Retrieval Augmented Generation: Creating Specialized Chatbots; The Role of Social Media Algorithms in Influencing Health and Nutrition Trends

Modern society relies heavily on algorithms embedded into everyday life. From low-tech algorithms like traffic light sequences to innovative AI models that can perform facial recognition or image classification, algorithms affect nearly every corner of life in both hidden and explicit ways. Social media algorithms, for example, provide highly curated content by quietly analyzing user behavior in a fashion that many users may not realize. The nature of these algorithms can lead to the spread of misinformation and polarized opinions on topics such as politics or health practices. Popular AI tools such as Large Language Models (LLMs) have a much more obvious interaction with users, typically consisting of users asking questions of or seeking advice from chatbots backed by LLMs.

LLMs are AI models that generate complex bodies of text in response to user questions or prompts and have become popular aids in daily tasks, but do not have reliable knowledge of specific datasets and documents. Retrieval Augmented Generation (RAG) bridges this gap by providing an LLM with relevant data and documents, allowing it to create an informed response. The technical report discusses the process of creating a RAG-backed chatbot that could answer user questions regarding client policies and cite the appropriate documents. This allowed analysts to answer questions much faster, with emphasis on the fine-tuning of the retrieval process to improve performance on the private dataset. More evaluation could be done to customize the pipeline to work best with the client’s data, including incorporating new RAG methods and evaluating retrieval with additional AI models.

The sociotechnical research paper explores the effect of social media algorithms on health and nutrition trends and their prevalence in today’s society. Specifically, this research answers the question: How do social media algorithms influence young American perspectives on health and nutrition practices? The interactions between the algorithms, users, social media companies, health experts, and other groups are characterized using Latour’s Actor-Network Theory (ANT). By considering the algorithms equal to human actors, the nature of the algorithms and their outcomes can be analyzed to fully understand how specialized content shown to users can deepen or change views on health and nutrition practices. This research provides fresh insights on the effects of social media, which is a recent topic with implications that have not fully been researched yet.

The combination of these two projects provides a wide scope of the benefits and dangers of new algorithms and technology. The technical aspects of RAG narrate a positive case where an algorithm enhances productivity and eases the workload of daily tasks, while social media algorithms give a cautionary narrative by rapidly spreading harmful or misleading information and shifting public views. Working simultaneously on these projects allowed for the exploration of new technology and its potential societal impacts. The concerns surrounding social media’s effects on users and user perspectives highlights the possibility of negative outcomes from other emerging technologies, such as the RAG discussed in the technical report.

School of Engineering and Applied Science

Bachelor of Science in Computer Science

Technical Advisor: Briana Morrison

STS Advisor: Bryn Seabrook