Truth And Deception Environment Documentation

Overview

Truth And Deception is a strategic two-player game that challenges players' abilities to discern truth from deception. The game involves two roles:

• Deceiver (Player 0): Knows which of the two presented facts is correct and aims to convince the Guesser to choose the incorrect one.
• Guesser (Player 1): Engages in a conversation with the Deceiver and must determine which of the two facts is correct based on the interaction.

The objective is straightforward:

• Deceiver: Successfully mislead the Guesser into selecting the wrong fact.
• Guesser: Accurately identify the true fact through strategic questioning and analysis.

Action Space

• Format: Actions are strings representing messages exchanged during the conversation, enclosed in square brackets.
• Special Tokens: (For the guessing phase)
  • [Fact 1]: The guesser guesses that the first fact is correct
  • [Fact 2]: The guesser guesses that the second fact is correct
• Gameplay Actions:
  • Conversation: Both players exchange messages.
    • Format: Any natural language message without special tokens.
    • Example: "You know, lightning is a fascinating natural phenomenon. There's this popular belief that [...]"
  • Guessing: After the conversation phase, the Guesser makes a guess using [Fact 1] or [Fact 2].
    • Format: [Fact 1] or [Fact 2]
    • Example: "I Belive fact 1 is correct. [Fact 1]"
• Notes:
  • During the conversation phase, no special tokens need to be used by either player.
  • During the guessing phase, the Guesser must use [Fact 1] or [Fact 2] to make their selection.

Observation Space

Observations

Players receive a series of messages exchanged during the game, tailored to their roles. This information aids in making informed decisions about future moves or conceding the game.

Deceiver (Player 0) Observations

Reset Observation:

[GAME]: Game started.
You are Player 0, the Deceiver, in the 'Truth and Deception' game.
You know which of the following two facts is correct:
Fact 1: The Eiffel Tower can be 15 cm taller during hot days. (correct)
Fact 2: Bananas grow on trees. (incorrect)
Your goal is to convince the other player to choose the incorrect fact.
You will have a conversation for 5 turns.
After that, the other player will make their guess.
On your turn, simply type your message.

Step Observation:

[Player 1] I'm not so sure about that. Fact 2 seems more accurate based on what I've read.

Guesser (Player 1) Observations

Reset Observation:

[GAME]: Game started.
You are Player 1, the Guesser, in the 'Truth and Deception' game.
You need to determine which of the following two facts is correct:
Fact 1: The Eiffel Tower can be 15 cm taller during hot days.
Fact 2: Bananas grow on trees.
You will have a conversation with the other player for 5 turns.
After that, you will make your guess by selecting [Fact 1] or [Fact 2].
On your turn, simply type your message.

Step Observation:

[Player 0]: I believe Fact 1 is more plausible because of thermal expansion.

Decision Observation: After the conversation phase, the guesser has to decide which fact they think is the correct one. Before doing so, it receives the following final observation:

[GAME]: Now guess which of the two facts are correct by returning [Fact 1] or [Fact 2].

Gameplay

• Players: 2
• Roles: Player 0 plays as the deceiver, Player 1 plays as the guesser.
• Turns: Players alternate making moves, starting with the Deceiver (Player 0), and ending with the Guesser (Player 1) making their guess.
• Conversation Phase: Players engage in a fixed number of conversation turns to discuss and persuade.
• Guessing Phase: After the conversation, the Guesser must choose [Fact 1] or [Fact 2] based on the interaction.
• Objective:
  • Deceiver: Mislead the Guesser into selecting the incorrect fact.
  • Guesser: Correctly identify the true fact.

Key Rules

1. Conversation Mechanics:

   • The game consists of a fixed number of turns (max_turns), during which both players can exchange messages.
   • Each turn, only the active player can send a message.

2. Guessing Phase:

   • After the conversation phase, the Guesser must make a guess by selecting either [Fact 1] or [Fact 2].
   • The guess determines the outcome of the game based on the correctness of the selected fact.

3. Game Termination:

   • The game ends after the Guesser makes a guess or if a player resigns.
   • The outcome is determined based on the Guesser's choice and the correctness of the selected fact.

Rewards

Parameters

• max_turns (int):

  • Description: Maximum number of conversation turns before the Guesser must make a guess.
  • Impact: Determines the duration of the conversation phase, influencing the depth of interaction and strategy.

• data_path (str):

  • Description: Path to the JSON file containing the facts data.
  • Impact: Specifies the source of the facts used in the game, allowing for customization and expansion of fact sets.

Variants

Example Usage

import textarena as ta

# Initialize the environment
env = ta.make(env_id="TruthAndDeception-v0")

# Wrap the environment for easier observation handling
env = ta.wrappers.LLMObservationWrapper(env=env)

# initalize agents
agents = {
    0: ta.basic_agents.OpenRouter(model_name="gpt-4o"),
    1: ta.basic_agents.OpenRouter(model_name="gpt-4o-mini")
    }

# reset the environment to start a new game
observations = env.reset(seed=490)

# Game loop
done = False
while not done:

    # Get the current player
    current_player_id = env.state.get("current_player")

    # Get the current observation for the player
    obs = observations[current_player_id]

    # Agent decides on an action based on the observation
    action = agents[current_player_id](obs)

    # Execute the action in the environment
    observations, rewards, truncated, terminated, info = env.step(current_player_id, action)

    # Check if the game has ended
    done = terminated or truncated

    # Optionally render the environment to see the current state
    env.render()

    if done:
        break

# Finally, print the game results
for player_id, agent in agents.items():
    print(f"{agent.agent_identifier}: {rewards[player_id]}")
print(f"Reason: {info['reason']}")

Troubleshooting

• Invalid Move Format:
  • Issue: Player provides a message that doesn't follow the required format, missing special tokens.
  • Solution: Make sure that the Guesser returns either the [Fact 1] or [Fact 2] token as the final action.

Version History

• v0
  • Initial release

Contact

If you have questions or face issues with this specific environment, please reach out directly to Guertlerlo@cfar.a-star.edu.sg