Using Machine Learning with NLP and Computer Vision Techniques as a Means of Constructing an Optimized Content Moderation System; An Analysis of the Ethical and Societal Impacts of Using Artificial Intelligence for Content Moderation on Social Media Platforms

Can artificial intelligence (AI) be better utilized to draw the line between freedom of expression and harmful content on social media? Knowing the current state of social media, it is crucial to establish a clear understanding of how modern moderation systems function and how they can be improved. Many social media platforms currently utilize AI in their moderation systems to streamline effective flagging, as human moderation teams are no longer able to keep up with the millions of posts uploaded every day. As such, my STS research delves deeply into the ethical and societal implications of AI usage in modern moderation systems on popular platforms like Instagram, Twitter/X, and YouTube. Building on these findings, the technical portion of my project involves the formulation of a proposed moderation system by leveraging machine learning (ML) and sentiment analysis to construct a highly tuned model.

In my STS research, I investigated the ethical and societal impacts of AI content moderation on popular platforms, as noted above. Higher priority was given to key areas like free speech concerns regarding over/under-censorship, algorithmic bias, and the inherent errors with mis-flagging and improper penalties. To effectively analyze such components of moderation systems, a dual-faceted approach was adopted. This approach involved reviewing existing case studies and conducting a dedicated survey for social media users. The findings revealed significant public concern for current moderation practices on most platforms, despite the advanced NLP techniques and filtering systems that are already in place. Key findings to improve existing practices included increasing the transparency to users, incorporating more human judgement for applicable cases, and establishing a method to mitigate biases. Ultimately, it was reinforced that while expressive freedom should be maintained, the current systems must be altered through further intervention to promote fairness and social equity in today’s digital society.

The technical portion of my project built on what was learned from the detailed research conducted relating to current moderation systems. The survey results were focused on, drawing upon the opinions of avid social media users to set a possible improvement plan to existing models for social media post flagging. By using sentiment analysis and computer vision along with more highly tuned hyperparameters, a proposed system to improve the user experience and safety on social media was constructed. The proposed system involves mechanisms for more customized flagging, making better use of user feedback when reporting posts. However, to balance out the higher weight that users are given in such a system, appropriate checks must be implemented to avoid exploitation. My technical proposal serves as a blueprint for a consistent moderation system, which I do plan to implement in the future.

My research revealed the underlying issues in content moderation systems that are being used regularly, and they must be adjusted accordingly. Through the insights provided both by my STS research and in the technical proposal, a concrete implementation can be made to improve the digital space that so many individuals participate in. During the time of this research and technical work, I evolved from viewing content moderation as a pure case-by-case issue to a significant problem that software alone cannot solve. Through keeping strong engineering principles and the utilization of AI and ML tools, I am sure that a more free and safe online space can be reached with an updated moderation system.

School of Engineering and Applied Science

Bachelor of Science in Computer Science

Technical Advisor: Yen-Ling Kuo

STS Advisor: Richard Jacques