Abstract
Artificial Intelligence is rapidly transforming healthcare, raising important questions about both its technical capabilities and its ethical implications. This sociotechnical synthesis examines this through both a technical report on automating image segmentation for cardiac MRI (CMR) data and an STS research paper on AI’s use in psychotherapy. Together, these works explore how AI can improve overall efficiency while also challenging trust and responsibility in the system they are put in place.
My technical report, Development of Fully Automated AI-Based Analysis Tools for Pre-Clinical Cardiac MRI, describes my group’s overall process in creating automated image segmentation of pre-clinical mouse Cardiac MRI (CMR) images. This project is important because these CMR segmentations are used in the research of heart disease, the current leading cause of death in the United States.
Our team created models for both CINE and DENSE MRI imaging modalities. The DENSE model utilized a previously created image processing pipeline in combination with a newly trained VGG16 UNET trained on mouse CMR images. The CINE pipeline was created from scratch using a classification model in series with a 3D VGG16 UNET image segmentation model. The classification model selects the time frame of both end diastole and end systole in the cardiac cycle. Following this the 3D VGG16 UNET model segments the images at the selected time frames. All these masks are then compiled on to a single .mat file that can be loaded back into the segmentation software.
These pipelines replace the traditional manual segmentation process that is both time-intensive and prone to inter-observer variability. This will greatly improve the quality and speed of research on heart disease. However, the implementation of such systems also introduces new considerations such as the over-reliance on automated outputs and the evolving role of scientists as mediators between the machines and their data.
My research paper, Between Understanding and Automation: The Ethical Frontier of AI in Psychology, takes a deep dive into these exact considerations. Currently, there is a shortage of psychological therapists, leading to a lack of mental health treatment. AI therapists have been proposed as a solution to this mental health crisis, but their implementation does not come without concerns. In response to these concerns, I critically evaluate whether AI therapists should play a role in psychological treatment.
To answer this question, I applied Actor Network Theory to analyze the system that these AI models are being introduced into, focusing on the interaction between the human and non-human actors. Key actors include patients, therapists, AI models, developers, and regulatory bodies, each of which plays a role in shaping the delivery of these AI services. From this, I identify that the entire network is responsible for the safe and effective implementation of this technology.
Following this, I use Principlism, utilitarian, and care-based ethical frameworks to evaluate the role of AI in therapy from different moral perspectives. Each framework highlights different priorities, such as maximizing the access of care for all or maintaining empathetic relationships. This results in differing, almost conflicting conclusions, but together they provide a more comprehensive view of the ethical trade-offs involved.
Ultimately, I conclude that AI therapists should not replace human providers but serve as a valuable resource in a hybrid model. This allows for the most patients to receive the quality treatment they need, satisfying utilitarian ethics, while still maintain the empathetic relationships that care-based and Principlism highlight. In addition to this, the ANT analysis demonstrated that the effectiveness and ethicality depend on the strength and oversight of the entire network. Therefore, responsible implementation requires maintaining human involvement, ensuring transparency in AI decision-making and establishing clear accountability across all actors in the network.
Separately, these two projects highlight both the potential and risks of integrating AI into healthcare, but together they emphasize the overall challenge of balancing automation with human judgement. While my technical project demonstrates how AI can significantly improve the efficiency and reliability in some aspects of healthcare, my STS research reveals the limitations of Ai in the context of empathy, trust and human connection. The comparison also emphasizes that the success of AI system is not just dependent on the model but the system it is in place. Ultimately, these projects reinforce the idea that AI should not be used as a replacement to human care but serve as a tool to increase human capabilities while maintaining accountability, ethical responsibility and quality of care.