Michigan Liar's Dice

Name: Michigan Liar's Dice
Author: University of Michigan

University of Michigan

DOCUMENTED

Published in academic literature

For:General Public & Enthusiasts

Free

Download on the App Store

App Summary

This application provides a game of Liar's Dice against an artificial intelligence opponent, serving as a public-facing demonstration of the Soar cognitive architecture. The associated research presents a case study that developed and evaluated a collection of AI agents using different combinations of probabilistic decision making, heuristic symbolic reasoning, opponent modeling, and learning. The authors demonstrate how this cognitive architecture can effectively integrate multiple forms of knowledge, using learning mechanisms to convert deliberate reasoning into tuned, implicit performance.

App Screenshots

Detailed Description

Functionality & Mechanism

Michigan Liar's Dice is a game-based simulation developed by the Soar research group at the University of Michigan. Gameplay sessions involve competing against one to three artificial intelligence opponents in the bluffing game Liar's Dice. The interface facilitates a complete set of bids, including pushes and exact calls. The system leverages the Soar cognitive architecture to power opponent behavior, which integrates probabilistic reasoning, symbolic heuristics, and opponent modeling to inform its in-game decision-making.

Evidence & Research Context

A published case study details the system's function as a testbed for integrating probabilistic decision-making and symbolic reasoning within a cognitive architecture.
The associated research describes the development and evaluation of a collection of Soar-based agents that utilize different combinations of probabilistic logic and heuristic knowledge.
The study demonstrated that the architecture's learning mechanisms could be used to convert deliberate probabilistic reasoning into implicit, tuned performance for the game agents.

Intended Use & Scope

This application is intended for the general public as a recreational game and for researchers as a demonstration of the Soar cognitive architecture. Its primary utility is to provide a challenging gameplay experience against a sophisticated AI. The application is not designed as a data collection instrument or a pedagogical tool for teaching principles of artificial intelligence.

Studies & Publications

1 publication

Peer-reviewed research associated with this app.

Development/Design Paper

A Case Study in Integrating Probabilistic Decision Making and Learning in a Symbolic Cognitive Architecture: Soar Plays Dice

Laird et al. (2023) · AAAI Fall Symposium Series Proceedings

Describes the research-driven development of this app

One challenge for cognitive architectures is to effectively use different forms of knowledge and learning. We present a case study of Soar agents that play a multiplayer dice game, in which probabilistic reasoning and heuristic symbolic knowledge appear to play a central role. We develop and evaluate a collection of agents that use different combinations of probabilistic decision making, heuristic symbolic reasoning, opponent modeling, and learning. We demonstrate agents that use Soar's rule learning mechanism (chunking) to convert deliberate reasoning with probabilities into implicit reasoning, and then use reinforcement learning to further tune performance.