Abstract
Over the past couple of years, the influence of AI has been expanding across the entire Internet. The new technological wonder of the decade, some might say. But perhaps it came too soon, as the greedy and lazy have already been using AI for their own benefit, harming others’ livelihoods at the same time. My two projects, the capstone project report and the STS research paper, both focus on the fast-arriving consequences of AI and what can be done about it. There have been many cases where outside parties replicate another person’s voice and make them say controversial statements, unfairly framing the victims for something they did not do and spreading deception online. The capstone project’s purpose is to prevent deepfake clips from spreading by being able to identify them. The STS research paper focuses specifically on AI generative art, its benefits and risks, and the consequences of AI art being so overused online. My reason for writing this paper is to point out the contradiction of generative AI’s original purpose: to promote and share art, when it is actually having the opposite effect. These two projects are connected because they are both focused on the negative effects of AI, the first is trying to solve an issue, and the second is underlying an issue itself.
Over the years, generative AI has been improving in its ability to generate and mimic voices. While this may have some benefits like bringing a passed loved one back to live in voice, it has even worse risks. Deepfakes can impersonate other people and frame them by saying something controversial on the orders of malicious users, surrounding the victim in hate and controversy they never deserved. The project’s purpose is to fight against these attacks by using AI to attack back. Working on a team with two others, our method was to combine two AI systems to create a viable AI deepfake detector, which could be used on a website. The combined AI system could train on a person or multiple people’s voices, and identify which ones were real and which ones were deepfake.
Even given our skills and the short amount of time during the semester, this was a challenging task. Our AI system was good at identifying the real voice and differentiating from accents of that same voice, but it was not effective enough at identifying deepfakes to be viable in the real world. But that was to be expected, so our main goal was to focus on creating a prototype that can prove a project like this could be done.
For my STS paper, I decided to focus on generative AI. I myself am an artist, so I was very passionate about this topic. My research question was, “How exactly are GenAI systems harming the art industry, and who is at fault?” I asked this because while the users of generative AI are the ones harming the real artists, they may not be the initial cause of this mess. Rather it could be the work of the large AI companies pushing AI onto everything. My methodology was focused on identifying all the actors involved with generative AI art, what their actions caused the problem, and why they were happening.
I learned that generative AI was first being trained upon millions of stolen artwork from other artists, pushed to a level of improvement that made it very hard to tell the difference between real art and AI art, and allowed non-artists to use this tool as a way to gain money and popularity. Worst of all, it broke a lot of trust in the art community, where users have to be skeptical if any art they are looking at is human-made or AI-generated. In conclusion, the multiple actions of actors have created a system of uncertainty and disruption in the art community, much like how other AI systems have been doing the same on the Internet.