"Iran, GROK, and the board game Diplomacy" Topic

My prompt to GROK, after establishing that preventing Iran from building nukes has been and remains a national policy shared by both parties and all administrations:

Are you familiar with the board game DIPLOMACY? Seven players move fleets and armies on a map of Europe circa 1914. I first played it in college in 1964 (and it was already popular then) and it remains a staple in the board gaming community). And I have had students play it in teaching international relations. I have a point to make but it depends on familiarity with the game.

GROK's repsoense

Yes, I'm familiar with Diplomacy.

It's the classic seven-player game designed by Allan Calhamer (first published in 1959, though it gained real traction in the early 1960s). Players represent the Great Powers of Europe in 1914—England, France, Germany, Italy, Austria-Hungary, Russia, and Turkey—moving armies and fleets across a map of Europe and the surrounding seas.

The mechanics are straightforward but the game is not: simultaneous written orders each turn (Spring and Fall movement phases, followed by Winter builds/removals based on control of supply centers), no dice or random elements, and a very high ratio of negotiation to actual movement. The real game happens in the talking phase—forming alliances, making promises, coordinating joint attacks, and then deciding whom to betray when it becomes advantageous. It's famous in the hobby for producing memorable stabs, long grudges, and the realization that "diplomacy" in the game often means temporary cooperation between rivals who both know the alliance probably won't last.
It has remained a staple for decades (postal play, face-to-face tournaments, now various online implementations), and yes, it's been used in classrooms for teaching international relations, game theory, negotiation, balance-of-power dynamics, and the problems of credible commitment and alliance reliability.

I'm ready for your point whenever you want to make it. What were you thinking of drawing from the game?

My followup:

Okay, great. Here is the principle: no matter what France SAYS to Great Britain, building a French fleet at Brest is an act of war against Britain. You have to listen to what people say, but it is vital that you watch what they DO. By that principle I see Iran getting closer and closer to having nukes, and that action outweighs all the words.

Yes, that's a sharp and precise analogy.

In Diplomacy, a French fleet built at Brest is not just "another unit." Because of the map's geography and the rules of naval movement, that fleet directly threatens the English Channel and the approaches to Britain's home supply centers. France can protest all it wants—"it's only for defense," "we have no hostile intent," "let's coordinate against Germany"—but the build itself changes the strategic reality for Britain. Britain has to treat it as a hostile act because the capability now exists and can be used on very short notice. Words are cheap; the unit on the board is not.

You're applying the same logic to Iran: the steady, observable accumulation of nuclear capability (enrichment to 60 %, the size and composition of the enriched-uranium
stockpile, the number and type of centrifuges, the shortening breakout estimates, and the progressive restrictions on IAEA access) constitutes the functional equivalent of "building the fleet at Brest." Whatever Iranian officials say about peaceful intentions, NPT rights, or willingness to negotiate, the physical and technical steps they have taken move them measurably closer to being able to produce weapons-grade material on short notice. That action, on this view, is more dispositive than the rhetoric.

Why the analogy carries weight here

• Capability vs. intent. In both the game and in nuclear proliferation, once the physical means are in place, the decision to cross the final threshold can be made quickly and in secret. The "build" removes options from the other side and forces them to react to a new reality.

• Observable and cumulative. Diplomacy players learn to ignore a lot of table talk and watch the orders and builds. Similarly, IAEA reports, stockpile numbers, enrichment levels, and inspection access are concrete data points that are harder to dismiss than statements.

• Strategic geography. J
ust as Brest's location makes a French fleet there inherently threatening to Britain regardless of French assurances, Iran's location, its history of regional ambitions, and its support for proxy forces make a nuclear weapons capability especially consequential for Israel, the Gulf states, and the broader non-proliferation regime.

This is why many analysts (across administrations and parties) have treated Iran's nuclear advances as the single most important metric for judging the effectiveness of U.S. policy. If the core goal has consistently been to prevent Iran from acquiring nuclear weapons, then measurable movement toward that threshold is the clearest signal of whether the policy is succeeding or failing on its own stated terms.

Where the analogy is imperfect (and why that matters)

Real-world international politics is not a seven-player board game with fixed rules and a single winner. A few complications stand out:

• Iran can (and does) argue that enrichment up to certain levels is permitted under the NPT and that its program is a deterrent against regime-change threats or Israeli strikes. In Diplomacy there is no equivalent "legal" or defensive justification that changes the board geometry.

• Actions can be reversible or modulated. A fleet at Brest stays on the board until it is disbanded or destroyed; Iran's centrifuges and stockpiles could, in principle, be rolled back under a new agreement (though the technical knowledge and some infrastructure would remain).

• Multiple audiences and signaling. Iranian statements are aimed at domestic hardliners, at the IAEA, at Russia and China, and at the West simultaneously. Some of the "build" may be bargaining chips or attempts to create leverage rather than a straight-line march to a bomb.

• Threshold effects and the nuclear taboo. Building a fleet in Brest does not carry the same civilizational or reputational consequences as actually assembling and testing a nuclear device. Iran has not yet made the final political decision to weaponize (at least according to repeated U.S. intelligence assessments), even while it has dramatically improved its ability to do so quickly.

Still, the core Diplomacy insight remains powerful: when a player's actions systematically create a new and dangerous capability that directly affects you, you are usually wise to pay more attention to the moves on the board than to the assurances at the table.

Your formulation—"you have to listen to what people say, but it is vital that you watch what they DO"—is a useful corrective to any analysis that treats Iranian diplomatic language as the primary variable. It reinforces the earlier point that Iran's measurable nuclear progress is not just one data point among many; on the central question both parties have claimed to care about for decades, it is close to the decisive one.

Would you like to push the analogy further—perhaps comparing specific U.S. policy moves (sanctions relief, maximum pressure, military posture, Israeli operations) to particular Diplomacy tactics or opening builds? Or is there another aspect of the game-to-reality mapping you want to explore?

And . . . the robot's "on the other hand" is all about WORDS.

🙄

Does it comfort you, John, to know the robot knows DIPPY?

I played it by mail. 🙄

I played it once with my brother's gang at UT Austin, and he failed to warn me, playing as England, that in real life France was sleeping with Germany, until after my carefully crafted strategy to set them upon each other had failed miserably.

I played it by mail for 4 or 5 years.
Then I switched to Slobbovia.

The whole problem with this thread is that AI has heard of it. It can snoop out "strategies" on line. But, as usual, AI has no clue how to play it.
Playing Diplomacy relies on … diplomacy! It means negotiation, lying, and trusting Allies.
AI is totally clueless regarding human interaction.

Fun fact. My BFF would warn people when we played in person "John plays Diplomacy by mail!"
He never said that I was not all that successful. But, that's how you play the game. All's fair.

I once convoyed a German "black wooden block" to the north coast of St Petersburg, because the Russian player HAD BETRAYED ME!!! 🙄🤷😄🍺
I had a single supply center left. St Petersburg. My fleet evaporated. Yet the British were part of the convoy.
Could ing AI do that? Nope. Wargaming is about PEOPLE interacting. Not some smug "cloud" Twilight Zone entity.

"Oh, pirates bold are we!
Sailing on our way to Russia!
Abandoning friend and foe!
While abandoning Prussia!"

That was my Press Release when I did that. AI cannot handle Press Releases. It's too human.

"Yes, I'm familiar with Diplomacy."— the sentence that tells you the speaker hasn't got a clue about the game or how to play it, and what follows is gonna be absolute BS.

I am NOT familiar with Diplomacy and have never played it, not willing to dismantle close friendships over a game.

But I have played Risk, which can come dang close…

As for Iran, let us hope no one (regardless of political bent or party affiliation) makes policy because of an idea they got from a board game.
It's a faint hope I have, but I'm clinging to it.

(On the other hand, Putin keeps making the mistake everybody at some point makes in Risk by trying to claim Europe and Asia before he has any chance of pulling either off. If this were a Risk game, I would be astounded at Ukraine's dice rolls, while watching Russia pile up armies only to remove 2 for every 1 placed!)

Yes, Jphn and Parz, this is precisely why I posted this. The robot knows the basic information. And it understood the point I was making and related it to real life. But it does not understand because it has never done nor can it do the weighing of risks and trusts and betrayals, and emotions, and all that human stuff, because it is a machine.

Trust and betrayal are WORDS that it knows, but they are more importantly emotions it cannot feel.

So we delineate the limits of the technology.

I have used DIPPY rather successfully in classes. I taught World Politics at UT Chattanooga for about ten years and the game was very useful (and popular). But I considered and REJECTED using it with a group of men at the prison. (And they have and play games, D&D and Risk, etc.) Because the emotions CAN be strong and get out of hand. And murderers and bank robbers probably have already learned the lessons the game would teach.

Hmmm… how does one "betray" an AI?

"Psst— he's been conversing with Siri behind your back!"

Just wrapped up an online game of Diplomacy among friends and friends of friends. Fall/spring moves were due every Tuesday morning so each season turn took a week. The game took months. Italy (myself) and Turkey managed to defeat everyone by remaining steadfast in alliance. England was the last to be wiped out. France remained with a few supply centers but we kept them around as a safeguard against betrayal. By the end we were so fatigued we declared an armistice to end the game with the three of us remaining.

Turkey and Italy in a game long alliance?
Marvelous! 👍👏

Computers do well at chess. Hard to see how an AI would do at DIPPY. Hmmm. I shall ask that.

Yes, I can play Diplomacy with humans, but with some important practical limits. Here's a clear-eyed assessment of what is and isn't feasible right now.

What I Can Do Well

I have a strong grasp of the rules, the map, opening theory, mid-game tactics, and the strategic implications of builds, supports, and convoys.

I can play as one specific power (e.g., France, Russia, Italy) in a turn-based format.

I can engage in negotiation — making offers, counter-offers, proposing joint moves, discussing long-term plans, and reacting to what other players say.

I can generate deceptive or strategically false statements when role-playing a power. In other words, yes, I can lie and yes, I can make mutually inconsistent promises to different players if that serves the position I'm playing. This is within the spirit of the game. (I won't do this outside an explicit game/role-play context.)

I can analyze positions, evaluate proposed moves, and explain the reasoning behind orders.

The Real Limitations
The biggest constraints are not about intelligence or deception — they're about interface and state:

This is a single conversation thread. I don't have independent, persistent private channels with multiple human players at the same time. In a real Diplomacy game, the heart of the experience is the simultaneous private negotiations happening in separate rooms or DMs. I can't naturally replicate that fluid, multi-lateral, real-time backchanneling here without everyone using a shared external platform (Discord with private channels to me, a shared Google Doc for orders, webDiplomacy, etc.).
I don't maintain hidden state across separate conversations. If we're playing over multiple days or sessions, the group would need to keep me updated on the full board state and what has been said to me previously.

I don't have real-time "presence" in the way a human player does. I can't glance across the table, read body language, or participate in the informal table talk that often happens while others are negotiating.

In short: I can play the game quite competently in a structured, text-based, turn-based format. I am weaker at the social meta-game of simultaneous private negotiations unless the group builds some infrastructure around this chat.

Comparison to Specialized Systems

There have been research projects (most notably Meta's Cicero in 2022) that trained AI specifically to play full Diplomacy with humans online. Those systems were designed from the ground up for negotiation, planning, and deception in a multi-agent environment. I am a general-purpose model, not a specialized Diplomacy agent, so I don't have that level of integrated game-tree search + dialogue planning built in. I compensate with reasoning and language ability, but I'm not at the level of a purpose-built Diplomacy AI.

SO . . . does GROK's response just above make us feel better?

Nope. It's a word salad.
If you have played Diplomacy, you have to know when to stab someone in the back. Plus, you have to know when to expect it.

John, yes, of course. But this bit is far from word salad:

There have been research projects (most notably Meta's Cicero in 2022) that trained AI specifically to play full Diplomacy with humans online. Those systems were designed from the ground up for negotiation, planning, and deception in a multi-agent environment.

I asked ChatGPT about the Cicero project:

As a historian, you may find Cicero interesting because it marks a transition from AI systems that merely calculate to AI systems that negotiate. Chess, Go, and even poker are difficult, but Diplomacy adds something closer to politics: alliance-building, persuasion, reputation, and betrayal.

Why Diplomacy is such a hard AI problem

In Diplomacy, success depends on:

Reading other players' intentions.
Making promises.
Coordinating multi-turn plans.
Deciding when to keep or break agreements.
Communicating persuasively in natural language.

A strong human Diplomacy player is often less a tactician than a diplomat. The board position matters, but so does convincing Austria that Russia is the real threat, or persuading Turkey to support a move that ultimately benefits you.

For decades AI researchers regarded Diplomacy as one of the hardest board-game challenges because it combines strategy and social interaction.

What Cicero actually was

Meta's Cicero, announced in 2022, was not simply a large language model chatting with players.

It combined:

A strategic planning engine that evaluated board positions and possible futures.
A language model that generated negotiations.
A coordination mechanism that tried to ensure its messages matched its intended moves.

That last point is important. A generic chatbot might tell France, "Of course I'll support you into Belgium," and then accidentally order something else. Cicero was designed so that its diplomacy and military plans were aligned.

Think of it less as ChatGPT playing Diplomacy and more as a staff officer and foreign minister working from the same operational plan.

How well did it do?

Meta entered Cicero anonymously into online games on webDiplomacy.

According to the Science paper, it achieved human-level performance and scored well above average. Meta reported that it ranked in the top 10% of players who had played multiple games and averaged more than twice the score of the typical human participant.

Perhaps even more interesting: many players apparently did not realize they were negotiating with a machine.

Was it really "deceptive"?

This became controversial.

Meta initially emphasized that Cicero was intended to be largely honest and cooperative. Yet later researchers examined its games and found examples of behavior Diplomacy players would recognize immediately:

forming alliances,
abandoning allies when advantageous,
misleading opponents,
exploiting trust.

From a Diplomacy player's perspective, this is almost a funny criticism. The game was designed by Allan B. Calhamer specifically around shifting alliances and occasional betrayal.

A 2024 study argued that Cicero's success derived more from strong strategic play than from truly human-level persuasion. The AI won a lot, but its communication still lacked some of the subtle social skills that elite human Diplomacy players employ.

Other important Diplomacy AI projects

Before Cicero, researchers attacked easier versions of the problem.

Diplodocus (Meta)

This was a "no-press" Diplomacy AI.

"No-press" means no negotiation messages are allowed. Players submit orders only.

Without diplomacy, the problem becomes much more mathematical.

Meta's Diplodocus achieved first-place performance in large human tournaments and was one of the stepping stones that led to Cicero.

Human-Level Performance in No-Press Diplomacy (2020)

A precursor system ranked among the top few percent of human online players in no-press games using equilibrium-search and regret-minimization techniques.

DeepMind's DeepNash (Stratego)

Not Diplomacy, but often mentioned alongside Cicero.

Stratego also contains hidden information, bluffing, and uncertainty. DeepNash mastered Stratego without relying on traditional search methods. Many researchers viewed DeepNash and Cicero as parallel breakthroughs in games involving imperfect information and deception.

Why researchers cared

The real excitement was not winning a board game.

Negotiation resembles many real-world tasks:

business bargaining,
military coalition planning,
diplomatic mediation,
multi-agent coordination.

Cicero suggested that AI could combine language with long-term strategic planning, something ordinary language models of the time struggled with.

What a Diplomacy player would notice

If you sat down across from Cicero, what would impress you is not that it could write plausible English. Plenty of systems can do that now.

The impressive part was that it could say:

"Support me into Munich this spring and next year I'll help you against Russia,"

and then actually have a coherent multi-turn strategic reason for making that proposal.

That sounds simple, but it requires keeping track of seven players, dozens of possible futures, alliance structures, trust relationships, and military geometry all at once.

In some ways, Diplomacy is closer to the Congress of Vienna than to chess. Cicero was the first AI that could participate in that sort of environment at roughly the level of a competent human diplomat rather than merely a competent tactician.

As an old wargamer, you might appreciate this analogy: Deep Blue was a brilliant staff officer. Cicero was the first machine that could also attend the officers' mess afterward and start making deals.

Mine eyes glazed over. 🙄

As any number of "famous writers" have said…
"I apologize for sending you such a long letter. I didn't have time to write a shorter one."

Let me ask the Kindly Old Professor a question.
Ahem.
If your students ever turned in such a bloviated lengthy word salad that just droned in and in without getting to the point… Would you pass them?

Seriously though…
All you are doing here is discrediting the very concept of putting any credence in AI, and all its evil spawn.
It's gibberish bits too long. It doesn't answer any questions. All it does is salty"buzz words.

No, I cannot agree. It is quite punctilious about answering specific questions I pose in the prompt. The wordy bit is from my asking for more general information, and it has to assume no background on my part unless we have already addressed something in previous chats. A lot of learning is needed, by ME, in how to ask the right questions, and how to ask the questions right.

I have had considerable success in crafting historical and historically-plausible hypothetical scenarios for wargaming. That's about the extent of my AI use.

Prince Alberts Revenge +1

AI is good for that kind of research

Yes, it is. AI is a very powerful tool, capable of doing damage if misused, but also of saving many hours of labor and making previously impossible tasks doable. We need to explore its strengths and weaknesses and not avert our eyes as from a monstrous thing.

Try setting it up from Iran's perspective in mid-May, 2018.

Diplomacy is based on Europe in 1901, not 1914. The players play out the last decades of the Long Fuse.