Abstract

Throughout the relatively short history of artificial intelligence (AI), there has been a significant concern surrounding AI’s ability to incorporate and maintain certain characteristics which were not inherently modeled out in its coding. These behaviors stem from the prominent usage of neural network AI, which can inherit human biases from the input data it receives. This paper argues for two possible avenues to combat these biases. The first is to rethink the traditional framework for neural network projects and retool them to be usable by a Generative Adversarial Network (GAN). In a GAN’s zero-sum game, two network techniques can combat discriminatory beliefs or incorrect values in manners unlike traditional networks, while not necessitating a completely new algorithm for neural network systems already proven effective. GAN technology is one approach for helping to solve the bias issue but confronting the humans behind the AI is just as important. Incorporating humanistic techniques such as unconscious bias training and participatory design into AI development further promote equitable AI by fostering communication between others. AI biases are merely reflections of human biases in a technological form, and any “bad” output data stems from bad output humanity has generated from itself. There cannot be a perfectly unbiased AI model, as there are no perfectly unbiased humans, and the influences of economies, politics, and other vested interests ensure this to an even larger degree.

Faculty Advisor/Mentor

Iria Giuffrida

Document Type

Paper

Disciplines

Artificial Intelligence and Robotics

DOI

10.25776/htk2-2h94

Publication Date

2021

Upload File

wf_yes

Share

COinS
 

Tackling AI Bias with GANs

Throughout the relatively short history of artificial intelligence (AI), there has been a significant concern surrounding AI’s ability to incorporate and maintain certain characteristics which were not inherently modeled out in its coding. These behaviors stem from the prominent usage of neural network AI, which can inherit human biases from the input data it receives. This paper argues for two possible avenues to combat these biases. The first is to rethink the traditional framework for neural network projects and retool them to be usable by a Generative Adversarial Network (GAN). In a GAN’s zero-sum game, two network techniques can combat discriminatory beliefs or incorrect values in manners unlike traditional networks, while not necessitating a completely new algorithm for neural network systems already proven effective. GAN technology is one approach for helping to solve the bias issue but confronting the humans behind the AI is just as important. Incorporating humanistic techniques such as unconscious bias training and participatory design into AI development further promote equitable AI by fostering communication between others. AI biases are merely reflections of human biases in a technological form, and any “bad” output data stems from bad output humanity has generated from itself. There cannot be a perfectly unbiased AI model, as there are no perfectly unbiased humans, and the influences of economies, politics, and other vested interests ensure this to an even larger degree.