Communication Effects when Playing with Humans in Multi-Round Games

October 06, 2021

Thursday, October 21st at 3pm

Advisor: Jacob Crandall

MS Thesis Defense for Tim Whiting

The main advantage of current AI algorithms - their ability to find efficient solutions to

coordination problems - is only applicable when the other person chooses to cooperate in the

way the AI expects. In the context of strategic encounters, many techniques have been used to

create an AI that cooperates effectively with opponents when their interests don’t align. In

simple strategic games, some of these learning processes do quite well. However, AI agents

don’t typically do as well when playing with humans as they do when playing each other. For my

Master’s Thesis, I will conduct a user study to determine how design decisions about verbal and

non-verbal communication affect the ability of a state-of-the-art algorithm to cooperate with