Data for development
01:00:36 Share

Welcome to the LSE Events podcast by the London School of Economics and Political Science. Get ready to hear from some of the most influential international figures in the social sciences.

Good afternoon everybody. Welcome to the LSE and welcome to the LSE Festival. So for those that aren't aware, the LSE Festival is a week-long affair here at the LSE and we're very excited to be able to put on a discussion today on the role of data for development. I'm very fortunate to be joined by three wonderful panelists.

I'll just start from the furthest to the right and to my left.

And we've just had a whole morning roundtable, so please excuse the formality I'll introduce with their first names as well. So Mulele is the Permanent Secretary of the Ministry of Finance in Zambia, and he's been contributing a lot to the use of data in governance in Zambia. And we'll hear a lot from him today on that.

Claire is from the Global Partnership on Data for Sustainable Development, and she has been, I think, a global forefront of thinking about how we

use data and piece in data to the places where it's missing. And then lastly, Natalia is really helping us think through about the tricky questions that face the use of data and AI in society. Think about the ethics and how we can build systems that achieve better outcomes but also do it in the right way. And so

Before I dive into our conversation today, I'd like to remind everyone that we are online, so we have participants joining remotely, and we will have opportunities to take questions online at the end as well. So what I'll do is I will open up with a few one or two thoughts, and then I'll be doing a quick interview style almost with each of our three panelists before opening it up to the room for plenty of discussion.

When we think about data today, we think of data as a commodity, a fuel, an infrastructure almost that can be used and harnessed for a number of means, whether that's for setting policies, guiding investments, or helping create strategies for the way forward. In developing countries, data can also be a mirror, however. It's something that tells us what we know, tells us what we're hiding, perhaps,

and also telling us what we choose to measure or not to measure in our countries. And what I hope we can do in our conversation today is to think through how we can use data meaningfully for inclusive development and to improve governance, to improve outcomes for society.

And so what I'm hoping to do is to create a bit of an arc here from the on-the-ground policymakers' perspective in Zambia through the global perspective from Claire and all the way to the frontier ethical considerations perhaps that we need to bear in mind when we talk about the use of data and technology in governance. So, Mulele, let me start with you. My first question will be

In your day-to-day business, the Ministry of Finance, you do many things: budget forecasting, national accounts, handling the debt, all of these different aspects. Where have you found that the data has fallen short? Can you give one or two examples in your work, you've just wished, "If only I had this piece of data, this information, I'd be able to make a better decision"?

Okay, thank you very much. So I think there's quite a lot we do as Ministry of Finance. As you've rightly pointed out, we make fiscal policy. We...

We basically decide how much to tax the citizens, whether direct taxes or indirect taxes. We make spending decisions. We also have to make policies in terms of the national plan. So Zambia has a vision.

2030 and to actualize that vision, it's a 25-year horizon vision, and we put in five-year development plans each year. I mean, every five years we put in five-year development plans to try to actualize that vision. And on an annual basis, we do a budget which tries to speak to the national development plans and try to achieve the outcomes of those plans.

So, I mean, just from that you can see that we really need a lot of data to try to achieve our long-term goals, our medium-term goals and our short-term goals. And in each one of them we really need quite a lot of data. So I think in tax policy we really require a lot of information. I think one of the things is that you can get the policy right

But taxes also, the administrative part of taxes, and to administer taxes also you need a whole different set of data and high quality information for that data. And I think this morning I did talk about

if you are charging, for instance, rental income tax, taxing those people that earn income through rentals, whether they've got commercial buildings or, you know, buildings where people are renting out. So you need quite a lot of data. You need to know the tenants there because in taxation most of the time

you find yourselves that it's mostly self-declaration that it's the landlord or the landlady that declares how much income they're making. But if you bring different aspects of data, as we are trying to work out now, is that you can still get

information on electricity consumption, water consumption, or any consumption of any utility, and it tells you the level of occupancy of a particular building or a particular complex and so on. And we all need that kind of data in order to just improve the compliance in that area. Coming to specific questions really

The data we really, really, really want to have and we think we should have within our hands is that of employment. Because largely what we're trying to do is to fight poverty and we believe that employment rates are very good and that gives us a good understanding of what happens. But to get that information, we've relied largely on surveys that happen maybe after every two years.

and there's some information gap that is missing. But now we're looking to now look at returns that are filed for tax purposes, returns that are filed also for rather the number of people that are contributing towards pension to gauge the level of employment. So

There are other proxies that we are now looking at having because that really helps to check whether our policy is going. So really, use of data is quite large, but we have some challenges in getting it, getting the quality of the data, making sure that it's standard, and making sure that it comes on time as we make our policy decisions. So it sounds like data plays an important role in all of the day-to-day processes that you do in government. Absolutely.

Now, Zambia has gone through macroeconomic forums recently, and I'm curious to hear whether it's in conversations internally on what the right reform is for the government or perhaps conversations externally with groups like the IMF. How has data fed into the modeling and the decision making and the wider planning that you undertake in the ministry on macroeconomic reforms?

So the past few reforms that we've had in the last few years, it's really trying to stabilize the economy. Zambia is in dire distress. We've reached a point around 2020 where

For every hundred pounds we get in domestic tax revenue, about 60 pounds was going towards debt service. So that was highly unsustainable. And also we needed about 43 pounds to just pay salaries to run government. And we needed much more resources to run the other services that government is supposed to give.

The government in 2021, when it changed, the priority really has largely been to try to sustain the levels of debt. And to do that, we've been trying to restructure the debts by elongating the maturities so that on an annual basis the outlays are much lower and that we are able to now

really look at what we need to do as government, try to ensure that we make the right investments in the social sector and provide really the services to our citizens. So to do that, we've had to go on an IMF program.

It's a 36-month program. It started in the year 2022 and is coming to an end this particular October. And we had three different sets of debts that we had. We had debt with bilaterals, governments such as the government of the UK. We also had debt in terms of bond, Euro bond holders, and another category were just commercial creditors. And we've had to do restructuring with those different groups.

We've reached a stage where we've done close to 92% of our debt has been restructured, and the annual outlets are much lower. And even in the program that we have, we are only supposed to pay about 14 pounds per hundred dollars that we collect in domestic revenue, which is quite sustainable. And that is what we've had to do. But of course, sometimes to reach that 14 pounds,

The denominator has had to increase in terms of the revenue base itself, and we're really looking forward to trying to either...

efficiencies in the way we actually collect these revenues. So I think data has played a very big role in this entire process of data structuring, making the justifications to different creditors, doing the debt sustainability analysis. The debt sustainability analysis really takes in quite a lot of data, macroeconomic data and other issues that we need to take. So it's

It's really been a use. I mean, in our day lives right now, really, we require a lot of use of this data to try to get the macros right. And that is where, basically, in the last few years, our big effort has been. And we've managed to do that. And in between doing that, we've managed to implement policies such as free education

And with that policy, there are 2 million kids who otherwise wouldn't have been in school and school right now. And I think that education really helps in terms of eliminating intergenerational poverty. We've managed to implement what we call a constituency development fund, which is really taking resources closer to the people and the people themselves make those decisions. And we've had to enhance that amount. But also in between last year, we were faced with a very devastating drought at Zambia.

which really affected food production and we've been self-sufficient for many decades, but for the very first time in many decades we were threatened with food insecurity and we had to import some food from neighboring countries.

And Zambia relies mostly on hydropower, and with very little rainfall, we lost about almost 6 million megawatt hours of electricity. We might do 19.3 million megawatts, and we went down to 13.3 megawatt hours. And that had a knock-on effect on a number of industries, and growth kind of slowed down. Rather, we had to grow slower than we actually projected.

I think that's a really nice story from how you first mentioned the need for, let's say, more microeconomic data on employment and now how that feeds all the way up to national accounts and thinking about macroeconomic topics like debt distress, etc. So my final question, Milele, for you is this data ecosystem. So you've mentioned how data can be valuable and how it's fed into these aspects, but what

We know, and I think you know more than I for sure, that there's often some challenges in accessing that data, or perhaps data gets stuck in some parts of the ministry or in a different department. How do you break these silos? And I'm aware Zambia is experimenting a lot with innovative approaches to try to incorporate a wider ecosystem of not only

sharing data but also using that data to generate evidence? Yeah. Well, I think there are basically two sources of data. You have data which comes from the national statistical system, which is made available pretty much to everyone. Then you've got administrative data which is collected by different sectors who collect it specifically for their own use.

but that data actually may also be useful to another ministry which at first hand may seem unrelated, but actually the data becomes very important in that aspect.

I think to try to break these silos, we've been collaborating with IGC and we have an evidence lab in Zambia. And this is basically trying to break some of these silos by enhancing the capacity of our public servants to analyze this data and how they can overlay the different data which is collected from other aspects. So that is something that we really must work on.

I mean, for instance, Minnesota Agriculture does collect its own data on, say, crop production and what really happens. But if you really dig down into that data, if you just ask a question to the farmer, their level of education, and you run some regression, you just discover that the more educated a household or family is, the higher their level of productivity. And this is an important aspect for the people that are in the education sector,

to understand how education affects food production. So I think it's very important that data is exchanged and that even the schooling system begins to react to some of these things. So if you look at disease patterns, you find a similar thing as well, that less

more educated people do not suffer from certain primary diseases because of the level of awareness those households have. So I think what is important is to really break these silos by making data available across everyone, by publishing most of the data, making it

really available even in raw form for others to do a lot of analysis, for researchers, for the people in the academic world, civil society, and governments to be able to analyze the data.

Wonderful. That's been a very helpful perspective from the front line. Let's move along this narrative arc and go to the global ecosystem. So, Claire, your organization, if I may say, kind of stitches together the supply and demand of data from many different groups, whether it's government, civil society, businesses, et cetera. From the work you've been doing, where do you see the biggest gaps? What's your diagnosis on, you know, in this space, this is what's missing, this would be great if we could have it?

Thank you, Tim, and thank you, Milele. That was really, really interesting. I mean, my organisation, the Global Partnership for Sustainable Development Data, was kind of born in 2015.

and it came out of the processes within the United Nations to negotiate the sustainable development goals. And there was a sense then that, you know, there was a huge amount of excitement then about all of the sort of new innovations in data, the sense that there was going to be, you know, all of these great new things that were produced and the satellites and mobile phones and all this stuff that was going to help to achieve the goals. So there was a real focus initially on the supply side.

And this sense that, look, there's all of these great things here. And, you know, what we need to do is kind of mobilize the supply side and make sure that those people in their, you know, garages and their university departments and all of the rest of it that are kind of building stuff are building the things that we need to solve the problems of sustainable development.

But I think very, very quickly it became evident that the problem really is not on the supply side because there is a huge amount of stuff being created and much of it is fantastic and underused.

The problem instead lay much more when it comes, and I'm talking now about very much in the same terrain as Malala was the sort of public sector data systems. This is the data that governments need to run the health service, education and so on. The issue was not a lack of gadgets and widgets. The issue was a lack often of demand, a lack of particularly political leaders

indicating that they want investment in this stuff, that they're prepared to put resources behind it, and a lack of the kind of institutional flexibility and incentives and all of the different hundreds of thousands of things that need to go into the chain to make it possible. So as we have...

gone that we're now 10 years into the Sustainable Development Goals. Our focus, and I think the focus of the community that we bring together, has very much shifted from the problem is supply to the problem is demand. Primarily, and I think that goes along with another way of framing that, is to say the problem has shifted from there's a technical challenge that we have to overcome to there's a political challenge that we have to overcome. That is actually the barrier that we're facing here.

So has the data revolution delivered? What lessons have you learned in the last 10 years? Well, it's interesting. The version of that question that you sent me in advance, just to let everybody in on the magic process here...

referred to a report that I wrote in the UN called A World That Counts that I was seconded into the Secretary General's office to write in 2014. And the question forced me to go back and read the report, which I hadn't done for a while.

And a few things occurred to me. I mean, first of all, I do think, you know, it held up fairly well. There was nothing in it that I found really embarrassing anyway, which is always a good start. But I think, you know, what really struck me when I was reading it is a few things. First of all, as we know from every revolution that has ever been, is always a few years off, you know. The revolution is always next week, next year.

And I think, you know, the data revolution is the same. We had a sense that something big was about to happen that was going to be great. And to some extent, you know, then we were talking about satellites and mobiles. Now we're talking about AI. But essentially, it's the same thing. There's some kind of technology that we're really excited about, which is going to deliver, but not quite yet.

So that was the first thing. And perhaps, you know, it's good to, you know, perhaps it's that sense of a sort of future which we're excited about, which does motivate us to do things that we need to do in the present. So not to denigrate that. But I think, you know, we also need to kind of think that, you know, as we think about AI, we have been here before with other ways of technology development. And then we need to sort of look at technology.

I don't know if there are any historians in the room, but what does that tell us about how change happens and the history of technological innovation? So that was the first thing, the revolution is always tomorrow. Whatever kind of technology you're thinking about. The other thing, reading about it and thinking about how we were then, that I think illustrates a problem that we haven't yet cracked, and it's massively frustrating that we haven't done this,

You know, I think at that point what we would think we had, you know, as I say, we had lots of great examples of small things that were happening. And that report, well, the counts has kind of peppered through with all kinds of like, oh, there's this, there's that, there's this project here, there's that project here. And then this very macro sense of kind of, and here's all the things that in some imagined future we could do.

And it's the gap between those two which remains, really, where we have, you know, we still have fantastic projects. We have some more of them. Perhaps they're at a slightly bigger scale, some of them, but still mostly projects. And we still have these huge problems that exist even more now. And we still are not bringing the technology to the scale of

that is needed to solve the problem, and particularly in government. Of course, for really good reasons, for accountability, for the use of public money, for trust, the public sector is and should be slow to innovate. We need a level of security in the public sector, but it's still incredibly slow, even bearing that in mind. So I think there is a sense that we haven't cracked down

the problem of scale and specifically of scaling through public sector institutions that was a problem then, is a problem now. You know, I think a lot of the things that are going on in Zambia are the way to do this. There's no single magic thing that we should be doing because if it was that simple, we'd already be doing it.

It's exactly that sort of thing. It's the political leadership. It's the slow work through public sector institutions, the legal changes, the investment in infrastructure, in capacity, the kind of insistence at the leadership level of doing things differently, of sharing data across institutions, all the things that people don't want to do. But I think that's the bit. It's the boring bit. You know, we've had all the excitement, but I think to realise that, we kind of have to do the boring bit.

So partnerships is in your name and it sounds like a lot of what you're saying really rests on establishing these partnerships. Can you say a few more words about that? Sure.

Sure. I mean, partnerships is a word that we reach for very often in this sector, not just in data, but in almost any sort of anything. And of course, in an era that we are all sitting here now with sort of declining resources, we reach even more for partnerships because there's less money to play with. So we need to sort of combine resources and find others who can...

who can bring Tibern out. If partnerships were easy, an organisation like mine wouldn't need to exist because what we do is help organisations to broker partnerships. If that was easy, we would all be doing something else. So, it's really difficult. I think it's become harder actually rather than easier, at least in the world of data for a number of reasons.

firstly connected to the funding squeeze. We are all in this sector, both in terms of national budgets and in terms of external budgets for ODA and so on, facing a funding squeeze that makes

the exact point at which collaboration is more urgent, it becomes harder because organisational incentives are all working against it. People are more jealous of the resources that they have, of the relationships that they have, less willing to share the limelight, sort of, you know, for many, many reasons, all of them bad, but all of them entirely understandable. It makes partnerships more difficult. I think potentially, and I would love to hear from anyone in the audience who has more experience of this, but...

it feels to me somewhat that, you know, as, you know, interested, you know, looking forward to hearing more about AI in a moment, but as

our sort of perception and the reality of the value of data has gone up. You know, people are more excited now, and institutions rightly are more excited now about the value, sort of both realized and potential of the data that they hold, the insights that it can give them in terms of, you know, how they can use it through AI algorithms and others, and in many cases, the commercial value of the data that they hold if they're able to form the right relationships.

as the reality and the perception of value goes up. So in some ways, the negotiation of partnerships actually becomes harder.

Because, you know, the sort of the sense of wanting to drive a good bargain of the opportunity cost of one thing versus something else of the sort of potential risks is getting higher. So I do think that just as our sense of the benefits that could be realized through partnerships has gone up with new technologies in the sense that the value of data is important, the political value.

of data has really gone up and all of those are good things. I think it does mean that it becomes somewhat harder, not impossible because there are lots of good ideas and it doesn't relieve us from the sort of necessity of doing it. We still have to do it. But I think it does mean that, you know, we have to think again about partnerships. And I think also...

Sometimes in the sector, you know, if we're driven, if we... One of the really interesting things about the world of data and technology is that it brings together a lot of different kind of parts of policy. So, you know, those of us who come into it from a sort of social good, sustainable development sense are motivated by one set of incentives and an understanding of sort of what the public interest is. There are many in this sector, for example, particularly in relation to AI, who come from a security background.

their incentives and sense of what is the public interest that they're trying to defend through their work in the public sector is somewhat different. Of course, a lot of these partnerships depend on

forming good relationships with private sector entities who have a huge amount to offer here, but there's also another set of incentives that's coming in. So I think understanding each other and trying to figure out a kind of way through that mixture of incentives, the more complex the landscape becomes, the higher the economic stakes as technology advances, the more difficult, but also challenging,

the more important that becomes. Terrific. Thank you, Claire. So we've started in Zambia, we've gone global. Now I'd like to turn to Natalia. So Natalia, your work is really on the kind of ethical and frontier lens, if you will, on AI and all these advances in technology. And it really focuses on how do we build trust and accountability in these systems. But why don't we start with the basics? Why is transparency still so hard? And why is that something we need to keep pushing for?

Because it's not a popular concept in the world where the most desirable developments in relation to AI are having more advanced models, are having different and multiple users of different AIs, essentially jumping on whatever is the next new thing in the world of AI. And I think transparency and ethics...

are an invitation to reflect on why are we using those systems in the first place? What are we using them for? What kind of problems are we trying to solve? And these are not questions that we can answer easily, unfortunately. And more practically speaking, in the government context, there are also a lot of resource constraints. So again,

to do, for example, what we did in the UK, which is establish a standardised way to publish information about how AI is being used in the public sector. This is a large operation. This took us a couple of years to think about what kind of information should be there, how do we present that to the public, how do we make sure that the people know that it's there, that they can engage with it. This costs a lot of money and it takes a lot of time, so it's not a popular idea necessarily to implement.

Then there is also the lack of political will. And actually, as Claire was talking about the barriers to data, I recognize a lot of the same problems, unfortunately. So we're talking about a slightly different area, maybe slightly different time, but the problems haven't changed that much, I think.

So there is also the lack of political will. There is a lot of political will across the world to deploy AI, but I think less so to be more transparent about it, to be more proactive about it, to be more ethical and measured in how AI technologies are being deployed. And I think there is also, unfortunately, a lack of public interest.

When it comes to algorithmic transparency initiatives, we really need the person to be at the end of that pipeline and to know that those kind of websites are there, that they can go and, for example, in the UK, you can go on gov.uk, you can check how the government here uses AI. And if people don't care about that, if there is not a lobby that will say to the government, please publish that information, we really need to know.

then there isn't an incentive for the government to do that. And there are so many different things. Governments are constantly under-resourced and there is always a lack of funding. So if there is insufficient public awareness, if we don't have that call from researchers or members of the public, then that's just another argument not to do it. So I think there is a lot of those kind of operational problems, but...

I think because of the way that AI works specifically, there is also an argument here that, well, why would we bother with transparency because AI is a black box, quote unquote, right? And this is something that I came across a lot when I was establishing those transparency measures in the UK.

And my answer to this would be that if we are using those tools to determine, to make decisions that can change people's lives, we really need to, even if the actual workings of the algorithm is a black box, we should publish everything else.

So we're in the policy process. This is embedded. How does it inform decisions? Who are the people who then take that information that's been AI generated and implemented? What are address mechanisms? It's essentially the less transparency and explainability we have when it comes to the model itself, the more we have to be transparent and explainable about everything else. But again, that's not an easy job.

Very interesting. In the UK, you've helped set standards on algorithmic transparency. So imagine Mulele invites you to Osaka next month.

From this process working on algorithmic transparency in the UK, what can we take to a context like Zambia or any other lower middle-income country? How much is transferable? What new aspects do we need to keep in mind when thinking about this? I'd love to. I'd love to do that, by the way. I think what's transferable is definitely the process that we've gone through. So we started...

And we started doing that in a very collaborative way. And we've actually asked the members of the public. So we had a deliberative study which asked people from all around the UK about what kind of information would you like to see published when it comes to the government use of AI? So I would say that kind of collaboration. So ask your ask your people, what is it that they care about? What is it that they want to see?

Another one is leveraging the knowledge that's already in the country when it comes to AI. So academics, civil society organizations, researchers, media, so people who will be using that as well. I think it's very, very important to invite them in and have that kind of conversation. And in terms of the content, I think there is a certain bare minimum of transparency that should exist.

be included in those transparency standards. So things like whereabouts in the government that's used, whereabouts that particular AI model is being used, what's the oversight, what's the accountability structure, what to do if something goes wrong, what kind of data was used to train it. So I think those kind of informations to me are very core. But I do think as algorithm transparency is a very new field, I think it should really be adapted to the local context. So

Are there any specific... Is there any specific information, for example, that's aligned with the kind of national values that Zambia has that should be published? And then what is it? Are there any specific areas that people care about as well? And I think...

It's really good to sort of take what we've done in the UK, what's being done in other places as an example and as a starting point. But I'm a strong advocate of making sure that it works within your unique local context, because obviously that's going to be very different. So sort of take it or leave it, but definitely ask people who will be the users of that.

Absolutely. And I think there's this narrative that there perhaps might be a risk in developing countries that they become AI policy takers, that they just take the technologies, including kind of the rules and standards with them and don't adjust them to their own context. I think some of the points you raised really are helpful in thinking about how we can make these systems applicable in these contexts, ethical in these contexts, but while still remaining globally interoperable.

My last question for you, Natalia, is have you written about the importance of data ethicists, not just data scientists? What makes a good data ethicist and how are they different?

Sure. So this is something that emerged when I was working in the government that was on the digital side of policy. And back then, we had many excellent data scientists. We had a lot of people who really knew the importance of data and the data skills.

But then sometimes what we've noticed was a little bit of a disconnect between what they were working on and the potential real world consequences.

This is the need that we've identified for a profession that's digitally literate and has a good understanding of data, but doesn't necessarily need to code, but actually in contrary brings that social science or philosophical perspective into that conversation. So this is how the role of the data ethicist emerged in the UK government. And I went through the process of formalizing it, doing a lot of research on what kind of skills that person should have.

And ultimately, the way I see it now, and it's called data, but really it applies to AI even more so than data. Maybe we should change it now to AI. I think it's really similar. But essentially, it's a person that convenes and creates space for conversations about...

what are the implications of using that particular algorithm? What is the worst that can happen? What are some of the unintended consequences of the work that we're doing with data? And some of the skills that that person should have are quite different from

your usual data scientists. So one of the things that we put in the government skills framework was for them to actually have a social science background. And I think that really is so important in the world of AI to be able to think critically, to be able to challenge, to be able to ask questions. Another one is communication. It's conveying complex ideas, translating between data people and policy people because, again, in governments, and I imagine we all sort of have that same

impression that sometimes those two worlds don't talk to each other. So there is a need for somebody to be that bridge. And also just knowing what are the latest developments in the AI world, in the data world, what's happening, what are the potential risks and how we can respond to them proactively. And by the way, this is not just something that should exist in the government context. There are a lot of data and AI ethicists in the private sector as well. And I think it's a really important profession in the coming years as well.

Hi, I'm interrupting this event to tell you about another awesome LSE podcast that we think you'd enjoy. LSE IQ asks social scientists and other experts to answer one intelligent question, like why do people believe in conspiracy theories or can we afford the super rich? Come check us out. Just search for LSE IQ wherever you get your podcasts. Now back to the event.

Wonderful, thank you. So I will go to one last lightning round question for all panelists. I had a boring version and a spicier version. I'm choosing the spicier version. In one word, max one sentence, AI, is it a leapfrog or a distraction? Lele, I'll start with you. It's a leapfrog. Leapfrog. Claire?

Oh, can I go somewhere in the middle? If I have to choose, I'd go Leapfrog, but I think really it's in the middle.

I'd say it depends. Wonderful. Thank you. So we have now a chance to hear from the room as well as from those online. So if you'd like to ask a question here, please do raise your hands and wait for the microphone to come to you. And equally online, please do submit some questions. I think we have a gentleman there in the back in the light blue shirt.

It's just really about the sustainability scaling up, increasing that scalability. How do you work with people who have experience in change from different disciplines and also the older generation? Because of people living longer, there's a whole pool of talent who are used to rolling these things out and scaling up and getting past the barriers. How do you incorporate that?

Wonderful. I should add, please, can you introduce yourself briefly? And also, if it's to any specific panelists, please. Sorry. I'm Jim Taylor, and I've got my own consultancy, basically. But I'm an expert on change and operating in a VUCA environment, which is actually what's behind what you're saying, driven by technology. Great. Would anyone like to briefly? OK, we'll collect three questions, perhaps. I think there was a gentleman here in the blazer up front, and then the lady in the white T-shirt in the middle.

Bill Anderson, I'm a data governance specialist working mainly in Africa. Just to pick up on the leapfrog issue, I mean as Claire said, there is a battle over resources. There always has been and it's far greater now. So to me on the leapfrog issue, I'd ask the panel, we at the moment are spending money on biometric IDs, digital payments, on data exchange, the foundations of digital public infrastructure.

Those are choices we make within the limited resources. We're spending money and investments in AI. We're not spending money on primary schools capturing the data of their students. We're not spending money on the capturing of health data in primary health care clinics. These are choices, and I think that within these choices,

The argument at the moment has been won by the techno-optimists who say that the leapfrog can work. And I ask the whole panel, do we need to fight this or is it going to work itself out? Great. Thank you. One more question.

Hi, I'm Julie. I'm a master's student here at the LSE studying social data science. And I know we've talked about AI a lot here on the panel, and I'm really curious, particularly in the context of public policy and sustainability, how often is AI referring to large language models specifically, which have really kind of launched the interest in AI, I think, in recent years versus machine learning more broadly? Great, thank you. So we have three questions.

One, broadly on managing change. Second, broadly on scarce resources. How do we think about prioritization? And then, is AI just an LLM or much more? I'll leave it open to whoever would like to start. Claire, please. All right.

So, I mean, in terms of how to incorporate talent, I think, again, you know, ultimately that is about creating the... I mean, it sounds to me like you're probably more expert at answering your own question than I would be, quite frankly. But it seems to me that, again, this is about the demand side. You know, we know there is lots of talent out there. The issue is creating...

the kind of space and the incentives and the leadership to use that talent in the place where you want it, which is whether it's talent or technology, essentially we're kind of talking about the same thing. So, but I think it is important that we do see talent in the round, not just as, you know, people like the majority of this audience, sort of very young, bright people with, you know, recently qualified, but also that, you know, this is the full spectrum. And I think that that is a point very well taken.

I mean, Bill, you know, in terms of the... I guess I would see the question, and this is something that we have discussed over many years and I'm sure will over many years to come, I guess I would see the question slightly differently. That I think...

At the moment, absolutely, you know, there is an emphasis in a lot of the investments on digital public infrastructure, on things like digital ID and payments and some of those foundational issues. But I think, you know, I would see, for example, setting up an effective digital information gathering and platform.

analysis system within the education sector as also a digital public good and something where governments need to invest resources in digital systems. So for me the question isn't should we invest in digital systems or in

something else. It's when we're looking at the huge range of investments in digital systems that governments want to do, what is the sequencing? What is the order? And how does that align with political parties? And I think for me, the other issue there is also making sure that these systems...

kind of don't don't exist in silos of each other you know and again this is something you and I've talked about and I should say you know when I asked my team when I was preparing for this and I said to my team what are the things that I should read a lot of Bill's work with the things that they said that I should read so everyone go and read Bill's work but um but you know I think um

Yeah, so for me, this is not about digital or not. This is about how do we do digital and doing it in a way which aligns with ultimately the goals of the sort of equity, public service and so on that we want rather than something else that may be more aligned, for example, with a sort of security agenda or something like that. Great. Malali, would you like to add? Okay, yes. I think Claire has clarified, I think, on the issue that Bea has raised. I think the...

I think in my country, for instance, I think this issue of biometric as well as e-payment platforms is very important. I guess, like I said, it's really what governments also want to achieve, what banks really want to invest in because they look at their clients. So when it comes to digital IDs, they find maybe the e-KYC is much easier and they can put up the money because that's the clientele which they have and not necessarily

the people in primary schools. So really the emphasis is like that, but also I think on the political side, the political leaders are also looking at votes and who votes and what age do people vote and that's where they want to make actually the investment themselves. So really

I think it's convenient for everyone to choose that thing. But I think, as Claire said, we still need to have systems which are digital even in primary schools so that we capture everything that we have and know who our citizens are and have all the aspects which we need to have.

Great. Natalia, perhaps on the last question on AI, is it just large language models? What is this AI beast? And how does that shape how we should think about the opportunity and the challenge? I'll get to that. But just a quick comment on the previous question, which I think is also relevant for this question, is that governments often just look at where they can save the most money with technology.

as little investment as possible. And I think this sometimes determines the areas that are being automated. And sometimes that's education, sometimes that's something else. So I think there is also a very pragmatic, very practical, how do we save money and time? And this relates to the last question, because again, if there is a way to have perhaps a less complicated, meaning cheaper,

AI or algorithmic tool, then that's something that's more likely to happen. And I think in terms of is it LLMs, is it a machine learning, is it any other

I think that really depends on what we're talking about, which government, what kind of context. I think it's quite hard to answer this question in the abstract because the definition of AI changes every season. So when I started in the field of AI ethics, we actually didn't really have that much machine learning to begin with. And what was being called AI was...

very elaborate spreadsheets or just very sort of algorithmic calculations that was AI back then then we moved on to machine learning then this became AI and all the different programs for example image analysis or image recognition or any kind of

biometric related processing as well. So then that was AI and I think now LLMs are the new AI in town. Next, I think agentic AI, but I would say the definition really depends on where we are on that journey. Where we are right now, I think we're still deep in the LLM space.

But where the LLMs are being used in the public sector, that really depends on the country, that really depends on the departments. I wouldn't want to, since we're talking about data, to throw in some sort of rough estimate because I don't actually think we have that kind of information at the moment.

So AI is just linear regression. Any further questions from the room? Yes, we have a gentleman here in black. And while the microphone is coming to him, I will read two questions from online. I will combine them. So Maria, who's an LSE alumnus,

working in the public sector. Maria is asking about the importance of data visualization. So we've talked about all these fancy tools, but how much value and what can we do to improve data visualizations to engage the public and for policymaking? And I'll attach Irene's question. So Irene is a visiting fellow at LSE.

And Irene is highlighting that we haven't discussed very much about data sense-making, or if I understand what Irene's asking, kind of teaching people to make sense of data, and perhaps these same visualizations that might be produced as per Maria's wish. So how can we encourage a better data sense-making understanding of data, and what specific things have you found that might be effective? So that's the first question, and then we'll hand over. Yeah.

First off, I'd like to thank all three of you for coming out to speak. It's been incredible to hear your thoughts on sort of how we use data in the world today. My name is Sid. I'm an exchange student here from the Elliott School at the George Washington University. My question largely pertains to the sort of rhetoric we have around AI and data. There's

and sort of relates to some of the other questions that other audience members have asked. There is a sort of fascination with the idea of artificial intelligence that has essentially turned it into a buzzword where we use it to describe LLMs and machine learning and just general algorithms despite the fact that YouTube's recommender system has existed since the early 2000s.

My question is sort of twofold. One, is this a problem? Is this something that we need to change in terms of how we discuss the usage of data and the sort of nuances within it? And if it is something that needs to be changed, how do we make that sort of rhetorical shift without necessarily losing the

the storytelling power that comes with treating it like the narrative it is, like saying we should use AI to create all these incredible potential tools for development. Great. Thank you. I think we have space for one more question. Yes, just right here.

Thank you very much. My name is Veronica, also LC alumni, now working at Smart Data Research UK, a recent UKRI investment in the use of smart data, which is any kind of data that comes after the interaction with devices. We have a sister program called Administrative Data Research UK, which is the same kind of environment to capture data for research and therefore for better decision making. This week, the new data use bill was announced, soon to become an act,

So I would love to hear your perceptions of this is go to the right direction in this tricky relationship between companies and data. How do we capture this data that companies hold for public good? That's something that we struggle every day in our program. How do we capture this data? How do we build partnerships and good relationships with data? How do we come with more carrots than stick with these new legislations? And I would love to hear your perceptions of this relationship and hopefully in the data bill soon to become an act. Thank you.

Great, thank you. So three questions. Data visualization, making sense of data, number one. Number two, the rhetoric around AI and data, do we need to change it? The pros and cons of the current narratives around it. And third, if you can comment on the data use actor bill, that would be terrific. If not, perhaps more widely on the role of ongoing legislation and what it's doing well, what it's missing, et cetera. I can take the second one.

Please go ahead. It doesn't need to be in order, so please start with the second one on the rhetoric. I think it is a problem, the narrative that we have around AI. And I think what we need is

a bit of a reality check on a global scale because i think ai has been described as this miraculous magical silver bullet solution to everything and and i think we're seeing a reflection of this in how we have ai companies just popping up every day pretty much and sometimes the impression that i'm under is that

Perhaps it's not problem driven, but it's solution driven. So we're coming up with this AI solution and then we're thinking, what is the problem that we can apply it to? And the thing that stops me personally in this hype, miraculous AI narrative is actually the environmental impact of it. The fact that this is something that we're not discussing as much.

But actually, AI consumes a lot of energy, a lot of space, a lot of natural resources. A lot of data centers need to be built. They need to be powered somehow. And I think...

if we take that whole picture of how we talk about AI and then what are the realities, the behind-the-scenes work that needs to happen for AI models to be used, then maybe we shouldn't actually be using ChatGPT to say, "What should I have for dinner today?" Because if you think about the amount of energy that's burning,

then really we can just make that decision ourselves and just leave AI to address some of the questions that human brain perhaps cannot tackle so effectively. So the way to get there, I think, is through being realistic about what AI can and cannot do and just being open about that it's not a solution for everything, that sometimes AI is not the best option. And...

And just maybe introducing some friction to this kind of narrative that AI will save the world, because I personally don't think it will. I think it has a lot of potential for good, but it also has a lot of potential for evil. So it all depends on how we use it.

Okay, well, let me have a go at the first. I mean, I think the data viz question is really interesting. I love a kind of data visualization. I think how data is presented is absolutely central to how it's used. You can't, you know, in a sense, we want to increase friction perhaps with AI, but we

must reduce any friction when it comes to sort of use of data and availability of information to inform people's decision-making and I think presenting data in a beautiful and easily understood format is Absolutely part of that the only but I would introduce into that which is not so much about which is not unique to data visualization But which I think is something that we often forget when we talk about data and it often is where

data where public trust is undermined and where we kind of lose people is the sense that all data comes with a confidence interval. Data is not real, the real world. Data is a representation of the real world. And as such, there's always an element of uncertainty. And I think we sort of confuse data with reality.

And therefore, when the reality turns out to differ, even in quite a small way from the data, it was like, oh, the data is rubbish. But it's not. It's just that we've misunderstood what it is. So I think, you know, that's a general point that we have to get across about all data. But data visualizations, you know, when they're done well, can sometimes reinforce that sense of absolute certainty about the data just by virtue of sort of what they are and what they look like. So...

I'm saying two slightly contradictory things, but I think somehow we have to do both of those at the same time. And then in terms of... I'm certainly not going to comment on the specifics of the legislation, but I do have noticed in the...

kind of nine or ten years or so that I've been involved in this world, that we have made what I think, you know, in my mind at least, is a really welcome shift from an assumption that any data sharing partnerships, but particularly public-private partnerships, have to be kind of separate. Each one has to be separately negotiated. We have to be super sort of cautious. You know, companies always have to be compensated and we have to, you know, to a sense that actually, no, you know what, some of this stuff

is really important in the public interest and it's okay for governments to just legislate on it and say, okay, these are the new rules now, so adapt your business models accordingly.

And I actually, you know, that's something that we really weren't, that wasn't, you know, that was very, felt very far away 10 years ago, but is not universally either in any individual country or within a country in any different sector. Of course, there's always going to be different approaches, but I have noticed a growing confidence among governments, I think, to sort of take legislative and regulatory approaches to data sharing. And to my mind, that's entirely welcome.

Great. Muthala, any thoughts? Okay, yeah, maybe just I think on data visualization. I think very important, of course, as government we have to communicate to citizens and how you present that data is very important. So in my country we've had to, and you know, you've got different strata of citizens. Educated ones, the others who are less educated, and how you package that is very important.

We've now started developing, even for the budget, what we call a citizen's budget. Basically, a whole budget is announced through a speech, maybe a two-hour speech made by the Minister of Finance with lots of information. But we tried to simplify that to the communities to try to say exactly what we're trying to say, but using just visuals.

Even numbers are really put in visuals and that makes it very easy, I think, for citizens to follow. So we've tried to use a whole range of how you actually communicate with the data. But I think the simpler you do it, the better, even for policy makers. And I think the traffic stop signs that we have, I mean,

Also just does help from a policy perspective to say, well, this is green, good. Well, not so good if it's amber. And red, perhaps I should focus much more my attention to try to fix things and to go. So I think visualization is equally important as Maria raised it. Great, wonderful. So we are at time. Allow me to close with a very short story from my experience at the IGC.

which I think captures a lot of the challenges and opportunities we might face. So in one of our countries, we spent five, six years meeting with many different officials trying to get access to some of their data to do what we thought was meaningful and helpful analysis. We sent letters. We met with the minister down to all the different bureaucrats.

And five years down the road, I go into a meeting and the individual asks me, so Tim, do you want the data in ASCII format or as an Oracle dump file? I was stunned. I was like, this is my moment. And I said, you know, anything, whatever you have.

And they responded, that's great to know, I have neither. So I just wanted to say there's this tension of all these on-the-ground challenges with huge technological advances, which is what we're trying to think a bit more about at IGC, and I think all these perspectives have been super helpful. So with that, let me start the round of applause. Can we all thank our panelists? Thank you.

Thank you for listening. You can subscribe to the LSE Events podcast on your favourite podcast app and help other listeners discover us by leaving a review. Visit lse.ac.uk forward slash events to find out what's on next. We hope you join us at another LSE Events soon.