Table of Contents
Fetching ...

Personalized Help for Optimizing Low-Skilled Users' Strategy

Feng Gu, Wichayaporn Wongkamjan, Jonathan K. Kummerfeld, Denis Peskoff, Jonathan May, Jordan Boyd-Graber

TL;DR

This work investigates how real-time AI guidance can aid humans in a complex, collaborative game. It introduces pholus, a personalized advisor that outputs both moves and messages tailored to a player's history in Diplomacy, and evaluates its impact through 12 online games with 41 players. Quantitative results show that move guidance positively affects performance, with the strongest gains when both move and message guidance are provided, especially for novices who can reach or approach veteran performance. The study demonstrates the potential of human–AI collaboration to accelerate learning in unfamiliar environments and outlines directions for modeling user intent and reducing overreliance while improving accessibility.

Abstract

AIs can beat humans in game environments; however, how helpful those agents are to human remains understudied. We augment CICERO, a natural language agent that demonstrates superhuman performance in Diplomacy, to generate both move and message advice based on player intentions. A dozen Diplomacy games with novice and experienced players, with varying advice settings, show that some of the generated advice is beneficial. It helps novices compete with experienced players and in some instances even surpass them. The mere presence of advice can be advantageous, even if players do not follow it.

Personalized Help for Optimizing Low-Skilled Users' Strategy

TL;DR

This work investigates how real-time AI guidance can aid humans in a complex, collaborative game. It introduces pholus, a personalized advisor that outputs both moves and messages tailored to a player's history in Diplomacy, and evaluates its impact through 12 online games with 41 players. Quantitative results show that move guidance positively affects performance, with the strongest gains when both move and message guidance are provided, especially for novices who can reach or approach veteran performance. The study demonstrates the potential of human–AI collaboration to accelerate learning in unfamiliar environments and outlines directions for modeling user intent and reducing overreliance while improving accessibility.

Abstract

AIs can beat humans in game environments; however, how helpful those agents are to human remains understudied. We augment CICERO, a natural language agent that demonstrates superhuman performance in Diplomacy, to generate both move and message advice based on player intentions. A dozen Diplomacy games with novice and experienced players, with varying advice settings, show that some of the generated advice is beneficial. It helps novices compete with experienced players and in some instances even surpass them. The mere presence of advice can be advantageous, even if players do not follow it.

Paper Structure

This paper contains 19 sections, 9 figures, 2 tables.

Figures (9)

  • Figure 1: pholus generates move and message advice based on the game state and the player's past messages. Initially, as Austria, the player considers the Balkan Gambit, assuming cooperation from Italy and Russia to capture Serbia and Greece. pholus suggests the Hedgehog, a more defensive opening. The player eventually adopts a synthesized strategy: forming an anti-Turkey alliance with Italy (Lepanto) while using the Vienna unit to defend against a potential Russian attack in Galicia. The final decision highlights altered moves.
  • Figure 2: Regression coefficients for advice settings and player skills to predict supply center gains. Not receiving any advice from pholus is slightly disadvantageous. Move advice has a positive correlation with player performance. Receiving both forms of advice has the greatest positive impact. As expected, not having previous exposure to Diplomacy is indicative of bad performance. However, with the help of pholus's advice, Diplomacy novices are on the same level as veterans and have the potential to defeat experienced players.
  • Figure 3: Regression coefficients for all features.
  • Figure 4: Human uses pholus's advice directly without modification.
  • Figure 5: Italy paraphrases the advice with the same underlying intention.
  • ...and 4 more figures