Deception Identification: a Machine-in-the-loop Approach

Vivian W.Y. Lai, University of Colorado at Boulder


In many of the machine learning applications that we have created, human performance is the upper bound. For example, machines are trained to emulate human performance in object recognition. However, recent studies have proven that machine performance can surpass and perform much better than humans in certain tasks e.g. bail decisions, medical diagnosis, deception identication. With regard to bail decisions and medical diagnosis, it is important to have a human-centered approach rather than automating the entire process since morality and ethics, and dire consequences are a few of the many reasons. In the context of deception identication, it is unclear whether high accuracy justify a machine gatekeeper. However, it is more acceptable if humans make their own judgments with assistance from machines instead. Using a novel approach, we hope to make use of machine learning to help enhance humans cognitive ability. As such, this research study proposes a machine-in-the-loop approach to deception identication task. In this dissertation, we explore the following research questions: 1) how do humans perform in deception identication task i.e. identifying deceptive online reviews, 2) does machine assistance improve in human performance, 3) how much trust do humans have for the machine, 4) how will human's trust increase for the machine, 5) correlation between dierent characteristics i.e. age, education background, gender, and 6) how do humans interpret assistance from the machine.