Podcast - How Meta Manages Misinformation in a Rapidly Changing World

Welcome to Beyond the Benchmark, the EFG podcast with Moz Afzal.

Moz Afzal:
Hi everyone. So a very special podcast Today I have Tom Bonsundy-O'Bryan. Tom, welcome.

Tom Bonsundy-O'Bryan:
Thank you so much. It's great to be here. Thank you for having me.

Moz Afzal:
Yeah, so for those of you who don't know, we actually missed Tom at our investment summit earlier this year because he had a family emergency, but glad to say everything was sorted out and things are on the up and up, so that's very, very good. So Tom, over to you. Just tell us about your history. How did you come into being a Head of Misinformation Policy for Europe, Middle East and Africa at Meta?

Tom Bonsundy-O'Bryan:
Yeah, thank you so much. I mean, there's probably a two hour version of the answer. I'll try and give you the two minute one. So yeah, I'm Tom, as you said, I'm the Head of the Misinformation Policy team covering Europe, Middle East, and Africa, here at Meta. Part of the team that advises company leadership on how we respond to address misinformation, sets the overall company strategy for how we tackle this. Very interesting but challenging topic as we'll talk about today. I joined Meta and entered the tech world going on five years or so. Now, more or less when Covid-19 pandemic was starting up, my background was not at all in technology. I was a diplomat in my past professional life. I worked for the UK government and the Foreign Office and with the United Nations as well. Had the privilege of working in some fascinating countries in Asia, Middle East and Africa, mostly countries which were experiencing violent conflict, civil war in one way or another. And I suppose I came to this job seeing that digital technologies were perhaps spreading in some of those countries slower later than they had here in the UK, here in Europe, and seeing some of the challenges associated with misinformation spreading online too. So stepped into this role to kind of work on that challenge directly from Meta or the Facebook side of things as it was then.

Moz Afzal:
So maybe a bit of background in terms of the number of people that work at Meta in terms of tackling misinformation and being, I guess the barrier that doesn't allow such information to come to the general public.

Tom Bonsundy-O'Bryan:
Putting an exact number on that is quite difficult. We have one 40,000 people working on what we might call online safety issues around the world, and we spent tens of billions of dollars on this issue globally over the past few years. We have all sorts of people with all kinds of very interesting different backgrounds from all around the world working on this stuff. People from Africa, from Europe, from the Middle East, from Latin America, from North America, you name it. More or less every part of the world. There's someone at this company joining hands working on this stuff. It's important to say, and we'll get into this, right, that there are also important limits that we put on our own role in tackling misinformation, funny enough, and we work with a wide array of external experts and stakeholders who are really on the front lines of tackling this stuff as well. So it's not just about who is sitting from the Meta side working on this problem.

Moz Afzal:
Yeah, it's a fascinating area and something that we read a lot about and obviously a lot of that information's kind of jumbled up with lots of other conjecture that comes along. I guess how do you find that balance, which I find quite interesting is I said how much information, I guess if you limit information, it is also I guess against the general principles, right? So finding the right balance is actually quite difficult and I guess you can never win right?

Tom Bonsundy-O'Bryan:
I think it's a fascinating topic. I mean, look, truth and falsehoods are concepts which predate certainly all of us and the internet and digital technologies and probably exist since the beginning of time, right? Humans have lied to each other or said things which aren't true for a variety of reasons, often by accident, repeating things that we thought were true around the cave fire or whatever. To take an example, I live in a town called Woking just outside London, and it's famous because the book War of the Worlds by H.G. Wells was written and set their stories about martians, aliens landing on planet Earth and it was serialised and broadcast on radio in the States actually. And folks at home listening in got really freaked out when they came home from whatever they were doing. This is the pre TV era, and heard this very realistic reenactment, not that they knew at the time it was a reenactment of aliens landing on planet earth and some people thought it was real. So with each kind of era of technologies all the way back as far as radio, and probably before it, this question of information, truth falsehood, the vast kind of grey zone again that we might get into today in the middle has been there and there isn't a single silver bullet solution either, right? Different stakeholders all have a role to play in combating this, and we have our role too, but no one has yet found a definitive solution, right? It's by definition, hard.

Moz Afzal:
Yeah, I suspect we never will. I think it's just one of those things where it is human nature and human being, right? It is what life is about. So let's go into a little bit more in terms of your role and exactly what you do. Maybe we step back a bit and say, what is misinformation? What is fake news? How do you define it? And of course, what type of risk does it post to us all?

Tom Bonsundy-O'Bryan:
Yeah, so we kind of think about it. There are these two terms that get thrown around often interchangeably in English, and they often don't translate perfectly into other languages. Some languages, I'm currently in Spain, like Spanish really only has a single term for these two concepts, but in English we tend to talk about misinformation and disinformation. And on the surface these two ideas look and feel similar. It's information fundamentally that is not true. Misinformation we tend to think about as being something that maybe someone shared without knowing it wasn't true. It was said that wasn't true, but they had no idea. They just thought something was interesting or funny or shocking and they said it to their friend or sent it in a group chat or whatever. And anyone can do that. I've done that. Even in this job. I make mistakes. Sometimes disinformation is quite different.

Disinformation we tend to think of as the intentional promotion of narratives or claims that aren't true often in a coordinated inauthentic way. And as a company, Meta has quite different approaches to those two problems. When we find disinformation networks, a network of accounts that are working in a coordinated way that all have fake names, fake pictures, obviously deceptive and are acting in that way, we remove them entirely, remove the entire account and the content that goes with it. Misinformation, the slightly more nuanced challenge of people getting stuff wrong sometimes and sharing it, but still potentially misleading other people is more complex. It requires a slightly more nuanced solution, and it's really that latter problem that my team and I work on.

Moz Afzal:
And just to go back to the former, in terms of disinformation, I guess for the uneducated around this sort of work, obviously you have kind of automated bots that just sending information. Does that information come from anywhere? Obviously we always read that it's coming usually from China or Russia or some of the countries that won't necessarily or use it in a more, let's call it a nefarious way very simply. Is that typically what you see or really it's just random kind of propaganda that runs through the internet and the internet systems?

Tom Bonsundy-O'Bryan:
Well, yeah, I'd recommend that folks check out our quarterly adversarial threat reports, that public transparent resource where you can see the entities of networks that we've removed for violating those policies. And I think some of what you say rings true, but it's also slightly more nuanced than that. I think transparency obviously is really important for people to have trust essentially. It's important for us that when people are using Facebook or Instagram or WhatsApp or Threads or whatever it might be, that they feel those places are spaces where they can share their views and opinions and information and at the same time that they can trust the information that they're seeing and reading there. It's also important for our business model, I sometimes get questions, accusations charged me that the company has some incentive in allowing misinformation to be spread on the platform. Fundamentally, this is a business which runs on revenue from advertising. And I can assure you that advertisers of all shapes and stripes absolutely don't want to see that advert next to that type of content. So we have a really strong business imperative both to be transparent about that and again, record those reports which are worth looking out, worth having a read for folks who are interested but also have clear policies.

Moz Afzal:
I mean, certainly that's something that I'll also take a look as well. So moving then onto your direct double clicking into your direct area, which is obviously misinformation. Maybe just go into a little bit more detail. What does your day-to-day work actually involve? Because shades of grey, it's very difficult for, and we've all done it as you said, I've done it many times where you share something and you thought afterwards, yeah, maybe I should have checked that source. And it's very easily done, right? Because everybody does it and sometimes people just do it automatically as you see, as you said earlier, something funny and they just send it on and don't really think about, well, was that real or not? Or even thought twice about it.

Tom Bonsundy-O'Bryan:
Yeah. And like I said, I have done that myself. It is so easy to do. We're all on the go or on our phones sometimes you don't click through or you share an image or a video that was sent to you by someone that you trust and pass it along. And it turns out not totally legit, right? It's so easily done, which is also what makes this a kind of interesting problem. Weirdly, I'm actually going to answer your question by kind of flipping it and telling you what I don't do is decide what's true or what's false. And it's really important to say that as a matter both of principle and practicality, no one at Meta is playing that role. We've consistently had feedback down the years from folks all around the world that this is not an appropriate role for any one single actor, let alone any one single company to be playing to effectively be policing what is true and what is false on the internet.

That is an enormous power and responsibility and not one that we hear folks saying they want us to hold. So that is not my job deciding what is true and what is false. My job is kind of writing the company's strategy to say, well, how then do we deal with this problem? And we have a three part strategy that my team and I are responsible for and we call it remove, reduce, inform. Three pillars. The first one remove. So we'll take down content entirely, remove it from our platforms where there's misinformation that could directly contribute to the risk of imminent physical harm or violence. So we see misinformation about a whole bunch of innocuous stuff. Is Lionel Messi going to be joining Arsenal? I don't think so. I don't think that's true. You didn't hear it here first. It's hard to see a connection to those kind of consequences I just described, but think back to some of the countries that I lived and worked in Afghanistan or Congo or Israel and Palestine, and you can see how in a context like that, certain types of misinformation, certain types of false claims, there could be that kind of risk.

So we remove misinformation that hits that bar that could directly contribute to imminent violence or physical harm. And we work with a global network of independent experts from those communities who help us to understand, to distinguish between the fake football transfer rumour, the kind of what is the innocuous less harm risk clearly versus what is the higher risk stuff. So that's pillar one, remove.

The second is reduce. So content, the independent fact checkers, we can say more about that in a minute outside of the US have determined isn't true content information, which isn't true. We don't delete that content but show it much lower in people's feeds, right? So let's say I share a piece of content that's been determined to have information in it that's not true to be misleading in some way. You could still go to my profile and find that piece of content, but you are vanishingly less likely to have seen it in people's newsfeeds.

And the third pillar is inform, this is really important. This is probably the most important of all three pillars. A label effectively on a piece of content that tells you the user, this piece of content contains information which might not be true effectively, words to words to that effect. And we're experimenting with different approaches to that kind of labelling and the ways in which it is determined that a given piece of content is true or not a different approach rolling out in the US very shortly, perhaps later in other parts of the world, world too. But the principle is the same and the underlying logic is the same. The way you combat misinformation is by helping people understand why that information is not true, a kind of remedy to that falsehood is true or more accurate context or information. And so in the rest of the world, again, outside the US, labels on pieces of content will take users of Facebook, Instagram, WhatsApp to content from fact checkers who said, look, this image has been shared. Someone says it depicts new electric buses in London. Random example, but actually it's not. This photo is from Paris. Those buses aren't coming to London anytime soon, et cetera, et cetera, whatever it might be. And so, sorry, slightly long answer that essentially is what my team does. We set the strategy for that and help with its operationalization too. Yeah.

Moz Afzal:
Okay. So now going into that fact checking versus particularly in an AI context where obviously a lot of content, big pictorial images, video, et cetera, can be doctor altered using artificial intelligence. How not do you actually check that through? What are the sort of technologies you're using to identify those and just recognise them, they are just kind of fake news?

Tom Bonsundy-O'Bryan:
Well, so I mean stepping back slightly, it's important to recognise that AI is an essential part of how you tackle these problems at scale. In a context where there's more than 3 billion people using these services online regularly. And obviously an enormous volume of content being shared in video or audio or pictures or text at any one time. AI is a very important part of how you action on content that violates our policies at scale. And again, we have community standards enforcement reports they're called that come out every quarter and they tell you how much content we've taken down across our different policies beyond just misinformation, right into areas like violence and incitement for example, or bullying and harassment and terrorist propaganda, dangerous organisations and individuals, whatever it might be. It also tells you the percentage that we removed by automation. And there I'm essentially talking about AI and across some policy areas, more than 90 or 95% of the content we've removed was detected by artificial intelligence.

Maybe at the moment someone posted it before it even went live. If we determined that video is the same one that violated our policy. So we've spoken about AI being a sword and a shield in the past in that clearly there are risks, but actually this is also in a very important part of how we keep folks safe online and enforce those policies. But for misinformation specifically, again, we've used AI for a long time to help tackle this problem. For example, when a fact checker determines that a given image or video is not authentic or contains misinformation, they can use artificial intelligence to automatically apply their label to other identical pieces of content for the sword side of things, for the risk, obviously last year was a very important year globally for elections. We saw an enormous percentage of the global population voting in elections and there'd been some kind of apocalyptic style warnings about what this new technology meant for the integrity of the way that we decide who runs our countries for the integrity of our democracies. Would we be able to discern fact from fiction with the walls, sort of melt around us not knowing what was real or false anymore.

And look, clearly people have used these technologies in deceptive ways in elections in the past, including last year, but that was a vanishingly vanishingly small proportion of even of the misinformation content that we saw across the major global elections last year. I think it was about 1% of the content that was fact-check, so that fact-checkers had determined was misinformation. Only 1% of that content was AI generated content related to elections or politics, a vanishingly small percentage of the overall content on our platforms. So I think that's one thing to say that it's not that that's forever going to be thus right? Things can change rapidly and dramatically, but to now that hasn't been manifested at quite the scale that some folks predicted. We have technical systems that we've developed in partnership with other industry stakeholders that help us detect, for example, when an image has been generated using artificial intelligence and when an image like that is shared on our platforms, automatically a little label pops up and is posted with the content on Facebook and Instagram saying AI info, just warning people that this content's been generated using AI, but it's very hard.

And then think anyone, we have to be transparent about that. Nobody yet has the perfect flawless technical solution to detect this stuff at scale. And it's a different technical challenge across image versus video versus audio. And people have different, again, sorry, long answer, but I think it's interesting, this whole new development of AI is throwing up interesting challenges in terms of what is useful notice for people. Initially when we launched this system of these automatic little labels for AI generated images, you think you'd catch some of that misinformation content and clearly it did. But it also picked up on people's images of their holiday on the beach where they, unbeknownst to them their iPhone was using AI when it was stabilising the brightness in the image. So what is useful notice if you have that AI info label on that kind of content, do people start thinking, did my friend not really go on holiday? They're not really on the beach. So we've had to calibrate and recalibrate a little bit with that. So where you draw the lines on this stuff, how you detect it, what is the right notice to give people, I think are ongoing dynamic challenges that the whole industry, all of us frankly, are going to face.

Moz Afzal:
And how much of, I guess the tools are you using, for example, Meta zone tools like Llama or any of these sort of things to work through? Is there a lot of internal technology or a lot of this stuff is actually bought in from outside?

Tom Bonsundy-O'Bryan:
Well, so like I said, for that work of saying what is an AI generated misleading video or audio or image that is work that fact checkers do, I'm not sitting here getting involved in arbitrating that process by design and each of our fact-checking partners, excuse me, have their own tools and ways of discerning that and editorial processes around that. They're independent from Meta. So we aren't directly involved in triaging that this automated system that I just mentioned for applying this AI info label to images that are shared on platform, we require folks to self-disclose, tick a little box when they're sharing an image or video that is AI generated, but for the one that appears automatically that is built using cross industry technical indicators essentially. So I guess the point to make is this is not something that we are doing or can do in isolation. This is a generational shift and a whole bunch of stakeholders from across society have a role to play in this. Yeah.

Moz Afzal:
Okay, great. So just stepping back a bit in terms of over the past, we've had a couple of wars, we've had terrorist attacks and so forth, pandemic over the last five years or so. Now, what are the lessons that you learn from each of those areas? How do you grow into those experiences and do you think you get better each time there's something that happens in another election?

Tom Bonsundy-O'Bryan:
Yeah, I mean absolutely. Like I said, we have tens of thousands of people literally working, especially during election time, just around the clock on these kind of challenges that naturally appear at those moments. And I think over many years and lots of practice, I've been at this company five years and have worked on more than a hundred global elections at this point. We have very well developed policies and processes and teams who are very good at responding to this kind of thing. On the misinformation side specifically, each election is different, each country is unique. I'd never want to suggest for a minute that it is the same challenges that appear, but actually I think our approach has worked well in responding to those challenges. And again, take something like AI, I think we were well prepared for responding to that last year, a very small percentage of the overall content on our platforms, a small percentage even of what got fact-check, but we have to keep an open mind for that learning. Each election, each crisis or conflict is unique and so we have to be ready for whatever comes up really.

Moz Afzal:
So in your job and what is your biggest fear? What is your personal biggest fear?

Tom Bonsundy-O'Bryan:
Oh goodness, that's a good question. I mean, in my personal life and professional life, I guess those are different questions. It is massively important to me on both those levels. I guess that we get this right, 3 billion people around the world use these services, these apps, which dating me slightly didn't really exist at the time that I left university. Just the global scale of these apps is really quite remarkable. And as I said at the beginning, places I have lived and worked all over the world, this is a very important tool service that people use in their lives in a whole bunch of ways. And so the responsibility to make sure that we get that response absolutely right, I think is something that naturally I'm very focused on. It matters to me a great deal that we get that right. And look, I've said today that I think we, we've had a great response to those things and I'm proud of the work that we've done.

The nature of this work is that naturally there will always be different viewpoints for any given policy, not just our misinformation policy, a fact checking approach or the community notes approach that's rolling out starting in the US today as we discussed, there will always be some people who think you're going too far and there will always be some people who think you don't go far enough, you're not doing enough. And in my experience that often doesn't land in a kind of typical or predictable way. Maybe one of the best things about my job is that I get to talk to a whole bunch of different people from all around the world, really, not just in the region that I'm responsible for mostly I suppose in Europe, the Middle East and Africa, and from Chad to the United Kingdom, to Yemen, to South Africa, to Lebanon, to France, to Spain. You'll hear very different perspectives from a whole bunch of different people on that question of what is the right way fundamentally to balance these two, often conflicting perhaps have always been conflicting in human history, values of safety on one hand and speech, free expression on the other. Those are two of our core values as a company guiding this work. And there will always be folks who disagree with where we land from a given piece of content to a policy overall. So those are some of the things that I'd say there that you fear and worry that people just get it all wrong and I guess alter the courses that you and others have to follow because of lack of understanding.

And look, I think we have work to do of course in helping people understand how we tackle these problems. I still often have that situation where I have conversation with folks and maybe they've used Facebook themselves or use Instagram themselves or whatever it might be, but learn something. When I share that insight into how our approach, how our strategy works, often the reaction actually I get is, oh wow, that is complicated. People think it's a question of just volition that if you really wanted to, you'd just be able to crack this problem. And when you I think walk through the practicalities of what it means to weigh those values, how do you give people access to free information to make sure that when they're using these services, they're consuming information that they can trust and rely on and at the same time, absolutely critically protect that space for free expression and opinions.

Moz Afzal:
Okay, so let's move on a little bit and then talk about you. And you've written this book, Football, War & Peace. Do you want to just go into a bit of the background to that and I guess some of the key learnings that you're portraying in that book.

Tom Bonsundy-O'Bryan:
And perhaps it's not quite as random a turn as it. So I'm a big football fan. I have been since I was a kid, but I'm a big Arsenal fan and as a kid, football and I guess Arsenal specifically were really the things that teed me up for a career in diplomacy and international affairs. I thought the way Arsène Wenger spoke Japanese and French and Italian and German was really cool. And when Arsenal had an away match in the Champions League playing Dynamo Kiev, I'd kind of say, oh, where is that? Football, actually, funny enough, was my first window on a world beyond the shores of the UK. And as my career sort of progressed and I lived and worked in these different places, I came to see that football actually plays a really interesting role right? In the sense that it's perhaps one of the single greatest sources of commonality, something in common that humanity has beyond maybe religion and some of those absolutely fundamental things.

Football probably comes quite quickly after in terms of just the number of human beings on this planet who are interested in it, follow it to varying degrees, have some affinity for it, et cetera. And so in some senses it has enormous potential to bring people together. Again, in some of the places that I wrote about in the book, Côte d’Ivoire, Ivory Coast, for example, football played an absolutely fascinating role in bringing together, and it helped that the Ivorian national team had some very good players like Didier Drogba, Yaya Touré and others, but they played a very important role as a national team in bringing together these warring parts of the country, north and south, Christian and Muslims together in support of this football team. Was it again, a silver bullet that single handedly ended the Civil War? No, but did it help? You bet. Did it have a positive impact? You bet, right?

 Look at other places like Lebanon for example. Again, another chapter in the book and football actually played a much more complex role, right? Football is inherently tribal, it's exclusionary. You are fans of one team and maybe you have another enemy team or team that you fight about. It is inherently a competition, right? There's a winner and a loser. It's not all kumbaya and hold hands and around the campfire, right? And in a place like Lebanon, a society with lots of different overlapping identities, I guess all societies do, but Lebanon in particular has many of them and a fracture history. You ended up with the Sunni team and the Shia team and the Christian and Mennonite teams, and for a long time during the Civil War, football matches were such a flashpoint for violence, for tensions between those committees that they shut the whole Lebanese Premier League down and then reopened it and kind of Covid lockdown like conditions with no fans in the ground. So the book really was just driven by that observation of the places I'd lived and worked in to say that, hey, this thing is very powerful, has such great potential to bring people together, but can also be this catalyst for some of the ugliest size of humanity.

Moz Afzal:
That's certainly very interesting and certainly well said. So maybe coming towards the end of this podcast, but there's something I just had to ask, and I'm sure many of our listeners have want to ask the same question. You could probably do a full disclosure as well. What is that most outrageous bit of fake news you've ever seen?

Tom Bonsundy-O'Bryan:
Wow, that's a good question. Wait, give me a minute to think about this.  I think I don't have a single wonderful example to come to mind. There's loads of them and I can send you some links for folks who want to troll through the archives. I mean, the thing that's challenging is that often people use, take satire for example, humour.

People say stuff that is literally not true, but it is understood by almost everyone as not intended to convey facts as intended to be funny. And so there's loads of examples of that, of satire and humour and people just saying dumb stuff that makes people laugh. That incidentally is like a real grey line. We don't allow fact-checkers to fact check satire, but that's difficult for them to discern what is something that is false literally, but intended to have comedic humour value and is actually a very important form of speech and voice, right? Versus something which maybe 10% of the population takes it literally and misunderstands it. I think it's actually the most interesting grey zone area. So I'll have to go and look up a couple of good examples there and get back to you.

Moz Afzal:
Yeah, I'm sure there's some probably really good football ones. I don’t know, a goalkeeper scoring a hat trick or something.

Tom Bonsundy-O'Bryan:
Sure. Yeah, there's definitely some good, I've given you one today, Lionel Messi coming to Arsenal, not true.

Moz Afzal:
We go. You never know. You never know. Well, Tom, listen, thank you very much for taking us through that. I think it's a topic that obviously is very, very important. It is very interesting in terms of how people perceive just not just the role of Meta, but others around it. It's a delicate subject. So I appreciate you opening up and giving us a little bit more information around it. And obviously we will certainly look out for those reports and understand them a little bit better than we already have. So thank you very much for doing that for us.

Tom Bonsundy-O'Bryan:
Thank you so much for having me, Moz. Appreciate it. Thank you.

Moz Afzal:
Thanks, so that wraps us up for the podcast today.

 

This podcast is provided for general information only and assumes a certain level of knowledge of financial markets. It's provided for informational purposes only and should not be considered as an offer investment recommendation or solicitation to deal in any of the investments or products mentioned herein and does not constitute investment research. The views in this podcast are those of the contributors at the time of publication and do not necessarily reflect those of EFG international or new capital. The companies discussed in this podcast have been selected for illustrative purposes only, or to demonstrate our investment management style and not as an investment recommendation or indication of their future performance. The value of investments and the income from them can go down as well as up, and investors may get back less than the amount invested. Past performance is not a guide to future returns, return projections or estimates, and provides no guarantee of future results.