Abstract
Sociotechnical Synthesis
With the recent development of Artificial Intelligence (AI) and Large Language Models (LLMs), there have been concerns with the safety of its adoption. My STS and capstone research connect to approach AI adoption by analyzing its impact in a highly sensitive field, and by providing an example of positive AI use respectively. My STS research paper analyzes how the black-box nature of AI impacts foundational trust structures in the medical practice sociotechnical network, and affects AI adoption in medicine. This research aims to identify failures with AI integration into the current medical system, allowing researchers to better define criteria for adopting medical AI. My capstone research demonstrates an example of a positive use of AI to create a news delivery platform called NewsDash, which sends tailored news articles to users via SMS as if receiving a text from a friend. NewsDash aims to leverage AI to promote consistent news reading, create a new way for readers to engage with content, and encourage discussion and critical thinking about the news people read.
News consumption has declined in the U.S in recent years, and popular news platforms fail to either tailor news to readers’ interests, notify readers effectively to remind them to read, or engage users in a way that promotes critical thinking and trains cognitive abilities. NewsDash aims to address these issues by providing an AI powered news delivery platform where users can register with their phone numbers, add any topics of interest to their account, and then receive SMS texts from NewsDash’s AI agent that contains an engaging hook message about news relevant to the user’s topics, including a link for users to read the relevant article. NewsDash distinguishes itself by sending texts like a friend, which users can reply to to have discussions about the news and find other related articles.
NewsDash successfully ingests news from all users’ topics daily, identifies users eligible to receive news at a given time based on their preferences, and writes an engaging hook to send to the user via SMS. NewsDash has not been fully deployed for external users to access, so its social impact has not been tested at scale. However, internal testing suggests that NewsDash meets its acceptance criteria for finding and delivering relevant articles, ultimately providing a new way for readers to think critically and uniquely engage with articles.
My STS research paper, Black Box: The Effects of AI in Medical Practice on Medical Decision Making, and Trust among Doctors and Patients, answers the question of how the black-box nature of large AI models impacts trust in the medical practice sociotechnical network, and how its impacts affect widespread AI adoption in medicine. The paper’s significance is in providing a framework for analyzing trust between actors in medical practice by structuring medical practice as an actor network involving patients, doctors, AI, and the surrounding infrastructure. My research then uses this framework to analyze how the black-box problem impacts axes of trust between actors in the network, specifically the Doctor-Patient, Patient-AI, and Doctor-Patient-Data trust axes.
I concluded that the black-box nature of AI inherently disrupts the trust axes that serve as the foundation for the medical practice sociotechnical network. The black box problem prevents doctors from providing explanations to treatment decisions made using AI, which violates patient’s trust and breaks medical ethics codes for effective diagnoses and treatments. Patients show distrust towards AI, viewing it as an agent that replaces doctors. The lack of centralized regulation on medical data use for AI cascades to failures in trust from doctors and patients to AI. My research finds that these three points of failure serve to create barriers to AI adoption in medicine, as defined by a previously researched AI adoption framework.
