those connections are very, very close, very tight and very coordinated for these efforts to be successful and and that's where we're focused, Tom, as Kyndryl is making sure that we're bringing that forward into the conversation, so that we're not just talking about the technology, but we're leading with the organization. We're leading with the governance conversations. Because if you don't have those, then invariably, you are highly likely to fail, regardless of how great the technology is.
Tom Rourke 20:52
Absolutely and actually, I could see Patrick as he listened to you. I'd really like to invite you to expand on that, because the question I had for you was just what did it mean for you as an experienced CIO in terms of the skills you to bring to bear? And you know, what were those key success factors?
Patrick Duverger 20:53
I totally agree with what Michael said. Lots of organization and ours also lack an AI strategy. So, when you need to start, end user doesn't know what can be done and what cannot. So that's why we started by a bottom-up approach. We started from technology to the use case. You cannot end with only technology. It needs to fit a use case, and that's why we partnered with the finance department and with the CFO, who was elected the first CFO of the year in France, thanks to this project. So, providing technology is necessary to avoid people to be connected to the reality, because with AI, lots of end users will try to to imagine that it can do anything. It cannot. It can automate things very well, but it cannot do anything. And as Michael said, we imagined that it will be perfect. Partnering with the end user is critical, but starting with the end user needs. If they don't know what AI can do and cannot do, it's very difficult also. That's why the first run helps us to elaborate an AI strategy in the organization.
Michael Bradshaw 22:47
That's a very, very key point, is getting it in the hands of folks so they can really start to understand it and then elaborate a strategy. Because oftentimes we think the strategy comes first, and then you go execute the strategy. And this is a case where you don't even really understand what the strategy could be, so you have to do a lot of education. And it's not just, again, technology. It's also on the business side.
Tom Rourke 23:09
The importance, Michael, of us having just clarity about the quality of the underlying data that we're going to apply these technologies to, and maybe the these technologies not being as tolerant to some of the quality issues that we've all endured over the years. It's a thing we've always needed to solve, but we now really need to solve it. And I guess, Patrick, if I could ask you about the, you know, what challenges did it give rise to in terms of the quality of your own data, particularly as you think about how you might expand this?
Patrick Duverger 23:37
For several years, everybody thinks that data is crucial. It's one thing to think that, and another thing to clean your own data, because people are working every day and there are no rules to set data. Not in a good way inside the software. So, it's very difficult to achieve a real data set that is clean, because currently we are just giving data to AI in order to have a result, but tomorrow, we will use the data of the organization to fine tune the model so that it will be more efficient, more specialized, and faster. So, quality of data is crucial. We had to clean manually with also some scripts, the data, the financial data of the city. And there are some use cases that you cannot avoid. When we speak about a school, we don't say "this school". We say the name of of the school. And the name of the school, the AI will not understand that it is a school. So, it is not a problem of crappy data. It's a problem of meaning. So, it's the way we use IT, the way we use software, will change. We will we need to be more explicit. So, I have no real way to solve the problem. The only way is to be very aware of the need, the crucial need, of cleaning data for the whole organization in order to do BI, business intelligence, and see interesting things in the dashboards in order to feed AI, in order to fine tune AI in the future. Data is a crucial point of any organization.
Tom Rourke 25:50
I just, as we kind of come to the second half or the final phase, and this being The Progress Report, I'd like to ask both of you to just look forward a little and, you know, in the context and based on the experience you've had so far, Patrick, what does progress look like for you as you look forward? You've had tremendous experience in learning some real lessons. But as you look forward to the possibilities for the application of AI in Antibes, what does progress look like for you over the next two to five years?
Patrick Duverger 26:21
Okay, the acquisition process is very difficult and complex in France, so we'll try to use AI to help people doing the acquisition. So, short and feasible step next year, then we will try to fine tune AI according to our expertise, like juridical low expertise and administrative complexity, we want to have our own AI that is able to answer to people needs.
Tom Rourke 27:03
Thank you, Patrick. And Michael, if I can turn to you as both somebody who has been a CIO for much of your career and is now advising and assisting other CIOs, particularly the application of AI. What does progress look like for you over the next number of years?
Michael Bradshaw 27:17
It really is going to come down to enterprise architecture elements, how we assess the environments that our customers are currently in, and how we work with them to build what is that journey for them into the future, and then how do we address and help them modernize their architecture without saying you have to clean sheet of paper everything before you can even start to do that. So, I think it's technically what we would call this, this is going to be the art of the brownfield transformation, right? And how we go from where we are to increasing capabilities and delivering business value along the way. And I think we're just on the cusp of seeing how this technology is going to enable us to do that. A lot of promising things that we're doing internally, we're working with a few customers, and I will be excited to come to Sophia Antipolis and help the team work directly with Patrick and the city to help surface some of these.
Tom Rourke 28:14
Patrick, Michael, thank you so much for a fascinating discussion today. I'm sure our listeners are going to be finding it very informative. My only personal regret is we didn't get to record it in person, in Antibes.
Michael Bradshaw 28:24
Great to be here, Tom, thank you.
Patrick Duverger 28:25
Thank you.
Tom Rourke 28:30
So, to our listeners. If you have enjoyed today's discussion, please be sure to share it with your colleagues and friends, and do be sure to subscribe for future recordings. Thank you for listening to The Progress Report.