Building Sovereign and Responsible AI Beyond Proof of Concepts

Okay. Sounds good. Okay. Well, this session will be all around that. So if we can have the next slide. So what we want to talk to you today is that there are so many different AI projects and AI pilots happening in the world. And a pilot is the same as a proof of concept. It’s an idea that you’re testing. And it’s a concept that you’re testing. And it’s a concept that you’re testing. And it’s a concept that you’re testing. to see if that idea is something that you can put into implementation later on. And I was looking at the stat of how many AI pilots are in the world, and that was very difficult to quantify.

But what I did find was that only 30 % of all the AI projects actually go into production. So what we’re finding in the world is that we have lots of different AI ideas, but really a difficulty in translating that into something real. And the point of this session and what I think is the point of the whole AI summit was that one of those reasons is because we don’t have trust. So if we can have the next slide. So if we think about trust, that could be an organization’s trust that the AI will work. It can be trust in us as individuals around how our data will be shared, the outputs that it will give us.

It could be trust in terms of the impacts that it will have on people and people’s lives. It could be trust in terms of jobs and how that will work. And with that, what we’re seeing is a lot of these AI projects are failing to consider that. And I don’t know if you’re familiar with the OECD AI Observatory, but they do a monitor where they essentially monitor all of the harms and all of the AI incidents around the world. And you can see that it’s been growing exponentially. In 2025 of December only, there were 600 different incidents in the world. So those are 600 different times that people were harmed or that there was some kind of AI hazard that was created through a pilot.

If we can have the next slide. It’s just to zoom in, so this is a little bit difficult for you to read now. But in that harms monitor, you can click on any of them and learn more about them. So some that I found, the first one is in Romania. AI was being used to clone people’s voices and then scam their voices. By making them think that they were in distress. As well, there was an example, I believe it was in Cairo. So there was a book fair, and a lot of the books there were actually produced using an equivalent of chat GBTs, using generative AI. But there was no humans included in that project, so the books were printed with the prompts and the AI instructions still in them.

So that created a lot of issue of creativity, and are these books generated by AI? Are they what we’re looking for? Is that what we thought we were buying? And then there’s several other examples happening all around the world where this is happening with facial recognition, for example. So using that at borders, and all of a sudden that might not work equal between different types of people. And all of these really build towards people losing trust in AI and being fearful of using it. So these are some examples, and we’ll kind of go into next what we can do about that. So. Next we’re going to look at. why do these proof of concepts fail and how do we shift from just experimenting to actually having impact.

So I can have the next slide. So I put here six ideas of what we’re seeing with the customers we work on is why proof of concepts are not working. The first one is between adoption and impact. So a lot of times we’ll have organizations that are working on AI and they’ve just thought about producing something but they haven’t actually thought about how will people use it. Will it have the goal that you’re hoping it to have? Or say, for example, I’m using a legal tool. Will it actually serve the purpose that I’m looking for? Will it require more work for me to actually review everything it’s doing? So there’s a gap there. The second is around governance failures.

So I’m not sure how many of you have thought about risk management. How do you identify all of the risks that are coming up? Who’s going to be accountable to solving them? That might be things like, is it treating people differently? Is it biased? It might be things around security, for example. And then there’s also a failure around misalignment. So between what you’re looking for in society, those might not be aligned. So if you’re, for example, prioritizing AI use to automate people, all of a sudden people are thinking, what about job loss? So there’s not really a link in value there, and that’s another reason. We’ve got three other challenges. The first one is sovereignty, which I think if anyone was around the summit today or this week, everybody was talking about sovereignty.

So questions around how do we maintain control? Who is responsible? If, for example, a foreign government decides to turn off that AI access, is that something we trust? Or how do we deal with that? we also have sustainability pressure so thinking about the carbon cost of using AI and lack of clarity around that and then change management is really all the people so if we’re thinking about these frontier firms where people are working with agents what does that work culture look like have we actually thought about how people use AI and have time to test it and practice with it have we thought about the relationship between people and AI and how that works as well so these are six quick concepts and if we can have the next slide is just a point to make is that when we’re considering a proof of concept we’re really just considering does it function we weren’t considering any of those other six things and if we want to scale AI we need to think about everything else so next slide so I guess the point of this session is really to think about how do we actually do that so what we have thought of is calling it AI and 4D so four dimensional the idea that you need to look at four different lenses to build trust in AI if we could have next slide and when we’re looking at that we’re thinking if you can look at all these four different lenses that’s really going to help you predict any harms or challenges that could come with the AI model and actually prevent them so that you can deploy and scale that AI there’s four dimensions that we’re looking at the first one is sovereignty so thinking about who controls it not just data but looking at all the security measures behind it where does the model come from who has access to it we’re looking at green so that’s sustainability can this scale without destroying our climate goals for example We’re looking at responsibility, so that is thinking about ethics and governance and bias and fairness and human -centered.

And then valuable, so is this project actually really going to deliver a real -world benefit to people? So next slide. This one, I think it might be difficult for us to create a poll, so what we’ll do is we’ll do it by hand instead. So if we can just go to the next slide. What I thought we could do before we give you more information of those 4D and how to apply them and break out into groups is we could just have some quick scenarios and test what your knowledge is of those themes already. So I’m going to give you an example, and then we’ll do a show of hands of who thinks what lens is missed here.

So this example is with a public health company. They’re using AI to read different x -rays and radiology scans. And the point of the proof of concept is to help triage different illnesses or different breaks, things that you might find in the scan, and reduce that backlog. So when they actually started modeling and rolling it out, the team realized that this required more compute than they needed. It would exceed, actually, the available power supply, so there was not going to be ability to use it consistently. And that, actually, there was a large demand on water because the GPUs needed to be cooled, and this is in a water -sensitive area. So that would be another challenge between people and the planet.

So this program failed, this hypothetical program failed, because it was financially and politically impossible to run. So who thinks that this is a problem because of sovereignty? Who thinks that this is a problem with sustainability? Yeah? who thinks that this is a problem with responsible AI and value. Yeah, I agree. So I mark this one as sustainability. I think it’s an example of the dynamics that we might have in the real world is we want to scale AI, do really great things, but actually we haven’t considered the power or the water usage that that has because we either don’t have the information or it hasn’t been something that’s been baked in the front to think about.

And we will give you some higher level into what this means and how to apply it in a moment. Okay, the next one. So we’ve got a second one. This is dealing with transport. So I think we’ve all dealt with traffic this week. We’re looking at this in this scenario here. It’s thinking about this project is to optimize traffic lights across the city and smooth congestion. But when they started implementing this project, it was only looking for average commute time. It was diverting traffic into lower income areas and pedestrian safety actually became worse. So while this met the technical triggers that it did reduce and optimize time, there was a lot of community backlash. So does someone want to tell me which one they think this is a failure of?

Sovereign and responsibility.

Yeah, we’ve got some sovereignty, we’ve got responsibility. I think this one is actually value. So here, what the ministry had thought was valuable, reduce overall time, is not what’s valuable to the people. What’s valuable to the people is that they want to have safety and walking. And what’s valuable to them is that you protect communities and you don’t have unbiased impact. Next one. So now we’re looking at justice. So here we’ve got a justice system. Our justice department is building AI. to triage different complaints from citizens and reroute them to the right legal body, so whether it’s the courts or a commissioner or something like that. In the pilot, it performed really well, but later when they started to prepare to deploy this into production, the team discovers that, one, the model is hosted offshore.

Two, they don’t have a lot of information on when the model will be updated, and they don’t have control over that. This government doesn’t. That different logic within the model could change based on updates that they couldn’t control, and that they can’t audit the logs. So what do we think this time?

Yes.

Okay, everyone is sovereignty. Sorry, did you say something else? A responsible AI? I think that could also be here, because they hadn’t thought of maybe all these risks beforehand. but I agree here especially when you’ve got a national organization they need to have control of the model and how it functions not being able to update it or audit it in such a sensitive area like justice is a real challenge so sovereignty is a challenge here and then last one okay so here we’ve got a social science agency and they’re using AI to determine who’s eligible for social benefits and the pilot showed that they were able to progress and reduce the time and have fewer manual checks but when they were actually doing this in real life the model wasn’t able to explain why it had made a decision so why it had allocated benefits to someone versus someone else there was no ability to understand how to appeal it so if you were rejected for example you couldn’t understand why that was and how to change that decision There was bias discovered between different groups, so age groups or ethnicity or gender.

It wasn’t applying it the same to equal to everyone. And there was no agreed process for how you would escalate if there was a problem. So this became very seriously harmful, and there was a lot of vulnerable citizens who could be impacted. So in this scenario, what do we think between responsible and value? Anybody else? Training data not accurate. Agreed. So I agree. I think this one is a good one of responsible but also valuable here. Responsible AI is thinking about bias. It’s thinking about fairness. It’s thinking about the data that you have. It’s thinking about all these. It’s thinking about all these harms up front and how you’re going to deal with them. And then equally with value, people need to see value of why they’re using AI in a public system.

And if it’s actually harming people, then it’s not necessarily a good use case. So far, everyone is doing good. I think we can move on. But what we wanted to go through now is how does this work in real life? What does this actually look like? And so I’ll pass to Omid. Can we have the next slide, please?

Right. So I think it’s clear, you know, having had this conversation and the contribution from yourselves, that it’s not so straightforward because there are different dimensions, and this is the point that Riz is making in terms of having to look at different angles. So over the last two days, or definitely the day before yesterday, I was going around in the summit hall, and I was asking everyone, because you see everywhere it says sovereign AI, sovereign AI. I was asking them, what do you mean by sovereign AI? And some people were talking about, oh, we need to have our data centers here. Somebody was saying, or our models need to be here. There were different kind of conversations in terms of what sovereign AI actually means in the context of AI and how it works and how it deploys and so on and so forth.

But the key thing is that ultimately it comes down to control. And my view is that it’s not even just about the organization, the sector potentially, or the nation, but also about the people. So where is your data? Who’s actually looking at your data? Why are they looking at your data? What will they do with your data? If you don’t have an understanding of that, the likelihood of you trusting that system is very low, and therefore it would be susceptible to failure. So it’s really, really key to understand the implications of data sovereignty, AI sovereignty, and so on. I mean, I was talking to one country… called Serbia, and they were saying that we have a view that we need to have control of our own environment, we’re building new large language models in our own geography, and we are going to have control over what we do.

And I think that’s the key thing. But the important thing is that if the trust is lost in terms of the sovereignty, the likelihood is that the system will fail. And I can assure you that if it’s not designed in at the beginning, you’re going to test this under a lot of pressure. You’re likely to be in a crisis as well, because when you don’t know if your health data is trained on somebody else’s data, or you’re using very commercially available large language models. then the thing is you’re actually beholden to those people and therefore you may not be able to achieve what you want to achieve as an objective. So it’s a really, really important dimension in terms of a successful deployment.

And all of the stuff that I’m going to go through here, whilst I’ve seen them through failure, but also they’re the recipe for success. So you can think of it in both ways. So if I could have the next slide, please. So green AI, I mean, this is kind of not dissimilar to what we had before in terms of cloud and green computing and the fact that unless you actually look at the environment, look at it from an economic viability of the system, ultimately what it means is that it’s going to cost a lot more and it won’t scale. And if it doesn’t scale and you cannot handle the data volumes and the amount of usage that you do, you’re going to have, the likelihood is that it would stop.

Now, in my mind, the approach to take here is to make sure you address both. And what happens is that addressing both the environmental effects as well as the cost actually work very, very nicely together. So we had a similar scenario before in how we deployed cloud services, and the same thing is translating to this now. So the more economic your system is, the more likely that it’s going to reduce less greenhouse gases as well. And as a result of that, you can sustain this system longer term. I mean, we all know people are building now massive data centers. Yesterday, there was, I think, a discussion around Microsoft building the new data center that consumes, as much electricity as all of Los Angeles, and Los Angeles is an enormous city.

So the environmental effects of what we’re doing are really key, and it has a direct link into the costs that are driven out of that as well. And I can again assure you, I think it does only take away, that if an AI system can’t scale sustainably, then it won’t scale at all. I’m pretty convinced of that. So we can move on. So the next one is responsible AI, and I think a lot of people here are familiar with that. In terms of governance, assurance, are we doing the right things ethically, is there bias in the system, all of those things fall under the responsible AI banner. And it’s really fundamental in terms of giving people that trust that Teresa was talking about in order to use the system in anger and kind of really link their kind of lifestyle to that, and so on and so forth.

And as you know now, there are all sorts of other systems now like the AI companions that kind of help you achieve different things, whether it’s weight loss or even provide you counseling and help you along in your life. But unless they’re done in an ethical way and an unbiased way, they’re not leading you down a particular path, they’re likely to fail as well. Now, one thing that I wanted to bring to attention, and yesterday, Prime Minister Modi was talking about this, which is really key as far as I’m concerned in the responsible AI area, is the human -centered design of AI. Because when you’re actually building an AI system, you need to have in your mind who you’re trying to help and how.

And what does this actually mean? And if you’re trying to do something, you have to have a clear vision of what you’re trying to do. And if you’re trying to do something, you have to have a clear vision of what you’re trying to do. to them when they start to use the system. So I think the example around the traffic management was a very good one because we all struggled over the last few days with the traffic. And if a system is put into place which does not take into account what the purpose of what they’re doing is, then it is likely to fail. I think the goal of the system itself as well is really key in terms of whether it gets the right sort of results or not.

So there are many systems where people don’t consider that and as a result of that, it becomes unusable by the people or it might have harms built into it as a result of that. But the last one, the last dimension is how valuable that AI is and what does it mean in terms of the outcomes and what the measures are and so on. So a couple of days ago I attended a session where we had a senior executive from UAE. They were talking about, as a country, what they’re trying to do. And it’s really key for us to understand what we’re trying to achieve. So they had a very simple kind of thinking in terms of what they were trying to do, which made what they were doing much more measurable.

So what was the intention for them? The intention was that there are about 12 million people in the United Arab Emirates. And they wanted to effectively be, rather than 12 million people, with the introduction of AI, do as much work as 120 million, almost like 10 times the size. And I think that actually is really, really key. Very simple reason as to why you’re doing what you’re doing and how you measure it. And what the value is. Now, if you actually think about that in the context of, say, India, in my opinion, that ambition doesn’t give India the value. So to create, I don’t know, lots of agents to replace people’s jobs or do more jobs, right, doesn’t actually have the right outcome because there’s already a lot of people here.

Why would you do that, right? So you have to think really carefully about what the value is of the system itself because without thinking about that, you end up building a system that you cannot measure the value of. And then ultimately what you would do is that it would just become a dead weight. Why do we have this at all? Should we be getting rid of it or not? So hopefully you kind of understand all of the aspects of the different areas. At Kainos, we deploy systems, AI systems, into production. So we see. A lot of these issues. And we are quite lucky because our customers, which are all the government departments. are actually very, very clued up as well in terms of what different aspects of what we’re doing are, and they see value in it.

So it’s not just about deploying the technology, but how is this technology going to affect the UK citizens and where we work in other countries like Canada, US, and so on, those countries respectively. So I think that was my last slide. I think I’m going to hand it over to you.

So we had originally intended to maybe do different breakout groups. The audience is quite small, so it’s up to you. We could either have everyone kind of have a few discussions and talk about what you think is the most challenging, or we could use 10 minutes if we want to do a Q &A, if people want to share their thoughts. Put your hands up if you want to go into a breakout group and discuss one of the concepts together. okay so we’ll do the second nobody voted for that so why don’t we, yeah we can have a discussion it’d be interesting to hear are you looking at these four challenges which do you think is the most difficult, which do you feel like you’ve solved and we can have a little discussion around that for a little bit introduce yourself

Hi there thank you my name is Ami Kotecha I’m co -founder of Amro Partners we are a real estate company and we are now getting involved in a data spin out my challenge is as follows, I’m one of the co -founders of the company as a leader I’m very keen of course that there’s AI adoption there’s upskilling etc. in the company and of course productivity challenges where we have them should be addressed using this technology. I feel like I am often left in the lurch to actually literally make all the decisions within the private sector environment whereas I think government needs to step in and make some of these decisions on our behalf in terms of model utilization, where we go, what we do with it.

I mean we are good experimenters so fortunately we are throwing capital at experimenting not every company can afford to do that or would want to do that because of the same sort of issues you mentioned right at the start which are aligned with just the fear of adopting something that is going to break your system or open yourself up to some kind of cyber attack etc so how do you see this sort of playing out in the next 6 months, 8 years 12 months because obviously the technology is moving really fast as to what role the government is going to play in saying this is safe to use and this is still experimental and you should worry about it?

It’s like, go ahead and do it. But then there’s a medium risk, a high risk would be something that would be like really critical infrastructure or something that’s impacting people directly. And if it’s a high risk, then there’s a load of different things that you need to do around transparency with people. There’s also prohibited use cases of how to use AI. So I think that’s one example where some governments are actually saying, this is what we’ve deemed safe. And if it’s not one of these uses, then we want to see a lot of other checks. In the UK, we have regulation that’s looking at third party suppliers right now. And if they’re critical to the infrastructure or not of the country, then there will be new requirements on AI as well, in terms of like the updates that go in transparency around models, explainability.

But then maybe you have the US approach where you don’t have regulation yet. So I think that’s one example. it really depends on the country. I think a lot of what we heard yesterday was around, you know, for India, thinking about ethical and responsible AI, but I don’t know if you have any regulation in place around that yet. I think, yeah, I think it’s very difficult otherwise for a private company because otherwise you’re fighting to who gets to the bottom, who’s the cheapest, who’s the quickest. And this week I was touring around with different businesses and everyone was thinking, how do we do agents? But no one was thinking about human -centered, ethical, responsible. So I think it does need to come from the government to have a base.

But I noticed that some are maybe more forthcoming with that than others.

I think just before, I just wanted to answer your question about the government. There is a data protection and data personalization law that was, you know, legislated last year. November 2025 is going to be legal in the next 27 October onwards. They are getting a time of, you know, around 18 to 24 months. After that, what you are saying, the addressing of how the data is handled by the person who is creating the data, who is like the person who is created, who is the principal or one who is the repository, all that rules are coming. But presently, I would say it is 0 .1 % of that responsible AI part which is happening. But over a period of these two years, the preparation is going to happen where it will slowly get into that mode, actually.

I was just going to say, so she’s a high flyer entrepreneur in the UK, actually. But I was just going to say, in my mind, right, there are a couple of things that we should really push the government to do, right? One is about smartness. Smart data. So they’ve been playing around with this for years and years. So we’ve got quite a lot of open banking applications now. But this can be extended way beyond open banking where different organizations can share data. Like, for instance, in the property market, you know, how do you go through the cycle of all the way from putting an offer in to conveyancing to, I don’t know, valuation to the end, right?

So that’s really critical. The other side of it is actually having trust in language models which are built within the U .K. itself, right? And I think most of the – even Serbia is doing that, right? French have already done it with Mistral. So there is a lot of examples of this, and that’s where the government can really help, and that’s what we should be lobbying them to do, in my opinion. Any other comment? Oh, yeah. Maybe behind you? Oh, sorry. You had your hand up. You had your hand up first. You go first, and then behind you next.

Yeah. Yeah. So my – I am building a agentic AI for vending machines. And I have been an entrepreneur in the corporate world. But before three years, I was just doing physical stuff, right? Doing products, innovation, the food and beverage sector. One of the challenges which I am seeing is how to build value at a platform level rather than an individual customer level, right? For example, if I offer this vending machine agentic AI for a PepsiCo, they would say don’t do it for Coca -Cola, right? Give it to us only and keep it with us. But UPI, for example, was not a master card or a visa card thing, right? It was for the whole country, right?

So how do you get that kind of attraction to build a platform instead of one very customized for a customer who might say that don’t give it to anybody else. So that is the key question that I am trying to address and I do not seem to find answers.

I agree I think that is a challenge in the corporate world I used to work at Microsoft and even there it was if you’re using our technology if we’re coming on a panel then we’re on a panel but we’re not having Amazon on the panel or Google on the panel with us but I think like you say that’s really figuring out what you have that’s so unique and that actually goes to the value lens I think is that if you have something that’s really valuable to people you make the case that it has to be shared but it is it is difficult if you’re building it with one customer first because that almost becomes their IP that they want to keep right so something that we are doing when we’re working on response by our projects is we’re looking at all this similarity of requests that come in and we’re sort of doing the work on ourselves in the background and then we’re taking elements that we need and exposing them to the different customers and that way we keep that IP but it is very difficult to get multiple customers on board if they’re all competing

yeah so for example I build a few IP in the area of sustainability like clean air clean water I sold to a company but that company is not commercializing it I don’t want to name that company because it didn’t want to commercialize it wanted to keep that technology right so that’s a big challenge that I am seeing in the corporate world that a company will buy another company but it won’t implement for the society or for the good right so that is the challenge that I am seeing how do you handle that because that is part of the responsible AI as well as the valuable AI part

yeah I think you’re right and I think you have your own kind of description of this problem but I was in the US a few months ago and I saw, I don’t know whether you’re familiar with SVB, but it’s basically Silicon Valley Bank, right? And they did a presentation to us where they were talking about where all the funds are going, right? And if you actually see what is going on in terms of this, I think it’s about a trillion dollars worth of investment. This investment is flowing only into a handful of companies. What those companies are doing, they literally are stifling everybody else, right? This is a commercial reality, right? But if I was to offer you some options, I would say there shouldn’t be just the IP.

You should be thinking about it more as a service that you could build layers on, right? So you may retain the IP or you may share the IP. It could be a co -created, whatever it is, but it’s got to have a service model attached to it because if PepsiCo buys X and then co -creates, Coca -Cola buys Y, why would they be buying it and how would you be able to build on top of that but you know it’s very very commercially challenging problem it’s been there for many years this is nothing new

as Shri said exactly like that UPI beat that right so today UPI compared to a Mastercard or a Visa in India everyone is using that right and there are applications which are attached to UPI whether it’s a Paytm or a Google or whatever Amazon Pay all of them are on the platform of API right so the question I had was why are IT companies for example Kainos right or an Infosys or an Accenture not looking at the platform approach and looking at the services approach where they can put their team manpower and run projects right so I see this as a challenge I have been talking to the top management of Infosys Accenture every time I go with the proposal they say just do it for a client you know and we will attach you as an expert I don’t want to do that I want to build a platform there is nobody who is really interested in building that sort of a business which is path breaking it takes longer time right like UPI it happened organically can these kind of initiatives happen inorganically that was the question

I think they are looking at both yeah so I think we should take one more question because we have very few minutes left so we can talk after and I want to get to the person behind you for his question as well and then we will do a quick back up

Good afternoon thanks for covering those areas in the lectures that were much needed to understand so you talked about sovereignty AI and then you talked about value or responsible AI so there might be few scenarios where while chasing sovereignty we might have to bypass value additions or responsibility for the citizens while the other way round also so can you discuss about those scenarios where you value sovereignty more than talking about responsible AI or value additions AI and the otherwise also and when they can be parallelly taken into account

So you’re asking around Responsible AI and valuable AI Where they link Where one might be more useful Than the other So where I see Responsible AI I think it can actually incorporate As a lens for everything But it’s much easier to think of it as separate I think Responsible AI can encompass five things So like ethics Trust Like bias and fairness Human centered Governance and security Where I think that distinguishes from value Is value Is looking beyond Financial growth So a lot of organizations You might work with Or when you think of many organizations In the world, they’re looking at how much money Will this save me Or how much time or how much productivity But I think valuable AI is looking at What goes beyond that And what’s the value of value And what’s the value of value And what’s the value of value does it actually create more well -being in people?

Does it give people time back with their families, for example, or other hobbies they want to do? Valuable is thinking about what’s the long -term benefit that this will have in terms of how we change society. Maybe it’s going to create a whole bunch of different jobs now in something else. So I think actually if you’re using responsible AI, it will create value. So I still think they go hand -in -hand, but that’s probably how I distinguish it. Is that your question? No. I’m not sure. Maybe Omid has an answer. You know also. Yeah.

So I think you’re saying what happens when you have to do a trade -off, right, between sovereignty and value. And I think this is a very good question, to be honest, right? Because when you – so, again, yesterday I was wandering around in the summit. I keep asking people questions about different things. Right. And one of the countries that I spoke to – they know that using GPD models or Claude and various other things is a quick route to building what they need to build because they’re there it is immediate and it can be done almost like without any issues at all but they’re taking the hard route so they’re saying actually we don’t want to do that because what if tomorrow we fall out with them as Europeans are falling out with Americans anyway so what happens if they turn off the systems, what would we do then so if you think about it in terms of the speed and the value is actually going with what you got but the more challenging thing which is the value the value is can we actually use this system for our citizens on an ongoing basis is that data something that belongs to us are the models aligned with what we are doing.

So they want to be able to enable their people in order to deliver the right outcomes. And that would not happen if they just outsourced their sovereignty to the U .S. So I think those are some of the very, very important factors that need to be explained. But ultimately, from a value perspective, three is a spot on. I think it’s about what is the value in terms of to the people that are going to use that system. So if tomorrow we found this fantastic system, like I give you an example, we’ve stopped multiple times in terms of the traffic because some VIP was coming out of somewhere, right? And then just literally closed the road.

So we’re sitting there for like half an hour and then we get going again. That’s happened, I think, three or four times so far. So if you were to build the system, you would be needing to think, you know, what is the value for those taxi drivers, all these public that are going around. and that’s the key thing that you need to be able to use AI to achieve, right? This needs to be measurable, it needs to actually help the people itself. So yes, it’s a very tricky trade -off.

I think the trade -off is really good, especially in sustainability as well. So a lot of times organizations might just think, how do we adopt AI as quick as possible, get people to use it as much as possible, but actually every query that you use has a sustainability impact. And so I think there’s a trade -off there because there might be a sustainability impact, but depending on where you are, you might value training people to use AI more, so you might be okay with that impact because it’s more about getting people comfortable with using it. But then if you are an organization that really values sustainability, you have really strong carbon goals or net zero goals.

then actually that might be the trade -off that you have. So I think one thing that we’re doing when we’re working with organizations is we’re getting them to make that very difficult decision of here’s high concern, here’s low concern. We map out all the harms that we can think of and all the principles and values that align to them and they can’t put any side by side. And they have to move all of them from high to low concern. And very quickly that makes you see what’s real for your organization and I’ve seen a lot of them put sustainability at the bottom, which to me is a little bit concerning, but it does start to think about really understanding your organization and how those trade -offs are going to play.

And that’s what we’re finding in the human one as well.

So just 10 seconds more, adding to yours, ranking low to high. So out of all these four, sustainability, sovereignty, responsibility and valuation, how do you rate them as low to high, all the four factors that you have covered in your paper?

how do you rate all this on a scale of how do I rate them I think that’s very difficult I think I’m putting response by AI at the top because I think it can actually it’s a cheat because it can kind of include sustainability actually and I think it will create value I think so then I would probably put sovereignty lower than that but obviously this year has maybe changed that geopolitically I think I still put response by AI at the top I’ll make that hard choice what do you say

I think I kind of agree and I think Modi Prime Minister Modi said this himself yesterday about human centered AI design is part of responsible AI a few days ago me and Teresa were talking to someone and they were describing a system now if you just indulge me for a couple of minutes let me explain the background of the system and then you see how it’s relevant so they were building a system for kind of the nursing or old people’s home so you may know that elderly get dehydrated and they forget to drink water and then that causes a lot of problems for them so they built a system where using AI and vision they were seeing if the elderly were having enough liquid in the day or not now that’s fantastic everybody says this is a brilliant idea but then you think about it they are monitoring those elderly both in the area where it’s common as well as where they may be in their bedrooms or whatever so that brings a challenge and then the other challenge was what about the people, the nurses who are actually hydrating them because that could become a negative effect on them because somebody might be saying you’re not doing your job right right and what about the family of the elderly, what about the impact on them, so I think it is really important to understand why we build the system who it affects how it affects them and what the long term benefits are which brings the value, right, this is why it’s four dimensions, none of these are independent, I think they’re all relating to one another, one shape or form

yeah so we’ll work towards wrapping up because I think we’re getting the time check can we but this one switched because it said 8 and then it said 17 and now it says 10 and then she told me I had 7 this one’s right, okay well let’s see if there are more questions yeah, but we have the takeaways and things to go through also, so I think we’ll wrap up and we can talk to people individually afterwards, can we skip through some of the slides, because we wanted to ask next one, next one next one Next one, I think. Okay, so we actually wanted to flip that question and ask this to you in the audience as well. Which one would be your top?

So of those four lenses, sovereign, green, responsible AI, which one do you think is an absolute must -have? And you can only pick one. And if this isn’t there, it’s going to derail the project. Shall I show of hands? Who says sovereignty is the most important? Who says that it’s green AI sustainability? Who says it’s responsible and then value? Some people didn’t vote. You didn’t vote back there. But I think it sounds like a lot of people are in the responsible and value is the most important. I think I agree. But I think what we wanted to come across is all of these need to come into play as well. Can we do the next slide?

Of that question, though, who has a responsible AI practice in place? Who uses a framework or anything like that? Anybody? Who has a sovereign AI policy in place? No? And who is looking at sustainability? None of us. So that’s a takeaway for all of us. I think we wanted to wrap up with how do I take this forward. So a couple points I want to make. The first is that we have taken a lot of the learnings and things you talked about in this and we’ve turned it into a white paper. There’s a link below, but we can share it with you. We’ve shared it on LinkedIn as well to talk about these learnings. And we wrapped it up to say for each of those themes, here’s eight to ten things that you could do if you really wanted to take sovereignty, green, sustainable, responsible, and valuable AI forward.

So please check out. I’m very happy to talk about that paper. I’m very happy to talk about it and give you insight. My thing, the key takeaway for us here is that no single dimension is the answer. I think that’s come out in the scenarios, in the conversations that we’ve had, and in how we’re prioritizing is that you can’t really have one, just one. You need all of them if you want to scale that project and really make it to production. The second point is on tradeoffs. So it was really good that that came up in the conversation, is that just being aware of the tradeoffs that you will have to make and having something in process to talk about why you made that decision is important.

I think a takeaway for everyone here is think about an AI policy, which talks about how you’ll use AI and what you will prioritize. Think about having a responsible AI framework, which is essentially all the questions and things that you want to be implemented across ethics or across trust, security. and then really think about how you can turn some of this into numbers. So what are the KPIs that you can actually look at for sustainability, for users, for ethics? Don’t just make them a concept of we will be ethical. Think about what that actually means for you and how you’re going to measure it. That’s important if you want to get funding, investment, and show the project is a success.

And then finally, think about how you can upskill your teams to understand these concepts and how you can incorporate diverse views. I think that’s probably the most important in building out the responsibility. So we will wrap up. This is to say if you want to get in touch with us, here’s our details. Find us on LinkedIn. Send us an email. We can have a couple minutes after this since I know there was two questions in the audience that we might not have got to. But otherwise, we hope this session was useful. If you want to give us feedback, here’s a bit. Bigger QR code. so if you want to stay in touch then fill out this, let us know if there’s anything we can improve on the session or any questions that you have, we’re super happy to hear that and otherwise just a big thank you for your participation and yes, hope you have a good rest of the day and a good weekend

Yeah, I was just going to say I think great questions there about trade -off and absolutely the right question to ask because none of these are unique Sorry, please, go ahead

Yeah Like you were talking about trade -offs, I just wanted to say okay, every model has their own aspects like pros and cons or as you say, different dimensions, I’ve got most of my answers from those questions but I just wanted to ask, if we’re building something and taking these aspects like okay, taking responsible AIA, valuable AIA on them. But if we are taking, we’ll be missing some aspects. As he said about responsibility, if we are taking accuracy and fairness, if we will take if it makes it easier to speak in Hindi, I understand. It’s okay, but fairness, okay. If we are doing Sorry, sorry. Again.

I think there’s no issue. You can ask us on email as well. It’s not an issue. We’re more than happy we’ll respond to you if you want to ask.

Just a question.