Building Trust using Algorithmic Transparency in Automation

As technological breakthroughs advance, humans will have more interactions and relations with artificial intelligence such as self-driving cars, voice assistants, and social media in their everyday lives. With the normality of using these algorithmic interfaces comes a greater need to implement transparent design so humans will trust and utilize these smart technologies. The present study sought to investigate the transparent design of algorithmic interfaces to cultivate trust in users. Participants were randomized in a computer generator and assigned to two experimental groups: no transparency and transparency. In the no transparency condition, participants were not provided information regarding the credibility of the automated agent; whereas, in the transparency group, participants were given extensive information on how the automated agent is qualified and reliable. Participants engaged in a pattern recognition task with the aid of a computer recommendation agent that produced a high-uncertainty environment where participants decided to consider or disregard advice given to them by an automated agent. Before submission of the final answer on each trial, participants received a recommendation from the computer recommendation agent and were able to stay with their initial response or switch their answer based on the recommendation. Shortly after they made their final answer, feedback revealed the correct answer. Participants filled out a self-reported trust report before and after the task. It is expected that participants who were in the transparency condition will report higher levels of trust. Designing for user trust through transparent algorithms is significant so that users use these interfaces to their highest capability. It is important to note that there is an optimal level of transparency that systems need to provide so that users can achieve a goal under a situation of uncertainty and vulnerability. Future experimental studies should evaluate varying levels of transparency to find the optimal level of transparency.

Hello and welcome, my name is Kayla Herrera, and I am an undergraduate psychology student concentrating in cognitive behavioral neuroscience. The research I will be presenting today investigates transparent algorithms employed to foster trust in technological users. My project diverges from Dr. Tyler’s Shaw’s HEART lab in the psychology department.

Our parents taught us as children to not get in a stranger’s car for obvious reasons such as strangers may not be trustworthy, we do not know their intentions, etc. Today, companies like Uber, Lyft, and other ride sharing options have normalized our parents fear of getting into the car of a stranger. It has become common for people to trust in applications like Uber because these applications successfully foster trust. Uber uses transparency or openness for their users in their algorithm by sharing a certain amount of information like type of car, driver’s name so that users trust their application. This ability to foster trust through using clear and honest algorithms is where my research interests lie.

As technological breakthroughs advance, humans will have more interactions and relations with artificial intelligence such as self-driving cars, voice assistants, and social media in their everyday lives. With the normality of using these algorithmic interfaces comes a greater need to implement transparent design so humans will trust and utilize these smart technologies.
I have included relevant research from a SeoClarity chart that reveals how Americans trust big tech companies from Google to Titkok. The results were fascinating by showing Google is the most trusted tech company whereas TikTok is the least trusted tech and social media brand in the US.

The present study sought to investigate the transparent design of algorithmic interfaces to cultivate trust in users. Participants were randomized in a computer generator and assigned to two experimental groups: no transparency and transparency. In the no transparency condition, participants were not provided information regarding the credibility of the automated agent; whereas, in the transparency group, participants were given extensive information on how the automated agent is qualified and reliable. Participants engaged in a pattern recognition task with the aid of a computer recommendation agent that produced a high-uncertainty environment where participants decided to consider or disregard advice given to them by an automated agent. Before submission of the final answer on each trial, participants received a recommendation from the computer recommendation agent and were able to stay with their initial response or switch their answer based on the recommendation. Shortly after they made their final answer, feedback revealed the correct answer. Participants filled out a self-reported trust report before and after the task. We are currently still investigating this study but I expect to see that the level of transparency will conclude to higher levels of trust in participants.

This is just the beginning of transparency algorithms research. It can be the turning point that educates companies, institutions, health care, and government facilities of the importance of transparency.

Lastly, I’d like to acknowledge the graduate mentors, professors, and colleagues that have patiently worked with me, believed in me to become a leader, and continued to support my future endeavors. Thank you.