Abstract
My overall problem is that there are negative effects on the environment and people from increasingly powerful artificial intelligence. The reason this matters is because currently, not enough research has been done to justify the widespread use of AI without knowing its full implications. My technical research relates to this problem by exploring how more efficient AI could be used in workforce tasks instead of just using the fastest model for it. This could allow for use of AI in hazardous areas while not contributing as much to the AI energy problem. My STS research relates by analyzing different problems seen already in AI systems, applying them to a power system, and trying to explore ways these issues could be solved. Essentially, I want to explore how an AI, if deployed without knowledge of the potential impacts, changes the lives of people and how knowing that should change AI implementation.
My technical problem is that people in shipyard environments still must do dangerous and fatiguing jobs that could be outsourced to artificial intelligence equipped robots. The main problem with this is that these people are very skilled at what they do, but it puts a lot of strain on them in the long run, leading to issues later in life. We looked at multiple papers on the effects of hazardous conditions and found that a promising solution could be in using autonomous robots for some of these dangerous tasks. In using a lightweight vision language action (VLA) model for our technical project, we found that using this type of model is promising but currently takes a significant amount of data in order to train it properly; even for a simpler task. There are a lot of other lightweight models that we didn’t use to test this, but these could yield better results in the future.
My STS research examines how an ethics of care based power system with integrated AI could be realized through analyzing the current issues seen in AI systems and seeing how these could be mitigated. I did literature review on multiple articles that investigated the effects of AI bias, data insecurity, and the lack accountability in AI enabled systems. My findings indicate that first, all of the measures to mitigate issues in these systems should be as transparent and implemented as uniformly as possible. Secondly, I gathered that, in order to have an ethics of care based power system, energy regulations and policies must be changed to account for the potential impacts of AI induced issues. These recommendations can also be applied to other AI based systems, but I was focusing on the impacts from an energy viewpoint.
From the technical side, we were partially successful in showing a more efficient AI could be used in hazardous environments, but there were limitations in how well our model could do a task and how quickly it could be trained to do it. I would recommend future researchers to try using different energy efficient models to evaluate how fast training could happen and how complex of tasks can it do well. From the STS side, I believe I have contributed to research on the impacts of AI and tried to provide some ideas on how AI could be used better and more carefully.
Finally, I would like to thank Keith Williams, Troy Harmon, Jeanu Joo, and Andrew Hwang for advising and working with me on the technical project; I could not have completed it without them. Also, thanks to Austal USA Advanced Technologies for financially supporting and providing most of the equipment for the technical project. I would also like to thank my family for their support and Caitlin Wylie for her advice and support throughout my sociotechnical writing.
