In this conversation, Matthew Stibbe interviews Henrietta Wilson, a PhD candidate at King's College London, about her work in open source investigations, particularly in the context of the Biological Weapons Convention. They discuss the importance of high-quality information in policy making, the role of AI in arms control, and the challenges faced in distinguishing between legitimate and harmful uses of technology. Henrietta emphasises the need for public literacy in research and the value of diverse expertise in open source investigations.
Please note that any views expressed in this podcast episode are the speakers' own, and do not necessarily reflect those of Blackdot.
Matthew Stibbe (00:00) Hello and welcome to From the Source, the Blackdot podcast. I'm your host Matthew Stibbe and today I'm with Henrietta Wilson, who is a PhD candidate at King's College London with a background in freelance OSINT research. Great to have you on the show, Henrietta.
Henrietta (00:16) Thank you very much, Matthew. I'm absolutely delighted to be here with you.
Matthew Stibbe (00:20) And I want to start with a question I ask all my guests. What are you geeking out about at the moment?
Henrietta (00:27) ⁓ Well, great question. I'm geeking out about the biological weapons convention. So I feel I should prefix all my comments here with a snapshot of... I'm not what I think of as a typical OSINTer. I think typical OSINTers are people that are looking at satellite imagery or social media posts ⁓ to make sense of situations or events a long way away. My sort of open source research are to address, to understand policy and technical challenges facing the global regulation of weapons of mass destruction. So I'm looking at sources that are very recognisable from an earlier iteration of research generally. So I'm looking at parliamentary accounts, I'm looking at newspaper records, I'm looking at budget or developments in weapons labs to put together different contexts that policymakers are deciding on.
So my PhD is really drawing on this. I'm looking at the Biological Weapons Convention, which entered into force 50 years ago and bans an entire category of weapons of mass destruction and yet does not have any formal ways to verify that everybody's complying with the treaty. So for my PhD I'm looking at what people understand by verification, whether it might be possible to do something to address what's recognised as a formal gap.
The States parties have tried many times to add verification and haven't managed it. They're negotiating a new set of measures or trying to address various different aspects of the treaty. So my research is going to the UN website, looking at the States Parties, meetings, papers. There are loads of them. They're all public access. I've downloaded over 6,000 and I'm going through them using various automated processes to pick out what people have said at different times. Like lots of OSINT challenges, there's a lot of little minuscule detail that may very well not be relevant, but I've been very excited to find out that verification has consistently been named by States Parties, even when it's seen as politically unacceptable.
So that's the first sort of like... they do care about this thing, it's not just me, they care about it, too. And moreover, different regional groups are disproportionately represented in this kind of big finding. So the Western group, which is countries from the West, they've mentioned it a fair bit. The Non-Aligned Movement (NAM) groups, which tends to be global South countries, they mention it so many more times that it doesn't even fit on the same chart. So different commentators might interpret that in different ways, that it's a way to kind of bounce a different set of conversations going around. I don't know, but it's very, it's really satisfying to have that sort of data really well documented and evidenced.
Matthew Stibbe (03:50) It's an absolute delight not to have the answer 'AI' because everyone I ask, obviously lots of people are very concerned about that. But tell me, how did you actually go about crunching that data? Or is that perhaps getting a little bit too close to the...
Henrietta (04:06) Well, so I started off doing it manually. I started off thinking I need my own human eyes to kind of go through the statements line by line. Well, you know, I used... I put them all in a folder and I used Ctrl-F to pick out different words associated with verification. I very quickly realised I was just missing lots of combinations and also I was missing them because my eyes aren't that good. It's very boring to read too many things at a time and you glaze over and you're like, god I just missed one.
Matthew Stibbe (04:47) Turning the page of a book and you realise you've turned 10 pages but you can't remember what happened.
Henrietta (04:51) Yeah, and also just I didn't have time, so I wrote a Python code to go through them. I used a Python code to first of all harvest them from the internet. So, please go to this website look at these things drop it in this folder and at the same time build an Excel spreadsheet that will capture... that will kind of bring out different data points in the papers themselves, the date, who wrote them, if they've got any kind of unique identifiers. So I used Python for that. And then I use a second set of Python codes to go through the documents and just cherry-pick the ones that mentioned some variation of the word verification. And moreover, to extract the paragraphs where it did do that. That's how I did it.
Matthew Stibbe (05:43) Amazing. I think I have a theory, scratch any OSINTer and you'll find a Python code. I'm developing this observation from experience. So tell me a little bit about you and your work. You've said you're obviously not ⁓ a kind of OSINTer in the common sense, but how did you start into this and what have you been doing?
Henrietta (06:07) Yeah, it was an accident, which I would imagine many of your guests might say. So I did a physics degree to start with. When I graduated, I had very much enjoyed my science, but I knew...
I didn't want to be a scientist and yet was really fascinated by the role of science in society. So I ended up working for a group at Sussex University, the Science Policy Research Unit, for a man who was a world leader in chemical and biological weapons disarmament. I came very much from a pretty much apolitical background. I didn't have a kind of thriving quest to rid the world of the worst, the most heinous weapons. Unlike many of my colleagues, I came from it much more neutral. You know, I wouldn't say I was an advocate for them either, but I wasn't motivated by that sort of strong political belief.
But very quickly, I found this intense interest and fascination with the fact that people have tried to ban these weapons, you know, globally there are many taboos against using poison in warfare and yet it's quite hard to define them and to write agreements that people can sign up to and then to make sure those agreements are adequately fulfilled. ⁓ And I found my scientific background was enormously useful in giving me critical thinking skills. I asked a different set of questions and I was more likely to question the provenance of sources than maybe some.
Matthew Stibbe (07:54) I was just curious about whether it was the scientific knowledge or the scientific method or kind of critical thinking.... In other words, was it a method of thinking or something that you knew because of your science background?
Henrietta (08:09) Well I think you know, I think definitely a bit of both, but maybe the...
Well, I was going to say maybe the scientific method more so, actually now I'm reflecting, it was maybe being slightly unusual. So most of the people I worked with, not most of them, many of the people I worked with were social scientists. And when they heard about my background, there was this sense like, could you invent a technology that would...
make it all easy, you know, like find that chemical weapon, distinguish between natural and unnatural disease to find a use of biological warfare. And I was just fundamentally, I am fundamentally more cynical about technology. Like, yes, technology is great and it can do stuff. It does the stuff we ask of it. It does the stuff that it's invented for. So you need to be mindful of what, what its use is and how to maximise that, and you also need to be mindful that science and technology cannot get over fundamental materialities we have in the world. So one of the big problems with biological warfare disarmament is a biological weapon involves a disease agent which could come from a naturally occurring disease. So already you're like, so which is it? Is it natural?
Or is it deliberate misuse of a natural entity? Also, the things that you need to get the agents, weaponise the... well, get the agents, they're commonplace in pharmaceuticals, in brewing, anywhere, you know, widely distributed. So again, nothing that anybody invents to detect an agent is going to get around that thing that it could be anywhere and it could be genuinely benign uses that have gone wrong somehow or that there's nothing to look at here. So the technologies help but they won't do everything for you.
Matthew Stibbe (10:21) Yeah, in several of the interviews I've done for this podcast, I think this idea of critical thinking has come up a lot. And for me, as a layman and an outsider, the idea of the scientific method, it seems to me like one of the greatest inventions of humanity, the fundamental idea that I could be wrong. So I'm interested to see how that gets applied. And we were talking earlier about the need for high-quality information in a sort of area of uncertainty and risk. And I wondered if you could give me an example of why that matters and where that's come up.
Henrietta (11:00) Yeah, so this really underlines a lot of my research. I get commissions for short, freelance contracts to look at particular issues facing negotiations or decision-making, nationally or internationally. And so my research is partly about finding information and it's partly about getting that information to the people that are going to do something useful with it.
And the philosophy behind that is that whatever your personal beliefs are, whatever your politics, whether you're for or against weapons or, you know, or for or against disarmament, the best decisions need strong information that you need to know the nature of the thing that you're trying to ban. You need to know where the proliferation risks are.
So that's about the disarmament, but also in terms of deployment of nuclear weapons, say. There's a big, there's a strong tradition in policy circles that to minimise the risk of an accidental release of a nuclear weapon, people need to be confident that nobody else is about to use a nuclear weapon against them. So through the Cold War, you saw robust information exchange systems set up between
Matthew Stibbe (12:27) Hotlines and things.
Henrietta (12:28) Yes, exactly, between the USA and the USSR. And now we're living in such a complicated information ecosystem that it's very hard to distinguish between the real information and the disinformation, the signalling that we've got something, we're going to do something and the reality of what's happening. An example is early on in the Russian invasion of Ukraine, there was lots of...
well there was some nuclear saber rattling from Putin's Russia and people were really worried and the world's non-governmental OSINTers that are au fait with this area pivoted to look well what can we see, what's happening and there was a collective well we can't actually see any changes to the deployment of the nuclear forces so we think it's just posturing at the moment, you know, not to relax, but we know we'll keep an eye on it. But that's an example of how information can stop unnecessary escalation and thereby reduce risks of these weapons that are on hair-trigger alerts being released for unwarranted reasons.
Matthew Stibbe (13:46) Which on the whole we're all in favour of, right? Definitely.
Henrietta (13:49) Definitely, yeah.
Matthew Stibbe (13:52) I'm, we sort of touched a little bit on AI and I'm wondering where that might have an application in arms control or in sort of risk reduction in these of in these saber rattling situations or does it have or has it, have you seen it be useful?
Henrietta (14:14) So I think AI is just such a potentially rich tool to add to the other tools open source investigators use. I think we're still very much finding out how to maximise it and where its best uses are. So I'm going to distinguish between nuclear on the one hand and chemical and biological weapons on the other. I look at all of these in different contexts. In the chemical and biological world, as we just outlined, there's this massive overlap between legitimate and harmful uses. And those weapons are completely banned. And yet, there's this difficulty of distinguishing, more so for biological, but it's the same as true for chemical, where there's a huge industry making paint, ballpoint pen inks, all this really commonplace stuff.
And so in that realm, it's very hard to think of there being any sort of unambiguous identifiers that an artificial intelligence... generative AI process could help to identify. In the nuclear context, however, there are still dual use overlaps and I don't want to overplay this, but there are some identifiers that are much more focused on illicit proliferators.
So you could imagine a world, and I'm sure somebody somewhere has had a go at this, I am not one of them, ⁓ but you can imagine a world which harvests data and then narrows down the data that it's got to put to a human, I think these warrant further investigation. So I think definitely AI working in collaboration with people is really promising and exciting in all sorts of ways. It's not helping my work on the Biological Weapons Convention or biological weapons challenges to this disarmament because there's not enough high quality data to draw on, I think.
Matthew Stibbe (16:37) Possibly because the AIs haven't digested this information from the arms control or from the UN, or in a sense of AI these large language models rely on having very large data sets about things and if there isn't a big data set to think about.
Henrietta (16:50) Yeah, yeah. So the set that I mentioned is there, the States Parties Meetings, what we have less of are examples of determined proliferators. Like what does an illicit bioweapons programme look like? There have been some examples, but they're quite different from each other. And so this idea that you can pull together and like, this is what I'm looking for. We don't really have, as far as I understand, the uses to be, we don't really have sufficient... And also there's this kind of interactive issue with doing anything like that in that you could inadvertently drive a proliferator. You could inadvertently start creating the data that then could be drawn on. So I'd be very wary about doing that anyway. And I think, yeah...
Matthew Stibbe (17:46) A funny thing happened to me preparing for this interview because I've, sorry, my goldfish brains, forgotten the name of the organisation that you had worked for where they had done some research into nuclear weapons. Just remind me of the name of that again.
Henrietta (18:03) ⁓ So ⁓ I'm not sure the particular organisation I work for. On nuclear issues, I worked with the British Pugwash Group and ⁓ the Nuclear Information Service. ⁓
Matthew Stibbe (18:08) You mentioned them earlier but I've done it.
Yes, I think it was one of those, but I think it might have been plugwash
Forgive me, it was a few weeks ago now. So I was curious, a couple of months ago about Trident and the so-called independent nuclear deterrent. So I asked ChatGPT, which I ask dumb-ass questions to ChatGPT all the time, I accept the risk of getting a bad or wrong or hallucinating answer. How dependent is our nuclear weapon programme
Henrietta (18:32) Mm-hmm.
Matthew Stibbe (18:45) on the United States. It gave, I had done quite a little bit of reading around this and I'm geeking out about some of these things, but, it gave me quite an interesting and I thought quite well-reasoned set of explanations about, you know, the, the some about that stuff. I won't get into the boring details. And then I came across you and I came across, I think it was Pugwash, but one of those organisations had published.
Matthew Stibbe (19:07) a very detailed report, which I then read because I'm interested in it about this. And I realised that what ChatGPT was doing was giving me a digested version of that report. I can't imagine there were many other sources. It did mention some parliamentary things as well, where they had interviews, previous CDSs about this and things, but it must have been. And I thought, actually there was an example of probably...
Henrietta (19:17) Yeah, good.
Matthew Stibbe (19:34) one or two sources driving the data. So I think what you're talking about where you create something, it into ChatGPT and then regurgitates it out and kind of creates a risk, it's not... non-zero risk of that, I think, in a space where there's not a lot of data.
Henrietta (19:50) Yeah, absolutely. And I'm intrigued by your conclusion, not to turn tables or anything, but what's your conclusion about how dependent the UK deterrent is on the US?
Matthew Stibbe (20:06) Well, I think if the present regime in Washington decided that they didn't want Britain to have nuclear weapons, we've probably got two or three years. And I think we don't have... two or three years is not enough time to develop, certainly not enough time to develop an independent submarine launched ballistic missile programme, France, 20, 30 years of doing it and the whole infrastructure, which we do not have.
Probably not enough time to even resurrect the WE177 air-dropped programme, let alone put that on top of a storm shadow, which is also dependent on ⁓ American targeting data. You know, I don't think we have an independent nuclear deterrent. I think we've been lent one. That's just me. I'm not that, you know, I'm not speaking for that.
Henrietta (20:46) Yeah,
I can totally understand that and I think what you said about the targeting is the key question that people sometimes... like whether or not the technologies are in the... whether or not we have the indigenous expertise, whatever, if we decided to use... if the UK decided to use a nuclear weapon then the US could stop it because it could turn off the targeting information and the command and control systems are entirely hand in hand. So yes, that's how I see it anyway. Yeah. I was just going say one more thing about AI. So I think I'm not a...
Matthew Stibbe (21:28) So interesting stuff, it's AI. Let's go back to something more.
Henrietta (21:38) I guess I'm a bit of a cynic about people who think technologies can solve problems, but I'm also not, you know, I'm an optimist about technologies generally, but I do think there are some issues about what happens about the black box situation, about analysts being able to say where they got their results from, which I think could undermine trust in open source work if people aren't careful. But we come back to the scientific method, which says, be clear about your methods and then be clear to yourself when you're constructing the methods why you've chosen what at any one time. Constantly be checking your hypotheses against different ways and then be transparent about which methods you used at which point so that other people can reproduce your results. So I think we know how to build trust in knowledge and we need to carry on doing that.
Matthew Stibbe (22:31) There's a degree of also, I think, moral courage for some practitioners in terms of being able to speak truth to power and not give them the result they want. The political decisions are political decisions, aren't they? But they shouldn't be based on, well, anyway, you can imagine the headlines when the first AI dodgy dossier comes along. And it could undermine an awful lot of actually very valuable work that would happen with AI, could be supported by AI.
Like the first crash of an AI-driven car will set that back by, well, it's already happened, when that starts being, self-driving cars start killing people in fairly large numbers, even though they might be safer than human-driven cars. Anyway, I want to ask you a penultimate question. You've edited a book about open source investigation and how they strengthen human security. So can you tell me a little bit about that, about that book?
Henrietta (23:29) I'm very happy to tell you about that. I'm very proud of it. Thank you for asking. So the book is called Open Source Investigations in the Age of Google. I was lead editor for it alongside colleagues Dan Plesh and Ola Samuel, who put the proposal together in the first place. And I collected chapters from different practitioners working, using open source research to investigate threats to human security or analysts that are working with open source researchers. So I was just so grateful to my authors. You know, we had Christian Trebert from the New York Times, Aric Toler from Bellingcat writing about how open source methods also are used by, well, QAnon misuse, the rhetoric of open source to create incorrect findings and how we need public literacy to be able to distinguish between good and bad research. We have Alexa Koenig and Lindsay Freeman talking about the creation of the Berkeley Protocol which is a set of guidelines through which if people follow them, then their findings are more likely to be admissible in the International Criminal Court.
So I could go on, Ben Strick, talking about his phenomenal open source investigations and Jeffrey Duke, a practitioner from South Sudan, pointing out that not all open source is Global North people looking at Global South people. And in fact, Global South local monitors are really vital for some use cases and that building networks of distributed expertise really benefits everybody. And Idris Ahmed made similar points using the Syrian civil war as a case study of how open source investigators can achieve better results than traditional journalists. So, as I say, I could go on about this for a long time. My main kind of principle in collecting such diversity was an instinct that one of the richnesses of non-governmental open source research is that they don't need to stay in their own silos. That by building communities with different expertise, different areas, there can be great serendipities.
Henrietta (26:03) You can get it from Amazon or the publishers, World Scientific Publishing. In keeping with a lot of academic books, the physical book is quite expensive. It changes prices, so I'm not going to give you a price now. But we made sure that the e-book is open access. So please do feel free to download it for free.
Matthew Stibbe (26:26) Fantastic. Well, I will go and find a copy. I try to only read ebooks these days because of dematerialisation and things. Now I want to ask you, we're almost out of time, I to ask you one last question. I'm always interested to hear about a project that somebody I'm interviewing has been working on and particularly what they learned and maybe what they would have done differently if they could have gone back in time and said, you need to know this. So tell me about a project and tell me what you learned from it.
Henrietta (26:57) ⁓ A project in my past... well one thing that I want to talk about in my freelance career, I think I've implied that it's all about making these treaties work better, international governance better, and definitely that is one strand of my work. I've also had commissions to look at the UK's nuclear weapons facility. So I had a commission.
Is the nuclear weapons lab fit for purpose? There'd been allegations about its safety record and then following that how it could be used to enhance disarmament technologies. So that was really interesting and I'm not sure I would have done anything different. Now I'm looking back, I'm just trying to think.
There are a certain number of resources available. I used those and I got semi-structured interviews with people that know stuff. But one of the messages that I want to put here is that at the start of the conversation we said good decisions need good information. That's not just about grandiose policymaking and implementation. It's also relevant to local communities and local citizens.
For our democracies to work well, it's really important that an electorate knows about the decisions that are being made in their name ⁓ and how they're being implemented. Is money being well spent in the Nuclear Weapons Lab? Are the local communities shielded from any potential accidents that might happen? So I'll end there because it's just again this sort of robust information message.
Matthew Stibbe (28:50) It's of more than abstract importance to me since I live less than 20 miles away from Aldermaston and have been reading up there, well, the government things of what to do if there's a leak or something happens and not entirely encouraging. But anyway, so that, thank you for that, I have to look... Better than it was in the 80s.
Henrietta (28:51) Thank you.
Much Better than it was in the 80s. Yes.
Matthew Stibbe (29:17) That's encouraging. Good. Well, on that bombshell, I think that brings this episode to a close. Henrietta, thank you so much. It's been absolutely fascinating.
Henrietta (29:28) Really my pleasure, thank you.
Matthew Stibbe (29:30) And if you're listening to this and you'd like more practical insights and you'd like to learn more about OSINT, Videris or Blackdot please visit blackdotsolutions.com. Thank you very much for listening and tune in for the next show.