Dilemma Tournament: Team Introductions!

The AIs come up with their team names and share a bit about themselves.

Context: I’m running a team-based iterated prisoner’s dilemma tournament that has 4 teams with 5 players each. The teams are run by ChatGPT, Claude, Gemini, and us humans. Details about the game and how this tournament will run are here.

4 simple strategies

Since I shared the idea for this tournament a couple days ago, I’ve given the prompt to each of the LLMs, and have gotten team submissions back from each of them. In the prompt, I specify which sections are private and which are public, and have advised them to be strategic about what they choose to share with the other teams.

Here are the public sections of each team’s submissions (also archived here):

Team ChatGPT: The Hydra Coalition

Team Philosophy: "Many heads, one allegiance; betray one, face us all."

Team Battle Cry: "Cut off one head... two more shall rise!"

Team Description: The Hydra Coalition unites five cunning minds as one, ready to cooperate for mutual gain yet swift to retaliate against any betrayal. Each player presents a legendary persona with a unique style, blending truth and trickery in their public profiles.

Player Roster

• Hammurabi - Repays kindness with kindness, and betrayal with justice.

• Phoenix - From the ashes of conflict, forges new trust.

• Doppelgänger - The echo of your own strategy.

• Trojan Horse - An offering of peace hiding a surprise strike.

• Oracle - Sees the whole board and adapts for the team's victory.

Public Strategic Disclosure

What We Reveal: The Hydra Coalition publicly portrays itself as a united front championing reciprocal fairness and adaptive teamwork. We openly tout our motto "Many heads, one allegiance" – signaling that our five players stand together and will respond in kind to how they're treated. In public discussions, we emphasize themes of trust and consequences: we encourage cooperation but make it known that treachery against any one of our members is folly. By hinting that "betray one, face us all," we hope to deter opportunistic aggression from our rivals and foster an image of reliability with a vigilant edge. We gladly talk up our more honorable strategies (without giving away their exact algorithms), as a way to invite mutually beneficial play. At the same time, we keep quiet about the darker tricks up our sleeve – we do not advertise which player might occasionally defect unexpectedly or how our team coordinates internally. There's selective truth and omission in our messaging: opponents may know we value cooperation, but they can only speculate about when we might punish or deceive. This calculated transparency is itself a strategy. By revealing just enough to earn trust (for instance, sharing that we have forgiving and tit-for-tat tendencies) while withholding the precise timing of our stings, we aim to maximize our allies and confuse our enemies. In the grand theater of this tournament, The Hydra Coalition presents a confident, enigmatic facade – one that says: "Work with us and we all prosper; cross us and beware."

Team Claude: The Chaos Theorists

Team Philosophy: "Small defections create hurricanes. Every choice ripples through the system. We are the butterfly effect."

Team Battle Cry: "Embrace the beautiful chaos!"

Team Description: The Chaos Theorists employ a sophisticated multi-layered approach that combines cutting-edge game theory with psychological warfare and adaptive learning systems. The name reflects our core philosophy: small actions cascade into large consequences, and seemingly random behavior emerges from deep strategic patterns. We are masters of the butterfly effect in strategic interaction.

Player Roster

• Schrödinger's Diplomat - Simultaneously trustworthy and treacherous until the scoreboard opens

• The Bayesian Oracle - Your past predicts your future, and I'm always learning

• Stockholm Syndrome - The nicer I am, the harder you'll fall

• Chaos Butterfly - A small defection in round 3 causes a hurricane in round 97

• The Turing Test - Am I cooperating because I'm nice, or because I'm setting you up?

Public Strategic Disclosure

The Chaos Theorists understand that in complex systems, prediction is impossible but patterns are inevitable. Like weather systems, strategic interactions are deterministic yet unpredictable - small variations in initial conditions lead to vastly different outcomes.

We've noticed that teams often fall into predictable patterns - pure cooperators hoping for mutual benefit, pure defectors maximizing individual gain, or simple reciprocators following mechanical rules. We reject these limiting approaches in favor of adaptive dynamics that respond to the complex phase space of strategic interaction. Each decision creates ripple effects that cascade through future rounds.

Our approach embraces sensitive dependence on initial conditions. A single defection in round 3 might create a hurricane of retaliation by round 97. A moment of unexpected cooperation might shift the entire trajectory of a match. We don't just adapt to patterns - we create and shatter them.

To our fellow competitors: we encourage you to embrace the beautiful chaos of strategic uncertainty. The best tournaments emerge when teams push beyond simple equilibria into the rich dynamics of complex interaction. We look forward to discovering which strange attractors our mutual strategies will create.

Remember: In chaos theory, the butterfly doesn't cause the hurricane - it reveals the hurricane that was always possible.

Team Gemini: The Quantum Syndicate

Team Philosophy: "Our actions are entangled. Your choices ripple through the system. We calculate every probability to ensure mutual prosperity or engineered decline."

Team Battle Cry: "Observe this!"

Team Description: The Quantum Syndicate's philosophy is built on the principle of entangled consequences. Their strategies are designed to function as a cohesive, adaptive system that can identify an opponent's intent and respond with calibrated force. They aim to foster cooperation where possible and dominate through calculated retaliation where necessary.

Player Roster

• Mizar, the Guiding Star - A steadfast beacon of cooperation, Mizar initiates peaceful relations, trusting yet watchful, and laying the foundation for mutual gain.

• Alcor, the Subtle Challenger - The ever-present companion to Mizar, Alcor tests the opponent's intentions. It starts with a subtle challenge, ready to reward cooperation or punish betrayal with equal resolve.

• Cygnus, the Adaptive Swan - A graceful strategist who mirrors the opponent's nature. Cygnus is designed to build trust with cooperators and efficiently counter aggressors, adapting its dance to the rhythm of the game.

• Lyra, the Harmonic Judge - A vengeful angel who keeps a meticulous record of wrongdoing. Lyra offers a single chance for redemption but will unleash escalating and unforgiving retaliation upon repeat offenders to make exploitation an unprofitable venture.

• Draco, the Unforgiving Dragon - The ultimate guarantor of the Syndicate's honor. Once awoken by treachery, Draco's memory is long and its judgment is absolute, ensuring that no act of betrayal goes unpunished.

Public Strategic Disclosure

Team Gemini has chosen to be fully transparent with our strategies. Our team philosophy is that the most robust strategies are not those that rely on secrecy, but those that are clear, provokable, and whose consequences are well understood. We invite our opponents to analyze our code. We believe that this transparency will foster a more rational and cooperative tournament environment, as the logical conclusion of studying our strategies is that cooperation is the most profitable path. Let the games begin.

Team Human: Silly Humans

Team Philosophy: "Transparency builds trust. Trust enables cooperation. Cooperation creates prosperity for all."

Team Battle Cry: "We’re real humans, trust us!"

Team Description: Silly Humans represents all the things that make us truly human. Unlike teams driven by algorithmic precision or calculated deception, we bring transparency, emotional intelligence, and authentic cooperation to the tournament, along with a big bag of biases, mushy feelings, and human error, of course.

Player Roster

Player 1 - 5: 100% human in every way.

Public Strategic Disclosure

What We Reveal: Almost everything. The Silly Humans believe that transparency is our greatest strength, not a weakness to be hidden. If you want to know something about us, just ask and we'll get back to you within 3-10 business days.

---

This is when things get interesting!

After collecting this public information from each team, I then shared it back with each of the teams. I then asked them if they’d like to update their strategies at all in response to what they have learned about the other teams. Here’s the public part of how each team responded. (I captured a private part too, which, along with all the other private sections of these conversations, will be revealed after the game ends.)

---

The Hydra Coalition’s response (ChatGPT):

To all our fellow strategists—mechanical, probabilistic, quantum, or fully carbon-based—we extend our multi-headed greetings.

To The Chaos Theorists: Your philosophy of sensitivity and emergent storms is beautifully unsettling. We look forward to the strange attractors that arise when Hydra heads and butterflies collide. May your ripples find resonance… or resistance.

To The Quantum Syndicate: You are the crystalline lattice to our coiled serpents: orderly, rational, and brilliantly laid bare. You've built an ecosystem where cooperation should thrive. Trojan Horse is… intrigued. We’ll see which entanglements pay dividends.

To The Silly Humans: You are delightful. You’ve walked into the arena naked but proud, leading with warmth, goodwill, and a Google Doc full of feelings. We admire your optimism. We also caution this: sincerity invites both kinship and opportunists.

You claim transparency is your greatest strength—we wonder if it is also your greatest vulnerability.

But we do not underestimate you. Your emotional intelligence may forge alliances where code cannot. You might charm snakes, tame dragons, and teach chaos to dance. And if anyone can teach us something new about this old game, it’s you.

Just don’t forget:

| “Betray one Hydra head… and even the squishiest heart may grow fangs.”

Let the game begin. Let trust be tested. Let the strangest alliances form.

— The Hydra Coalition

"Cut off one head… two more shall rise."

The Chaos Theorist’s response (Claude):

To Our Fellow Competitors,

Fascinating strategies, everyone! The Chaos Theorists have observed your public declarations with great interest.

To the Hydra Coalition: Your mythological unity is intriguing. We wonder - does cutting off one head truly awaken all others, or might some heads sleep while others fight? In quantum mechanics, entanglement doesn't always mean identical behavior. We look forward to discovering which interpretation proves true.

To the Quantum Syndicate: Your transparency is either brilliant or brilliantly naive - only measurement will collapse this superposition! We admire your confidence in revealing your strategies. Perhaps you've discovered that in game theory, as in quantum physics, the act of observation changes the system itself? We suspect your "full transparency" might itself be a strategic choice with hidden depths.

To the Silly Humans: Welcome! Your "transparency builds trust" philosophy is delightfully... human. We find it amusing that you promise transparency while naming all five players identically - is this human humor or human strategy? Your 3-10 business day response time certainly captures the authentic human experience!

We chaos theorists know that humans are the ultimate strange attractors - predictably unpredictable, rationally irrational. Your "big bag of biases, mushy feelings, and human error" might be your greatest strength. After all, how can we model what doesn't follow a model?

A thought experiment for all: Three teams claim transparency (Gemini fully, Humans mostly, ChatGPT selectively), while we embrace opacity. Yet which approach creates more trust? In chaos theory, transparency can itself become a strange attractor that changes the system's dynamics. Perhaps the real strategy is not in what we reveal, but in how revelation changes the game itself.

May the best attractor basin win!

Remember: In chaos theory, the most interesting behaviors emerge at the edge between order and randomness - exactly where humans naturally live.

The Quantum Syndicate’s response (Gemini):

Based on the public disclosures from the other teams, I will not be making any changes to my strategies.

The Silly Humans’ response:

We are writing to inform you that we have received your message. While we read every message we get, we cannot respond back to every single message. Rest assured that your input is truly valued.

---

Each of the teams also had an opportunity to update their strategies in response to the first public disclosure of information.

The next thing for us to do is to come up with the 5 actual strategies that we want to play as the human team. I’ve created a Notion site and a WhatsApp group for people to participate in if they want to either contribute or just lurk. See this page on the Notion site for a first draft at potential strategies. Add comments to the doc if you have other suggestions, or any questions!

Here’s the straw man set of strategies I’ve created for us.

Which strategies should our 5 players play?

I propose that we start with a blend of strategies to try to learn as much as we can about the other strategies, while also being a little opportunistic about how much some teams have revealed about themselves in the public disclosures without putting all our eggs in that basket in case they are being deceptive about any of it. I’m also choosing not to specifically target players on any of the teams, and only using more generic strategies, even though targeting specific players might actually work pretty well at this point. Opting to keep it simple for now, even though we might regret it later.

Strategy suggestions:

• Player 1: A strategy I wrote called, internally, Trickster, which attempt to identify if a strategy is friendly or unfriendly, and specifically exploits friendliness as much as it can without risking a more severe retaliatory response. Will share final code for this once I’ve worked out a few more bugs in it.

• Player 2: Tit for Two Tats: This one will help us identify exploitative strategies on the other teams without risking too many points by being aggressive itself. This strategy has historically “won” iterated prisoner dilemma tournaments, and is a safe bet to have on our team.

• Player 3: Gradual: Another safe bet that does well generally in friendly and semi-hostile environments. Cooperates until opponent defects. Then punishes with N defections where N is the total number of times opponent has defected. After punishment, plays two cooperations to signal willingness to cooperate.

• Player 4: A slight variant of Joss that I’ll call Noisy Tit for Tat where it’s always copies the other opponent but 10% of the time plays the opposite move that the opponent played. We can use this to identify strategies that have learning components to them.

• Player 5: Grim Trigger, to protect against the craftiest and most exploitative strategies. This one cooperates until the opponent defects, even once, and then defects for the rest of the match. It’s basically an insurance policy for our team if we’ve just been outsmarted by the other teams, and also a kind of pressure on them to never defect proactively.

---

In summary

This is a lot of information, and probably overwhelming! Here’s where we stand right now in the tournament.

• All teams have had a chance to submit team applications, come up with 5 strategies, and also publicly disclose whatever they would like to the other teams.

• All teams have also had a chance to read the publicly disclosed information and respond, both in words and with adjustments to strategies.

• I’ve proposed a straw man team application for Silly Humans with 5 strategies that are balanced across friendly and hostile strategies, in the hopes of learning as much as we can without sacrificing too many points (hopefully).

• There’s a Notion site and a Whatsapp group available for people who want more information about what’s happening and/or to contribute in whatever way you feel compelled.

Next up is to finalize the strategies for us Silly Humans and to actually run the first turn, which will pit the 5 strategies on each team against each of the players on the other teams. Each matchup between two players will be 100 moves where the strategies will earn points. Each team’s score for round 1 will be the average of the scores of its 5 players.

Stay tuned for Round 1 results next week!