MARTHA Speaks: Implementing Artificial Intelligence and the Theory of Mind for Intuitive and Inferential Communicative Acts Involving Deeply Nested Layers of Rationality
Session Number
F14
Advisor(s)
Piotr Gmytrasiewicz, University of Illinois at Chicago
Location
A-121
Start Date
28-4-2016 8:50 AM
End Date
28-4-2016 9:15 AM
Abstract
The theory of mind is, perhaps, one of the most important psychological constructs in humans today, as it allows us to record, understand, and predict the desires, intentions, beliefs, and goals, of other individuals or agents. The purpose of this project was to implement an assistive agent that we call MARTHA, or the Mental-state Aware Real-time Thinking Assistant, in order to simulate various deeply nested layers of user and self-induced thought. This assistive Artificial Intelligence agent is able to process a combination of social norms/implications and user motivations in order to provide the user with an optimal course of action. The action with the greatest utility is the one that is selected. In order to determine the most intuitive communicative acts, I implemented a recursive searching algorithm for each nested layer of thought that determined all possible plans of action and their hypothetical consequences. The general scenario involved the interaction between two cars or people. I implemented four practical applications of this scenario, and the most useful response was provided by MARTHA 100% of the time. These results are a positive indication of the wide range of pragmatic situations in which MARTHA could be extremely useful.
MARTHA Speaks: Implementing Artificial Intelligence and the Theory of Mind for Intuitive and Inferential Communicative Acts Involving Deeply Nested Layers of Rationality
A-121
The theory of mind is, perhaps, one of the most important psychological constructs in humans today, as it allows us to record, understand, and predict the desires, intentions, beliefs, and goals, of other individuals or agents. The purpose of this project was to implement an assistive agent that we call MARTHA, or the Mental-state Aware Real-time Thinking Assistant, in order to simulate various deeply nested layers of user and self-induced thought. This assistive Artificial Intelligence agent is able to process a combination of social norms/implications and user motivations in order to provide the user with an optimal course of action. The action with the greatest utility is the one that is selected. In order to determine the most intuitive communicative acts, I implemented a recursive searching algorithm for each nested layer of thought that determined all possible plans of action and their hypothetical consequences. The general scenario involved the interaction between two cars or people. I implemented four practical applications of this scenario, and the most useful response was provided by MARTHA 100% of the time. These results are a positive indication of the wide range of pragmatic situations in which MARTHA could be extremely useful.