Delivering Trustworthy Electoral Oversight Artwork

Living With AI Podcast: Challenges of Living with Artificial Intelligence

This podcast digs into key issues that arise when building, operating, and using machines and apps that are powered by artificial intelligence. We look at industry, homes and cities. AI is increasingly being used to help optimise our lives, making software and machines faster, more precise, and generally easier to use. However, they also raise concerns when they fail, misuse our data, or are too complex for the users to understand their implications. Set up by the UKRI Trustworthy Autonomous Systems Hub this podcast brings in experts in the field from Industry & Academia to discuss Robots in Space, Driverless Cars, Autonomous Ships, Drones, Covid-19 Track & Trace and much more. If you have any questions or suggestions for future topics, please email us at contact@tas.ac.uk - You can find out more on our website at www.tas.ac.uk#LWAIP #TASHub

All Episodes

Living With AI Podcast: Challenges of Living with Artificial Intelligence

Delivering Trustworthy Electoral Oversight

July 10, 2024 • Sean Riley • Season 4 • Episode 7

Democratic elections should be both free and fair, but also that citizens believe that the elections are free and fair. It is also essential that the institutions in charge of managing elections are trusted in the delivery of these goals.

Regulators struggle to keep up with technological developments, and many are under-resourced and time-poor. The integrity of elections is under attack, so a focus on getting these things right is more important than ever.

The "Delivering Trustworthy Electoral Oversight" project investigates a natural language processing model that can classify and analyse election spending returns in near real time. Without good quality information, concerns about electoral integrity foment so this project aims to address that issue.

Project website: Delivering Trustworthy Electoral Oversight – UKRI Trustworthy Autonomous Systems Hub (tas.ac.uk)

Podcast guest:
Sam Power, Senior Lecturer in Politics, University of Sussex

Podcast production by boardie.com

Podcast Host: Sean Riley

Producer: Stacha Hicks

If you want to get in touch with us here at the Living with AI Podcast, you can visit the TAS Hub website at www.tas.ac.uk where you can also find out more about the Trustworthy Autonomous Systems Hub Living With AI Podcast.

Podcast Host: Sean Riley

The UKRI Trustworthy Autonomous Systems (TAS) Hub Website

Living With AI Podcast: Challenges of Living with Artificial Intelligence

Season: 4, Episode: 7

Delivering Trustworthy Electoral Oversight

Episode Transcript:

Sean: Welcome to Living with AI, the podcast from the Trustworthy Autonomous Systems Hub. Today we’re going to be talking about one of the task projects, this is an episode entitled Delivering a Trustworthy Electoral Oversight. I’m your host, I’m Sean and we’re recording this today on the 25th of April 2024. So, if you’re listening to this years in the future and the Robot Overlords have taken over then you know what’s going on and why we tried to make the elections free and fair.

Our guest today is Sam Power. Sam, could you introduce yourself please?

Sam: Yeah, sure. So, my name is Sam Power. I’m a senior lecturer in Politics at the University of Sussex. And I study all things to do with elections, political financing and the way in which we can perhaps make our elections function slightly better than they do already.

Sean: Brilliant stuff. Well, thank you so much for joining us today on the podcast. Obviously, the overarching theme of the podcast is about AI and trust. Where, somewhere, where trust is so, so important at elections. So, tell us a bit about the project, then. You know, what was the problem you were trying to solve, I suppose.

Sam: Yeah, so, you’re right in the sense that- One of the primary things we’re concerned about as political scientists, as political researchers is trust, public trust because democracy depends on the public believing that politics works for them. And that the democracies that they live in, if indeed they do live in democracies, and are free and fair. And there’s many ways to think about this, but one of the key ways is at elections.

And if you live in a system where you don’t think the elections are being run in a free and fair way, and you don’t trust in the electoral process you can start having pretty big societal problems pretty quickly. Indeed, even if you live in a dictatorship, and in many dictatorships they have elections from Russia to North Korea you have elections. Now we wouldn’t call them free and fair elections but they have elections. And one of the problems if you’re a dictator, and you hold an election, and if you rig it slightly too much, then people can really start having problems with the fact that these elections are not fair and they can start an uprising.

That said, in democracies, in very well established liberal democracies like the United Kingdom this problem of trust can be a real issue because if people don’t think that elections work for them, if people don’t trust in the electoral process, then they might not necessarily take to the streets. There might not be civil uprisings, but what you can get is significant levels of apathy. Turn out can go down. People can just switch off from politics altogether. And at that point maybe extremists can start getting a foot hold in the political system, but also politics stops working for people because people just stop being represented, and stop feeling represented.

So, what our project was trying to do was trying to think about whether we can use certain AI techniques to help electoral management bodies. These bodies that are set up to ensure that elections are free and fair. And whether we could use some of these advances in technology to both help them ensure that these elections are free and fair, but also that people would see that some of these new and novel techniques that might be used, can be used for good. Can be used to benefit politics, because there’s a lot of concern about AI and its effect on elections, but if we can show perhaps some case studies where it can be beneficial, perhaps we can, not necessarily improve trust, that’s the dream but not necessarily cause a decline in trust in the political system and in elections in particular.

Sean: Well I think when we think about this and trusting elections we’re often perhaps thinking about the count and who voted and, you know, there’s all sorts of kind of conspiracy theories about dead people voting and all this side of it.

But you had a focus on spending as well didn’t you, electoral spending. And if we go back a few years and the Brexit referendum I think there were some sort of large cases to do with Vote Leave and their spending, and some questions about that. How did you approach this in terms of the project?

Sam: Yeah, precisely. So, our project is looking at a very, very specific element of the electoral process. And, actually, the way in which we oversee elections, particularly in the UK. And public concerns around that specific element and what it is, is spending.

So, in the UK you can donate as much money as you like to political campaigns. There’s no cap. If you’re very, very rich and you want to give a political party five million pounds you can do that and people have. We can park that as whether that’s a concern and talk about that at any other time.

What we do have is quite strict spending limits on what money you can spend at an election. And, indeed, you know a lot of reporting around the way that money is spent at elections, and significant levels of concern about certain campaign activities. So, you’re right in the 2016 referendum there were concerns about the way the online advertising functioned, micro-targeting in particular. But more broadly there can just be wider concerns about where the money that people donate to political parties, where and how that money is being spent.

And one of the principles that we have in our electoral system is one of transparency, that a way to alleviate that concern is that people can see precisely who is spending money on what, when and how.

The problem that we have is that there’s quite a lot of money sloshing around in the system and it’s quite hard to both provide people with the information of who is spending money on what and how. And it’s quite hard for the electoral commission, who are the body that oversee elections in the United Kingdom to track where the money is being spent in a way that does not comply with electoral law. And it will take years, frankly, to actually uncover any wrongdoing.

So, you talked about Vote Leave there, that was 2016, the referendum happened, the result happened and then there were questions about certain ways in which money was spent and money was donated that were uncovered in 2017/2018, long after the result had happened and long after the processes to enact the referendum were in place.

So, the problem that we have, or the problem that the electoral commission has is (a) how can they create some kind of very clear information ecosystem, if you will that provides voters and the general public with information that is useful about what spending is going on. And, above and beyond that, can we create a system in which the electoral commission can start to track wrongdoing and misbehaviour as an election is occurring rather than a year after the event.

So, what we think and what we are doing on our project is trying to work with tools that can, at least, you know, I suppose- Not the short term but our modest goal is to be able to create a system in which we can better categorise this information and present it to voters using AI techniques. And our more ambitious goal is to move towards a process in which real time analysis of election spending can occur, such that the electoral commission, as any- I’m going to be diplomatic here, as any non-compliant behaviour might be happening, if they can sanction and prosecute that as it occurs rather than well down the line after an election has happened, and the dust has settled.

Sean: I think that’s really interesting. It reminds me a little bit of- I’m going to mention football here, I’m not a massive football fan but, you know, you have these questions about questionable, maybe a referee or VAR decisions on the pitch and maybe they get questioned after the match but the match result is not changed is it. And that’s the same with the election isn’t it. So, having something that’s happening in real time sounds like a really important kind of approach. Is that something you’re going to sort of work on going forwards?

[00:09:43]

Sam: Yeah, precisely. So, the aim of the project is to really bridge that gap. Is to, ideally, get to a situation and to use these tools such that rather than complaining about the result of a match that happened last season, that seemed to have an unfair outcome, to actually be able to do it at the time. And to make a judgment on it at the time. Because this is a problem for not just the UK, but democracies the world over. So, ideally the tools that we work with and that we build will not just be of use to the UK. We’re working with the UK as a case study and we’re presenting our work to the UK electoral commission. But the dream, if you will, is to come up with some kind of really useful tools that are applicable in lots of different countries because, as I say, this is a democratic issue. This is not just a UK issue.

To return to your football analogy, it’s not just in the Premier League that people complain about unfair decisions after a game has happened and nothing can be done about it, it happens in every league across the world.

Sean: Yes, absolutely, grassroots and up, absolutely. So, something I was thinking about, obviously, you’re talking about using AI to inform some of these projects. And, obviously, we have a problem with kind of bad information going in causes bad information coming out, like this so called garbage in, garbage out. Is that something that you have encountered while you’ve been working on the project?

Sam: Yeah, in many ways, actually, this is what inspired the project in the first place because- So, I’m a political scientist, I’m not a data scientist, I’m not an engineer, I’m not a computer scientist, we have those people working on the project. So, initially my thought was oh well what I want to do is track election spending, for example, and then think about the different AI tools that might be applicable, long before we started this project.

The initial project was actually one in which me and a research team went through invoices that were presented to the electoral commission at the 2019 general election. If you spend over £200 at a general election you have to provide an invoice of what that spend was on. So, I think we sort of lost our minds. It was during COVID so everyone lost their minds a little bit. We decided to look at every single invoice, there’s about 7500 of them, just to see what the invoices looked like and to manually categorise the different types of spend that were happening. And what we found was, as you say, actually quite a lot of garbage information, frankly.

A lot of the invoices you just couldn’t tell what the spending was done on. They either were devoid of information, as in they were blank. They were either so blurry that you couldn’t read what was written on the invoice. Sometimes they just had a very, very general description. So, you know, you’d have an invoice for £500,000 or £50,000 and it would say for spending on the general election, which doesn’t really tell anyone anything about anything, right. And if we have a system where, in theory, it’s transparent, in practice it’s not.

In the first instance we do actually have a genuine compliance problem here. And what we have is an electoral commission that quite rightly doesn’t have the time, perhaps, to do what we did, which is a four person research team over a year and go through every invoice forensically.

So, in the first instance what we thought that any kind of AI tool could do, would be a very simple oversight mechanism, where it would try and read information and be able to actually seek out the garbage information. So, you get kind of the garbage in and then a red flag comes up, pings over to the electoral commission and it says this invoice that in theory is supposed to say something is blank or it can’t be read by a machine, maybe it can be read by a person, you might want to have a look at this or maybe you need to get in touch with these people and they need to provide us with more or better information.

So, the very first inspiration of it was to just create some kind of quite simple extra cog in the machine to try and filter out the garbage information that we were seeing. And it was a significant amount of garbage information.

So, we wrote a number of academic articles and more policy focused articles about this initial project. And we showed that about £1 out of every £10 spent at the general election, slightly more than that was, at least in terms of the [s/l stand 00:15:22] that was invoiced, garbage information couldn’t work out what it was.

Sean: I can imagine like this, you know, £50,000 invoice saying sundry or general expenses. And, I think that’s fantastic that you can potentially flag that up for someone to answer that question straightaway and say, right, what was this.

But, also, just going back to what you said about the kind of over £200 we saw a lot of this kind of micro-billion for things like social media advertising. So, I suspect there’s all sorts of loopholes people exploit, but, you know, lots and lots of under £200 bills. I’m again imagining the worst and thinking of lots of £199.99 invoices going through, is that something you spotted or- ?

Sam: So, one of the challenges of social media, at least in terms of election spending is how cheap it is. So, it might not even be, I suppose what you would call, let’s say malicious compliance in the sense that you get a lot of invoices for under £200, I mean you do get a lot of spending under £200 and you do get a lot of spending on social media under £200, but it is just the fact that it is so cheap.

The reason why I think political campaigning has moved online to such a degree is precisely that it’s a very cheap and efficient way of campaigning. Not necessarily the most effective way of campaigning but it’s incredibly hard to get enough people to knock on every door in every constituency in the country. So, if you can have a relatively sophisticated targeted campaign at very specific kinds of people that you think you need to turn out, and it’s going to cost you a tenth, if not less, of a poster campaign, of leaflets, of, you know, volunteer campaigners that you don’t have to go and knock on doors, then it’s inevitable that you end up doing that.

And to return to your point about the invoices that aren’t very clear, this is again something that we want to work with the electoral commission to move towards, although, you know, it’s perhaps more lobbying on our part because we think it would be a useful thing to start considering, which is actually perhaps moving to some kind of standardised form of invoicing so that (a) lots of invoices look the same, and they can actually be easily read by a machine, for example. And, (b) you can begin to sort of self-categorise activity.

And we don’t think that this is an overly ambitious aim, because a lot of the things that we do- I’m not going to call it our everyday life, but in life, has this kind of semi-automated aspect to it. The example that I always use, because I have two children who are relatively young, so it’s the only thing that I talk about these days, is- One of them is 12 weeks old, one of them is about two years old, and we both- About two years ago and about a month ago applied for Passports for these children. Quite hard to get a good photo of the baby that is Passport compliant. But when you do it, you upload your picture, a little sort of clock turns up on the Passport website and it says this photo is no good take another one. And it takes you about an hour because you need to get a good photo of the baby and it’s incredibly hard. But eventually you get there and it says this photo’s good. Upload it, your Passport is going to come in a month, however, long.

So, I don’t see any reason why that sort of system, which just says actually this invoice is no good it can’t really be read is the first step in an online system, and then perhaps you have little drop down menus that say, we’ve got all these different categories, we think this is this category are we right, if not, why not, what category is it out of this drop down menu. And it just makes it easier for campaigners to upload an invoice and also much easier for the electoral commission to kind of auto-sort this spending and invoicing as we go along rather than doing it all after the event.

[00:20:07]

Sean: And, obviously, yes, and then flags up any anomalies for someone to actually take a good look at. I mean, I run a business, I use a similar sort of system for my expenses, you know, you take a photo of a receipt, it gets analysed and if it can’t read it properly it’s flagged up. Is part of the problem here getting through kind of layers of bureaucracy, because the tech side of it feels like it’s pretty straightforward. But implementing it and actually getting it to be used, that’s surely the bigger problem, isn’t it?

Sam: Yeah, so there’s two primary problems. Bureaucracy is a good way of putting it. Yeah, so it’s bureaucracy but it’s, I suppose, more to the point it’s a political reform, effectively. So, when you are engaging with political reform, it’s necessarily slightly more sensitive because it involves politics and it involves our democracy. So, you really do need to think about every element of the reform that you’re doing and you need to persuade people that it’s the right thing to do and that it’s a priority at that time.

So, there’s an element to it that it’s not just a bureaucratic issue but it’s a political issue that any sort of large scale major changes to the way in which we report electoral standing, for example, that probably needs legislation. So, it’s not just a change that can be implemented easily because enough people think it’s the right thing to do, you need to convince the right people that it’s the right thing to do and then engage in that reform process.

And, you know these things are- There are a lot of considerations that need to be taken in to play. So, I don’t think it would necessarily add to the administrative load, if you will, of political parties and campaigners, but it’s really important that it doesn’t because one thing that I think people don’t necessarily understand about our elections is that even at the highest level of political parties these things- And campaigns are often run by volunteers, who are engaging in the democratic process, who might be the treasurers of their local party. Who, maybe, you know, were the last ones sitting down when the music stopped. And they’re doing good work and really engaging with our democratic process. So, we don’t want to- Well we certainly don’t want to criminalise them, and we certainly don’t want to add to their administrative load rather than remove it.

So, that is a key challenge that we engage with, in implementing it, it seems relatively simple and it seems perhaps even relatively common sense, but firstly, you know, you’ve got to create the legislation in the first place. And then, secondly you’ve got to ensure that the reforms work for, particularly, volunteer campaigners.

And then, above and beyond that we’re talking about trust here, right, and the project is called delivering trustworthy electoral oversight. You’ve got to bring the public along with you as well. If you start talking about the automation of oversight of elections, that’s probably going to raise some alarm bells. And trust and perceived trust is as important in this process as in any because even if the process is perfectly sensible and works fine, but if people don’t believe that it does then it’s no good because this is elections that we’re talking about. So, all of these things are a challenge to be considered.

Sean: Yeah, absolutely, I mean, you know in this post-truth world we need to be whiter than white. But also you mentioned those volunteers, ideally you want to make their life easier. If this can help them do their job rather than- And also then adds in that layer of accountability then that’s a good thing surely.

Sam: Yeah and that’s the aim. And I still believe that- I suppose if you introduce a new system there will always be a slight challenge that, well it’s not the way that we’ve always done things. So, there’s always a sort of- I am as guilty of this as anyone an inherent small “C” conservatism of but I quite like the way that we do things even if it’s maybe massively inefficient, it’s just the way that I’ve always done it. But, if you do it and you do it correctly then it can ease the burden. If it’s just a case of having this really, really well functioning online system that then perhaps- That once it’s done, it’s done, I can see that it would ease the administrative burden rather than add to it.

Sean: Is the project still ongoing or have the conclusions been drawn, you know, tell me a bit about that?

Sam: The project is ongoing. So, what we’ve been doing over the past year is developing these AI tools primarily around machine learning and natural language processing. We’ve been taking all of the invoices, both the ones that we have already manually coded and taking them from different elements of the database to improve the model to create a level of confidence that we are happy with that some kind of auto-sorting categorisation model works, that this AI tool works. And at the same time thinking about, and presenting, the different applications of what this model may do to the electoral commission.

So, the project is at the stage where we’re quite confident in the tool, we need to continue to improve it with various annotations and continued processes of improving the tool. But we’re quite confident with the broad architecture of it if you will and we’re confident that we’ve got a number of pretty good suggestions to move forward with the electoral commission.

So, this is the TAS Hub project, many members of the team have also got some funding from Responsible AI UK, to conduct an international partnership, particularly, with scholars at Johns Hopkins University in the USA, just outside Washington DC and their applied physics laboratory who have been doing similar things with the USA Federal Elections Commission database.

So, we’re pooling that knowledge at a series of workshops, the first one will be in Washington DC, to think about how we can build out just from the UK, context. The primary challenge there, we’ve been talking about challenges, is we think we’ve got a tool that works and we think that we’ve got a tool that works for other Anglophone countries. We’re not quite sure what to do in countries where people don’t speak English, and there are a lot of countries where people don’t speak English. So, we’ve got people coming over from Brazil to that workshop who have a very similar, at least, system where- A transparency system and are working with these kinds of tools as well.

So, what we’re trying to do is create an international community, if you will, with a follow on project to really push this beyond just the UK context. So, we’re quite confident in the TAS Hub project that we’ve got something that works with a very specific element of electoral oversight. We’re then trying to build that out internationally. And then think about wider parts of the democratic system as well. So, are there ways in which AI can be beneficial in different elements around democracy.

Another challenge that we see is- There’s a resourcing skills gap in terms of actually ensuring that the tools that we’re creating can be embedded and can be embedded in the long term, because they do need people with specific qualifications around natural language processing, but around much broader qualifications in data science. And with the best will in the world governmental organisations can’t pay as much as the private sector, by, you know, about five times. So, it’s actually very hard to get the talent to engage in the kind of work that we’re suggesting.

[00:29:59]

Understandably, people might well believe in the democratic system, but if they can’t get paid enough and they can get paid significantly more elsewhere, it’s actually very hard to get a significant pool of people with the right skills in the right places. So, one of the things that we’re thinking about and trying to work with is how can we either (a) convince them, (b) get the resourcing there, but both of those things are incredibly hard, so, (c) is there a more collaborative approach that can be taken. Is what we need a much more significant collaboration between academics, between the private sector and between government and governmental organisations. Because, you know, our team have the skills, and I use our team there not me, our team have the skills which can be implemented and we’ve been working on it with good funding. And there’s certain incentives for us to do that, you know, it’s good to get funding if you’re an academic, it’s good to do good work. So, you know, we can be persuaded to do something like that.

Is there a case to then really build out and start working with industries. Start working with people like Microsoft and Facebook and incentivise them to help with some of this stuff and to second people to some of these organisations. And then can we get people like the electoral commission and wider government on board in this collaboration. So, that’s, I suppose, a bit of a pipe dream, but it’s a real challenge that we’ve seen. And it’s something that’s hard to overcome and could actually, I suppose, in many ways cripple any ideas that we have if you don’t have the people to actually implement them in the places they need to be.

Sean: Sam, I really appreciate you sparing the time to be on the podcast today. And if people do want to collaborate, if Mark Zuckerberg’s listening then I’m sure we’ll be able to put all the details in the show notes for you.

Sam: Yes, well, thanks so much for having me, it’s lovely to talk about this kind of thing. And, yeah, my whole thing is I want people to work with us on this. I want people to get in touch and I want people to collaborate. You know, we’re not precious about what we do in that sense, it’s not a closed shop. So, if anyone listening, from Mark Zuckerberg to, you know, someone in their room that finds this kind of thing interesting and think that they can help, please do get in touch because I’d love to hear about it and love any help that we can get.

Sean: Brilliant, thanks again, Sam.

Sam: Thank you.

Sean: If you want to get in touch with us here at the Living with AI Podcast, you can visit the TAS Hub website at tas.ac.uk where you can also find out more about the Trustworthy Autonomous Systems Hub. The Living with AI Podcast is a production of the Trustworthy Autonomous Systems Hub, audio engineering was by Boardie Limited and it was presented by me, Sean Riley.