American Diplomat Podcast: AI - Demon Taking Over, or Tool for Diplomacy?
By: Pete Romero, Dan Spokojny | June 20, 2024
What does expertise look like in diplomacy with recent rapid changes in technology? Will AI replace diplomats? What is AI good at, and where is it dangerous? Large language models are great for brainstorming but can they replace relationships and judgment? Tune into Dan Spokojny, former diplomat, now Ph.D. and CEO of the nonprofit FP21, while he breaks it down for us.
PETE: Welcome to American Diplomat, our stories behind the news. I'm Pete Romero… [today] we have Dan Spokojny, who is our guest, and he came to my attention because he wrote an interesting article in the Foreign Service Journal, being one of our sponsors, and I wanted to invite him on the show just to talk a little bit about diplomacy and artificial intelligence.
Artificial intelligence has really been in the news lately, particularly with the Pope and the Vatican, with the Pope making a major speech to the G7 in Italy last week about the dangers of artificial intelligence and the need for countries, governments to regulate it, to look into it very seriously, but also an interesting article about the Vatican being a consultant to Microsoft, which I thought was pretty interesting on artificial intelligence.
But again, today, we've got Dan Spokojny. We're going to talk to him about artificial intelligence and the diplomatic toolbox. Welcome, Dan. How are you?
DAN: Ambassador, it's great to be here. Excited to chat with you about this.
So, listen. Diplomacy and artificial intelligence. You were in the Foreign Service. Now, you're up to your elbows in artificial intelligence, getting a degree in it. Tell me, scroll back a little bit, and tell me about your background.
Yeah, sure. Yeah, I come by this conversation honestly, I guess. I worked on the Hill and then joined the Foreign Service and was in government for a little more than 10 years in total. And through that entire time, I kept asking myself, how do I be more effective at this job? Something I know that's been discussed on this podcast before.
Sometimes training and how to be an expert in diplomacy and foreign policy is less than many of us would be desired. And so I actually took leave and then officially resigned from the Foreign Service to go pursue a PhD in political science. I spent a lot of time with statistics, dabbled in artificial intelligence, machine learning, and started thinking big about what expertise looks like in diplomacy, especially in light of all these rapid changes in tech, artificial intelligence being prominent among them. So I've been thinking about this issue for many, many years, and I want what's best for diplomacy and for foreign policy.
So Dan, okay. So you're in artificial intelligence. You've got a decade of experience in diplomacy. How do these two things match up here? I mean, there's a lot of discussion about artificial intelligence taking away jobs. Okay. In this case, we're talking about diplomats. Is artificial intelligence going to somehow elbow out human beings that call themselves diplomats?
Is it going to? Yeah, it might. Should it? I don't think so. Not as where we are now.
Let's think carefully about what artificial intelligence is good at and what it's not good at. And I should mention, I think one needs to be very careful about the messenger often in artificial intelligence and whether they've got a stake in the conversation going one way or another. There's a lot of people who have a lot of stock options right now who are pitching how incredibly game-changing artificial intelligence is going to be.
So what is artificial intelligence good at and where does it fall short? I think often when we're talking about artificial intelligence today, we're referring to the chat GPT-like technology. And different companies have different versions of this. This is known as a certain type of artificial intelligence called a large language model. And large language models, they're very good at generating text. They search their sometimes massive, almost unthinkably massive archives. Their training set, is often what it's called. And they generate patterns of speech that they've seen from the internet. Usually it's the internet from written archives.
And when you ask them a question, they go find very similar wording and similar questions that have been asked in the past or similar discussions. And they'll feed you back statistically related information and knowledge words on those topics. So the artificial intelligence is very, very impressive, almost startlingly impressive at producing human-like speech and thought and ideas.
But I think we need to be careful because often those ideas, if you scratch beneath the surface, aren't particularly novel or new or even have vetted against true expertise. So my top line argument is, particularly for LLMs, for these large language models like chat GPTs, we need to still have an expert in the loop who can validate whether any output is going to be right or useful or not. So this makes it great for brainstorming.
It makes it great for knowledge management and finding an old idea or an old document or a history of a topic. It makes it, I think, less good if you're thinking about trying to replace the relationships that diplomats try to build, the judgment, the complicated political, ethical, causal judgment that's required for good policymaking. To suggest that an AI is going to displace a policymaker, a decision maker at this stage is mere science fiction and potentially even dangerous.
Why do you say dangerous? I'm intrigued by that.
There's dangers at kind of two levels. The first danger is that we just give it bad... we receive bad information.
If you take an expert out of the loop and our policy decisions are going to be, if not entirely decided, but largely steered by the computer, it's going to steer you towards solutions that sound plausible, that were popular answers on the internet, but that's no way to govern. That's no way to decide a policy that may affect war and peace, or to shape a relationship, or even craft a good, carefully calibrated message.
Every day you have diplomats sending very carefully calibrated messages in response to events in the world and say, here's where the American government stands. Here are our values. Here's where we want to urge you in your challenges to move. That sort of thing, it's dangerous to get it wrong because the consequences could be to add to risk or conflict.
And now there's another danger at an even higher level that many very prominent scientists and artificial intelligence experts talk about, which is this existential risk. That if you give machines too much power, and there's not enough, as the Pope was talking about, and there's not enough safety protocol built in, the machines may do something crazy, not because they're evil, but just because they don't know any better.
A popular example is if you program a robot to be your paperclip maker, and all of a sudden it starts gobbling up all the resources on earth to make more and more and more paperclips, and nobody forgot to say, hey, stop making paperclips, you've got a machine all of a sudden that may be doing really disastrous things.
So I think that there's some long-term, really existential-size risks to letting this technology go unchecked.
So Dan, let me ask you a question. You and I both know that when you're talking about the formation of a foreign policy, in this case, the United States foreign policy, we can take a fictitious country. Why don't we do Venezuela, for example.
That policy usually starts in a conversation between a couple of people at state, moves into a bigger discussion, moves into an interagency meeting where you have the State Department and affected other agencies, maybe Treasury, maybe Justice, maybe the intelligence community, et cetera. This is all in the process of formulating what it is we're going to do, what it is we're going to propose, what we're going to ask allies to sign on to, whatever it might be in the case of Venezuela.
And then when there is this unanimity, then we launch. But let's go back to that meeting, that interagency meeting where everything is discussed. When you talk about artificial intelligence, could artificial intelligence be good for that kind of meeting? This is before you launch. This is before you actually get out there with your talking points and your demarche and your visits and all of that stuff.
I think the answer is yes, if it's used very thoughtfully and carefully. So let's consider what the purpose is of this often convoluted, bureaucratic, lengthy process.
It's to get expertise from a variety of different perspectives. You want your intelligence people in the room with your military people, in the room with your economic sanctions people, with the trade people, with the diplomats. You want the human rights officials to be thinking about the same issues as the counterterrorism people are thinking about.
So in an ideal world, what that process does is bring out, extract all of the best expertise from across a very wide variety of competencies that are all examining the same country. Sometimes that works well, as you know better than I, Ambassador. Sometimes that process can be arduous and it can distort decision-making and it can leave behind a lot of really great evidence.
So if one suggests that you could turn over all of this information to artificial intelligence and say, tell me what our policy should be on Venezuela. I think maybe a lot of people are asking that question of chatGPT these days. What should we do on Ukraine? How do we defeat Vladimir Putin? It's not going to be very good at answering that question.
And I don't find it very useful at that sort of big picture question, unless you're just looking for interesting brainstorming ideas. Instead, where I do think it is interesting is to ask it questions about the knowledge that we already possess. So list to me all the diplomatic agreements, the history of diplomatic agreements we have with Venezuela, or we have with Venezuela and all of its neighbors. List for me, or help me understand the trading relationships that Venezuela has with other countries on earth, or find me some about Venezuela's human rights record. This sort of task is really extraordinary for computers. It can search through a vast amount of information, find stuff that sounds right.
Is it going to get everything? No, but it's going to get a lot more than even a long-serving ambassador to Venezuela is going to be able to hold in their head. The human brain is just not capable of holding the amount of knowledge that the internet is holding. So if we're using it for knowledge search, I think that that's a really wonderful application of artificial intelligence and could save a lot of time and maybe turn up some information and knowledge in the process that we would otherwise miss.
Well, you know, it's interesting. We just had a show with the deputy assistant secretary from the Bureau of Stabilization and Conflict, and they have just been tasked with big data, acquiring big data on crises as they've unfolded in the last 65 years, I guess really since the end of the Cold War or the end of World War II, and kind of amalgamating all of that in one place, in one home, so to speak, so that diplomats have access to all of that and not just necessarily what's on the headlines going forward at any given moment. And they've put a lot of resources behind it and they're accumulating that data as we speak.
And quite frankly, as a ex-diplomat, I think that that's a very good thing. You can't hope to keep in your brain, even if you are the best expert in the State Department on a subject, you can't just keep it all in there the way you would like to have it. So if that's the case, okay, where do the people come in?
Well, you know, a starting point for this conversation could be a question like, what is artificial intelligence? And my favorite definition is artificial intelligence is, particularly the intelligence part, is kind of anything we don't know how to do yet. It's always one step ahead of us. So when we invented our first calculators, that was artificial intelligence.
The fact that you could punch in a number into this little computer and it would give you the right answer on a math problem, that's incredible. Today, we don't think of calculators as artificial intelligence. It's just a basic little tool.
We barely think of Google Maps as artificial intelligence, though that application is pretty extraordinary and central to many of our lives. So what can we do with conflict data? Data like that can help us understand trends. It can help us compare between different countries at different time periods and understand the tools that will work to mitigate conflict.
It will help us understand what the factors are that seem to predict when conflict is going to spiral. If we want to do things like the Bureau that you mentioned, the Bureau of Conflict and Stabilization Operations, like mitigate future genocides or mass atrocities, you need to get in ahead of when the killing starts. If the killing starts, you're already too late.
So what happens right before the killing starts? Computers and algorithms may really help us find those patterns. Yet those algorithms need to be programmed by somebody. They need to be understood by somebody.
You need to ask the right questions of the data and you need to make really hard choices about how to present the data. Most of those choices cannot be done by artificial intelligence very effectively. So let me ask you a simple question.
How do you code whether a country is at war or not? There's no simple yes or no answer to that. When I was in the hallways of academia, a lot of people would say, well, it doesn't really matter. You choose the definition that you need for your purposes.
And I think that that's analogous to how we think about making decisions without data. We try to think about, well, what are the countries that are comparable to this? If I'm working on Venezuela, what are other historical circumstances that I think are relevant here? And what historical circumstances do I feel like are irrelevant? Okay. Is Venezuela like pre-cultural revolution China? Is that a good case study? Maybe in some ways and not at all in other ways, but it depends on the question that we have to ask.
So my article in the Foreign Service Journal was just one of many that all discussed this topic. The whole edition of this month of June was on artificial intelligence and I recommend all the articles. Each of the people who was writing was trying to suggest a different way of using data and all of them mentioned data, which is so interesting, of using and preparing data for their purposes.
Whether they're working on conflict or international organizations or public diplomacy, they want good, clean data that people use in their day-to-day work effectively. And doing that is going to require choices from experts about what we consider relevant. And then what do you do with that information once you get it? Okay.
So it turns out that we know this factor is very likely to precede a civil war. And now we see this thing and now we're really worried Venezuela may be tipping into really ugly violence that could infect the whole region. So what do we do about it? I think those types of policymaking questions, and is it an American priority? Do we have an interest in investing time and money and potentially putting people in harm's way to try to mitigate this issue? That's something that artificial intelligence is never going to be able to tell you.
That's an ethical choice. It's about our values. It's about what it means to be an American and that's always changing, but that's up to us, not the computers.
Right. Well, I wonder if there's, if you're going to use artificial intelligence and I think the answer to that is most certainly yes. I mean, nobody is going to just walk away from it.
Like, you know, I mean, I've got friends who still have flip phones, but you know, few and far between. So, I mean, anyway, the thing about it is it's going to be used. It just seems to me that it cries out for someone who's an expert and knowledgeable enough, not only to be able to take on whatever it is AI feeds you, but to wade through it all and decide what's useful and what's not.
And it just seems to me that you've got access to an enormous amount of data, some of which may be highly useful in, in, in making a decision or planning a course of action diplomatically. Okay. But more than ever, it seems to me, you're going to need experts to be able to wade through this massive big data.
So I don't see how it's going to eliminate jobs. In fact, it probably will create more jobs on the diplomatic front and the analytical front.
I think that’s right. There may be places in our, in our broader economy that are more vulnerable than others.
That's not really my expertise, but at the state department. Okay. You know, when I, so when I served at embassy at the, at the consulate in Karachi, Pakistan, we had what was called motor pool and we had a limited number of cars and security officials to drive us to our meetings.
There were more people than there were cars. So we had a guy with pen and paper who was distributing schedules and, and, you know, working on the map and trying to match people with cars, with officers to their meetings at their appropriate time. This is a very complicated process.
It was very inefficient. Often you were left without a car when you needed it. That's the type of activity that, you know, I don't know if that guy's job is secure.
And, you know, maybe that's going to be replaced by software at some point, but yet for every displaced task, that's probably better done by a computer. There may be two others created of the person who needs to record, you know, record, I mean, you know––as an officer in a foreign country, you're going to meetings all the time. And it'd be really great if you could pull up the dossier: Tell me all the past cables that have been written by this guy that I'm about to go meet. What does he like? What political parties has he been in? What, you know, what, what have been the challenges and the opportunities that we've faced in the past? Often we don't know that information. We can't find it.
So you build your relationship and you have the knowledge in your head and that's important. But if you could go back and have somebody who could help present to you all this old data, and it was good data that you really trusted and exactly what we did, where, with whom, and when there's a lot of jobs in that. I've talked to some folks who think that, that the today, the old conception of office management specialists, our OMSs, should probably be shifted towards more knowledge management capacity these days, that these should be folks who help us capture all of this vital information about the world, put it into a computer so we can very quickly access it and analyze it next time.
But that seems, that seems like an interesting shift of, a shift in response to the technology rather than just throwing up our arms and saying, it's going to take our jobs, which doesn't seem, doesn't seem quite, you know, sensible.
When it gets to actually going out there and making your pitch, your demarche, your approach, whatever you want to call it to a foreign dignitary, do you see the use of artificial intelligence and or video meetings taking the place of that, that kind of human contact?
I don't know. I think, again, there's ways that being better informed in advance of your meeting, of your demarche, of your pitch is, is important.
I think we have a lot of work that we should be doing to make our demarches more effective and to really study when these are working and when they're not. But the tools of diplomacy often are, they're about relationships. They're about trust.
Asking an intransigent political leader to go out on a limb for you, despite threats to his leadership, his country, maybe even his life or her country or her life, that's not going to be done because they look at the data and say, ah, it does look like this will improve my economy by two percentage points in the GDP. They're going to do it because you look them in the eye and you say, hey, listen, as ambassador, I am going to stand with you. The three previous ambassadors are going to stand with you.
And the three next ambassadors are going to stand with you and your country and your people. And you can count on us. And if they believe you, then they'll go with you.
And if they don't believe you and they don't trust you, it doesn't matter what you say. So that trust is not, that's not a function that can be, the stories about being kind to their family or doing them a favor. Those things, those things give us so much leverage in diplomacy to help us nudge something across the finish line.
And I don't, I don't think that's going to be replaced by anything.
Well, you know, I'm just thinking back at my career, Dan, and I remember, I recall there was a time back in the nineties when Ecuador and Peru were at war over a disputed part of the jungle and they called it the rumble in the jungle, but it was a serious conflict. And we, as the guarantor countries, there were four of us, Brazil, Argentina, Chile, United States, we were able to separate the two, the two forces and to have a temporary ceasefire only to see that whole ceasefire start to disintegrate.
And the forces moving closer to one another and trading pot shots. And it looked like it was going to start all over again. And the president of Ecuador at the time was a friend of mine from the time that I'd been there. And we had a pretty good relationship. And I went down there because it looked like they were going back to war.
And he told me, you know, I had my foreign minister call. I've had my secretary call, uh, uh, the president of Peru, Fujimori wasn't returning my calls and that sort of thing. And I said to him, and it was just like, I pulled it out. I don't even know where, but, but I said to him, you know, Fujimori is a very lonely man. He only listens to his brother and his daughter and pretty much everybody else's kept at arm's length. And he's a pretty lonely guy. He was running for reelection at the time. I think if you just call him personally with you on the line and talk to him, you might get through.
In fact, he did, he got through and they were able to avert a re-initiation of, of, of the conflict between the two, between the two sides. It's that kind of thing that there's, I don't see how artificial intelligence could give you that.
I'll give you another example. I had retired from the foreign service and I was out at, out in the Middle East to try to get an American out of, out of prison. And there were bogus charges. It was in Dubai and the charges were bogus, but they were, you know, they, the sheik wanted to kind of shake them down, you know, that sort of thing, and, you know, put up a huge, a huge fine. He did nothing wrong. And they're trying to get him released from jail and nothing seemed to work.
And I went to the so-called prime minister, who's the assistant of the sheik. And I said, you know, one of the things that's so painful for my client is that he thought that he was really a close friend and ally and, you know, basically had a working personal relationship with you. And the hardest thing for him to accept is that somehow that's gone.
And the guy fell for it and, you know, ultimately released him from jail. And years later they became business partners again. But it's the kind of thing that, that, you know, you, sometimes you pull it out of your pocket in desperation, but there's just no way artificial intelligence is going to give you that.
I agree. I think that's right. Yeah.
And I think about, so I was a political officer and, and, and really took a lot of pride in, in reporting and being able to understand what was happening in the countries that I was in. And, and I remember the good political officers that I tried to model myself afterwards would, the first meeting was never, that was never the meeting where you got the information. That was the meeting where you showed them who you were. You would take them out to dinner. You would talk about your family. You would find shared values in places that seemed so different.
And yet, you know, the, the, the, the big picture differences between, between all of us humans walk in this round planet of ours is, is pretty small. And, you know, it's the second meeting, it's the third meeting. It's the, it's the run in at the restaurant someday when all of a sudden they confide in you and say, you know, I really think that you understand these issues.
And, you know, let me tell you something, let me give you the juicy gossip. Or when they call you in the middle of the night and they say, hey, something's going on. I need, you know, I need help. I need to speak to somebody I trust.
I've got so many stories of that happening, you know, middle of the night phone calls or, or being pulled into a, yeah, you know, why don't you come on, come on in with us? You know, I'm, I'm going to meet with the president right now. Why don't you, why don't you come with me? I think, you know, just, just sit and, just sit and listen. I think you'll enjoy this.
And those relationship issues are, are just so vital. They're just so vital.
And, and we should say, you know, I don't think that they're prioritized enough today. Artificial intelligence aside, I think, I think diplomacy is not playing the role that it should, that it must within our national security bureaucracy. And I hope this, you know, advancements in tech, like artificial intelligence help, help empower diplomacy, not disempower it.
I think that's the lens through which I view success here.
Well, Dan, this has been super interesting. There is a whole lot more to this as, you know, as the days progress and we move forward even closer to, to artificial intelligence and the benefits and the liabilities, obviously.
But we generally go out on a song and I'm going to give you first shot at picking that outro song that would kind of exemplify where you're coming from.
I'm going to think of one and send it to you. I, you know, this type of thing. I got to go ask, I got to go ask artificial intelligence what they recommend.
Okay. So l asked Google Gemini, what song I should pick to introduce the podcast about Al and diplomacy. And it suggested “Everything in its Right Place” by Radiohead.
This song has a sense of order and control contrasting with the unpredictable nature of artificial intelligence. And Radiohead famously wrote a lot of songs about technology and trying to maintain humanity in the face of technology. So I think that that's quite appropriate.
Thank you, Dan. Really appreciate it. We will play that one.
