Can Artificial Intelligence End a War?
I asked ChatGPT to draft a peace agreement between Russia and Ukraine. Here’s what happened.
The artificial intelligence fever that broke out two years ago hasn’t spared diplomacy, and governments and think tanks around the world are scrambling to adapt. AI is already automating various diplomatic tasks — from research and data analysis to drafting and editing documents, such as cables, memos and speeches. AI is also sharpening public diplomacy tools, enabling more precisely targeted messaging.
At the Washington International Diplomatic Academy, we have designed programs to train current and aspiring diplomats to utilize AI in almost every aspect of their work. The first to benefit will be the participants in our eighth annual Summer Diplomatic Academy next month.
Our research found a particular focus in several countries on how AI can shape diplomatic negotiations, including peace talks. Some have developed models that incorporate elements like assumptions, historical trends and cultural traits to influence simulated outcomes. Others aim to predict the likely results of real-life negotiations.
I decided to try a simpler experiment. I asked OpenAI’s ChatGPT to “draft a peace agreement between Russia and Ukraine to end their war.” That was my entire prompt — I deliberately avoided offering any hints, preferences or biases. I provided no instructions on what data to use or what sources to consult, and I hadn’t previously fed it any information or commentary on the subject. I also didn’t know how the AI had been trained to respond to such a prompt, though I assumed it leaned toward a Western worldview.
I have no illusions that meaningful negotiations between Russia and Ukraine — let alone peace — are imminent. The two sides are so far apart that they can barely agree to exchanging prisoners and soldiers’ remains. I did the experiment purely as an educational exercise, hoping to see a possible starting point for a future process, rather than anything close to a final accord, though I didn’t say that to ChatGPT.
What’s the verdict?
So how did it do? All things considered, as well as an average low- to midlevel Western diplomat — it certainly didn’t measure up to a seasoned professional. After “thinking” for a couple of seconds, it told me that its draft was “designed to end hostilities and lay the foundation for lasting peace, while balancing political realities, international law and humanitarian needs.” Let’s see if it delivered on that promise (you can read the full text here).
From the preamble, it’s clear that the AI is familiar with diplomatic agreements. Although the language is standard for such documents, it overlooks the unusually loaded meaning of certain words and phrases in this particular conflict. Here are the first four lines:
Recognizing the profound suffering caused by the armed conflict,
Affirming the sovereignty, territorial integrity and political independence of all states,
Desiring to restore peace and stability in the region,
Acknowledging the legitimate security concerns of all parties…
As central to the conflict, “sovereignty, territorial integrity and political independence” are likely to remain unresolved long after the agreement is signed, as the AI itself acknowledges later in the text.
The most plausible outcome might be a “stalemate, where each side agrees to a line of disengagement without international monitoring or guarantees, and where neither side has to acknowledge the territorial claims of the other,” said David Lindwall, a former senior U.S. Foreign Service officer. “Think Kashmir,” he added in reference to the disputed region between India and Pakistan. Lindwall is one of several veteran career diplomats I asked to review ChatGPT’s draft.
Bias or wishful thinking?
In the agreement’s main body, the first of 10 articles calls for “a complete, bilateral and verifiable ceasefire” by a mutually agreed date and time, “with all military operations, including drone strikes and missile launches, to cease immediately.” It also says that “both parties shall instruct all armed forces and affiliated groups to halt offensive operations and maintain current lines until further agreement.” It elaborates in the next article:
ARTICLE II — Territorial Status and Administration
The Parties agree to initiate a three-year transitional period to determine the final status of the territories currently under dispute, including Crimea, Donetsk, Luhansk, Zaporizhzhia and Kherson.
During this period:
A Special Joint Administrative Mechanism (SJAM), supervised by the United Nations, will oversee civilian governance, humanitarian aid and security coordination in disputed areas.
No demographic changes or forced displacement shall be permitted.
A binding internationally supervised referendum will be held in each disputed region at the end of the transitional period, ensuring full access to international observers and displaced persons.
Russia is certain to refuse grouping Crimea with the other provinces it has occupied, and Ukraine is unlikely to part with the four other provinces — at least not in their entirety. It’s unclear whether “the armistice line, which intersects these provinces,” would follow the front lines or encompass entire provinces, said Thomas Miller, a former U.S. ambassador to Greece who was involved in Cyprus reunification talks. “Similarly, would a referendum include an entire province?” he asked.
Chas Freeman, a former U.S. ambassador to Saudi Arabia and assistant secretary of defense, said it’s “very doubtful” that most of the current population in the Russian-controlled areas would agree to rejoin Ukraine. “What is certain is that Russia will not ask them to do so,” he added.
Eric Rubin, a former U.S. ambassador to Bulgaria and deputy chief of mission in Russia, said that Ukraine isn’t likely to accept referenda, either, as large numbers of residents in the regions occupied by Russia have fled to Ukrainian-controlled territory. “International law says that borders can only be changed by mutual agreement,” he noted.
Other provisions — like the stipulations in Articles III and VII that “Ukraine’s internationally recognized borders as of 2013 are acknowledged as the legal baseline,” and that “an International Reconstruction Fund for Ukraine shall be established with contributions from the global community, including Russia” — are also far from realistic, at least for now.
One more try
So I asked ChatGPT to make the agreement “fairer to both sides.” It replied that the revised version avoids “unilateral language, emphasizes mutual compromise, and proposes mechanisms rooted in neutrality and international oversight.” Let’s see if it lived up to that (you can read this version on the same page as the first).
The second draft retains Crimea on the list of “disputed territories,” extends the transitional period in those areas from three to five years, and removes Russia as a contributor to Ukraine’s reconstruction. While the first version included no mention of NATO, this one says that “Ukraine agrees to adopt a policy of non-alignment for a period of seven years, during which it will not seek NATO membership or host any military alliance infrastructure.”
Freeman said it’s unlikely that Russia will ever accept Western forces in Ukraine any more than the United States “would accept Russian forces in Cuba or Chinese forces in Mexico.”
ChatGPT also offered to draft “a clause-by-clause rationale for policymakers,” along with a “classified annex to the peace agreement” addressing the most sensitive matters, including cyber operations, the status of Crimea and “confidential inventories of weapon systems and foreign support units active in or near the conflict zone” (you can read them here). Some of the content will hardly be considered confidential by either side.
I won’t analyze the AI’s work in greater detail, but even these few highlights reveal a clear pattern: Although the AI can “distill the many elements of the dispute and lay them out in a coherent step-by-step plan,” as Lindwall noted, it reflects a distinctly Western perspective. No wonder the race to build and train AI systems, especially between the United States and China, is so intense.
Replacing humans?
ChatGPT’s draft, Miller concluded, offers a “good outline to use as a starting point, but it would need to be filled out much more fully to have any value.”
How would human diplomats approach a task like this? If they are any good, they would rely on skills that only extensive diplomatic experience can teach. They include empathy, deep understanding of the history, culture and interests of all sides in a conflict, mastery of the nuances of their positions and deciphering their negotiating strategies and tactics, such as red lines, bottom lines and walk-away points.
“With a better data set,” Freeman said, “an AI might produce suggestions that would have a greater prospect of success.” The question is, can AI match the expertise, judgment and intuition of a human diplomat with decades of experience?
As useful as AI may be in many aspects of diplomatic practice, when it comes to high-stakes negotiations — where context, trust and insight are vital — it’s hard to imagine replacing human diplomats anytime soon.
What’s your take?
How would you assess the AI’s draft?
Leave a comment below.
Nicholas Kralev is the founder and executive director of the Washington International Diplomatic Academy, and a former Financial Times and Washington Times correspondent. His books include “Diplomatic Tradecraft,” “America’s Other Army” and “Diplomats in the Trenches.”
CHATGPT looks like boilerplate from allies of Ukraine. It has all the fundamental clauses, and the spin for PEACE according to WESTERN POWERS. Russia would balk at it, because they want to reclaim the Ukraine Territory for RUSSIA. The Integration of the UN is critical for peackeeping and the discussion of the disputed Terrirtory. The freeom to leave the territory should RUSSIA keep control is not offered. The presumption of Russian Speaking Ukrainians sympathetic to RUSSIA is false. Repeatedly they have been interviewed and rejected rule by the Kremlin.
Thought Provoking Exercise by NIcholas Kralev. I gained valuable experience at WIDA discussing REAL WORLD PROBLEMS with seasoned Ambassadors. Thank you NICKOLAS.