Artificial intelligence has already beaten humans in “Diplomacy”—or at least in the classic board game of the same name.
In “Diplomacy,” which dates back to the 1950s, players negotiate and forge secret alliances to conquer territory. It is a game of strategy and deception, in which participants need to anticipate their opponents’ moves and intentions to outmaneuver them. For decades, scientists cut their teeth trying to teach computers how to master it—until researchers at tech giant Meta trained an AI system to compete in the online version of the game.
Not only did just one of 82 players say they suspected that the system, named “Cicero” after the Roman statesman, was not human. The software also scored “more than double the average score” of its human competitors, according to the researchers.
The results, released late last year, stunned the AI community—and the triumph of machine over man created ripples in the world of real diplomacy. It came as an eye-opening moment to many diplomats, illustrating that part of their work, like that of other knowledge workers such as lawyers, translators, and journalists, can increasingly be automated with AI software.
This could of course help diplomats focus on the key aspects of their work. But it will also make access to cutting-edge AI, or the lack thereof, a factor determining success or failure on the diplomatic stage. And it is catching many in the field off guard, raising questions about when decisions can be delegated to machines, and when they shouldn’t. “Many of us haven’t fully grasped what’s coming,” said one German diplomat, who asked not to be named because he wasn’t authorized to speak publicly.
The Rise of AI Diplomats?
Contrary to what TV series like Netflix's "The Diplomat" suggest, a significant portion of a diplomat’s daily work involves relatively mundane tasks. Particularly among the lower ranks, diplomats often find their days filled with time-consuming and repetitive assignments. They spend hours searching for potential partners in their host country; monitoring local news and social media; or compiling reports to send back home.
Already today, AI can do much of that work in a fraction of the time. In addition, AI systems can analyze troves of texts, images, and videos so large that human diplomats could never read or watch all of them. This allows them to uncover hidden patterns, identify potential conflicts, and anticipate future trends that would have gone unnoticed without the programs.
Increasingly, AI can also be used to automate more sophisticated jobs. Take international negotiations and the preparations that precede them, for example: Ahead of difficult talks, diplomats often spend weeks combing through vast amounts of scientific, political, and cultural research to understand the interests of their counterparts.
Yet AI systems can already do much of this footwork for them—and more: Programs trained in game theory, for example, can generate tailored proposals for how to approach negotiations. They can simulate different scenarios for how talks will unfold, allowing diplomats to prepare various responses in advance. And even during the actual negotiations, “hagglebots” can continue to feed negotiators with advice on what to do or to say.
AI-powered Propaganda
At the moment, the use of such tools in everyday diplomacy is still the exception. But that is widely expected to change in the coming years. Many diplomats hope that this will help them free up time for other tasks. That time, they say, is much needed to address one of today’s most significant challenges to diplomacy, which AI is expected to supercharge: online disinformation.
Throughout history, and long before the advent of digital technology, nations have sought to shape public opinion abroad. Often, that involved spreading deliberately misleading or false narratives. Debunking and refuting those narratives, both in the open and behind closed doors, has always been a part of diplomats’ jobs.
Today, however, they are facing a different beast than just a few decades ago: In the pre-internet era, disinformation campaigns were costly and required extensive expertise and planning. But over the last two decades, the rise of social media has made it easier and cheaper than ever to disseminate them.
With the recent release of systems like ChatGPT, Bard, and Midjourney, which can create convincing text or images from scratch, it has also become cheaper and faster to manufacture this “fake news.” As a result, diplomats are anticipating a surge in AI-generated disinformation in the coming years.
"We haven't witnessed the 'big bang' yet, but one can envision countless scenarios of how it might unfold," said another German diplomat, echoing what he described as widespread concerns among his colleagues.
There is good reason to be worried: For Russian diplomats, for instance, spreading disinformation has long become "part of the job," as the AP news agency put it. That is why an understanding of how AI-generated content can be abused for disinformation and how to identify it is becoming an essential skill for diplomats. Well-aware, foreign offices worldwide have either developed or are in the process of establishing guidelines on how to do that.
AI Literacy 101
But this alone will not be enough to prepare diplomats for the age of AI. Equally important is that they have a solid understanding of how AI systems function today, where their potential lies, and where their limits are.
Suppose diplomats use generative AI software to write the initial draft of a graduation speech or official condolence. Not only do they need to keep in mind that these systems—as convincing as the texts they spit out may be—will fabricate facts whenever they can’t find an answer in the data they were trained on. They also need to be aware that the underlying machine-learning technology, if unmonitored, has a tendency to replicate and often magnify biases and discrimination from the analog world.
Such “AI literacy” will be ever more important, as diplomats will increasingly use software to help them with complex tasks such as developing policy recommendations—and it should be taught as part of the standard training diplomats receive before joining the corps.
The Limits of AI
What's next? What other tasks that diplomats currently perform could, at some point, be delegated to AI? And where will technology hit its limits, if anywhere?
"Diplomacy is fundamentally about persuading other people," the German diplomat said. He stressed that at its core, diplomacy is often more art than science, relying on empathy, building personal relationships, and coming up with new and creative solutions to unprecedented problems. "I can't envision machines doing that," he added.
Indeed, AI systems have so far largely struggled to imitate such distinctly human traits, let alone replicate them. But that also can change, and the rapid progress with which AI has conquered increasingly intricate board games illustrates how fast that could happen.
Back in 1997, when AI first triumphed over a human champion in chess, games like "Diplomacy"—which relies on "soft intelligence," on sensing your opponent's thoughts and guessing what they believe you will do—still seemed elusive. A decade later, computers solved checkers; less than another 10 years later, in 2016, the Chinese game of “Go.” Now, AI has cracked "Diplomacy."
When Meta unveiled the results, Andrew Goff, a three-time “Diplomacy” world champion who was involved in the research, specifically pointed out the system's extraordinary "ability to communicate with empathy and build rapport."
Where AI Shouldn’t Replace Diplomats
In light of this, it is time for countries to draw red lines for where to use and where not to use AI in diplomacy. Even if the most intricate tasks of diplomats can be automated at some point, that does not mean they should. Rather—as helpful as AI tools are for speeding up processes—they should be considered as just that: tools that assist rather than replace humans.
World politics are no board game. Decisions made by diplomats can sour or save international relations. They can escalate or calm political tensions, move markets, and they have very real consequences for real people. That is why in certain areas, AI should never replace human judgment: Computers should, for example, never be solely in charge of making decisions that could have life-changing consequences for individuals. In these cases, humans should always have the final say.
At the same time, countries should carefully examine any software before using it in critical areas of diplomacy such as negotiations—not least because of another factor that has so far received little attention in the discussion: Most of the models like “Cicero,” which will likely provide the basis for the next generation of AI diplomacy tools, aren’t the work of governments or research groups at universities. Instead, they are developed behind closed doors, inside the fundamental research divisions of the world’s largest and richest tech corporations, from the US to China.
As they develop the systems, companies encode their ideals and ethical standards for what "good diplomacy" looks like into the programs. Countries need to check whether these align with their own before employing them. It is, however, an open question to what extent they will be able to do so.
In a bid for transparency, Meta—to its credit—released the code and model underlying its "Cicero" system after its triumph in "Diplomacy." It’s fair to assume that not all companies worldwide will be willing to do that.
Janosch Delcker is the chief technology correspondent at Germany’s public international broadcaster Deutsche Welle, where he covers the intersection of politics and technology.
